The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
Intermediate
Targets and
Indicators for
Monetary Policy
A Critical Survey
Federal Reserve Bank
of New York
Intermediate
Targets and
Indicators for
Monetary Policy
A Critical Survey
Federal Reserve Bank
of New York
CONTENTS
Foreword
£. Gerald Corrigan
i
Introduction
Richard G. Davis
1
Possible Roles for the Monetary Base
Ann-Marie Meulendyke
20
Monetary Aggregates as Intermediate Targets
John Wenninger
67
Liquid Asset Measures as Intermediate Targets and
Indicators for Monetary Policy
Gabriel S.P. de Kock and Lawrence J. Radecki
109
A Review of Credit Measures as a Policy Variable
Lawrence J. Radecki
183
Targeting Nominal GNP
Spence Hilton and Vivek Moorthy
232
Interest Rates as Targets and Indicators for
Monetary Policy
Charles Steindel
274
Commodity Prices as Intermediate Targets and
Information Variables
Spence Hilton
305
Possible Roles of the Yield Curve in Monetary Policy
Arturo Estrella and Gikas Hardouvelis
339
The Use of Dollar Exchange Rates as Targets or
Indicators for U.S. Monetary Policy
Charles Pigott and Christopher Rude
363
Optimal Monetary Policy Design: Rules vs.
Discretion Again
A. Steven Englander
421
A Review of Federal Reserve Policy Targets and Operating
Guides in Recent Decades
Ann-Marie Meulendyke
452
FOREWORD
For nearly two decades after the Federal Reserve-Treasury Accord of 1951, U.S. monetary policy
was viewed, both by its practitioners and by others, as mainly a matter of "leaning against the wind"-of tightening money market conditions when inflationary pressures threatened and easing them when
the economy weakened. The 1960s, however, saw a gradual revival of interest in the possible role of
the money stock in cyclical fluctuations and, especially, in the longer run behavior of prices. At the
same time, the economy was experiencing a gradual, though irregular, acceleration in the trend rate of
inflation, a development that was to reach its climax in the late 1970s and early 1980s.
Against this backround, the Federal Open Market Committee began to experiment more
systematically with internal objectives for money and credit growth beginning around 1970. This was
done in the belief that such targets could provide a better focus for the potential impact of policy on
the economy and, in particular, on the longer run trend of inflation. Subsequently, a 1975
congressional resolution called on the Federal Reserve to report formally to Congress on its "plans and
objectives" with respect to growth in money and credit. This requirement was reaffirmed and refined
in the Humphrey-Hawkins legislation of 1978. Thus came into being a set of "intermediate targets" for
monetary policy—intermediate, that is, between instruments such as open market operations and the
discount rate and the broad objectives of policy with respect to economic performance.
The potential usefulness of such "intermediate targets" seemed-and continues to seemsubstantial. At least in principle, targets for growth infinancialaggregates such as money and broad
measures of credit can provide an appropriate way of structuring the intermediate and longer run
strategy of monetary policy, improving both the process of internal decision making and the
communication of policy objectives to the Congress and to the public at large. Longer term objectives
for money and credit growth can have an especially important role to play in ensuring that shorter
term decisions are consistent over the long run with stability in the value of the nation's money, an
objective that is uniquely a concern of the central bank.
The decade of the 1980s just passed, however, proved to be a particularly difficult period for the
effective use of money and credit targets. A major wave of financial innovation and deregulation and a
sharp reduction in the rate of inflation and the accompanying decline in interest rates, together with
I
other factors not completely understood, have produced major changes in the relationship of the
various measures of money and credit to economic activity. The Federal Reserve has found it
necessary to adapt to these changing conditions by reducing or eliminating the use of some financial
measures as intermediate targets while increasing emphasis on others and, more generally, by relying
importantly on an eclectic and judgmental approach to policy making. Moreover, given the apparently
impaired usefulness of the more traditional measures of money and credit as policy targets and
indicators, some have sought to find new measures, such as commodity prices and the yield curve, to
serve as guides for policy.
In these circumstances, it seemed useful to us at the Federal Reserve Bank of New York to review
the evidence on the potential value of a range of financial measures as intermediate targets and/or
indicators. The results of this review are contained in the studies that make up the present volume. The
papers survey the vast literature that has grown up around these various measures while undertaking
new research where that seemed to be needed. We hope that the results, as collected in one volume,
will prove useful to all those concerned with issues of monetary policy.
Not surprisingly, given the experience of the 1980s, each of the measures examined in these
studies is found to have serious limitations. Thus some of the measures are clearly unsuitable as
intermediate targets, if only because they cannot be readily controlled with existing monetary
instruments. Moreover, in varying degrees, all the measures prove to have only an imperfect relation to
the broader economic measures policy seeks to influence.
Despite the problems that have emerged with the traditional money and credit measures in the
1980s, however, the Federal Reserve remains committed—not only as a matter of law but also as a
matter of philosophy—to the formulation of money and credit objectives that will serve as a general
guide to its longer run intentions. Paths for the downward adjustment of the longer run trend of money
and credit growth are an essential element in any coherent long-run strategy for eliminating inflation as
a significant factor in our economic life. The experience of the 1980s provides an important reminder,
however, that such an approach to monetary policy must retain a great deal of flexibility, including the
need to look at a wide variety of financial and nonfinancial indicators in framing judgments about
policy. In any case, the Federal Reserve will undoubtedly continue to study the evolving performance
of money and credit measures in an effort to find which of them may prove most suitable for framing
the strategy of monetary policy in the conditions of the 1990s.
E. Gerald Corrigan
July 1990
ii
INTRODUCTION
Richard G. Davis
Over the years, a broad array of mainly financial variables has been proposed for use in
formulating and implementing monetary policy. This collection of papers examines the potential value
of these various measures as intermediate targets and/or indicators of monetary policy. It includes a
review of the Federal Reserve's evolving approach to the use of policy targets and operating guides in
the postwar period. It also contains an analysis of the recent academic literature on the theory of
policy rules that is relevant to the potential usefulness of intermediate targets.
Systematic analysis of monetary tactics and strategy in light of the relationships among policy
instruments, a broad array of monetary and financial variables and measures of economic performance,
began to expand rapidly in the late 1950s. Over the subsequent decades the subject has generated a
large body of literature. One early source of motivation for this work was monetarist criticisms of the
Federal Reserve's post-Accord procedures. In these procedures, the behavior of the money stock
played, at most, only a limited role. Another impetus to the literature on monetary tactics was
progress in modeling the markets for reserves and money. This work provided far greater analytical
and quantitative detail on the connection between Federal Reserve actions and the response of the
reserve and money aggregates than had previously been available.
Continuing controversy over the appropriate role of money stock targets sustained and
intensified interest in the question of intermediate targets and their implementation in the 1960s and
1970s. Interest in the subject was especially intense in the period after the October 1979
announcement of a change in operating procedures designed to improve the implementation of targets
for the monetary aggregates. By the early 1980s, signs of an emerging breakdown in the existing
relationships of the money measures to the economy generated suggestions that money stock targets be
augmented or replaced by broad measures of liquid assets and credit. As the extent of the shift in the
relationship of all these various measures to GNP became more apparent, however, research interest in
their use as intermediate targets or indicators waned and their role in policy making diminished. More
recently, as discussed in the relevant papers in this volume, some interest has been expressed by
economists and policy makers in possible roles for nominal GNP and/or for market measures such as
l
commodity prices, the yield curve, and foreign exchange rates as policy targets and/or indicators.
In general, however, confidence that there exist financial measures that can replace in part or in
whole a basically judgmental, pragmatic, and eclectic approach to policy seems currently (1990) at a
rather low ebb. Virtually without exception, the results reported in this volume support such a
skeptical attitude. Nevertheless, as argued below, the issue is far from closed. Indeed, interest in the
problem of devising and implementing "intermediate" guides for policy is likely to prove a hardy
perennial in the years ahead.
I. Some Terminology
One product of the debate on these issues has been the development of a useful and reasonably
settled vocabulary to discuss them. One can imagine a spectrum of economic measures that has, at
one end, the "ultimate targets" of monetary policy. These almost always include the price level and
real output and sometimes also include the behavior of the balance of payments and the foreign
exchange value of the dollar.
At the other end of this spectrum are the "instruments" of monetary policy. These include open
market operations, the discount rate, and in earlier periods, required reserve ratios and Regulation Q
ceilings on deposit interest rates. Just one step along the spectrum beyond these instruments are
"operating targets," measures that can be controlled with a rather high degree of precision through
manipulation of the policy instruments. Potential operating targets include measures such as
nonborrowed reserves, the nonborrowed monetary base, and short-term money market rates, most
notably the federal funds rate. Borrowings from the discount window clearly also constituted a
potential operating target under the system of lagged reserve accounting that prevailed between 1968
and 1984, since the Trading Desk could take required reserves as predetermined within any reserve
averaging period. Even under the present system of approximately contemporaneous reserve
accounting, most people would probably still want to count borrowings as a potential operating targetthough to achieve it in any given reserve maintenance period means that the Desk must correctly
estimate required reserves in the current period as well as market factors supplying reserves and the
levels of excess reserves.
"Intermediate" measures, whether considered as "targets" or as "indicators," are variables that,
as the term suggests, are intermediate between (1) the instruments and operating targets that are
capable of rather tight control and (2) the ultimate target measures that can only be influenced
indirectly. Measures of the money stock are perhaps the classic examples of such "intermediate"
2
variables, but as noted, the list includes other broad financial aggregates, such as credit extended to the
nonfinancial sectors, as well as market measures, such as the foreign exchange rate, that are thought to
be significantly influenced by movements in the operating targets.
Some measures, as discussed in more detail in the appropriate papers, are a little harder to
classify. Thus, for example, short-term interest rates are usually treated as operating targets but may
also be treated as intermediate targets. Conversely, the monetary base is most often discussed as an
intermediate measure but sometimes, more controversially, is viewed as a potential operating target.
Nominal GNP is sometimes treated as a potential intermediate measure, at one step removed from its
ultimate target components of prices and real output.
The various intermediate measures may have the potential to serve as intermediate "targets"
and/or as intermediate "indicators" of monetary policy. "Targets" are, obviously enough, objectives
the Federal Reserve seeks to achieve over some time period with some degree of precision and under
some particular set of circumstances. The concept acquired legislative status with the 1975
congressional resolution requiring the Federal Reserve to report on its "plans and objectives with
respect to the growth of the monetary and credit aggregates over the coming year," language that was
repeated in the Humphrey-Hawkins legislation of 1978. The concept of an "intermediate target" seems
to imply that to qualify, a measure should be (a) reasonably subject to control by the Federal Reserve
through adjustment of its operating targets and (b) reasonably closely (that is, predictably) related to
the ultimate targets or, in practice, at least to nominal GNP. Consequently, the papers in this volume
examine the various measures considered as potential intermediate targets from both points of view.
The concept of intermediate measures as monetary "indicators" is a bit more complicated
because it is sometimes taken to mean a measure of the stance of monetary policy and is sometimes
interpreted as an indicator of current or future developments in the economy. In much of the earlier
literature (early 1960s), the term was interpreted in the sense of indicators of the stance of monetary
policy—that is, as measures that could provide, in some sense, an index of monetary "ease" or
"restraint." The attempt to pin down such an index produced, and indeed continues to produce, a
certain amount of confusion and ambiguity. Consider, for example, a situation in which the Federal
Reserve is using interest rates as an operating target and has no intermediate target objective for the
money stock. Suppose on entering a recession, the policy makers progressively lower their interest
rate target, but, owing to the recession-induced decline in the demand for money, the money stock
falls (probably along with a drop in total reserves). Measured in terms of intentionality, policy has
clearly "eased," because the declining short-term rates are, at least in large part, the direct result of
3
policy decisions to ease. But if one believes that it is not intentionality but rather the impact of policy
on the economy that matters, and if one also believes that this impact is best signaled by the money
stock, then in this instance, the declining money stock indexes not an "easing" but a "tightening" of
policy.
This may be a terminological problem in the sense that one may want to talk about an indicator
of policy intentions or an indicator of policy impact and these may not be the same thing. But the
distinction between measures of intention and measures of impact, if they are in fact different, may
also raise an econometric issue: how to decide which intermediate measure, if any, should be treated
as "predetermined" for estimating purposes. In this example, the money stock or short-term rates?
In any case, the recent technical literature has tended to focus on intermediate "indicators"
(sometimes, in this context, also called "information variables") not as measures of the stance of
policy, but as measures of the present or prospective state of the economy. This is the sense in which
the term is generally used in the present volume. To be sure, there are places in the literature where
the two senses of a monetary "indicator" are conflated. For example, a rise in commodity prices or a
steepening of the yield curve may be taken as indicating both that the prospects are for rising inflation
in the future and that policy has been "easy" or, perhaps, "too easy."
Clearly, the main requirement for a good intermediate indicator of the state of the economy is
that it be reliably (predictably) related to the current or prospective behavior of ultimate goals such as
inflation and/or real output. In practice, statistical tests have often been couched in terms of the
relationship of the measure to nominal GNP.
A question arises whether a measure that has proved to be a good indicator in this sense can
then be used as an intermediate target while still continuing to be a good indicator. It has sometimes
been asserted that when a financial aggregate such as the money stock becomes an intermediate target,
presumably chosen in part because of its good indicator properties, these properties will then be altered
(for the worse?) by the very fact of its targeting by the authorities. This may or may not be a problem
with respect to financial aggregates that are treated both as intermediate targets and indicators. It
clearly could be a complicating issue, however, for such market measures as commodity prices,
interest rates, and the foreign exchange rate. Knowledge in the market that the behavior of the
measure is being used by the authorities to make policy decisions is very likely to alter that behavior.
Partly for this reason, proponents of these latter measures have generally advocated them for only a
single purpose: for example, commodity prices and the yield curve, simply as indicators; interest rates
and the dollar, either as indicators or as operating or intermediate targets, but not both as indicators
4
and as targets.
II. Is There a Case for Intermediate Targets?
It is clear that a coherent monetary policy requires a decision on operating targets. It is equally
clear that "indicator" measures providing advance information about the current or prospective state of
the economy are, almost by definition, of value. The usefulness of intermediate monetary targets,
however, has always been more controversial. No measure selected for such a role will ever be
perfectly predictably related to the ultimate targets that matter. At least some uncertainty, some shortterm instability, and some longer term drift in the relationship of any intermediate target to final
objectives seems inevitable.
It has therefore been argued that the use of intermediate targets will result in suboptimal
decisions. Policy makers will adjust their operating targets, not directly in terms of the settings most
likely to achieve their ultimate objectives, but instead in terms of the settings most likely to achieve
the intermediate target. According to this line of thought, intermediate measures such as the money
supply may be useful, at best, as variables that may shed light on (1) the current state of the economy,
perhaps because of more prompt reporting, or (2) the economy's prospective future state, because of
leading indicator properties. On the other hand, their use as intermediate targets is likely only to
produce poorer control over ultimate targets than if instruments were adjusted directly in terms of
objectives for these latter targets.
The logic of this criticism of intermediate measures as targets seems impeccable. Nevertheless,
it misses the heart of the case for the use of such targets, a case that encompasses a much wider range
of considerations. This broader case envisions a number of potential benefits from the use of
intermediate targets. It has been argued, for example, that intermediate targets can usefully provide a
means of communicating the central bank's intentions to the public. Moreover, such targets can
provide a form of central bank accountability.
The ultimate target measures may not be well suited for these various purposes. Thus, as
discussed in the paper on nominal GNP targeting, there may be real problems in having an
independent central bank set or announce goals for ultimate targets. Equally to the point, actual
economic performance over any given period is subject to many important influences in addition to
monetary policy. Hence it may be quite inappropriate to judge the success of this policy by the actual
performance of the economy—that is, by the ultimate target measures-given the role of nonmonetary
influences. By contrast, an intermediate target—a goal for the rate of money growth, for example—can
5
be judged in advance for its probable consistency with acceptable economic outcomes. Moreover, it
can be used to judge, ex post, whether the central bank's day-to-day decisions have been appropriate
to achieving its intermediate target objective. Moreover, the existence of an intermediate target,
defined over a time period such as a year, can be useful to the central bank as an internal check on the
appropriateness of the shorter term settings of its operating targets.1
But there are other fundamental arguments for the use of intermediate targets-provided suitable
targets can be found. Thus it has generally been argued that over the long run, monetary policy can
only affect nominal magnitudes. Its longer run influence over real growth, real interest rates, and
employment mainly reflects its success or lack of success in achieving an environment in which
economic decisions can be made with a minimum of concern and uncertainty about price level
instability. If this view is correct, the appeal of intermediate targets in providing a "nominal anchor"
for policy decisions is fairly clear. Such targets can provide, in principle, an indication that the longer
run thrust of policy will be consistent with longer run goals for price behavior. In principle, at least,
any one of the various monetary and credit aggregates could, if used as intermediate targets, provide
this kind of "nominal anchor" for policy—as could nominal GNP.
Another role that has been suggested for intermediate targets is in dealing with the potential
conflict that may exist between short- and long-run optimal policy, an issue known as the "time
consistency problem" in the academic literature and in more informal discussions as the "credibility
problem." A conflict between short-run and long-run optimizing can arise from the fact that in the
short run, the monetary authorities can probably engineer some extra real output, at least up to a point.
They can only do this, however, through an expansionary policy that yields more inflation than is built
into the public's expectations. According to widely accepted theory, increases in wages and prices that
are more rapid than expected will "fool" the public into supplying more labor and goods under the
mistaken impression that the higher wages and prices represent higher real rewards.
In the short run, there may be pressure on the central bank to seek output gains through such
"surprise" inflation. But once the public comes to recognize that the policy makers are operating in a
way that accelerates inflation, the public will anticipate this acceleration. Put simply, the attempt to
boost output through policies that create surprise inflation will be self-defeating. Over time, the public
will catch on, and the higher inflation will no longer be a "surprise." Inflation that is anticipated will
^ e s e various arguments were cited in a speech, "The Contributions and Limitations of 'Monetary1
Analysis," given by Paul A. Volcker in September 1976 and most recently reprinted in the 75th
Anniversary issue of the Federal Reserve Bank of New York's Quarterly Review, May 1989.
6
have no power to induce higher output. Thus, over the longer run, the effort to induce higher output
through excessively stimulative policies will fail. Output will be no higher than it otherwise would
have been—trending at its potential rate over time—but the rate of inflation will be higher. Thus on
balance, stimulative policies that seem attractive period by period will, over the longer run, simply
result in higher inflation without any output gains-a result desired by no one.
An intermediate target, publicly announced and faithfully adhered to, could, in theory, avoid
this kind of outcome. It could do so by effectively tying the hands of the authorities, preventing them
from yielding to the temptation to seek short-run output gains in a process that over the longer run
only guarantees higher inflation. Probably the best known prescription for using an intermediate target
in this way is the constant money growth "rule" or, in some versions, money growth targets that settle
down to such a rule after some period of accommodation to disequilibrium initial conditions.
Of course a monetary growth rule also has potential disadvantages. Thus while it may ensure
reasonable long-run price stability, it makes no provision for accommodating shocks—whether from
supply or demand-and thus may achieve long-run price stability only at the expense of unsatisfactory
shorter run outcomes for both output and prices. It might be possible to design a more complex
monetary growth rule that allows money growth to adjust to such short-run disturbances, but in a
predetermined way that still prevents the authorities from seeking short-run output gains at the expense
of higher average inflation. However, monetary rules that embody such automatic response features
may themselves create credibility problems—as discussed in the paper in this collection that reviews
the "time consistency" literature.
It has to be emphasized that all these various potential virtues of intermediate targets—improved
accountability, improved communication with the public, provision of a nominal anchor, and
prevention of short-run decisions that serve merely to raise inflation over the longer run—can be
achieved only if suitable target measures exist. As noted earlier, "suitable" in this context means
measures that are "sufficiently" controllable and "sufficiently" stable in their relation to the ultimate
objectives. But this is not an all or nothing matter. No intermediate target will be perfectly
controllable, even over a year. And no measure will be related in a perfectly predictable way to the
ultimate targets. At least some slippage on both counts is inevitable. On the other hand, even if there
is some slippage, the benefits derived from intermediate targeting may, over the longer run, outweigh
the costs that arise as a result of this slippage. Clearly it is a matter of more or less-that is, a
question of how much slippage can be expected from the use of intermediate targets, on the one hand,
and how much one values their potential longer run benefits on the other. Typically, individuals most
7
concerned with long-run inflation results have tended to minimize the problems with intermediate
targets, while those most concerned with the shorter run real output consequences have tended to
worry most about these problems.
in. Evaluating the Candidates
Eight papers in this volume examine individual candidates or groups of candidates—for
example, the multiple measures of money and credit—as potential targets and/or indicators of policy.
While the papers differ somewhat in organization and emphasis, they all touch on certain common
issues. These include (1) the theoretical basis for believing that the particular measures in question
might be useful targets or indicators, (2) the statistical evidence for believing a relationship to ultimate
targets exists and evidence for the stability of any such relationship, (3) issues of central bank control,
and (4) the question whether the measure, even if not used as a formal target, might be useful in a
subordinate role. For example, the paper on interest rates considers the possibility that even if interest
rates make little sense as an intermediate target, upper and lower bounds for real short-term rates
might nevertheless be useful as "constraints" on settings for an interest rate operating target.
As noted earlier, the mostfrequentlyadvocated measures for intermediate targeting over the
past three decades have been the various measures of the money stock and the monetary base and,
more recently, liquid assets and various credit measures. The statistical results for these measures
form a vast literature varying in method, sophistication, periods covered, and conclusions. This
literature is summarized and evaluated in some detail in the individual papers in this volume. It may
be useful here, however, to give some crude sense of the problems that developed for these measures
in the 1980s.
Charts 1 to 6 show the departure from trends of the GNP velocities of a number of potential
intermediate target measures in the 1980s. These departures are clearly large in all cases—greatest for
Ml, the monetary base, and nonfinancial credit; less for M2, M3, and liquid assets. These departures
from past experience are fairly easy to explain in some cases. Thus the velocity of narrow money
(and the monetary base) almost certainly fell because of declines in inflation, nominal interest rates,
and hence the opportunity cost of holding these measures. Explanations in the case of the broader
measures that internalize the effects of market interest rate movements seem less clear.
In any case, the same pattern of major departures from earlier postwar relationships is evident
in Table 1, showing regressions of growth in nominal GNP on current and lagged growth in these
various financial measures. As the error measures suggest, equations estimated on data from 1960 to
8
Chart 1
Velocity of Monetary Base (GNF/Monetary Base)
Level
4th
quarter
1970
1972
1974
1976
1978
1980
1982
1984
1986
1988
Chart 2
Velocity of M1 (GNP/M1)
Level
4th
quarter
4.5
1970
1972
1974
1976
1978
1980
1982
1984
1986
1988
Chart 3
Velocity of M2 (GNP/M2)
Level
1970
1972
1974
1976
1978
1980
1982
1984
1986
1988
Chart 4
Velocity of M3 (GNP/M3)
Level
4th
quarter
i i i i i i i i i i i i i i > i
1970
10
1972
1974
1976
1978
1980
1982
1984
1986
1988
Chart 5
Velocity of L (GNP/L)
Level
1970
1972
1974
1976
1978
1980
1982
1984
1986
1988
Chart 6
Velocity of Nonfinancial Debt (GNP/Debt)
Level
1970
1972
1974
1976
1978
1980
1982
1984
1986
1988
Table 1
Summary Statistics from Reduced Form Equations
1960-H to 1979-IV
Y = 2.49 + 1.18M1
(2.39)(6.29)
Y = 1.43 + 0.85M2
(0.99)(5.19)
Y = 1.57 + 0.75M3
(1.07)(5.07)
Y = 0.37 + 0.93L
(0.25)(5.69)
Y = 1.00 + 0.87Debt
(0.59)(4.54)
Y = 2.72 + 0.98Base
(2.34)(5.33)
Average
Error
Actual-Predicted
for 1980-89
Average
Absolute Error
2.05
-4.13
5.72
6.16
3.92
1.86
-0.38
3.55
4.48
0.23
3.94
1.86
-0.55
3.40
4.49
0.30
3.78
2.01
-0.97^'
3.67i'
4.54i'
0.29
3.80
2.08
-2.65i'
4.35i'
4.77i/
0.27
3.85
1.92
-2.40
3.99
4.41
R2
SEE
0.34
3.66
0.24
DW
RMSE
Notes: All equations regress the growth rate of nominal GNP on the current and four lagged growth rates of the
financial aggregate. Figures in parentheses are "t" values. L represents the Federal Reserve Board's measure of
liquid assets.
V 1980-1 to 1989-m
Table 2
Summary Statistics from Reduced
Form Equations
1981-1 to 1989-IV
Y = 5.85 + 0.16M1
(4.17)(1.03)
Y = 7.21 + 0.01M2
(2.88X-0.02)
Y = 8.05 + 0.06M3
(3.16)(-0.21)
Y = 9.46 + 0.21L
(3.33X-0.21)
Y = 10.80 + 0.30Debt
(2.85)(-0.90)
Y = 0.37 + .99Base
(0.12)(2.56)
R2
SEE
0.06
3.67
1.36
0.02
3.76
1.31
0
3.79
1.33
0.06i'
3.69±;
1.38i7
0.07i'
3.661'
1.5 li'
0.16
3.48
1.51
DW
Notes: All equations regress the growth rate of nominal GNP on the current and
four lagged growth rates of thefinancialaggregate. Figures in parentheses are
"t" values. L represents the Federal Reserve Board's measure of liquid assets.
y 1981-1 to 1989-m.
12
1979 do a very poor job in estimating GNP growth in the 1980s. And as Table 2 shows, similar
equations estimated over data from 1981 to 1989 have almost no explanatory power, with coefficients
that are not significant (indeed usually negative!) for all measures except the monetary base. This
kind of result makes clear the reasons for the growing disillusionment in recent years with the
potential of these measures as intermediate targets or indicators.
The other property required of a potential intermediate target (as opposed to indicator) is of
course controllability, and that poses a different set of problems. Some of the broader measures, such
as total liquid assets and aggregate credit, are clearly not closely related to Federal Reserve operating
targets. They can perhaps only be controlled indirectly-that is, by first controlling GNP! The
narrower measures such as Ml and M2 have clearly retained substantial interest rate sensitivity for
horizons out to a year or so because many of their component own-rates respond only slowly to
changes in market rates. As a result, growth in such measures may be rather sensitive to changes in
interest rate operating targets. That very sensitivity is itself a problem, however. Thus it makes the
GNP outcome associated with any successfully achieved growth rate target for the aggregate quite
sensitive to shifts in the demand for goods and services~at least over periods out to a year or so. Use
of such targets, therefore, either will imply a wide range of uncertainty about the GNP outcome
associated with a given money growth rate or will make it seem desirable to define targets in terms of
broad ranges of growth rates. This latter approach, however, clearly weakens the usefulness of such
targets for many of the purposes they are designed to serve.
The breakdown in earlier relationships between financial aggregates and nominal GNP in the
1980s has lead a number of academic writers to suggest that the velocity problem could be "solved" by
targeting nominal GNP directly. Such an approach has some attractive features. A long-run nominal
GNP growth rule set in light of the expected trend growth in real output would establish a "nominal
anchor" for policy. Adhered to as a "rule," with or without automatic feedback mechanisms, it would
solve the short run/long run inconsistency problem and would satisfy the other objectives of an
intermediate target. In the short run, adherence to a nominal GNP target would automatically offset
both the price and the real effects of demand shifts and would split the impact of supply shifts
between real output and prices.
Unfortunately, the nominal GNP approach appears to have equally large problems. First, it
"solves" the velocity problem only on the assumption that there exists a way of accurately achieving a
nominal GNP target with the means at the disposal of the central bank. Obviously the method of
choice would not be through intermediate money targets, for that would simply reintroduce the
13
velocity problem. A different option would be to aim at GNP targets through constant resettings of an
interest rate operating target. Clearly success with such an approach is far from assured.
A second difficulty is that to get a handle on final objectives with a nominal GNP target, you
need to have a predictable relationship between nominal GNP and output in the short run—that is, you
need to be able to predict the price/output split resulting from a given GNP result, at least to the extent
that you have short-run output objectives. But of course this problem is shared with other potential
intermediate targets such as money growth rates.
A third difficulty is that a fixed nominal GNP rule could, under quite plausible conditions
examined in the GNP paper in this volume, generate problems of dynamic instability in the path of
real output in the face of supply shocks and, apparently, in the face of prior misses in hitting the
nominal GNP objective. Such problems can be avoided by resetting, on a discretionary basis, the
nominal GNP target year by year. But this approach would create a very uncomfortable situation for a
central bank that is "independent within the government." Year-by-year settings of GNP targets would
come very close to setting year-by-year targets for real growth and inflation. Such a situation could
well create pressures for precisely the kind of short-term optimization that produces the worst of all
possible worlds over the long run-that is, it would seem to maximize the risks of creating the kind of
"time consistency" problem cited earlier. Overall, it seems quite possible that as a practical matter,
discretionary nominal GNP targeting could result in worse inflation outcomes than might exist in the
absence of any intermediate target at all. In summary, the nominal GNP route appears, on closer
examination, to be no panacea for the problems created by the velocity instabilities of the 1980s.
The remaining measures examined in this volume, commodity prices, the yield curve, and the
foreign exchange value of the dollar, have generally been proposed as intermediate indicators that
might be used to guide settings of the operating targets, rather than as intermediate targets themselves.
Because these markets are often regarded as efficient in incorporating relevant economic information,
they could perhaps signal changes in the economic outlook very quickly.
In the case of commodity prices, it seems likely that these prices could in fact be targeted
through direct open market operations in commodity markets. Indeed, that is just what a "commodity
standard," whether defined in terms of a basket of commodities or a single commodity such as gold,
would involve. Instead of such operations, it has been suggested merely that commodity prices may
represent sensitive advance indicators of changes in general inflation rates that can be used to signal
the need to tighten or ease the conventional operating targets.
The results surveyed in the paper on commodity prices included in this volume suggest that
14
movements in commodity price indexes do have a marginal contribution, but only a marginal
contribution, to make in forecasting inflation. As leading indicators of turning points in broad
movements in the overall inflation rate, commodity price measures, suitably averaged and smoothed,
do have some predictive value. But they have also at times given false signals of turning points in the
general inflation rate. In the case of correct signals, moreover, their lead times tend to be rather
variable and there appears to be little relation between the magnitude of commodity price movements
and the magnitudes of subsequent movements in overall inflation. On balance, it appears that
commodity prices may be reasonable additions to the items the central bank "looks at" when it surveys
the prospects for inflation. They do not, however, add much to more conventional methods of
assessing the outlook for inflation.
Another market measure that has been proposed as an indicator but not as a target of monetary
policy is the yield curve. In this case, the question whether the measure is to be thought of as an
indicator of the stance of policy or an indicator of the future course of the economy is somewhat
ambiguous. And this ambiguity is directly related to the theoretical assumptions underlying the
attention sometimes given to the yield curve for either or both of these roles. Belief in the indicator
properties of the yield curve appears to rest on the expectations theory of the yield curve-the view
that longer term rates should be regarded as (weighted) averages of the market's expectations of the
future course of successive short rates. While this theory has considerable intuitive appeal, empirical
tests of its validity over the years have produced mixed results.
Even if the expectations theory of the yield curve is accepted as correct, moreover, the
theoretical implications of particular yield curve configurations, as interpretations both of monetary
policy and of prospective economic performance, appear to be ambiguous. This ambiguity stems in
part from another widely accepted theoretical premise—that nominal interest rates reflect the sum of a
real rate and an inflation premium that allows for the expected rate of inflation over the life of the
instrument. Thus an upward rising yield curve, for example, could imply either that the market
expects real short-term rates to rise in the future or that it expects the rate of inflation to rise.
Against this background, the paper on the yield curve in this volume points out that it would be
very hard for a central bank to interpret the significance of, for example, a steepening of the yield
curve on the basis of theoretical considerations alone. Such a steepening could mean that the market
expects inflation to accelerate, which the central bank could interpret as a need to tighten.
Alternatively, it could mean that the market expects a rise in the productivity of capital and hence a
rise in real short-term rates and an acceleration of real growth. This cause of a steepening in the yield
15
curve might or might not imply a need to change the settings of operating targets, depending on
circumstances. A third possibility is that the steepening reflects a market judgment about the future of
monetary policy itself-that is, that policy is expected to tighten, real short rates to rise, and therefore,
quite possibly, real growth to slow. So a central bank looking at a change in the yield curve must try
to sort out its possible meanings and then must decide what implications, if any, the change may have
for policy.
Despite these interpretive ambiguities at the theoretical level, the yield curve paper gives some
fairly concrete results. It suggests, for example, that the Federal Reserve does have significant power
to affect the yield curve by changing the federal funds rate as an operating measure. Since no one
proposes the yield curve as a target, however, this is of rather limited significance. But the paper goes
on to suggest that the yield curve, simply as an empirical matter, has proved to have significant
forecasting value for both real output and inflation, even in the presence of other forecasting variables
such as short-term interest rates, the leading indicators, and the consensus of economists' forecasts.
One has to wonder, however, how this forecasting value might be affected if the yield curve were to
become a major forecasting tool for the authorities and if the market were to become aware of such a
development and were to respond accordingly. A kind of "two-person game" situation between the
market and the authorities might greatly distort the behavior of the yield curve relative to what it
would be in the absence of a belief in its indicator significance.
Finally, the increasing sensitivity of the U.S. economy to international developments has led to
growing interest in the use of the foreign exchange value of the dollar as a guide for U.S. monetary
policy. However, the paper on this topic emphasizes that, because of important differences between
exchange rates and more traditional variables, the systematic use of the dollar's value in U.S. monetary
policy operations is likely to be highly problematic and would almost certainly raise considerations
beyond those traditionally incorporated in U.S. policy deliberations. In particular, manipulation of
policy instruments to regularly counter or "target" dollar movements could be destabilizing for the U.S.
economy under a wide range of circumstances and could require a significant degree of international
policy coordination. The paper does suggest that exchange rates can play a useful role as policy
indicators but generally only under fairly limited circumstances. At times, for example, foreign
exchange market conditions have proved helpful in gauging market perceptions and the likely reactions
to policy changes. Beyond these circumstances, however, the evidence raises considerable doubts
about the reliability of exchange rates as regular indicators of underlying inflation pressures or the
monetary policy stance. Accordingly, on present knowledge, the case for upgrading the role of the
16
dollar in U.S. monetary policy formulation appears questionable.
IV. A Future for Intermediate Targets?
The cumulative effect of the papers included in this volume is to leave one impressed with the
limitations of all the various measures, certainly as intermediate targets and, for the most part, even
simply as indicators. But if all potential intermediate targets have problems, it is also important to
recall the many ways in which policy conducted without any such target is itself less than satisfying.
In practice, conducting policy without reference to intermediate targets means setting operating targets
directly in line with changing assessments of the likely outcomes for the ultimate goal variables.
Perhaps most often, this will mean adjusting some money market rate in line with changing projections
of the future behavior, under assumed paths for such a rate, of prices and real output.
The difficulties of this approach to policy making are numerous. Perhaps the most obvious
problem is the need to assess correctly the future state of the economy under alternative assumptions
about settings of the operating targets. Note that it is the future state of the economy that matters
given the universally acknowledged existence of significant lags in the impact of policy on output and
prices. While there is substantial evidence that experienced macro forecasters can improve
significantly on naive extrapolative procedures in projecting the future, it is also clear that forecasting
remains as much an art as a science. Macro forecasting normally reflects a blend of reliance on
econometric models, interpretation of incoming information (both statistical and "anecdotal") on the
current state of the economy, and the selective use of an array of leading indicator measures. While
such forecasting is clearly useful—indeed, absolutely necessary given the lags of policy's impact—it is
also obviously fallible. As a further complication, policy decisions must be based on multiple
forecasts, implicit or explicit, that are conditional on multiple alternative settings of the operating
variables under consideration.
The well-known limitations on the ability to forecast raise the risk, moreover, that policy
makers will find themselves putting undue weight on the current state of the economy despite the
acknowledged importance of lags in the process. And of course absent intermediate targets, policymaking procedures do not readily lend themselves to an "objective," quantitative way of
communicating the intentions of policy to the public or of evaluating its success after the fact. Even
more serious, an approach that relies on setting operating targets in light of projected future economic
outcomes fails to provide a "nominal anchor" for policy and does nothing to solve the conflict between
period-by-period and long-run optimizing in policy making.
17
So we have a real tension here. On the one hand, intermediate targets, if suitable ones exist,
have the potential for improving the overall performance of monetary policy, especially over the
longer run. But, to repeat, "suitable" means not merely controllable, but sufficiently tightly related to
ultimate goals that slippages can be ignored and thus the forecasting problem bypassed. The
experience of the 1980s has left serious doubts that such "suitable" target measures do in fact exist.
Faced with this tension, the Federal Reserve has in practice compromised. It has continued to set
intermediate targets for money and credit aggregates—as, indeed, it is required to do by law—but it has
defined these targets in terms of rather wide ranges (generally 4 percentage points for annual growth
rates). Moreover, on occasion the Federal Reserve has felt free to allow even these wide ranges to be
violated when it has appeared likely that the targets could be achieved only at the expense of inferior
economic outcomes~or at least outcomes that are "inferior" within the one-year time horizon of the
current targeting process. The target measures have been given more attention when at the top or
bottom of their ranges, with particular attention focused on the behavior of M2. In summary,
intermediate targets have continued to exist, but only as rather wide ranges and without any clearly
defined means of connecting them with day-to-day or month-to-month operational decisions. Even
under these circumstances, the extant intermediate targets have had some value in providing a
"nominal anchor"—though one that tends to drag a bit—and have provided a means of connecting, if
somewhat loosely, short- to intermediate-run objectives with the longer run objectives for price
performance. Nevertheless, it is apparent that their usefulness for these purposes falls far short of
what, at least in theory, could be provided by more formal adherence to a satisfactory intermediate
target.
The broad appeal of the intermediate target concept is such that interest in it seems bound to
persist. Research on the subject has continued, both within and without the Federal Reserve System.
In particular, some economists at the Federal Reserve Board have developed evidence to suggest that
long-run M2 velocity may have retained enough stability to make M2 behavior a useful indicator of
the longer run behavior of inflation. In the meanwhile, it is possible that after the major shocks to the
monetary aggregates (and possibly also to broad credit measures) from financial innovation and
deregulation in the 1980s, these aggregates may settle down to a pattern of behavior that, if changed
from earlier decades, has nevertheless again become predictable enough to be useful. The future role
of intermediate targets in the policy-making process is certainly likely to depend in part on such
potential developments. But given the short-run slippages that would inevitably persist between
intermediate targets and ultimate objectives even under the best of circumstances, the future role of
18
intermediate targets probably also depends on the weight that is given to the objective of long-run
price stability. It is in the context of such an objective that the potential usefulness of intermediate
targets is particularly clear.
19
POSSIBLE ROLES FOR THE MONETARY BASE
Ann-Marie Meulendyke
The concept of the monetary base has a long history. In essence, the monetary base consists
of currency issued by the monetary authority, primarily the Federal Reserve System in this country,
and reserve balances held with the monetary authority by depository institutions. The most commonly
used measures of the monetary base make an adjustment for reserve requirement changes (see
appendix for definitions). Several names have been applied to what is now most often called the
monetary base. A number of writers have used the term "high-powered money."1 Gurley and Shaw
(1960) introduced the term "outside money."2 These names for the monetary base, perhaps, are more
suggestive of why the concept has played a role in policy discussions. The monetary base consists of
monetary instruments that can function as reserves of the banking system and are obligations of the
government or the central bank. It is readily observable with a short lag. Furthermore, many
economists have argued that it should be feasible to control the monetary base because its components
are on the Federal Reserve's balance sheet. In addition, because of regulatory and behavioral linkages
between the monetary base and various monetary aggregates that have been proposed as intermediate
targets, the base's behavior has the potential to affect broader monetary and economic variables.
Three different roles have been proposed for the monetary base. The oldest use for the
concept has been as an analytical device to explain relationships between gold flows and central bank
actions and the behavior of money, credit5 and prices. The second suggested use of the monetary base
is as an operating target of policy, most commonly as a means of achieving desired behavior of an
intermediate target such as a monetary aggregate. The third type of proposed use of the monetary
base is as an intermediate target or as an indicator to guide the central bank in its efforts to achieve
its long-run goals of sustainable economic expansion and price stability, especially the latter.
*See, for instance, Burgess (1936), Friedman (1959), and Lothian (1976).
*They describe "outside money" as an obligation of the government (including the central bank) that
is outside the control of the private sector. In contrast, "inside money" is as an obligation of the private
sector.
20
The analysis of the possible roles for the monetary base has proceeded on more than one level.
In concept, it has appeared in the context of models of money demand, business cycle determination,
and the appropriate responsibilities of a central bank. The monetary base has had particular appeal to
those economists who have pushed for limiting monetary expansion to promote price stability.
Proponents of the quantity theory of money, the classical long-run neutrality of money, and rational
expectations have all tended to give prominence to the notion of a long-run anchor for the price level.
Many of them have supported the use of the monetary base in that role.
Other analyses have focused on more narrowly technical issues associated with using the base
in one or another of its potential roles. Topics considered have included the definition of the monetary
base, details of its controllability, and institutional and regulatory questions. Overlapping the
conceptual and technical sides of the issue have been the questions whether the monetary base can be
exogenously determined and whether estimated relationships between the base and intermediate or
ultimate policy goals would be sustained if serious efforts were made to control the base. Many of the
debates are affected by the time period considered relevant for achieving the goals. In some cases,
failure to clarify assumptions on that question has complicated developing an understanding of the
issues.
This paper examines the three proposed uses of the monetary base concept, considering both
conceptual and practical aspects. It evaluates the components of the monetary base and the extent to
which the base has the properties attributed to it. The paper then reviews the evidence from
experiences with targeting the base or a related reserve measure.
In summary, the monetary base concept appears to be useful as a descriptive tool. Considered
as a possible operating target, the base would seem to present some problems if efforts were made to
control it directly over time periods shorter than one to two quarters. Considered as an intermediate
operating guide, the base may be of some value, especially as a warning signal of potential inflationary
policies, but the generally weak relationships between the monetary base and ultimate policy goals
suggest that it should be used cautiously. A review of experience in Switzerland and West Germany
with the use of measures similar to the monetary base as intermediate targets suggests some problems
but also a measure of success.
I. The Monetary Base as a Descriptive Device
Early in the twentieth century, economists were making analytical use of the monetary base
concept, although not by name. In 1911, Irving Fisher ([1911] 1963) described the role of currency as
21
a circulating medium and as a major source of reserves to the banks. The other form of reserves was
interbank deposits; small banks generally held balances at large banks. Fisher described changes in
currency and reserves as the primary source of growth or decline in the quantity of money because, in
normal times, the ratio of currency to deposits would be determined by payment practices and customs
that did not change easily.3 Furthermore, Fisher argued that prudent banking practice and the high
cost to a bank of being unable to meet withdrawals would determine the ratio of reserves to deposits
within a fairly narrow range. Fisher anticipated that disturbances to the ratios would occur during
periods of transition to new institutional arrangements. He also expected the ratios to shift gradually
over the business cycle in response to sticky adjustment of prices and interest rates.
Randolph Burgess (1936), an officer at the New York Federal Reserve's open market desk in
the 1920s and 1930s, described in 1936 what then seemed to be a common view of the monetary base
(which he called high-powered money) as follows:
It may be said, from one point of view, that there are in any country two kinds of money, and
for the sake of giving them names they may be called high-powered money and low-powered
money. The central bank deals in high-powered money, the money which constitutes bank
reserves. Historically, this high-powered money has been closely related to a country's basic
reserves of gold and currency, though the specific form of this relationship shows wide
variations under different banking laws. . . . When the amount of high-powered money
increases, the amount of low-powered money tends to increase also, but in multiple relation to
the high-powered money, (pp. 5-6)
Burgess presented high- and low-powered money as descriptive concepts to help explain how
the purchase or sale of gold or securities would lead to growth or shrinkage of commercial bank
deposits. Provision of high-powered money by the central bank makes it possible for the banks to
create more low-powered money according to what has now become a familiar textbook multiplier
model.4 In that model, the creation of deposits is constrained by limits on the ratio of deposits to
reserves. The ratio is partially determined by banking practices, as Fisher had suggested, but under the
Federal Reserve, the relationship has also been importantly influenced by the reserve requirement
ratios that are applied. The relationship further depends upon the ratio of currency to deposits.
These relationships can be expressed in equation form as M=mB, where M is the monetary
aggregate in question, B is the monetary base, and m is the multiplier linking them. In addition, M is
3
Fisher defined money as currency and coin, but this analytical work was based on such money plus
bank deposits.
4
See, for instance, Mayer, Duesenbury, and Aliber (1987, pp. 161-70).
22
defined to equal C+D, where C is currency and D is deposits, and B is defined to equal C+R, where R
is reserves in the form of vault cash or balances at the Federal Reserve. The multiplier can be
expressed as a function of ratios linking deposits, reserves, and currency as follows: Let k=C/D and
r=R/D. Then m=M/B=(C+D)/(C+R). Dividing the numerator and denominator by D and substituting
gives m=(k+l)/(k+r).
If underlying behavioral characteristics serve to make the ratios k and r stable, then the
monetary base multiplier can be a useful device for relating central bank actions affecting reserves to
the behavior of money. In that case, a change in the monetary base would be associated with a
proportional change in money. If either of the ratios, k or r, shiftsfrequently,however, the multiplier
will not be stable. Under these circumstances, money and the monetary base will not move closely
together. Furthermore, the level of the reserve-deposit ratio, r, will have other effects. A low ratio
implies that a low weight is given to deposits relative to currency in the monetary base, since only that
fraction of deposits represented by reserves is counted. Many economists over the years have
expressed unease over the implicit weighting scheme that gives relatively heavy weight to currency,
because they believe that the behavior of deposits has a major role in influencing economic activity.
Nonetheless, during long stretches of time, the two ratios have been sufficiently stable that many
analysts have downplayed the worries about the low weight given to deposits.
The most dramatic exception to that pattern of stability occurred in the 1930s, at the time
when Burgess was writing. Bank failures in the early 1930s periodically raised the demand for
currency and made the ratio of currency to deposits highly variable. Moreover, by the time Burgess
published his book, excess reserves had risen to unusually high levels; the ratio of reserves to deposits
was higher than it had previously been and rather variable. Hence, it is not surprising that Burgess did
not consider the possibility of using high-powered money as a guide to policy.
Henry Simons, writing in 1934 and observing the problems associated with fractional reserve
banking, proposed that the variation in the multiplier could be eliminated if 100 percent reserve
requirements were imposed. Under that arrangement, currency and deposits would receive equal
weight and shifts between them would have no effect on the relationship between the monetary base
and money. If the banks held 100 percent reserves against deposits, then r would presumably be
around 1. The multiplier would be approximately (k+l)/(k+l) or 1 regardless of the ratio of currency
to deposits. Simons recognized that his proposal was radical. Nonetheless, he believed it would be
feasible to separate institutions providing deposits from those providing credit.
Milton Friedman (1959) supported Simons' proposal although he believed it was necessary to
23
address its primary weakness. Issuing deposits subject to 100 percent reserve requirements would be
costly. Because the banks could not use those deposits as a source of funds for making loans and
investments, they would have to charge large fees on deposits to cover the overhead costs associated
with them. Alternative instruments that were exempt from the 100 percent reserve requirement would
have to fund credit extension. Consequently, those instruments could pay interest instead of incurring
charges. Given the relative attractiveness of these alternative liabilities, they would probably evolve
into something that closely resembled money. Broader forms of money would then consist of deposits
subject to very high and very low reserve requirements, and the high and stable ratio of reserves to
deposits that the proposal was designed to achieve would be lost. Friedman recommended paying
interest on required reserves to mitigate the cost disadvantage associated with deposits subject to the
high reserve requirement. Nonetheless, the instruments funding credit extension would have to be
priced attractively, and some risk would remain that they would take on some of the characteristics of
money.
A less radical proposal for handling the problem of minimizing the variation of the ratio of
reserves to deposits was to make reserve ratios uniform according to the nature of the deposit rather
than the characteristics of the issuing institution. In the years since most of the literature on this
subject was written, steps have been taken to make the ratios more uniform with respect to transaction
deposits. However, time and savings deposits are subject to low requirements, and beginning in the
1980s, personal time and savings deposits have been exempt from requirements. For money definitions
that include some of these deposits-M2 and M3—shifts between transaction and nontransaction
deposits will affect the ratio of reserves to deposits and hence the monetary base multiplier. Whether
the existence of low and differing reserve ratios for various types of deposits interferes with the
usefulness of the monetary base in either of the roles proposed remains controversial.
One proposal has been made to create a monetary base measure that includes artificial reserve
ratios higher than those that actually prevail. To a limited extent, the "St. Louis" definition of the
monetary base in effect does that since it uses a 12 percent reserve ratio against transactions deposits
(see appendix) rather than the actual reserve ratio, which has recently been around 8 percent. The
German concept of the central bank money stock (discussed below) also uses a higher than actual
reserve ratio. In neither case was the step taken explicitly for the purpose of raising the weight on
deposits. The results fell out of efforts to minimize distortions imposed by changes in the level and
structure of reserve requirement ratios.
24
II. The Monetary Base as an Operating Target
The case for using the monetary base as a tool in the policy process was made in the 1960s in
a series of books and articles promoting the control of a monetary aggregate as a means of achieving
the longer run monetary policy goals of price stability and an expanding economy. In this role, the
monetary base is, in the terminology most commonly used in this literature, an operating target. The
base would be controlled not for its own sake but in order to achieve desired behavior of another
variable, usually a monetary aggregate.
By the 1960s, when monetary aggregate targeting became a popular objective, sufficient
stability had returned to the ratios of reserves to money (Chart 1) and currency to money (Chart 2) to
make the arguments for such an approach to targeting plausible, although many analysts remained
skeptical. (See the paper by John Wenninger in this volume.) Milton Friedman (1959) proposed
targeting a monetary aggregate and suggested that manipulating high-powered money might be a
reasonably effective way to attain the monetary targets. Karl Brunner and Alan Meltzer (1968) made a
similar proposal and explored several aspects of the multiplier relationship.
In several articles that appeared in the St. Louis Federal Reserve Review, Andersen and Jordan
(1968a, 1968b) and Burger, Kalish, and Babb (1971) developed in some detail the suggestion to target
the monetary base. The St. Louis articles presented multiplier relationships between the monetary base
and Ml similar to those described above, although the relationship was more elaborately drawn.5
While the economists promoting control of the monetary base recognized the problems
associated with the differential weights given to currency and deposits, they doubted that these
difficulties would prove to be serious in practice. At the time they were writing, reserve requirements
were high enough to discourage banks from holding large quantities of excess reserves. The
economists believed that the presence of binding required reserve ratios would make the ratio of
reserves to deposits relatively stable as long as adjustments were made whenever the Federal Reserve
changed the specified ratios. Building on Fisher's arguments, they expected payment conventions and
the absence of banking crises to provide stability to the ratio of currency to deposits. Empirical
analysis of the data covering the 1950s and 1960s generally gave some support to their expectation.
Burger, Kalish, and Babb presented such an examination and recommended a control procedure in
5
A variety of multipliers have been computed that make separate allowances for transactions deposits
and time deposits. Johannes and Rasche (1979) built an elaborate multiplier model that has been regularly
updated for Shadow Open Market Committee meetings. See also Rasche and Johannes (1987).
25
11
[•
10.5 h
r
^
\y\
10
H
9.5
i
i
Chart 1
Total Reserves as a Percentage of M1
8.5
\- i
h
|
7.5
l i i i l i i i i i i i l i n l u i l n i iiilnil i i i l i i i l i i i l i i i l i i i l i i i i i i i lui
1960
1965
1970
1975
i n l i i i h I I I I I I I M I " i l i i i l i i i l i n JJJ
1980
Note: Reserve data are seasonally adjusted and are adjusted for changes in
reserve requirements. M1 values are seasonally adjusted. Ratios are plotted
quarterly.
Chart 2
Currency as a Percentage of M1
30
h
28
L
26
h
•
24 h
r
•
i
22
20 h
I
n lI .InI lI ,I 111
MIM
I I JI M . L M I M . I M J M . I M ,
iiiiiniiiii . . . I M . I > H I I M I . I I I I I I I n I I I I I I M I I n I I n I 11 I n M I I n I I I I I I I I I I I I I I I I 11 I I I I I I I I I I I 1 I I,11
1955
1960
1965
1970
Note: M1 values are seasonally adjusted. Ratios are plotted quarterly.
1975
1980
1985
1989
which the monetary base would be targeted to achieve desired growth in Ml. They proposed
estimating the multiplier from recent behavior of its constituent ratios. Then they compared their
model forecasts of the multiplier with the actual values of the multipliers. They found the errors to be
small in size and observed that they were not cumulative. They concluded that if the proposed
monetary base targets were achieved and the multipliers were the same as those that actually occurred,
then money growth would have deviated only slightly from a smooth path. From these results, they
argued that following their procedures would produce a reduction in unwanted variation in money
growth.
The Burger, Kalish, and Babb proposal would have required the central bank to control the
monetary base. Most of the supporters of the proposal argued that the central bank could control the
monetary base since the base consisted of items on the central bank's balance sheet that could be
observed with at most a one-day lag (Balbach 1981). Others attacked the proposal, suggesting that the
consequences of trying to control the base would be undesirable because the observed multiplier
relationships on which such proposals were based were estimated when the monetary base was, in fact,
determined endogenously. The critics predicted that the relationships would break down if the base
were actually controlled. To explain the issues involved, it is helpful to review the potential methods
for controlling the monetary base.
As the appendix indicates, the monetary base can be viewed as consisting of three elements:
currency in circulation, total reserve balances of depository institutions at the Federal Reserve, and a
factor that adjusts for the effects of changes in reserve requirement ratios. The adjustment factor does
not introduce control problems, except possibly briefly at those few times that the Federal Reserve
changes reserve requirement ratios.
Currency, the biggest component of the base, has been issued by the Federal Reserve to
depository institutions on demand. The Federal Reserve debits the reserve accounts of a depository
institution when it issues the currency. Limiting currency issuance would be a sharp break with a
tradition that extends back to the beginnings of the Federal Reserve. Most of those proposing control
of the base did not contemplate a limitation on Federal Reserve issuance of currency, but instead
advocated offsetting undesired currency movements with increases in or restrictions on the provision of
total reserves through open market operations (Burger, Kalish, and Babb 1971). The Federal Reserve
would have precise knowledge of the amount of currency it had issued, and therefore would know the
size of offsetting adjustments in reserves needed as soon as any unwanted currency movements took
place.
27
In practice, the process is not so simple. Total reserves, the factor that would need to be
manipulated to achieve a desired level of the monetary base, consists of reserves provided by open
market operations—nonborrowed reserves-and reserves obtained at the discount window—borrowed
reserves.6 If, for example, the Federal Reserve attempted to control the monetary base by offsetting
undesired expansion of currency through a reduction in nonborrowed reserves, depository institutions
might borrow at the discount window to obtain the reserves lost, lifting total reserves back to the level
where they stood before the reduction in nonborrowed reserves occurred. Consequently, it would not
be possible to control the monetary base even though it would be possible to control what is often
referred to as the nonborrowed base—the monetary base minus borrowed reserves.
Whether depository institutions would in fact borrow at the discount window to offset the
reduction in nonborrowed reserves would depend upon their demand for total reserves. Depository
institutions hold total reserves in order to meet reserve requirements on average over a reserve
maintenance period (which has been two weeks long since 1984, but was one week long between 1968
and 1984, when much of the discussion of controlling the base took place). Some depository
institutions may also choose to hold reserves in excess of requirements if day-to-day flows through
reserve accounts are large relative to their reserve requirements.7 The cost of close management of
reserves to avoid being unexpectedly short may exceed the interest lost from holding some excess
reserves.
If the total reserve component of the monetary base target is to be achieved, a decline in
nonborrowed reserves cannot be offset by an increase in borrowed reserves. A reduction in
nonborrowed reserves can only result in an equal decline in total reserves if the lower level of
nonborrowed reserves is consistent with existing demands for reserves to meet requirements—because
excess reserves are larger than depository institutions desire-or if depository institutions adjust loans
and deposits to reduce required reserves by the full amount of the reduction in nonborrowed reserves.
In order to lower required reserves, deposits would have to fall by a multiple of the desired
decline. Whether such sharp adjustments to deposits over a short time period are feasible has been
technically, the Federal Reserve cannot, through use of open market operations, achieve desired
nonborrowed reserve levels with precision because there are a number of factors on its balance sheet, such
as Treasury cash and Federal Reserve float, that it does not control and can only observe after the fact.
Most of the time, forecasts of those factors are sufficiently accurate that average errors in reserves for a
reserve maintenance period are small. Occasionally, however, they are large.
7
In equation form: total reserves = nonborrowed reserves + borrowed reserves = required reserves +
excess reserves.
28
extensively debated. Those favoring short-run control of the base generally argued that depository
institutions could make quick adjustments to deposits if they were given the incentive to do so. They
believed that any additional nonborrowed reserves provided would quickly be lent or invested, setting
in motion a multiple expansion process that would raise deposits until the unwanted excess reserves
had disappeared. In a parallel fashion, they believed that depository institutions would respond to a
shortage of nonborrowed reserves by contracting loans and deposits until required reserves had shrunk
to the point that they were consistent with existing supplies of nonborrowed reserves. They argued
that regulations that weakened the money multiplier process were serious impediments to such
adjustments. In particular, they cited the relatively low discount rates (or alternatively the variable
spread between the discount rate and the federal funds rate) and lagged reserve requirements that
interfered with banks' ability to adjust required reserves within a reserve maintenance period.8
(Between 1968 and 1984, there was a two-week lag between the reserve computation and maintenance
periods. Since 1984, required reserves on transactions deposits have been quasi-contemporaneous,
with a two-day lag at the end of the period.)
Another impediment to monetary base control was the way the discount window functioned.
Many of the supporters of base control suggested that the discount window be closed or that the
discount rate be set high enough to ensure that borrowing would be costly. The latter option would
not make the monetary base precisely controllable, since banks still could borrow reserves in excess of
those consistent with the target, but the high cost of doing so would discourage borrowing and make
the monetary base approximately obtainable. The severe restrictions on borrowing would be presumed
to limit deviations between the monetary base and the nonborrowed base.
Other analysts claimed that removing the impediments would not solve the problems of
controlling the base. They argued that depository institutions could not easily make the large
adjustments to loans and investments or to deposits needed to achieve the required reserve levels
consistent with the desired monetary base. They reasoned, first, that each individual institution would
have difficulty gauging whether an excess or shortage of reserves relative to its requirements resulted
from a deliberate policy adjustment, from a poor distribution of reserves in the banking system—either
among depository institutions or over the maintenance period~or from a Federal Reserve error in
predicting reserve balances on the day. Only the first of those sources would induce depository
institutions to consider taking the steps needed to change required reserves; changes in reserves from
^ornton (1982) discusses these arguments.
29
the latter two sources would be quickly reversed, and banks would attempt to maintain existing levels
of reserves and deposits. Each individual depository institution would have trouble discovering the
true reserve picture from observing its reserve levels because flows of reserves are huge relative to
average reserve balances. Many large banks routinely experience reserve positions that within a day
range from being overdrawn to being substantially in excess, even when overall reserve levels are
consistent with requirements.
Second, even when a depository institution correctly perceived a policy-based reserve shortage
or excess, its most likely response would be to adjust its loan and deposit rates and fees gradually.9
There are costs to making price changes because they have to be disseminated and because they may
complicate efforts to achieve good relations with customers. In addition, customers would not respond
immediately to the changes in pricing unless the short-run interest elasticity of demand for deposits
was very great. But most observers have perceived the short-run interest elasticity of deposits to be
relatively modest. In that circumstance, sharp swings in interest rates would have to occur to achieve
the necessary changes in deposits. A depository institution facing a reserve shortage or excess might
try to pass it to another institution by adjusting its activity in the federal funds market. If all
depository institutions did that simultaneously, the federal funds rate would show large short-run
movements.
Some observers argued that those large swings in rates would induce instability in the demand
for money through a complex lagged adjustment mechanism. Thus, money would not respond as
hoped to a reserve excess or shortage, but would exhibit a series of overshoots and reversals. Both the
targeted monetary aggregate and interest rates would be subject to undesirable short-run volatility.
To the extent that the supporters of strict monetary base targeting addressed this issue, they
suggested that depository institutions would respond by building up their normal holdings of excess
reserves. Once the institutions held enlarged levels of excess reserves, they could allow excesses to
vary in the short run whenever required reserves and nonborrowed reserves were inconsistent. Those
excess reserves would serve as a shock absorber to mitigate the short-term fluctuations in interest rates.
If excess reserves became large and variable, those opposing monetary base targeting argued,
the reserve-deposit ratio would become variable and the multiplier would cease to have the relatively
stable relationship observed in the historical data generated when excess reserves were small and the
'Demand deposits cannot pay explicit interest. Depository institutions may pay interest implicitly by
extending services for which they do not charge. Since rates on NOW accounts were decontrolled in 1986,
interest paid on them has been adjusted very slowly if at all.
30
monetary base was not a target. If the multiplier were unstable, controlling the base would not
achieve desired money targets.10 For the most part, the proponents of base targeting rejected these
objections. They believed that the multiplier was structurally stable because of mandated reserve
ratios and underlying customary behavior patterns and institutional structures. Hence, they anticipated
that the ratios would either maintain or quickly return to their norms (Brunner and Meltzer 1983). If
the ratio of currency to deposits is determined by the structure of payment practices, then currency will
be induced to shift in the same direction as the change in deposits, taking some of the adjustment
burden off reserves.
One issue rarely mentioned in the debate that probably deserved more attention was the
appropriate time period over which it was desirable to achieve control of the monetary base. Very
short-run control of the monetary base was never advocated for its own sake. Instead, it was seen as a
means to achieve a desired path for money and in turn for economic activity and prices. Many of the
supporters of a monetary base approach to money control argued for short-run targets for two reasons.
First, they were afraid that if deviations were permitted, the central bank would allow misses of
sufficient duration to affect adversely economic activity and prices. Second, they believed close
control was feasible at low cost.11 Their critics had a range of views with regard to the first point
but, as indicated above, disagreed with the second. Many of the operational and stability concerns
associated with monetary base targeting would be reduced if the time period for achieving the target
were lengthened beyond the single reserve maintenance period cited in most of the control proposals.
Presumably, the monetary base could be controlled with only modest errors over a one- to two-quarter
horizon if nonborrowed reserves—or the federal funds rate—were manipulated to bring the monetary
base back on track once evidence developed that it was deviating significantly from a desired path.
Technical control should be easier than it is for Ml or M2 because the base does have a closer
relationship to reserves and can be observed accurately with no lag. The more meaningful question is
whether the monetary base can be controlled in a way that does not introduce short-run instability to
interest rates and money demand. Without actual experience, it is difficult to answer that question.
10
Carlson (1981) describes the two viewpoints.
"Thornton (1983), generally a supporter of monetary base targeting, recognized that very short-term
control was not feasible because loan and investment decisions, under any reserve accounting scheme,
would not be closely linked with current reserve levels. Thus, the adjustment would not occur instantly.
He did not think that the delays would be a problem because banks would adjust over meaningful periods.
His arguments on this point are not typical of the literature supporting base targeting.
31
Nonetheless, the longer the control horizon, the more likely the answer will be that control of the base
without unacceptable instability in rates and money demand is indeed possible.
Although most proposals concerning the monetary base made since the early 1980s have
focused on its use as an intermediate target rather than as an operating target, one exception has been
the work of McCallum (1987, 1988a, 1988b, 1988c). Influenced by the breakdown of the relationship
between Ml and GNP in the 1980s, he suggested two modifications to the common monetary targeting
proposal. First, he suggested that nominal GNP could serve as an intermediate target as long as it was
chosen to ensure that inflation could not be rapid. Second, he advocated an adaptive policy rule that
contained the means for the procedure to recover when underlying relationships between the base and
GNP shifted as a result of deregulation or other developments. (See the paper by Spence Hilton and
Vivek Moorthy in this volume.) McCallum suggested that the monetary base could serve as the
operating instrument. He seemed to argue that the base could be controlled directly on a day-to-day
basis, although he only advocated a quarterly average growth target. McCallum did not discuss his
assertion that the monetary base could be controlled. His equation for quarterly target growth of the
monetary base consists of three terms.12 The first is a constant, equivalent to the desired trend
growth in nominal GNP-in his preferred specification, 3 percent at an annual rate. The second term
subtracts from the constant the average increase in monetary base velocity over the previous four
years. This term introduces a gradualresponseto changes in monetary base velocity so that the model
will adapt to changing trends. The third term provides for a partial response to deviations of GNP
from its constant growth target. McCallumrecommendeda 25 percent per quarter adjustment factor in
response to such deviations.
in. The Monetary Base as an Intermediate Target of Policy
The monetary base has been proposed as an intermediate target in place of a traditional
monetary aggregate in order to resolve the possible short-run control problems associated with such an
aggregate or to treat the monetary base itself as a narrow monetary aggregate. The idea that the
monetary base could be considered a monetary aggregate developed gradually. Lothian (1976) found
that the monetary base showed a more consistentrelationshipwith net national product than did a
broad monetary aggregate across groups of twenty-nine or forty countries (using two different
12
In equation form: bt - b^ = 0.00739 - (l/16)[xt_, - x ^ - bM + bM7] + .25 (x\x - xM), where b is the
log of the monetary base and x is the log of GNP. An asterisk indicates a target value.
32
formulations). He posited that differential interest rates, inflation, and regulations such as interest rate
ceilings and reserve requirements had a larger influence on the demand for deposits than they did on
the demand for non-interest-bearing central bank monetary assets.
Drawing on the econometric techniques used to evaluate monetary aggregates as potential
intermediate targets, analysts have tried to test empirically the ability of the monetary base to serve in
the role of intermediate target for the United States. The traditional approach has been to estimate a
model to see how well it predicts the behavior of economic activity or prices. Although one may
question whether using a model built to accommodate the observed characteristics of one monetary
variable is appropriate for evaluating monetary variables with a different set of characteristics, the
approach has been widely used. Differences in methodologies and assumptions have yielded
conflicting results.
A. Studies in Which GNP Is the Dependent Variable
A number of authors have estimated "St. Louis-type" models, substituting the monetary base
for Ml or M2. Models of this type were developed at the St. Louis Federal Reserve in the late 1960s
by Andersen and Jordan and updated and modified in subsequent years. In these models, a measure of
nominal income (generally GNP) is regressed on a measure of money and a measure of fiscal actions
(often high employment expenditures). Initially the models were estimated in first difference form,
although later the variables were typically specified as rates of change. Sometimes variables have
been added to allow for the effects of strikes and other shocks, such as the two large increases in oil
prices in the 1970s. A series of studies were made in the late 1970s and early 1980s that compared
the models' performance when the monetary base was used as the monetary aggregate with the
models' performance when Ml was the aggregate. The models were estimated with data covering the
previous twenty to thirty years. The models generally achieved better fits for Ml than for the
monetary base, although the differences ranged from trivial to very large depending on specification
and time period. A few of the authors offered simulations after the estimation period that tended to
provide predictions of similar quality for the monetary base and Ml. Andersen and Karnosky (1977)
used a simplified version of the St. Louis equation from which they omitted the fiscal variable
(because the sum of coefficients on the fiscal variable was usually insignificant). They estimated their
model from 1952 to 1961, using quarterly data, and then extended it year by year through 1975.
(They did not report their regression results.) In each case, they used the estimated model to simulate
the next four quarters' GNP. They found that in the equations using Ml, the errors had modestly
33
lower variance than the errors in the equations using the monetary base. However, the equations using
the monetary base achieved lower mean errors and mean absolute errors. The authors concluded that
it would be worthwhile to consider using the monetary base as an intermediate target because its
forecasting performance was only slightly worse than Ml, and it was easier to control.
Davis (1979-80) ran similar simplified St. Louis model regressions, which also omitted the
fiscal variable. His regressions used a sample period from 1961 to 1978 and two subperiods. He
found significantly better fits for Ml than for the monetary base, using as measures of goodness of fit
either R2 adjusted for degrees of freedom or the standard error of the estimate. (Table 1 contains
highlights of reported results for this and other studies.) He also estimated similar equations using
either the currency component or the reserve component of the monetary base. The limited
relationship he found between the monetary base and GNP seemed to derive from the currency
component, not the total reserve component.
Cullison (1982) argued that the monetary base equation in Davis* formulation was misspecified
because it contained the current value of the monetary base, which was largely endogenous over the
sample period. He also criticized the choice of sample period, arguing that the starting and ending
points should coincide with the same stage in the business cycle to allow for the cyclical pattern in the
coefficients. His equations used 1959 as a starting point and ended in 1969, 1973, and 1979 (all cycle
peaks). His results show much less of a gap between Ml and the monetary base, if judged by the
values for adjusted R2. His two formulations, with and without the current value of the monetary base,
give similar results.
Gambs (1980) estimated somewhat different models from those used by Davis. He made his
estimates over a longer time period (1953-78) and included a variable for high employment
expenditures, an approach closer to the traditional St. Louis model. The monetary variables, especially
the base, explained larger portions of the variation in GNP than they did in Davis9 studies, but Gambs
still observed a sizable differential between Ml and the monetary base.
Hafer (1984) made a study using a sample period running from 1960 to 1980. He used a
more elaborate version of the St. Louis model that included high employment expenditures and special
factor variables for oil price impacts, strikes, and wage and price controls. He allowed coefficients to
vary over time. With GNP as the dependent variable, he achieved better overall results, presumably
because additional variables were included. He found that the version using Ml showed only a
slightly better fit than the version using the monetary base.
Benjamin Friedman and Kuttner (1989), using data available in 1988, reestimated equations
34
Table 1
Nominal GNP Equations: Selected Model Results
Davis (1979-80) and (1990)
GNP, = a + I&nvi for i = 0.....4.
1961-1 to 1978-IV
1961-1 to 1969-IV
1970-1 to 1978-IV
(Davis 1990)
1981-1 to 1989-IV
Financial Measure
R2
SEE
R2
SEE
R2
SEE
52
SEE
Ml
0.31
2.96
0.23
2.25
0.19
3.64
0.06
3.67
M2
0.25
3.07
0.27
2.19
0.08
3.88
0.02
3.76
MB-B
0.08
3.41
0.01
2.56
-0.06
4.16
0.16
3.48
Total reserves
-0.02
3.59
0.19
2.31
-0.11
4.26
-
-
Currency plus nonmember
bank cash vault
0.11
3.35
0.13
2.40
0.01
4.03
-
-
Cullison (1982)
GNP, = a + E^im,.! for i = 0....4.
1959-IV to 1969-IV
1959-IV to 1973-m
1959-IV to 1979-IV
R2
SEE
R2
SEE
R2
SEE
Ml
0.360
0.0066
0.364
0.0072
0.371
0.0078
MB-SL
0.230
0.0072
0.280
0.0076
0.290
0.0083
Financial Measure
GNP, = a + Iib,.iMB,.i for i = 1,2.
1959-IV to 1969-IV
1959-IV to 1973-m
1959-IV to 1979-IV
Financial Measure
R2
SEE
R2
SEE
R2
SEE
MB-SL
0.270
0.0070
0.270
0.0077
0.260
0.0085
35
Gambs (1980)
GNP, = a + Zjbim,.! + I^E,., for i = 0.....4.
1953-1 to 1978-IV
R2
SEE
Ml
0.41
3.74
1.18(6.53)
MB-SL
0.31
4.04
0.94(5.67)
Financial Measure
5^b,.,
Hafer (1984)
GNP, = a + p, XJwnn + fc I , . ^ + Z, for i j = 0.....4
1960-1 to 1970-11
1970-m to 1980-IV
Financial Measure
R2
SEE
Ifo.;
R2
SEE
2^),.,
Ml
0.479
2.320
1.142
(3.31)
0.469
2.992
1.361
(2.92)
MB-SL
0.468
2.345
0.955
(3.42)
0.460
3.018
2.191
(2.99)
Friedman and Kuttner (1989)
GNP, = a + Ijbim,.! + Z^E,., for i = 1.....4.
1960-n to 1979-m
1960-n to 1986-IV
1970-1 to 1986-IV
Financial Measure
R2
R2
52
Ml
0.32
0.11
0.02
MB-B
0.23
0.10
0.02
M2
0.27
0.19
0.06
36
Board Staff (1988)
GNP, = a + 2ibini,.i + EftE^
for i = 0.....8 (or less).
1961-n to 1979-IV
R2
SEE
Ml
0.21
3.4
MB-B
0.06
3.7
Ml-A
0.19
3.4
M2
0.18
3.5
Financial Measure
Stone and Thornton (1988)
GNP, = a + ZibiiiVi + IjdjGNP,.j
1961-1 to 1980-IV
Financial Measure
W
for i = 0,...,4 and j = 1,2.
1961-1 to 1987-11
R*
1981-1 to 1987-H^
W
Ml
0.356
0.159
0.348
MB-SL
0.230
0.156
0.422
MIA
0.334
0.142
0.468
M2
0.199
0.179
0.295
Total reserves
0.204
0.132
0.320
- For this sample period, i = 0,1,2 and j = 1.
37
Naive Model
GNPt = a + 2jd,GNPt.j for j • 1,...,7 (first two columns)
for j = 1,...,4 (third column)
1961-1 to 1980-IV
1961-1 to 1987-11
52
0.069
R2
0.109
1981-1 to 1987-11
R2
0.244
Notes: t-statistics are in parentheses. Notation changed from that of the author to make variables consistent.
GNP
m
E
MB-SL
MB-B
Z
=
=
=
annual rate of change in nominal GNP
annual rate of change infinancialmeasure
annual rate of change in high employment
expenditures
= annual rate of change of monetary base-St Louis
measure
= annual rate of change of monetary base-Board of
Governors measure
= energy and strike variables and price control
dummies.
similar to those of Gambs for 1960 to 1979. They found a similar differential in the performance of
Ml and the monetary base, but generally weaker relationships to GNP, presumably because of the
different sample period.
In the early 1980s, the seemingly dependable relationships between the various monetary
measures and GNP underwent substantial shifts. Two developments appeared to be primarily
responsible. The first was the deregulation of interest rates applied to a variety of types of deposits.
Such deregulation made it more attractive to hold wealth in money form and led to a downward shift
in the velocity of money. The second development was the fundamental reduction in the trend rate of
inflation, which also lowered velocity.
It was suggested that deregulation would have less effect on the monetary base than on Ml,
38
since neither component of the base paid interest in the earlier years or in the 1980s.13 Nonetheless,
the monetary base was somewhat affected by deregulation since induced changes in transactions
deposits affected its total reserves portion. While the currency component of the base might remain
stable, any instability in total reserves would complicate targeting and control. Presumably, the
reduction in the trend rate of inflation, which increased the demand for money generally, would affect
the demand for the monetary base as well as Ml and M2.
In any case, as the 1980s unfolded, the data strongly suggested that structural shifts had in fact
occurred. The shifts were most dramatic for Ml, as the charts of income velocity in the introductory
essay in this volume illustrate, but they also occurred in the monetary base velocity relationship. It
would be hardly surprising in the face of such patterns that extending the observed log-linear
relationships underlying the St. Louis equations would produce unsatisfactory results. Nonetheless, a
number of authors pursued such an approach.
Friedman and Kuttner (1989) ran the same model cited above, extending the sample period to
1986. For sample periods running from 1960 to 1986 and 1970 to 1986, the proportions of the
variation in GNP that were explained by either Ml or the monetary base were very low.
Staff members at the Board of Governors (1988) estimated reduced-form equations that
included a monetary variable and a measure of fiscal stimulus for the period from 1961 to 1979. The
sum of the coefficients on the monetary variable was constrained to equal one to be consistent with
long-run neutrality of money. The estimates using the monetary base were considerably poorer than
those for Ml, Ml-A, or M2. The staff then simulated the models for the postsample period from 1980
to 1988. The simulations showed large errors for all measures.14 The monetary base showed bias,
but the error statistics were better than those on the standard narrow money definitions and similar to
those for M2 and a shift-adjusted version of Ml-A.
Stone and Thornton (1988) used rates of change of various money measures and lagged values
13
Gambs (1980) presents the argument without endorsing it. The Shadow Open Market Committee
recommended monetary base targeting, partly to deal with deregulation.
14
Summary statistics for Board staff simulations for 1980-1 to 1988-IH:
Mean absolute error
Root mean squared error
Bias (mean error)
SH=shift-adjusted
Ml
6.1
7.4
-4.5
Base
4.8
5.6
-3.3
Ml-A
5.1
6.7
-.2
MI-ASH
4.5
5.7
-1.4
M1SH
6.1
7.5
-4.1
M2
4.3
5.5
-1.6
M2SH
3.9
5.1
-1.2
39
of GNP as independent variables and obtained results that were generally similar to those obtained by
others over the initial sample period of 1961 through 1980. Extending the period of estimation to 1987
weakened the explanatory power but greatly reduced the differences among monetary measures.
Furthermore, out-of-sample simulations from the first estimations, over the period from 1981 to 1986,
showed considerable deterioration with respect to both Ml and the monetary base.
Stone and Thornton went on to reestimate the models, using data generated during and after
the structural shifts described above. When the authors reestimated the model with 1981 as the
starting date (and reduced the lags to gain degrees of freedom), the values for adjusted R2 actually
improved compared with those obtained for the earlier period, and the monetary base outperformed all
of the other measures except MIA, to which it was a close second. They argued that the monetary
base could appropriately play a role in the policy process.
In the introduction to this volume, Davis offers updated versions of some of his earlier
equations, now estimated over 1981-89. He achieves the same ordering as Stone and Thornton, with
the base improving and Ml deteriorating relative to the earlier period. But the values for adjusted R2
and the coefficients are very different from those found by Stone and Thornton. While the time
periods are not identical, the main source of the difference is the inclusion or exclusion of lagged
values of nominal GNP with the independent variables. (The effects of alternative specifications are
discussed below.)
Milton Friedman (Darby et al. 1987) performed a slightly different exercise and reached a
similar conclusion. He calculated eight-quarter moving standard deviations of the quarterly change in
income velocities of Ml, M2, and the monetary base, both before and after the structural shift of the
early 1980s. He found no increase in the size of the standard deviations of these measures in the
1980s, suggesting that velocity had not become more unstable. He claimed that monetary base
velocity had the smallest standard deviation of the three measures, although the differences among
them were not dramatic. (He presented his results in a graph. Visually, Ml appears to have
deteriorated somewhat, while it is hard to tell about the other measures.) I repeated the calculations he
described using data covering 1972-80 and 1982-89. The first time period is shorter than that
examined by Friedman, who began his exercise with 1961. The second time period is longer;
Friedman's sample ended in 1986. My calculations show some deterioration for the standard
40
deviations of Ml and M2 velocity and some improvement for the monetary base.15
B. Studies in Which Prices Are the Dependent Variable
Another pair of studies (including one cited above) examined the impact of the choice of
monetary variable on the rate of change in prices rather than on nominal income. Unfortunately, these
studies only examined the relationships through the 1970s. Compared with the GNP equations over
similar sample periods, the monetary base did relatively better, and in one of the studies it came out
ahead of Ml.
Hafer substituted the rate of change in prices for GNP in equations that included longer lags
than his GNP equation. He observed a stronger relationship between the monetary variables and
prices than between the monetary variables and GNP. Ml still outperformed the monetary base, but
the difference narrowed (Table 2).
Fama (1982) used a different approach from the others, and concluded that the monetary base
was superior to Ml as a monetary measure. Fama developed a model based on Fisher's work and
various rational expectations models in which the long-run quantity theory was assumed to hold. He
posited that the rate of change in the demand for real money was positively related to the rate of
change in anticipated real activity and negatively related to the nominal interest rate (assuming money
does not pay interest). He then made the assumption that real economic activity is determined outside
the money sector (in keeping with the long-run neutrality of money hypothesis).
From these assumptions, he built a model to test the performance of alternative definitions of
money. He ran a number of variants of the model, estimated from 1954 to 1976, first using annual
data and then quarterly or monthly data. Inflation was expressed as a function of the monetary
variable and nominal interest rates (both of which were expected to have positive coefficients) and as a
function of actual and anticipated economic activity (expected to be negatively related). When the
three-month Treasury bill rate proved insignificant, Fama dropped it. He used current money and
money lagged one period. Initially, he tried money with a lead as well, on the assumption that it
might serve as a proxy for the expected course of monetary policy, but dropped it when the
15
Standard deviations of eight-quarter moving averages of logs of velocity of various monetary
measures:
Mi
Mi
MB-B
MB-SL
1972-80
0.02333
0.01388
0.01578
0.01663
1982-89
0.02587
0.01614
0.01238
0.01252
41
Table 2
Price Equations: Selected Model Results
Hafer (1984)
P, = Eibimt.i + Z, for i = 0.....20
1960-1 to 1970-n
1970-ffl to 1980-IV
Financial Measure
W
SEE±'
2,b,.|
R2
SEE^
2,bt.,
Ml
0.737
0.896
0.942
0.681
1.278
0.972
MB-SL
0.671
1.002
0.848
0.504
1.593
0.749
Fama (1982)
P, = a + £,b,nit, + EjCjA^
Financial
Measure
Economic
Activity
Measure
Ml
for i = 0,1.
*?
R2
SEE
rr*
rIP
0.79
0.0130
0.42
0.11
Ml
rGNP
0.80
0.0126
0.50
MB-SL
rIP
0.91
0.0083
MB-SL
rGNP
0.95
DD
rIP
DD
rGNP
£.b,
j -
-0.03
0.92
-1,0,1
0.12
-0.07
0.90
-1,0
0.06
-0.08
-0.01
0.71
-1,0,1
0.0064
0.38
0.07
-0.12
0.72
-1,0
0.65
0.0167
0.52
0.22
0.02
0.85
-1,0,1
0.65
0.0166
0.55
0.21
0.00
0.82
-1,0
Notes: Notation changed from that of the author to make variables consistent.
P
m
Z
Ml
MB-SL
A
rIP
rGNP
DD
=
=
=
=
=
=
=
=
=
annual rate of change in prices (Hafer: GNP deflator, Fama: CPI)
annual rate of change in financial measure
energy and strike variables and price control dummies
annual rate of change in Ml
annual rate of change in monetary base-St. Louis measure
annual rate of change in economic activity measure
annual rate of change in real industrial production
annual rate of change in real gross national product
annual rate of change in demand deposits.
-;SEE
%rk
= standard error of the estimate.
= residual autocorrelation coefficient (k = 1,2,3).
42
coefficients proved insignificant. He measured economic activity alternately with industrial production
and real GNP. Current economic activity was included along with activity with a one-period lag and a
one-period lead. The leading variables (which had highly significant negative coefficients) were
justified as proxies for anticipated real economic activity. Expecting objections to the use of leading
variables, he also proxied anticipated activity with a stock price index, which gave a poorer fit but had
the same sign and did not substantially change the coefficients of the other variables.
He ran the various models, alternately including the (St. Louis) monetary base, Ml, or demand
deposits as the monetary variable. In each model variant, the base dominated both Ml and demand
deposits as measured by adjusted R2, the standard error of the regression, and lagged residual
autocorrelations. He reran some of the models using quarterly and monthly data and achieved the
same ordering of results. Hence, he concluded that the monetary base was the appropriate monetary
variable to follow to achieve a price goal. His model, which is complex, does not appear to have been
updated to include the 1980s. McCallum cited Fama's work as justification for using the monetary
base as a target variable. It is unfortunate that he did not offer an update of the price relationship.
C. Empirical Estimates of GNP and Price Relationships
To gain some additional insights into relationships using the rate of change in nominal GNP as
the dependent variable in the 1980s, I estimated a variety of regressions similar in formulation to many
of those reported here. Two time periods were used, 1972-80 and 1981-89. To allow for the observed
structural shifts in the early 1980s, it was necessary to use relatively short time periods for this type of
estimation. Hence, the results must be interpreted very cautiously. The exercises suggest some ratios
arising from sensitivity to the specifications used. Including or excluding lagged GNP or the
contemporaneous monetary variable made sometimes sizable differences in the results, as did
specifications of the lag structure (Table 3).
Results can be judged by values of adjusted R2 or by the plausibility of sums of coefficients
(which for most of the monetary variables might be greater or less than one but which should not
differ dramatically from one). In all specifications and under either definition of the monetary base, the
base performed considerably better in the later period than in the earlier period. Indeed, it generally
showed a better relationship to GNP in the later period than Ml showed in the earlier period. Ml was
the only monetary variable tested that showed reasonable results in the earlier period. The
deterioration in Ml as a predictor of GNP between the 1970s and the 1980s is dramatic if judged by
sums of coefficients but modest if judged by error statistics. The M2 equations do not appear to be
43
Table 3
*
A Summary of Selected Regressions Using Nominal GNP as the Dependent Variable
Time Period of Estimate: 1972-80
Unconstrained
Mon Variable
with Lagged
GNP; Contemp
& Lagged Money
Unconstrained
Mon Variable
with Lagged
GNP; Lagged
Money Only
Poly Distr Lag
Poly Distr Lag
Not Tied;
Not Tied;
with Lagged GNP; with Lagged GNP;
Contemp & Lagged Lagged Money
Only
Money
Poly Distr Lag
Poly Distr Lag
Not Tied;
Not Tied;
No Lagged GNP;
No Lagged GNP;
Contemp & Lagged Lagged Money
Only
Money
Poly Distr Lag
Tied Far End;
Poly Distr Lag
Tied Far End;
with Lagged GNP; with Lagged GNP;
Contemp & Lagged Lagged Money
Money
Only
Poly Distr Lag
Tied Far End;
No Lagged GNP;
Contemp & Lagged
Money
Statistical Measure
M1
AdjR2
St err regress
Sum coef
.38
3.61
1.23
.07
4.43
.62
.41
3.51
1.28
.15
4.22
.74
.40
3.55
.87
.11
4.33
.09
.37
3.62
1.33
.17
4.17
.78
.38
3.62
.95
M2
AdjR2
St err regress
Sum coef
.05
4.47
.58
.08
4.4
.56
.09
4.38
.56
.08
4.38
.55
.11
4.33
.49
.11
4.34
.44
.11
4.32
.57
.11
4.32
.56
.13
4.27
.50
St err regress
Sum coef
-.1
4.81
-.03
-.1
4.83
-.46
-.13
4.88
-.19
-.08
4.77
-.46
-.10
4.81
-.26
-.84
4.78
-.50
-.1
4.80
-.18
-.11
4.82
-.53
-.07
4.74
-.25
Total reserves
AdjR2
St err regress
Sum coef
-.11
4.84
-.15
-.07
4.75
-.15
-.08
4.77
-.15
-.07
4.75
-.15
-.03
4.65
-.15
-.03
4.65
-.13
-.05
4.70
-.15
-.04
4.67
-.14
-.01
4.60
-.15
-.06
4.72
.02
-.09
4.8
-.67
-.02
4.64
.06
-.09
4.80
-.67
.03
4.52
.06
-.04
4.69
-.71
.01
4.57
.04
-.07
4.74
-.65
.06
4.45
.04
-.08
4.76
1.37
-.11
4.84
.41
-.04
4.68
1.39
-.11
4.84
.42
-.01
-.06
4.73
-.01
4.60
1.37
-.08
4.76
.44
.02
4.54
1.09
Currency
AdjR2
Mon base—Board
AdjR2
St err regress
Sum coef
Mon base—St Louis
AdjR2
St err regress
Sum coef
Note: Polynomial distributed lags are third order.
4.60
1.15
.20
Table 3 - Continued
Time Period of Estimate: 1981-89
Unconstrained
Unconstrained
Mon Variable
Mon Variable
with Lagged
with Lagged
GNP; Lagged
GNP; Contemp
and Lagged Money Money Only
Poly Distr Lag
Poly Distr Lag
Not Tied;
Not Tied;
with Lagged GNP; with Lagged GNP;
Contemp & Lagged Lagged Money
Only
Money
Poly Distr Lag
Poly Distr Lag
Not Tied;
Not Tied;
No Lagged GNP;
No Lagged GNP;
Contemp & Lagged Lagged Money
Money
Only
Poly Distr Lag
Poly Distr Lag
Tied Far End;
Tied Far End;
with Lagged GNP; with Lagged GNP;
Contemp & Lagged Lagged Money
Money
Only
Poly Distr Lag
Tied Far End;
No Lagged GNP;
Contemp & Lagged
Money
Statistical Measure
M1
AdjR2
St err regress
Sum coef
.31
3.15
.13
.29
3.19
.21
.37
3.00
.11
.31
3.13
.23
.23
3.31
.04
.10
3.58
.22
.21
3.37
.03
.19
3.40
.22
3.46
0
M2
AdjR2
St err regress
Sum coef
.23
3.34
.31
.20
3.39
.44
.25
3.27
.31
.20
3.37
.43
.12
3.54
.32
.05
3.69
.47
.13
3.53
.25
.19
3.40
.41
.05
3.69
.26
Currency
AdjR2
St err regress
Sum coef
.43
2.87
1.62
.44
2.84
1.66
.41
2.89
1.71
.43
2.85
1.66
.43
2.85
2.00
.43
2.84
2.16
.32
3.11
1.77
.45
2.81
1.64
.36
3.02
1.86
St err regress
Sum coef
.19
3.41
-.04
.09
3.61
-.02
.12
3.54
-.05
.09
3.60
-.01
.05
3.70
-.11
-.05
3.87
-.07
.14
3.51
-.05
.06
3.67
-.01
.06
3.68
-.11
Mon base—Board
AdjR2
St err regress
Sum coef
.31
3.15
.85
.31
3.16
.94
.21
3.35
.90
.31
3.14
.91
.17
3.44
1.01
.17
3.45
1.25
.17
3.45
.78
.30
3.16
.99
.13
3.53
.92
Mon base—St Louis
AdjR2
St err regress
Sum coef
.36
3.02
.71
.36
3.04
.81
.36
3.02
.76
.35
3.04
.80
.30
3.17
.89
.22
3.34
1.11
.24
3.30
.70
.37
3.00
.78
.24
3.31
.81
Total reserves
AdjR2
t*
.16
Chart 3
Currency as a Percentage of Nominal GNP
4.8
4.6 h
4.4
©
a.
4.2 h
4 h
38
1970
1975
1980
1985
1S
Note: Nominal GNP and currency values are seasonally adjusted. Ratios are
plotted quarterly.
well specified in either period. They are poorer predictors than other work on M2 might suggest16
When currency was placed in the equations, the values for adjusted R2 were much higher in the later
period than in the earlier period-when there was no apparent relationship. During the 1970s, currency
represented an expanding share of Ml (Chart 2) but declined relative to GNP (Chart 3). In the 1980s,
currency's share of Ml showed no trend, while currency as a percent of GNP rose modestly. The
estimated coefficients for currency seemed implausibly high in the later period. The equations did not
pick up a significant relationship between total reserves and GNP in either period. The behavior of
total reserves was different in the two decades, with total reserves falling sharply as a share of Ml in
the 1970s and more slowly in the 1980s (Chart 1). Movements in total reserves responded to both the
behavior of reservable deposits and the shifts in the reserve ratios arising from sources other than
16
See Wenninger (1990).
46
formal changes in requirements. Using the monetary base, which combines currency and total reserves,
in the equation produced more plausible coefficients than using either currency or total reserves
separately. These results give some support to the notion that the monetary base might have a role to
play as an intermediate target or at least as an indicator. The important role of currency in the
relatively good results achieved for the base in the 1980s means that one would want to know more
about why currency behaved as it did and whether it was likely to continue to do so.
When prices were used as the dependent variable in empirical work covering the 1980s, the
short estimation period was even more of a handicap than it was for nominal GNP. Results obtained
for the earlier years suggested that the monetary variables affected prices with lags that were on
average significantly longer than those observed for nominal GNP. Furthermore, the strong results
obtained by Fama and Hafer were achieved with relatively complex models incorporating rather
specific views of the relationships among economic variables and would be hard to replicate.
Nonetheless, while recognizing that the exercise was subject to many pitfalls, I estimated
several regressions with rates of price change as the dependent variable.17 Prices were defined
alternately with the GNP deflator or the consumer price index (CPI). The two price measures
produced some differences but a generally similar ordering of results. Table 4 presents selected results
for regressions that used the implicit deflator as the dependent variable. Most of the equations also
included lagged values of real GNP. Those without lagged real GNP generally had insignificant
values for adjusted R2 and less plausible coefficients. (When real GNP was included, the monetary
coefficients could be expected to be modestly lower than one.) With the notable exception of
currency, all of the monetary measures showed considerably poorer results for the 1980s than they did
for the 1970s if judged by adjusted R2. For the monetary base, the coefficients were more plausible in
the 1980s than in the 1970s, although very low when real GNP was omitted. They nonetheless were
more plausible than those on currency. The results for the 1980s were disappointing, although the
short time period and the number of changes in money demand occurring in the early part of the
decade were probably contributory factors.
More broadly, one could ask how much light any of these exercises, those with nominal GNP
or those with a price index as the dependent variable, shed on the issues they are supposed to address,
namely, the relative merits of these measures as intermediate targets of monetary policy. The basic
17
To gain a few more observations, 1980 was chosen over 1981 as a starting date in these regressions,
although the earlier date may have introduced more distortions from the transition period.
47
Table 4
00
A Summary of Selected Regressions Using the Implicit Deflator as the Dependent Variable
Time Period of Estimate: 1972-80
Unconstrained
Mon Variable
with Contemp &
Lagged Real GNP;
Contemp & Lagged
Money
Poly Distr Lag
Tied Far End; with
Lagged Real GNP;
Contemp & Lagged
Money
Poly Distr Lag
Tied Far End; with
Lagged Real GNP;
Lagged Money
Only
Poly Distr Lag
Not Tied; with
Lagged Real GNP;
Contemp & Lagged
Money
Poly Distr Lag
Not Tied; with
Lagged Real GNP;
Lagged Money
Only
Poly Distr Lag
Not Tied; No
Lagged Real GNP;
Contemp & Lagged
Money
Poly Distr Lag
Not Tied; No
Lagged Real GNP;
Lagged Money
Only
Statistical Measure
Ml
AdjR2
St err regress
Sum coeff
.38
1.88
.78
.35
1.92
.68
.35
1.93
.65
.39
1.86
.76
.41
1.83
.77
.32
1.97
.53
.33
1.95
.53
M2
AdjR2
St err regress
Sum coeff
.52
1.65
-.08
.58
1.54
-.08
.49
1.71
-.29
.59
1.54
-.01
.54
1.62
-.10
.54
1.62
-.36
.50
1.69
-.39
St err regress
Sum coeff
.34
1.94
.90
.38
1.87
.63
.36
1.91
.84
.36
1.91
.64
.34
1.94
.84
.26
2.06
.92
.23
2.09
1.17
Total reserves
AdjR2
St err regress
Sum coeff
.17
2.17
.02
.19
2.15
.06
.20
2.13
.07
.16
2.19
.06
.17
2.17
.07
.02
2.36
.13
.05
2.32
.13
.57
1.57
1.64
.62
1.48
.59
1.53
1.92
.58
1.55
1.94
.48
1.97
.60
1.50
1.96
1.72
2.45
.46
1.75
2.42
.56
1.58
1.92
.56
1.58
2.24
.56
1.58
2.22
.54
1.61
2.24
1.61
2.22
.29
2.01
2.75
.30
1.99
2.73
Currency
AdjR2
Mon base—Board
AdjR2
St err regress
Sum coeff
Mon base—St Louis
AdjR2
St err regress
Sum coeff
Note: Polynomial distributed lags are third order.
.54
Table 4 - Continued
Time Period of Estimate: 1980-89
Unconstrained
Mon Variable
with Contemp &
Lagged Real GNP;
Contemp & Lagged
Money
Poly Distr Lag
Tied Far End; with
Lagged Real GNP;
Contemp & Lagged
Money
Poly Distr Lag
Tied Far End; with
Lagged Real GNP;
Lagged Money
Only
Poly Distr Lag
Not Tied; with
Lagged Real GNP;
Contemp & Lagged
Money
Poly Distr Lag
Not Tied; with
Lagged Real GNP;
Lagged Money
Only
Poly Distr Lag
Not Tied; No
Lagged Real GNP;
Contemp & Lagged
Money
Poly Distr Lag
Not Tied; No
Lagged Real GNP;
Lagged Money
Only
Statistical Measure
M1
AdjR2
St err regress
Sum coeff
-.06
2.71
-.17
.10
2.49
-.11
.07
2.54
-.19
.08
2.53
-.11
.04
2.58
-.17
-.03
2.67
-.26
-.02
2.66
-.28
M2
AdjR2
St err regress
Sum coeff
.25
2.74
-.01
.05
2.56
-.36
.08
2.52
-.06
.03
2.60
-.04
.05
2.56
-.06
-.04
2.69
-.21
-.03
2.67
-.13
Currency
AdjR2
St err regress
Sum coeff
.76
1.56
2.44
.64
1.58
2.32
.63
1.60
2.24
.65
1.55
2.40
.65
1.56
2.34
.42
2.00
2.19
.43
2.00
2.23
St err regress
Sum coeff
.46
2.32
-.06
.31
2.18
-.03
.32
2.17
-.04
.29
2.22
-.03
.30
2.20
-.04
.12
2.47
-.02
.09
2.51
-.04
Mon base—Board
AdjR2
St err regress
Sum coeff
.15
2.43
.70
.23
2.31
1.12
.15
2.43
.90
.21
2.35
1.11
.12
2.46
.91
-.08
2.74
.45
-.08
2.74
.40
.39
2.46
.54
.21
2.34
.77
.12
2.47
.67
.19
2.38
.79
.10
2.50
.70
-.08
2.74
.21
-.09
2.74
Total reserves
AdjR2
Mon base—St Louis
AdjR2
St err regress
Sum coeff
s
.21
model from which all of them were drawn was built to capture the behavioral characteristics that had
been observed for Ml in the 1950s and 1960s. At the time, none of the components of Ml paid
explicit interest, and interest sensitivity of the demand for money did not appear to be very large. The
relationships have all changed in recent years, with short-run interest rate sensitivity increasing
considerably, especially for Ml. M2 has been sensitive to relative interest rate moves all along,
although the patterns have changed as deregulation of rates has spread. Furthermore, since each of the
monetary measures has different characteristics, it is doubtful that the same model would prove
optimal for each of the monetary variables. Thus, there is no reason to believe that models developed
in the 1960s to explain Ml behavior would reveal the correct relationship or even the correct ordering
among monetary variables over the 1980s. It is possible that better relationships between the
aggregates and the target variables could be found than any of these equations have discovered.
IV. Does the Concept of the Monetary Base Make Sense?
Benjamin Friedman (1988), commenting on McCallum's article (1988a), questioned the logic
of using the monetary base as either an indicator or as an intermediate target. He had two basic
criticisms. First, a large portion of the monetary base consists of currency, and the demand for
currency is not well understood. He noted that currency outstanding as of the end of 1986 amounted to
almost $900 for every U.S. citizen, far in excess of what surveys suggest most people hold.18 Some
currency is used for illegal transactions that are not captured by U.S. GNP statistics. A portion is used
for transactions in countries where the local currency is not stable or is not freely convertible.
In addition, he simulated McCallum's equations for 1963-87, substituting alternately currency
and total reserves for the monetary base. On the basis of root mean squared errors, currency
outperformed the base while total reserves produced very large errors.
Friedman did not try to explain the poor relationship between total reserves and economic
activity. One possibility was that the ratio of total reserves to M2, the aggregate with the best track
record in the 1980s, might have been unstable. As Chart 4 shows, however, the ratio of total reserves
to M2 was reasonably steady. Indeed, in the 1980s, the ratio of total reserves to M2 showed less
quarterly variation than the ratio of total reserves to Ml (Chart 1), even though a large share of M2
deposits were not subject to reserve requirements. Friedman's poor results for total reserves may have
reflected variations in the ratios of currency to deposits and reserves to deposits that were not picked
18
The figure was around $900 million (not seasonally adjusted) at the end of 1989.
50
Chart 4
Currency and Total Reserves as Shares of M2
c
2 5
Q>
Total reserves as a
share of M2
I , ., I ,, , I , , ,I
i •,, i , , ,
11 1 1 1 1 1 1
_L , I , , , I . . . I , • , I
1970
1975
1980
Note: Reserve data are seasonally adjusted and are adjusted for changes in
reserve requirements. M1 and currency values are seasonally adjusted. Ratios
are plotted quarterly.
1
1985
1989
up by the model. Some formulations of the St. Louis equations reported in the preceding section give
slightly better results for total reserves than Friedman found (although they are still poor). Thus, on
the basis of limited evidence, it is hard to make a choice between currency and the monetary base.
The validity of either currency or the monetary base as money measures was debated
implicitly in the early twentieth century. The question was asked, Should deposits be considered
"money," or should money be defined to consist only of gold and other government obligations?
Irving Fisher and other economists argued persuasively that deposits performed essentially the same
payment and accounting services as currency and coin. Therefore, ignoring them would lead to a
serious understatement of existing monetary services. The subject was mostly left to rest until the
1970s.
Lothian (1976) presented a rationale for a narrow money measure such as the monetary base
or currency. He argued that all financial assets provided a mix of money- and bond-type services.
51
Currency and non-interest-bearing reserves have little opportunity to provide bond-type services.
Deposits, on the other hand, may pay interest either explicitly or implicitly, and the institutions
accepting them may offer bond-type services to depositors. In an environment with variable
regulations and inflation, the mix of money- and bond-type services provided by deposits may vary
more than the mix provided by the non-interest-bearing monetary base. Consequently, Lothian
concluded that the base was likely to be a better proxy for monetary services when deposit
characteristics differed (over time or across countries). Lothian's cross-country empirical results were
very similar for the monetary base and currency.
A reasonably strong relationship between currency and economic activity might be plausible.
Currency is convenient for carrying out certain classes of transactions. If those transactions represent a
relatively steady share of economic activity, then a close relationship might be expected. One might
anticipate, however, that currency would perform a gradually declining share of monetary services
whenever the number and sophistication of substitute means of payment grew. During the 1970s and
early 1980s, rapid inflation stimulated efforts to find substitutes for non-interest-bearing currency. The
pattern over that period did show a decline in currency relative to GNP (Chart 3), although currency
was a growing share of Ml. Countering that tendency in more recent years has been slower average
inflation and the increased ease of obtaining currency from automated teller machines, developments
which have raised the value of currency as a provider of payment services. The irregular but rising
pattern for currency use during much of the 1980s may in part reflect those factors.
Nonetheless, Friedman's criticisms of currency as a potential intermediate target deserve
attention. The use of U.S. currency for transactions that are not included in U.S. GNP statistics
because they are illegal or take place in other countries would seem to be a serious problem only if the
proportion of currency used for such transactions was variable. Limited direct information about
variability over long periods of time is provided by two Board staff studies (Avery et al. 1986,1987).
The Board made surveys of cash holdings for transactions purposes by American adults in 1984 and
1986. Both surveys could account for about 11 to 12 percent of the cash outstanding. The 1987
article reported that the staff discovered a survey of cash holdings undertaken in 1944. That survey
"accounted for a proportion of circulation currency that was remarkably consistent with the 1984 and
1986 data" (Avery et al. 1987, p. 191). Considering the numerous changes in payment practices and the
rise of the illegal drug trade between the 1944 survey and the two more recent surveys, the similarity
is surprising. In 1944, currency represented a share of Ml similar to its share in the mid 1980s, but it
constituted a much larger share of GNP-on the order of 10 percent. (Of course 1944 was also a year
52
of turmoil, and people in the war zones would most likely have sought to hold U.S. currency.) In any
case, over short time periods U.S. currency can undergo significant unpredictable variations when
developments abroad provoke large movements of currency outside the United States.
If one should wish to make currency the centerpiece of policy, one confronts the problem of
how to control it. As long as currency is provided on demand, it is not directly controllable. Control
would have to be achieved indirectly by influencing demand. Variables that act upon the growth of
nominal economic activity, such as real interest rates and reserve provision, would influence currency
over time. The authorities, therefore, would have a greater chance of success if they used currency as
an intermediate indicator rather than as a target. They could use a procedure that adjusts for trend
changes and large shipments of currency abroad.
V. Experiences with Monetary Base Targets
The United States has no direct experience with monetary base targeting, but the nonborrowed
reserve targeting procedures employed between October 1979 and the autumn of 1982 have some
features in common with a nonborrowed base operating target. In addition, Switzerland and West
Germany have used measures similar to the monetary base in a role closer to an intermediate target
than an operating target.
A. Reserve Targeting in the United States
Under the procedures followed between 1979 and 1982, the primary operating target was
nonborrowed reserves. The procedures focused on intervals three weeks to five weeks in length, thus
encompassing more than a single one-week reserve maintenance period. By averaging reserves over
several weeks, the procedure avoided forcing offsets to very short-run deviations of reserves from the
desired path—deviations which were thought likely to be self-reversing. In many ways, the procedure
resembled one that could be used to target the nonborrowed base. No underlying growth rate
objective was set for nonborrowed reserves (or any other reserve measure) over quarterly or annual
time periods. Instead, the staff derived the targets for nonborrowed reserves by estimating the volume
of total reserves believed to be consistent with desired Ml. Under the procedure, the staff made
estimates of currency and of reserve ratios expected to be associated with desired Ml, two steps which
were also recommended for developing an operating target for the monetary base. The nonborrowed
reserve operating objective was derived from the total reserve target by subtracting an amount of
borrowing believed to be consistent with interest rate levels that in turn would be consistent with the
53
desired money growth. (In practice, the FOMC set the initial allowance for borrowing at a level close
to that most recently in effect. It occasionally raised or lowered the allowance modestly if money
growth had been more or less rapid than desired.) The operating targets were for nonborrowed
reserves rather than total reserves because, as indicated earlier, existing institutional arrangements did
not permit the Federal Reserve to control total reserves~or the monetary base-even if it had wanted to
do so.
The degree of similarity between targeting a nonborrowed reserve or nonborrowed base
measure and a total reserve or total base measure depends upon the way banks respond to the reserve
shortages or excesses. As noted in the earlier discussion, differing views on that response mechanism
lie at the heart of the debate about the controllability of the monetary base. Limited information from
the 1979-82 experience suggested that the process of adjusting Ml deposits to changes in reserve
availability began fairly quickly but continued over a period of several months. Thus, the
nonborrowed reserve targeting procedure produced reserve and money adjustments in the desired
direction within a reasonable time period, but the adjustments were not by any means instantaneous.
Even though the procedure allowed for some modest smoothing of the federal funds rate, the
funds rate varied considerably more, on a day-to-day and week-to-week basis, than it had before the
procedures were introduced (when deliberate actions had been taken to smooth the rate day to day).
Some of the rate variation may have reflected specific features of the institutional reserve accounting
structure, such as the length of the reserve maintenance period, the monetary authority's reaction when
depository institutions failed to meet requirements, the presence of lagged reserve requirements, and
the rules of access to the discount window. But much of this rate variation probably stemmed from
the lags that must inevitably be part of the banks9 adjustment process in a system involving many
institutions, large flows, and a complex set of motivations for bank pricing policies.
The other notable feature of the period was that Ml growth slowed on average and came
closer to its desired targets over extended periods of time, but it experienced considerable variability
from month to month and even quarter to quarter-more than it had before. The variability is not
completely understood, but the lags in the banks' response system could very easily have produced a
damped oscillatory response in Ml to the changes in reserve pressures. Some of that variation might
have been reduced through modifications to the targeting procedures, but there are limits to how much
reduction could have been achieved in the face of delays in the adjustments undertaken by the
commercial banks.
It appears that the nonborrowed reserve targeting procedure followed was able to bring Ml
54
back on target when it veered off and to prevent an off-target trend. Nonetheless, it did introduce a
substantial amount of short-run variability to both Ml and interest rates.
Monetary base targets abroad
For practical evidence using the monetary base as an intermediate target, one must look
outside the United States. This section reviews the experiences of Switzerland and of West Germany,
two countries that have made considerable use of monetary base targets. The Swiss National Bank
(SNB) began targeting the monetary base in 1979, after having targeted Ml from 1973 to 1978. As
described by Kohli and Rich (1986) and Rich (1987), the SNB considers the monetary base to be an
intermediate target. The base is believed to be reasonably well linked over time with price stability
and approximately controllable within a year's time frame through adjustments to reserves. The SNB
sets annual target rates, which are chosen to be consistent with moving as close to its long-run goal of
price stability as is considered feasible. The main restraint is that with heavy dependence on
international trade, exchange market conditions must also be considered. The SNB does not view
short-run manipulation of output and employment as appropriate activities for the central bank. It
attempts to achieve price stability in a way that causes as little fluctuation in output as possible.
Rich and Kohli reported that the SNB judged its experience with monetary base targeting to
have been largely successful, with a couple of notable exceptions. The SNB had its first problem with
monetary base targeting in 1979. A strong exchange rate and trade pressures meant that the SNB was
unable to shield the country successfully from the inflationary impulse of that year's oil price shock.
It temporarily set aside its monetary base target that year, permitting an overshoot, but then accepted
shortfalls in the next two years that offset the overshoot. Then in 1987, strength of the Swiss franc in
the exchange markets and concerns that such strength was damaging the export sector led the SNB to
try to discourage franc appreciation by pursuing a somewhat more expansionary policy than it would
have preferred.
During the 1980s, the SNB ran into some problems with shifting money demand (Belognia
1988). In contrast to most other western countries, Switzerland experienced a significant fall in the
demand for Ml early in the decade, apparently in several steps, and a more modest reduction in the
demand for the monetary base. As the SNB observed the shifts, it lowered its targets for the monetary
base. Because of these shifts, the SNB increased the attention paid to money market rates in order to
obtain early indications of possible money demand shifts. In 1988, the SNB introduced revised
liquidity requirements and a new interbank payments system. Both factors acted to lower the demand
55
for Ml. In response, the SNB deliberately undershot its base target in that year and again in 1989.
Undershooting was also deemed consistent with handling the inflationary impulse that had developed
in response to the earlier accommodative posture.
Targets for the monetary base apply only to the year as a whole. The growth of the base is
monitored during the year. Deviations from the target are tolerated if special factors suggest that they
are likely to be self-reversing. Otherwise the degree of reserve provision is adjusted to ensure that the
base will be brought back on target. Rich and Kohli reported that the SNB had the ability to control
the monetary base within what had been judged to be acceptable limits. In most years the actual
figure was within 1 percentage point of the target, and often it was closer. A miss of 1 percentage
point or less is not believed to do serious damage to the drive for price stability. The exceptions have
been years when a deliberate decision was made to miss the target because of an observed special
factor.
The SNB achieves the targets by adjusting reserve availability through temporary repurchase
operations against foreign currencies (known as swaps) and by placing limits on the commercial
banks' access to the discount and Lombard windows. The banks are required to report in advance any
planned large takedowns of discount or Lombard credit, giving the SNB room to offset the reserve
effects. This restriction in borrowing access has led the banks to hold more than minimal levels of
excess reserves. It has also led to occasional wide fluctuations in the overnight money market rates on
the last day of the month, when reserve requirements must be met, although these fluctuations have
little or no effect on longer term rates. Variations in excess reserves arise in the adjustment process,
since banks cannot instantaneously adjust their deposits when the base level is changed. The SNB has
judged monetary base targeting to have been largely successful, both because unplanned extended
misses have been small and because average price inflation has been considerably lower than in other
countries. Control of the base was credited with limiting the potential for inflation. Others have
challenged that view, suggesting that the determination to control inflation and a conservative fiscal
policy established a climate where almost any technique could have worked (Wenninger 1983).
Although the policy climate was obviously helpful, monetary base targeting has served as a visible
sign of the central bank's efforts to contain inflation. Consequently, monetary base targets in
Switzerland appear to have played a useful part in bringing about the desired results.
Like Switzerland, West Germany appears to provide an example of relatively successful use of
the monetary base as a policy target. The West German Bundesbank targeted a measure called the
central bank money stock (CBM) from 1975 through 1987, setting annual target ranges. The measure,
56
still in use today as an indicator, closely resembles the monetary base. CBM consists of currency plus
compulsory cash reserves held by the banking system at constant January 1974 reserve ratios, and thus
is essentially a monetary base measure adjusted for reserve requirement changes. Articles by
Bundesbank staff raised objections to equating CBM with the monetary base (Dudler 1980, 1982).
The articles indicated that Bundesbank policy makers disliked identifying CBM with the base for two
reasons. First, they wanted to distance their approach from that of using CBM as an operating target;
when CBM targeting began, most of the discussion of the monetary base in the literature focused on
direct short-run control. Second, they viewed CBM as a proxy for a broad money measure, M3. M3
consists of currency plus sight deposits and time deposits of less than four years and deposits at
statutory notice. Throughout the time the Bundesbank targeted CBM, data were available much more
promptly for CBM than for M3. Most of the time, the two measures tracked each other's behavior
fairly closely. The Bundesbank manipulated interest rates when it observed CBM going off track, and
although it did not set a formal target for M3, it assumed that efforts to bring CBM into line with the
target would also lead to appropriate behavior for M3 (Deutsche Bundesbank 1985).
In 1988, the Bundesbank changed its monetary target to M3 (Deutsche Bundesbank 1988). In
making the change, it again pointed out the similarities between longer run movements of M3 and
CBM. It indicated that both aggregates had shown fairly stable positive relationships to the growth of
nominal production potential and negative relationships to interest rates. Movements in both measures
led movements in economic activity. The change in the target variable was made because CBM had
been tending to move in an exaggerated fashion, overstating swings in money growth rates in
preceding months. At the time, currency was expanding disproportionately because interest rate levels
were low. Because CBM's movements were often dominated by fluctuations in currency, CBM was
responding too much, in the Bundesbank's view, to movements in interest rates. In other words, CBM
was not serving as a good proxy for M3. The Bundesbank also cited some past episodes of
divergence, but reported that those episodes had generally not lasted overly long.
Another motive for changing the target was some discomfort with the abstract nature of the
reserve component of CBM. Actual required reserve ratios were very different (and generally lower)
in 1988 than they had been in 1974. Furthermore, several components of broad money were not
subject to minimum reserve requirements. Nonetheless, the data delays that had previously led the
Bundesbank to target CBM rather than M3 continued after 1988, and for this reason the Bundesbank
retained CBM as an early indicator of changes in M3 growth.
The Bundesbank over the years took its targets, whether for CBM or M3, seriously. It was
57
often successful in achieving them, but it also experienced a number of misses both planned and
unplanned.19 In choosing its annual monetary target, the Bundesbank made estimates of the growth
rate of the target variable likely to be consistent with desired nominal GNP. It took account of what it
considered to be unavoidable inflation, likely prospects for real economic expansion, and forecasts of
velocity. Consequently, it varied the targeted growth rate range for CBM significantly from year to
year as its estimates of appropriate economic activity and velocity changed. Sometimes it chose to
miss a target during the year because of conflicts with desired developments in the foreign exchange
markets or because of unanticipated shifts in the behavior of velocity. The Bundesbank was generally
pleased with this flexible use of monetary targets. Like Switzerland, West Germany generally
achieved inflation rates below those in most other industrialized countries. It is of course true that the
anti-inflation policy in West Germany, as in Switzerland, had strong internal support that greatly
increased the likelihood it would succeed. Nonetheless, the CBM targets made a contribution to
achieving the price objective.
Appendix: Definitions of Monetary Base and Reserve Measures
The Monetary Base
There are a number of possible approaches to defining the monetary base. Two definitions are
widely available, one prepared by the St. Louis Federal Reserve, which pioneered the empirical work,
and the other prepared by the Board of Governors. The base concept can be developed from either the
sources side or the uses side of the balance sheet. The results should be equivalent, although details of
timing and treatment of seasonal adjustment issues can introduce slight differences.
The St. Louis staff has approached the construction of the base conceptually from the sources
side. It describes the base as consisting of Federal Reserve credit-holdings of securities in the
portfolio, loans by the discount window, and other balance sheet items, as well as the gold stock,
Special Drawing Rights, and Treasury currency. Several categories of liabilities are subtracted,
namely, Treasury and foreign deposits at the Federal Reserve, Treasury holdings of currency, and
certain miscellaneous items. When staff members actually construct what they call the source base,
19
See articles entitled "The Economic Scene in the Federal Republic of Germany" in various issues
of the Monthly Report of the Deutsche Bundesbank.
58
however, they define it as currency in the hands of the public plus required reserves plus excess
reserves. In other words, they define the source base in terms of its uses, although they treat vault
cash contemporaneously rather than in lagged form. For dates before 1984, when contemporaneous
reserve requirements (CRR) were introduced, the staff has allowed for the lag structure on reserve
requirements.
The Board staff defines the base in terms of its uses, as consisting of currency and total
reserves held by depository institutions. More specifically, its definition of the base before adjustment
for reserve requirement changes consists of (1) total reserves, (2) required clearing balances and
adjustments to compensate for float at Federal Reserve Banks, (3) the currency component of the
money stock less the amount of thrift institutions' vault cash holdings normally included in the
currency component of the money stock, and (4) the excess of current vault cash over the amount
applied to satisfy current reserve requirements at insitutions not having required reserve balances.
For dates after the introduction of CRR in February 1984, the Board measures currency and
vault cash over the two-week computation period ending Monday, two weeks and two days before the
two-week reserve maintenance period begins. Before CRR, all components of the monetary base other
than excess reserves were seasonally adjusted as a whole, rather than by component, and excess
reserves were added without seasonal adjustment. Since CRR, the seasonally adjusted series has
consisted of seasonally adjusted total reserves, which include excess reserves on a not seasonally
adjusted basis, plus the seasonally adjusted currency component of the money stock plus the remaining
items seasonally adjusted as a whole.
The major difference between the two definitions arises from the adjustment procedures to
handle changes in reserve requirement ratios. The Board's "break adjustment" method allows for
discontinuities arising from regulatory changes in reserve requirements by taking the new ratios,
calculated for different classes of institutions, and applying them to the actual historical deposit series
for each banking group. The annual indexing of low reserve tranches is applied to the reserve ratios
as if the change were phased in over the whole year instead of being applied in the first maintenance
period of each year. It includes the 3 percent requirement on nonpersonal time and savings deposits
but ignores the requirement on Eurodollar deposits.
The St. Louis reserve adjustment magnitude (RAM) employs a different approach. Instead of
using current reserve ratios to define the monetary base, the RAM uses hypothetical reserve
requirements. For years since 1980, the RAM has set the reserve ratio on transactions deposits equal
to 12 percent, the marginal reserve requirement on such deposits. (The actual average reserve ratio
59
has recently been about 8 percent because there are zero and 3 percent reserve tranches.) The RAM
measure assumes a zero reserve requirement for other types of deposits, even though some of them are
actually subject to a 3 percent ratio. For years before 1980, it uses a ratio equal to member bank
deposits and only applies the ratio to such deposits. It applies separate ratios for time and savings
deposits. Interim techniques were used between 1980 and 1987 when new reserve requirements were
being phased in. For a more detailed description of the techniques of reserve adjustment, see Gilbert
(1980, 1984, 1987).
Studies have suggested that it makes an important difference whether the base is adjusted for
reserve requirement changes (Haslag and Hein 1989). However, the different techniques of adjusting
for reserve ratios are of limited importance over most time periods. The St. Louis base measure is
higher than the Board base since it uses a higher reserve ratio, and it thus gives more weight to
deposits relative to currency. For brief time periods, the movements of the two bases often differ.
However, the differences in rates of growth are generally slight for periods of a quarter or more.20
Other Measures
Total reserves consist of reserve balances with Federal Reserve Banks, excluding required
clearing balances and adjustments to compensate for float, plus vault cash held during the lagged
computation period by institutions having required reserve balances at Federal Reserve banks, plus the
amount of vault cash equal to required reserves during the maintenance period at institutions having no
required reserve balances.
Nonborrowed reserves consist of total reserves less those reserves borrowed by depository
institutions from the Federal Reserve's discount window.
Borrowed reserves consist of those reserves acquired from the Federal Reserve's discount
window. For some purposes, extended credit borrowing by banks in financial difficulty is treated as
part of nonborrowed reserves.
Required reserves consist of reserves that depository institutions must hold against deposits.
Requirements are based upon deposit levels during various computation periods according to the type
of deposit and the size of the institution.
Excess reserves consist of total reserves not needed to meet reserve requirements.
^ o n e of the studies cited in the text reported results for both base measures in the regressions. In
the equations I ran, the measures of significance differed somewhat for the two base measures, although
the sums of the coefficients were generally similar.
60
Bibliography
Andersen, Leonall C, and Jerry L. Jordan. 1968a. "Monetary and Fiscal Actions: A Test of Their
Relative Importance in Economic Stabilization." Federal Reserve Bank of St. Louis Review,
November, pp. 11-24.
. 1968b. "The Monetary Base-Explanation and Analytical Use." Federal Reserve Bank of St.
Louis Review, August, pp. 7-11.
Andersen, Leonall C, and Denis S. Karnosky. 1977. "Some Considerations in the Use of Monetary
Aggregates for the Implementation of Monetary Policy." Federal Reserve Bank of St. Louis
Review, September, pp. 2-7.
Avery, Robert B., George E. EUiehausen, Arthur B. Kennickell, and Paul A. Spindt. 1986. "The Use of
Cash and Transactions Accounts by American Families." Federal Reserve Bulletin, vol. 72, no. 2
(February), pp. 87-108.
. 1987. "Changes in the Use of Transactions Accounts and Cash from 1984 to 1986." Federal
Reserve Bulletin, vol. 73, no. 3 (March), pp. 179-96.
Balbach, Anatol B. 1981. "How Controllable is Money Growth." Federal Reserve Bank of St. Louis
Review, April, pp. 3-12.
Bank for International Settlements, Monetary and Economic Department. 1980. The Monetary Base
Approach to Monetary Control. September.
Banque Nationale Suisse. 1988. "Monetary Policy in 1989." Bulletin Trimestriel, no. 4 (December),
pp. 271-72.
. 1989. "The Monetary Policy of the Swiss National Bank in 1989 and 1990." Bulletin
Trimestriel, no. 4 (December), pp. 291-92.
Belongia, Michael. 1988. "Stability of Swiss Money Demand: Evidence for 1982-1987." Banque
Nationale Suisse Bulletin Trimestriel, no. 1 (March), pp. 68-74.
Board of Governors of the Federal Reserve System. 1988. The Monetary Base. Staff Study,
December.
Brunner, Karl. 1961. "Some Major Problems in Monetary Theory." American Economic Review,
vol. 51, no. 2 (May), pp. 47-56.
. 1980. "The Control of Monetary Aggregates." Controlling Monetary Aggregates in, pp. 1-65.
Federal Reserve Bank of Boston, October.
61
Brunner, Karl, and Allan H. Meltzer. 1964. "Studies on Money and Monetary Policy-Some Further
Investigations of Demand and Supply Functions for Money." Journal of Finance, vol. 19, no. 2
(May), pp. 240-83.
. 1968. "Liquidity Traps for Money, Bank Credit, and Interest Rates." Journal of Political
Economy, January-February, pp. 1-37.
. 1981. "Time Deposits in the Brunner-Meltzer Model of Asset Markets." Journal of
Monetary Economics, vol. 7, no. 1, pp. 129-39.
. 1983. "Strategies and Tactics for Monetary Control." Carnegie-Rochester Conference Series
on Public Policy, no. 18, pp. 59-104.
Brunner, Karl, Alex Cukierman, and Allan H. Meltzer. 1983. "Money and Economic Activity,
Inventories, and Business Cycles." Journal of Monetary Economics, vol. 2.
Burger, Albert E., Lionel Kalish HI, and Christopher T. Babb. 1971. "Money Stock Control and Its
Implications for Monetary Policy." Federal Reserve Bank of St. Louis Review, October, pp. 6-22.
Burgess, Randolph. 1936. The Federal Reserve Banks and the Money Market. New York: Harper and
Brothers.
Cagan, Phillip. 1982. "The Choice among Monetary Aggregates as Targets and Guides for Monetary
Policy." Journal of Money, Credit, and Banking, vol. 14, no. 4 (November), pp. 551-686.
Carlson, John B. 1981. "The Monetary Base, the Economy, and Monetary Policy." Federal Reserve
Bank of Cleveland Economic Review, Spring, pp. 2-13.
Christiano, Lawrence J. 1986. "Money and the U.S. Economy in the 1980Y. A Break from the Past?"
Federal Reserve Bank of Minneapolis Quarterly Review, Summer, pp. 2-13.
CuUison, William E. 1982. "Money, the Monetary Base, and Nominal GNP." Federal Reserve Bank of
Richmond Economic Review, May-June, pp. 3-13.
Darby, Michael R., William Poole, David E. Lindsey, Milton Friedman, and Michael J. Bazdarich.
1987. "Recent Behavior of the Velocity of Money." Contemporary Policy Issues, vol. 5, no. 1
(January), pp. 1-33.
Davis, Richard G. 1979-80. "The Monetary Base as an Intermediate Target for Monetary Policy."
Federal Reserve Bank of New York Quarterly Review, Winter, pp. 1-10.
. 1980. "The Monetary Base as an Intermediate Target for Monetary Policy in the United
States." The Monetary Base Approach to Money Control, pp. 214-29. Basle: Bank for
International Settlements, Monetary and Economic Department, September.
Deutsche Bundesbank. 1985. "The Longer-Term Trend and Control of the Money Stock." Monthly
Report of the Deutsche Bundesbank, vol. 37, no. 1 (January), pp. 13-26.
62
. 1988. "Methodological Notes on the Monetary Target Variable M3." Monthly Report of the
Deutsche Bundesbank, vol. 40, no. 3 (March), pp. 18-21.
Dudler, Hermann-Josef. 1980. "Money-Market Management, Supply of Bank Reserves and Control of
the Central-Bank Money Stock." The Monetary Base Approach to Money Control pp. 52-77.
Basle: Bank for International Settlements, Monetary and Economic Department, September.
. 1982. "The Implementation of Monetary Objectives in Germany-Open Market Operations
and Credit Facilities." Central Bank Views on Monetary Targeting, pp. 18-29. Federal Reserve
Bank of New York, May.
Fama, Eugene F. 1982. "Inflation, Output and Money." Journal of Business, vol. 55, no. 2, pp. 201-31.
Fellner, William. 1982. "Criteria for Useful Targeting. Money versus the Base and Other Variables."
Journal of Money, Credit, and Banking, vol. 14, no. 4 (November), pp. 641-60.
KisHyr, Irving N. [1911] 1963. The Purchasing Power of Money. Reprint. New York: Augustus M.
Friedman, Benjamin M. 1988. "Conducting Monetary Policy by Controlling Currency Plus Noise-A
Comment." Carnegie-Rochester Conference Series on Public Policy, no. 29.
Friedman, Benjamin M., and Kenneth N. Kuttner. 1989. "Money, Income, and Prices After the 1980s."
National Bureau of Economic Research, Working Paper Series, no. 2852. February.
Friedman, Milton. 1959. A Program for Monetary Stability. Millar Lectures, no. 3. Fordham
University Press.
Friedman, Milton, and David Meiselman. 1963. "The Relative Stability of Monetary Velocity and the
Multiplier in the United States, 1897-1958." In Commission on Money and Credit, Stabilization
Policies. New Jersey: Prentice Hall, pp. 165-268.
Gambs, Carl M. 1980. "Federal Reserve Intermediate Targets: Money or the Monetary Base?" Federal
Reserve Bank of Kansas City Economic Review, January, pp. 3-15.
Garfinkel, Michelle R., and Daniel L. Thornton. 1989. "The Link Between Ml and the Monetary Base
in the 1980s." Federal Reserve Bank of St. Louis Review, September-October, pp. 35-52.
Gilbert, R. Alton. 1980. "Revision of the St. Louis Federal Reserve's Adjusted Monetary Base."
Federal Reserve Bank of St. Louis Review, December, pp. 3-10.
. 1984. "Calculating the adjusted Monetary Base under Contemporaneous Reserve
Requirements." Federal Reserve Bank of St. Louis Review, February 1984, pp. 27-32.
. 1987. "A Revision in the Monetary Base." Federal Reserve Bank of St. Louis Review,
August-September, pp. 24-29.
63
Goodhart, Charles A., M.D.K.W. Foot, and A.C. Hotson. 1979. "Monetary Base Control." Bank of
England Bulletin. June, pp. 149-59.
Gurley, John G., and Edward S. Shaw. Money in a Theory of Finance. Washington, D.C.: Brookings
Institution, 1960.
Hafer, R.W. 1984. "The Monetary Base or Ml? Results from a Small Macromodel." Temple
University Journal of Economics and Business, vol. 36, no. 1 (February), pp. 85-93.
Hafer, R.W., Scott E. Hein, and Clemens J. Kool. 1983. "Forecasting the Money Multiplier:
Implications for Money Stock Control and Economic Activity." Federal Reserve Bank of St. Louis
Review, October, pp. 22-33.
Hall, Thomas E. 1988. "McCallum's Base Growth Rule: Results for the United States, West Germany,
Japan, and Canada." U.S. Department of State, Bureau of Economic and Business Affairs, Planning
and Economic Analysis Staff, Working Paper Series. December.
Haslag, Joseph H., and Scott E. Hein. 1988. "Evidence on the Two Monetary Base Measures and
Economic Activity." Federal Reserve Bank of Dallas, Research Paper. December.
. 1989. "Reserve Requirements, the Money Base, and Economic Activity." Federal Reserve
Bank of Dallas Economic Review, March, pp. 1-15.
Johannes, James M., and Robert H. Rasche. 1979. "Predicting the Money Multiplier." Journal of
Monetary Economics, vol. 5, pp. 301-25.
Kohli, Ulrich, and Georg Rich. 1986. "Monetary Control, The Swiss Experience." Cato Journal, vol. 5,
no. 3 (Winter), pp. 911-26.
Lindsey, David E. 1980. "Control of the Monetary Base and Reserves - The U.S. Case." The
Monetary Base Approach to Money Control, pp. 205-13. Basle: Bank for International Settlements,
Monetary and Economic Department, September.
Lindsey, David E., et al. 1981. "Monetary Control Experience Under the New Operating Procedure."
New Monetary Control Procedures-Federal Reserve Staff Study, vol. 2, February.
Lothian, James R. 1976. "The Demand for High-Powered Money." The American Economic Review,
vol. 66, no. 1 (March), pp. 56-68.
Mayer, Thomas, James S. Duesenberry, and Robert Z. Aliber. 1987. Money, Banking and the
Economy. 3d ed. W.W. Norton and Company.
McCallum, Bennett T. 1987. "The Case for Rules in the Conduct of Monetary Policy: A Concrete
Example." Federal Reserve Bank of Richmond Economic Review, September-October, pp. 10-18.
. 1988a. "Postwar Developments in Business Cycle Theory: A Moderately Classical
Perspective." Journal of Money, Credit, and Banking, vol. 20, no. 3 (August), pp. 459-71.
64
. 1988b. "Robustness Properties of a Rule for Monetary Policy." Carnegie-Rochester
Conference Series on Public Policy, no. 29, pp. 173-204.
. 1988c. "Targets, Indicators and Instruments of Monetary Policy." American Enterprise
Institute Conference on Monetary Policy in an Era of Change, October.
Osborne, Dale K. 1986. "Velocities of Ml, and the Money Base: A Correction of Standard Formulas."
Federal Reserve Bank of Dallas Economic Review, January, pp. 10-24.
Poole, William. 1988. "Lessons on Monetary Policy from the 1980s." The Journal of Economic
Perspectives, vol. 2, no. 3 (Summer), pp. 51-72.
Rich, Georg. 1987. "Swiss and United States Monetary Policy: Has Monetarism Failed?" Federal
Reserve Bank of Richmond Economic Review, vol. 73, no. 3 (May-June), pp. 3-16.
Rich, Georg, and Kurt Schiltknecht. 1980. "Targeting the Monetary Base~The Swiss Approach." The
Monetary Base Approach to Money Control, pp. 150-70. Basle: Bank for International Settlements,
Monetary and Economic Department, September.
Roberds, William. 1989. "Money and the Economy: Puzzles from the 1980's Experience." Federal
Reserve Bank of Atlanta Economic Review, September-October, pp. 20-35.
Robertson, D.H. 1948. Money. 4th ed. Cambridge: Cambridge University Press.
Schiltknecht, Kurt. 1982. "Switzerland~The Pursuit of Monetary Objectives." Central Bank Views on
Monetary Targeting, pp. 72-79. Federal Reserve Bank of New York, May.
. 1983. "Targeting the Monetary Base: The Swiss Case." Improving Money Stock Control:
Problems, Solutions, and Consequences, chap. 2, pp. 67-96. Boston: Kluwer-Nijhoff.
Simons, Henry C. 1948. Economic Policy for a Free Society. Chicago: University of Chicago Press.
Stone, Courtenay C , and Daniel L. Thornton. 1988. "Selected Aggregates as Intermediate Policy
Targets: Some Initial Evidence." Federal Reserve Bank of St. Louis Research Paper, February.
Tatom, John. 1980. "Issues in Measuring an Adjusted Monetary Base." Federal Reserve Bank of St.
Louis Review, December, pp. 11-29.
Thornton, Daniel L. 1982. "Simple Analytics of the Money Supply Process and Monetary Control."
Federal Reserve Bank of St. Louis Review, October, pp. 22-39.
Tinsley, Peter A., et al. 1981. "Money Market Impacts of Alternative Operating Procedures." New
Monetary Control Procedures-Federal Reserve Staff Study, vol. 2 (February).
Von Hagen, Jurgen. 1988. "Alternative Operating Regimes for Money Stock Control in West Germany:
An Empirical Evaluation." Weltwirtschaftiches Archiv, vol. 124, pp. 89-107.
65
Wenninger, John. 1983. Discussion of a paper by Kurt Schiltknecht. Improving Money Stock
Control: Problems, Solutions, and Consequences, chap. 2, pp. 91-96. Boston: Kluwer-Nijhoff.
Wood, John H. 1984. "The Search for a Monetary Policy Rule in an Uncertain World." Federal
Reserve Bank of Dallas Economic Review, September.
66
MONETARY AGGREGATES AS INTERMEDIATE TARGETS
John Wenninger
The Federal Reserve began to use monetary aggregates as intermediate targets in the mid1970s. Increasing recognition of the drawbacks to relying on interest rates and the growing need to
formulate a long-run strategy for containing inflationary pressures were the main reasons for adopting
monetary targets. The experience gained in using the monetary aggregates in the policy process was
in many respects the precipitating force behind much of the work that has been undertaken over the
years on the general topic of using intermediate variables to attain ultimate policy goals. And the
difficulties encountered in using monetary targets provided much of the impetus to search for
alternative intermediate variables for policy such as total credit and liquid assets initially and
commodity prices and the yield curve later on.
This paper reviews the recent experience with using monetary aggregates as intermediate
targets. It begins with a brief overview of the improvements in the policy process that were expected
from formulating policy with monetary targets and the many practical problems that developed in
using the monetary targets.
The second section presents some of the theoretical arguments for and against using money as a
target. A simple IS-LM model is used to illustrate the conditions under which money would be a
good target variable as well as the conditions under which money could be readily controlled using the
monetary base (as opposed to using interest rates).
The third section of the paper provides some perspective on the stability of the money-income
and the base-money relationships by presenting simple extrapolations of velocity and money multiplier
trends. These simple techniques show that prior to the mid-1970s a strong empirical case for using
monetary targets could be made, particularly for Ml. Both velocity and the money multiplier seemed
to follow stable trends. Since the mid-1970s, however, the velocity link has become considerably less
stable, and there is also some evidence that the relationship between Ml and the base has deteriorated
as well. For the broader aggregates, while the multiplier relationships were not as tight as those for
Ml even before the mid-1970s, the changes in velocity relationships have not been as pronounced as
those for Ml. These results suggest that the broader aggregates are currently better suited for policy
67
purposes.
The fourth section confirms these conclusions using money-income and base-money reduced
form equations. The final section explores the reasons behind the recent instability in Ml relative to
economic activity by focusing on the effects of innovation and deregulation on the public's demand
for money. The way banks adjust the rates on various types of deposits as market rates change in a
deregulated environment appears to have made the use of Ml as an intermediate variable more
problematic than the use of the broader aggregates.
The first appendix to this paper describes the breakdown in the relationship between narrow
money and inflation. This appendix also contains a discussion of recent research on using M2 in
combination with the level of unused capacity in the economy as a long-run leading indicator of
inflation (often referred to as p*). The second appendix examines the operating procedures the Federal
Reserve has used over the years in attempting to keep money on target.
I. Circumstances Leading to the Adoption of Monetary Targets
By the mid-1970s, the problems with using nominal interest rates as policy guides when
inflationary expectations were shifting became increasingly apparent. Increases in inflation and
inflationary expectations could cause real rates to fall even while nominal rates held steady or rose.
Although real rates were difficult to measure, they were generally accepted as the important variables
for spending decisions. It was also recognized that interest rates could give the wrong impression
about the posture of monetary policy over time if interest rates were changing because of a shift in
aggregate demand. To indicate more clearly whether policy had eased or tightened, policy makers
should pay attention both to the movements in interest rates and to measures of bank reserves and/or
the money stock.
Around that same time, it also appeared increasingly important to establish a longer run strategy
for lowering inflation.1 Control of the money supply, because of its presumed long-run relationship to
inflation, was viewed as a key variable to attain price stability over time. In addition, monetary targets
came to be seen by many as an approximate means to communicate longer run policy objectives to
Congress and the general public, to establish central bank credibility and thus lower inflationary
*For a discussion of the monetary aggregates as practical guides for policy, see Paul Volcker, "The
Role of Monetary Targets in an Age of Inflation," Journal of Monetary Economics, vol. 4, no. 2 (April
1978), pp. 329-39.
68
expectations, and ultimately to hold the central bank accountable for its policies. Finally, monetary
targets could provide a more enduring standard against which shorter run policy decisions could be
evaluated over time. Monetary targets might also expedite policy decisions because some empirical
work had suggested that movements in money led movements in nominal GNP and inflation by
several months. Overall, many analysts believed that monetary targets could provide a nominal anchor
for policy, both for internal operational purposes and for external communications.
Monetary targets, however, were far from a panacea for policy. Establishing monetary targets
also required dealing with many practical problems. These problems included:
—choosing the monetary aggregate or aggregates to be targeted and the appropriate definitions
of the monetary aggregates during a period of innovation and deregulation;
—establishing the time period over which targets should be set, howfrequentlythey should be
reset, how much they could reasonably be changed from year to year, and whether the previous
target value for money or the actual value should be used as the base for the target in the next
period (base drift);
—deciding how quickly a deviation in money from target should be corrected to avoid
overshooting the target and needing to reverse policy;
—determining whether reserves or short-term interest rates would be better instruments to attain
monetary targets;
-exercising judgment in interpreting deviations in money from target as economically
significant or insignificant; and deciding whether the policy process should focus only on
money or on a broad range of indicators to ensure that money was giving the correct signal;
—deciding how to identify and then deal with shifts in the trend of velocity and the demand for
money, particularly when inflation and inflationary expectations dropped sharply following a
period of monetary restraint, thereby changing the relative demands for financial and real
assets.
II. Theoretical Considerations in Using Money as a Policy Guide
The theoretical reasons for and against using monetary aggregates as intermediate targets were
debated long before many of the practical problems mentioned above became apparent. Readers
already familiar with these arguments can skip to Section in where the empirical results begin.
At a theoretical level, two models that represent extreme and opposite possibilities can be used
to illustrate the circumstances under which the money supply would or would not be useful as an
69
intermediate target. In the first model, money plays a key role as an intermediate target. Causation
flows from an exogenously determined supply of high-powered money (reserves and currency) to the
money stock, with a stable or predictable money multiplier serving as the link. The money stock, in
turn, determines the level of nominal spending, this time with a stable, predictable velocity providing
the link. The observed correlation between money and income occurs because money induces changes
in nominal spending. In addition, since the demand for money is viewed as being stable and relatively
insensitive to changes in interest rates, changes in exogenous expenditures will not cause income to
deviate from the desired path. Hence, managing the monetary base in such a way that money grows at
a moderate rate would also achieve a noninflationary path for nominal spending. In other words,
monetary control is the key to preventing inflation in the longer run.
In the other model, monetary aggregates as intermediate targets or indicators would have a
much reduced role, if any. In this model, causation runs from shifts in exogenous spending to changes
in total nominal spending and then to increases or decreases in the quantity of money demanded. The
changes in the quantity of money demanded in turn lead to changes in the demand for the monetary
base, changes which are passively accommodated by the central bank. In other words, money is
endogenously rather than exogenously determined, and the correlations between money and income
and between money and the monetary base result from changes in spending that induce changes in the
demand for money and the demand for the base. Even if the central bank attempted to restrain the
monetary base and therefore money, the high interest rate sensitivity of the demand for money and the
instability in the money demand function would make control of the money stock as an intermediate
target a difficult exercise. In this model, money might serve as an early indicator of changes in
spending (only because money data are available before GNP data), but there would be little value in
controlling money because keeping money on a steady path would not prevent changes in autonomous
expenditures from affecting GNP.2 In this model, inflation would not occur so much because of
changes in the money supply, but rather because of changes in autonomous expenditures that pushed
aggregate spending beyond the economy's capacity to produce. In other words, a sensible fiscal policy
2
For good textbook presentations of alternative approaches to stabilization policy, see Thomas
Mayer, James Duesenberry, and Robert Aliber, Money. Banking and the Economy (New York: W.W.
Norton and Company, 1987); and Lawrence Ritter and William Silber, Principles of Money. Banking
and Financial Markets (New York: Basic Books, Inc., 1989).
70
Table 1
Basic IS-LM Modef
(1)Y =-cr + X
(2)M = -ar + bY + Z
(3)M = R/m + dr + K
M = narrow money supply
r = the interest rate
Y = income
Z = money demand shift
X = autonomous expenditures
R = monetary base
K = money supply shift
m = money multiplier
a,b,c,d = structural parameters
Reduced Form Multipliers
JL
Y =
+
r =
M =
+
JL
c
a+d+bc
a+d+bc
a+d
a+d+bc
1
m(a+d+bc)
1
a+d+bc
1
a+d+bc
b
a+d+bc
a+bc
m(a+d+bc)
+ a+bc
a+d+bc
a+d+bc
bd
a+d+bc
c
m(a+d+bc)
+
X
-'Prices are assumed constant in the short run so that nominal and real values are the same.
would be the key to price stability.3
In practice, however, causation in the economy does not flow in just one direction, as portrayed
in these extreme models. The basic IS-LM model can be used to illustrate some of the complexities
facing policy makers in deciding whether to use money as an intermediate guide. The first table
shows the basic IS-LM model and the derivation of the reduced form multipliers.
3
For overviews of the theoretical arguments for intermediate targets, see Edward Kane, "Selecting
Monetary Targets in a Changing Financial Environment," Money Policy Issues in the 1980s
(Proceedings of a symposium sponsored by the Federal Reserve Bank of Kansas City, August 1982),
pp. 181-206; Benjamin Friedman, "The Value of Intermediate Targets in Implementing Monetary
Policy," Price Stability and Public Policy (Proceedings of a symposium sponsored by the Federal
Reserve Bank of Kansas City, August 1984), pp. 169-91; and Richard G. Davis, "Monetary Aggregates
and the Use of 'Intermediate Targets' in Monetary Policy," New Monetary Control Procedures, vol. 1,
Board of Governors of the Federal Reserve System Staff Study, February 1981, pp. 1-44.
71
For the Federal Reserve to control money and GNP with a great degree of precision, it would
be necessary to know the structural parameters of the model (a, b, c, and d) and the source and size of
other exogenous shocks to the economy (K, Z, and X) in order to offset their impacts on money and
GNP. If the Federal Reserve cannot anticipate and offset these shocks, both money and GNP will be
affected. For shifts in the expenditure function (X) and shifts in the supply of money function (K), the
impacts on money and GNP will have the same signs. Therefore, by looking at money, the Federal
Reserve can at least discern whether it needs to contract or expand the supply of reserves to stabilize
GNP. However, for money demand shifts (Z), the multipliers have opposite signs and the Federal
Reserve could destabilize GNP by attempting to keep money on target when the demand for money
changes.
Finally, even if the demand for money were stable and the Federal Reserve only needed to
respond to shifts in the expenditure function (X), stabilizing money would not necessarily stabilize
GNP entirely. Only in the special case when the interest elasticities of the supply of and demand for
money (a and d) were equal to zero would keeping money on target ensure that the desired
performance of GNP would be attained. In this case, neither money nor GNP would be affected as a
result of shifts in the expenditures function (X). In the other extreme case, however, when the interest
elasticity of the demand for money (a) would be very large, shifts in autonomous expenditure would
have small effects on money but relatively large impacts on GNP. Money might therefore stay quite
close to target even while GNP deviated substantially.
Clearly, even a very simple model underscores the need for caution in using money as an
intermediate target. Unanticipated shifts in the demand for money, of course, appear to be the most
serious difficulty that could arise. In this case, a reduction (increase) in the demand for money,
without any change in the supply of money, could create inflationary (recessionary) pressures. In
addition, a very large money demand interest elasticity could also be a significant drawback to
monetary targeting. An increase (decrease) in autonomous expenditures could create inflationary
(recessionary) pressures, by pushing GNP growth beyond (below) its potential, without appreciably
more (less) rapid money growth.4
4
For more background, see William Poole, "Optimal Choice of Monetary Policy Instruments in a
Simple Stochastic Model," Quarterly Journal of Economics, May 1970, pp. 197-216; John Wenninger,
"Financial Innovation, A Complex Problem Even in a Simple Framework," Federal Reserve Bank of
New York Quarterly Review, Summer 1984, pp. 1-8; and James Tobin, "Monetary Policy: Rules,
Targets, and Shocks," Journal of Money, Credit, and Banking, November 1983, pp. 506-18.
72
In general, the following conditions would favor monetary targeting when the monetary base is
the policy instrument:
—a low interest rate elasticity in the supply of money function (d) and small values of shifts in
the supply of money (K). These conditions would tighten the linkage between the base and the
money stock. The ratio of the money stock to the base (the money multiplier) could be used to
determine how much monetary base the Federal Reserve should supply to attain its monetary
objectives.
—a low interest rate elasticity in the money demand function (a) and small values of shifts in
the demand for money (Z). Under these circumstances, the money stock would largely
determine the level of income, and any shifts in the expenditures function (X) would mainly
affect interest rates. Money and income would move together in a predictable way (stable
velocity), and velocity could be used to calculate the target path for money consistent with a
desired GNP objective.5
III. Predictability of Velocity and Multiplier Trends
This section uses extrapolations of velocity and money multiplier trends to illustrate the
predictability of these trends with straightforward and transparent techniques. The analysis reveals that
for Ml, in particular, the predictability of these monetary relationships has deteriorated considerably.
The recent breakdown in the money-income (GNP-M1) relationship has been well documented
in numerous other studies.6 To illustrate the timing and relative magnitudes of this breakdown, the
5
It is, of course, also possible to control the money stock using the interest rate. In this case, the
reduced form for money is M=(-a-bc)r+bX+Z. When M is controlled using the interest rate, shifts in
money demand (Z) and autonomous expenditures (X) would have larger impacts on M than when the
base is used, but shifts in the supply of money function (K) would not have any effect. Hence, the
choice of instrument—the base or the interest rate—would depend in part on the relative sizes of the
shift parameters in the IS-LM model.
6
For overviews of the breakdown from the monetarist and nonmonetarist perspectives, see
Benjamin Friedman, "Lessons on Monetary Policy in the 1980s," The Journal of Economic
Perspectives, Summer 1988, pp. 51-72; and William Poole, "Monetary Policy Lessons of Recent
Inflation and Disinflation," The Journal of Economic Perspectives, Summer 1988, pp. 73-100. Some
analysts, however, do not agree that a breakdown in the relationship of money and GNP occurred in
the 1980s. They find that an alternative definition of Ml that excludes NOW accounts has tracked
(continued...)
73
second table shows the annual errors in forecasting the growth of Ml velocity (as well as the
velocities of M2 and M3) as an average of the previous ten years. This approach allows changes in
the trend of velocity to be gradually reflected in the forecast without putting too much weight on any
given year. While any such naive approach to forecasting velocity is somewhat arbitrary, it is
sufficient to show the relative stability of these velocity trends over time. Indeed, monetary targets
appealed to policy makers because the techniques used to project velocity could be simple and could
easily be explained to Congress and the public. That is, elaborate econometric models were not
required for selecting the monetary targets and making them understood as long as the econometric
evidence suggested a low interest rate elasticity and stability in the demand for money function.
From the early 1960s to the mid-1970s, the average error in forecasting Mi's velocity was
zero, while the root mean squared error (RMSE) was about 2 percentage points. Errors of this size
were interpreted as meaning that the monetary authorities should use a range about 3 percentage points
wide rather than setting a single numerical target for Ml. From the mid-1970s to 1988, however, the
average error was -1.5 percentage points and the RMSE more than doubled to 4.4 percentage points.7
6
(...continued)
economic activity fairly well. For more detail, see Michael Darby, Angelo Masarco, and Michael
Marlow, "The Empirical Reliability of Monetary Aggregates as Indicators," Economic Inquiry, October
1989, pp. 555-85. For a wide range of views on the behavior of velocity in the 1980s, see Michael
Darby, William Poole, David Lindsey, and Milton Friedman, "Recent Behavior of the Velocity of
Money," Contemporary Policy Issues, vol. 5, no. 1 (January 1987), pp. 1-33.
7
Unusual movements in the actual Ml velocity statistics relative to expected velocity occurred in
two subperiods of the interval from the mid-1970s to the late 1980s. The first was 1972 to 1978,
when velocity (except for one year) was unusually strong, most likely as a result of increased
emphasis on cash management as nominal rates rose to extremely high levels. For more detail, see
Stephen Goldfeld, "The Case of the Missing Money," Brookings Papers on Economic Activity, 3:1976,
pp. 683-739; and Thomas Simpson and Richard Porter, "Some Issues Involving the Definition and
Interpretation of the Monetary Aggregates," Controlling Money Stock HI, Federal Reserve Bank of
Boston, October 1980, pp. 161-233. Other analysts argued that the demand for money had been stable
but the function had not been properly specified. See, for example, Michael Hamburger, "Behavior of
the Money Stock: Is There a Puzzle?" Journal of Monetary Economics. July 1977, pp. 265-88. The
second period of instability was from 1982 to 1986, when velocity declined sharply. This decline
probably stemmed from the sharp drop in interest rates at a time when the deregulation of deposit
interest rates had made the demand for Ml more interest sensitive, at least in the shorter run. For
more detail, see John Wenninger, "Responsiveness of Interest Rate Spreads and Deposit Flows to
Changes in Market Rates," Federal Reserve Bank of New York Quarterly Review, Autumn 1986,
(continued...)
74
In the mid-1980s, the extremely large errors prompted the Federal Reserve to drop its target range for
Ml.
The velocities of the broad aggregates in the earlier time period were not as predictable as Mi's
velocity, with RMSEs 1/2 to 1 percentage point greater than for Ml. However, the velocities of the
broader aggregates did not become as unpredictable as Mi's in the later period. For M2, the RMSE
increased from 2.5 to 3.3 percentage points, whereas for M3 the velocity forecast errors declined
slightly to 2.5 percentage points. This change in the relative predictability of the velocities of Ml and
the broader aggregates was a major factor in persuading the Federal Reserve to place greater emphasis
on the broad aggregates during the 1980s and to cease setting targets for Ml. 8
Table 3 contains some additional information on velocity growth since 1961. The left side
shows maximum and minimum values for velocity growth on an annual basis, while the right side
shows the maximum and minimum values on a ten-year moving average basis. Whether calculated on
a year-to-year basis or on a ten-year moving average basis, Mi's velocity looks the least stable. On a
ten-year moving average basis, M3's velocity looks the most stable, while on a year-to-year basis
M2's velocity appears the most stable. Again, when looked at in this light, the broad aggregates
appear to be better suited for policy purposes than Ml. Not only are the RMSEs smaller on average,
but they also have narrower ranges of extreme values.
Next, we turn to the stability over time of the relationship between the money supply and the
monetary base.9 Some analysts have raised serious questions about using the base as a policy
7
(...continued)
pp. 1-10; and Courtney Stone and Daniel Thornton, "Solving the 1980s' Velocity Puzzle: A Progress
Report," Federal Reserve Bank of St. Louis Review. August-September 1987, pp. 5-23. Also see
references cited in footnote 4.
^ e r e was also evidence that the demand functions for the broad aggregates had been more stable
than the demand for Ml. For more detail, see John Wenninger, "Money Demand—Some Long-Run
Properties," Federal Reserve Bank of New York Quarterly Review. Spring 1988, pp. 23-40.
*We use the St. Louis adjusted monetary base (adjusted to neutralize the effects of changes in
reserve requirements) because it has been used extensively in earlier empirical work and is available
for a long time period. For a detailed review of the relationship between Ml and the base focusing on
the effects of deregulation, see Michelle R. Garfinkel and Daniel Thornton, "The Link Between Ml
and the Monetary Base in the 1980s," Federal Reserve Bank of St. Louis Review. September-October
1989, pp. 35-52.
75
Table 2
Errors in Predicting Velocity
(Using Simple Velocity Trends)
GNP/M1
GNP/M3
-10.8
1.8
2.7
0.6
-1.4
-0.8
-0.9
3.2
3.9
-2.5
2.3
2.9
-1.2
-3.7
-0.3
4.5
1.0
-2.0
-3.4
0.9
6.2
0.6
0.5
-0.5
-6.4
-2.1
1.1
-1.9
-4.0
4.5
1.5
-0.2
-1.9
-1.5
-1.6
2.6
4.0
-2.9
2.1
6.1
-3.7
-4.1
-0.7
1.0
-0.8
1.7
-0.8
0.4
3.3
-0.4
1.4
-2.0
-5.7
1.4
-0.8
-0.2
-2.9
4.1
1.6
1961 to 1988
Mean =
RMSE =
-0.7
3.5
0.1
2.9
0.0
2.7
1961 to 1974
Mean =
RMSE =
0.0
2.1
0.5
2.5
-0.1
3.0
1975 to 1988
Mean =
RMSE =
-1.5
4.4
-0.4
3.3
0.1
2.5
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1.7
1.0
-0.4
-1.6
2.8
1.8
-3.9
-1.0
0.1
-3.2
-0.3
0.7
3.2
-0.5
2.5
0.6
0.6
2.9
-2.0
-1.1
0.2
-8.9
GNP/M2
-2.8
0.8
-7.2
Table shows errors resulting when each year's velocity growth (fourth quarter to fourth quarter) was predicted as
the average growth rate over the previous ten years.
76
Table 3
Range of Values for Velocity Growth
(1961 to 1988, in Percent)
Yearly
(Fourth Quarter to Fourth Ouartert
Minimum
Maximum
Range
Ten-Year Moving Average
Maximum
Range
Minimum
Ml
-9.4
6.2
15.4
0.0
3.8
3.8
M2
-5.4
4.3
9.7
-0.9
1.0
1.9
M3
-6.1
5.3
11.4
-0.3
-1.3
1.0
instrument.10 Over the years, other reserve aggregates such as total reserves, reserves on private
deposits (RPDs), nonborrowed reserves, and the nonborrowed base have been studied. In addition,
some analysts who accept the notion that it would be desirable to control money have left open the
question whether it should be done through a reserve aggregate or through a short-term interest rate.
When controlled with an interest rate variable, the money demand function often has been viewed as
the reduced form equation for money.11 Whether some reserve measure or interest rate should be
used in practice depends upon the properties of the money demand and supply functions, that is, their
interest rate sensitivity and relative stability. The Federal Reserve has at different times used both
short-term rates and reserves aggregates in its attempt to control money (Appendix 2 examines this
topic in more detail). The empirical work in this paper focuses only on the monetary base because
other papers in this volume will deal with alternative reserve measures and interest rates in the policy
10
Benjamin Friedman, "Conducting Monetary Policy by Controlling Currency Plus Noise,"
Carnegie-Rochester Conference Series on Public Policy, no. 29 (1988), pp. 205-12. The main problem
with the base is that it consists of 75 percent currency and 25 percent reserves, and the Federal
Reserve has usually provided an elastic currency. It is not clear that the Federal Reserve could control
the entire base as an instrument simply by adjusting the relatively small reserves portion.
1
Barnes L. Pierce and Thomas D. Thomson, "Some Issues in Controlling the Stock of Money," in
Controlling Monetary Aggregates II: The Implementation (Proceedings of a conference sponsored by
the Federal Reserve Bank of Boston), September 1972, pp. 115-36. See also Richard G. Davis and
Frederick C. Schadrack, "Forecasting the Monetary Aggregates with Reduced Form Equations,"
Federal Reserve Bank of New York, Monetary Aggregates and Monetary Policy, October 1974,
pp. 60-71.
77
process.
Tables 4 and 5, similar in format to Tables 2 and 3, respectively, assess the predictability of the
money multiplier (ratio of a monetary aggregate to the monetary base) using the naive model of a tenyear moving average as the projection of the growth rate of the multiplier for each year. Like
velocity, the money multiplier would be a much more useful policy tool if relatively simple techniques
could be used to predict its performance, not only in terms of clearly expressing the intent of policy
but also in terms of attaining monetary objectives. Not all those who would advocate using the
monetary base would agree with such an approach, but some have pointed out the relative ease with
which the trend in the multiplier might be predicted.12
In Tables 4 and 5, the annual data (fourth quarter to fourth quarter) are used to correspond to
thefirameworktypically used for setting the longer run monetary targets (see footnote 12 for references
on short-term multiplier forecasts). Table 4 shows that the Ml multiplier was the most predictable
over the 1961 to 1974 period, with a RMSE (1.1 percentage points) about half as large as the RMSEs
for the M2 and M3 multipliers. However, over the 1974 to 1988 period the RMSE for the Ml
multiplier more than doubled and the RMSEs for the multipliers of all three aggregates were in the 2.0
to 2.S percentage point range in the later period. Hence, it would appear that the links between the
monetary base and all three monetary aggregates are currently not very tight.13
Table 5 shows the range of values taken on by these multipliers. On a year-to-year basis, the
Ml multiplier shows the most extreme values, whereas on a ten-year moving average basis, the
12
Some earlier analysts went beyond simple extrapolations of past trends to estimate the money
multiplier. In addition to using lagged values of the multiplier, they used interest rates and seasonal
dummy variables. But their studies, unlike this paper, were not aimed at establishing the long-run
trend relationship between the base and Ml. That relationship was taken as established. Rather, they
were concerned with forecasting the money multiplier a few months ahead to achieve precise monetary
control over the short ruii. See, for example, Albert E. Burger, "Money Stock Control," in Controlling
Monetary Aggregates II: The Implementation (Proceedings of a conference sponsored by the Federal
Reserve Bank of Boston), September 1972, pp. 33-55. For a more recent empirical analysis
concluding that the money stock is controllable using the base, see James Johannes and Robert Rasche,
"Predicting the Money Multiplier," Journal of Monetary Economics, July 1979, pp. 301-25. Our effort
here to extrapolate annual trends is more in the spirit of the work of Anton Balbach, "How
Controllable is Money Growth?" Federal Reserve Bank of St. Louis Review, April 1981, pp. 3-12.
13
For a detailed explanation using the money multiplier framework, see Albert E. Burger, "The
Puzzling Growth of the Monetary Aggregates in the 1980s," Federal Reserve Bank of St. Louis
Review, September-October 1988, pp. 46-60.
78
Table 4
Errors in Predicting the Money Multiplier
(Using Simple Multiplier Trends)
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
Ml/Base
M2/Base
M3/Base
-0.1
-1.6
-1.1
-1.2
-0.9
-1.2
0.5
0.9
0.4
1.0
0.7
-0.7
-1.9
-1.5
-3.5
-0.9
-2.4
-2.5
-2.7
3.4
2.1
-2.6
-4.3
3.2
3.9
0.5
-3.3
-1.9
-1.4
2.7
-0.9
1.0
-1.4
-1.0
1.8
1.3
0.0
-1.1
-1.0
-3.6
-0.4
-2.3
-5.5
0.0
3.9
2.5
-0.6
-0.2
1.0
-1.2
-2.8
-1.3
-0.3
0.8
0.0
0.7
0.3
2.0
1.5
1.7
-1.1
4.5
0.8
-2.6
-0.7
0.1
-3.2
-0.8
-2.4
-2.0
1.2
1.0
-0.6
-0.9
-2.3
4.5
-1.5
-2.5
0.5
-2.8
-1.8
-2.1
-2.1
1961 to 1988
Mean =
RMSE =
0.2
2.0
-0.7
2.2
-0.6
2.2
1961 to 1974
Mean =
RMSE =
-0.6
1.1
-1.1
2.2
-0.4
2.5
1974 to 1988
Mean =
RMSE =
0.6
2.5
-0.5
2.4
-0.8
2.0
6.6
Table shows errors resulting when each year's multiplier growth (fourth quarter to fourth quarter) was predicted
as the average growth rate over the previous ten years.
79
Table 5
Range of Values for Multiplier Growth
(1961 to 1988)
Yearly
(Fourth Quarter to Fourth Quarter)
Minimum
Maximum
Range
Ten-Year Moving Average
Minimum
Maximum
Range
Ml
-3.5
6.6
10.1
-1.3
0.8
2.1
M2
-2.5
5.5
8.0
0.7
4.2
3.5
M3
-2.2
7.5
9.9
1.8
4.4
2.6
broader aggregates show more extreme values. If the objective is to identify the aggregate most
susceptible to control, the predictability of the relationships between the base and the three definitions
of money would not suggest a clear-cut decision for narrow or broad aggregates.
In general, based on the large size of RMSEs and the extreme values taken on at times for both
velocity and multiplier growth, it would appear that considerable slippage can be expected in both the
links between the amount of high-powered money supplied by the Federal Reserve and the ultimate
objective variable, GNP. In terms of change since the 1960s and early 1970s, the data clearly suggest
that the Ml relationships have undergone the greatest deterioration.
IV. An Econometric Evaluation
Much of the early impetus for monetary targeting using the monetary base as the instrument
came from reduced form equations relating (1) GNP growth to current and lagged values of Ml
growth, and (2) Ml growth to current and lagged values of monetary base growth.14 This section
assesses the stability of these econometric relationships by estimating these two relationships for
successive ten-year periods, dropping and adding one quarter's observation in each successive
"For some early analysis of these relationships, see Leonall Andersen and Jerry L. Jordan,
"Monetary and Fiscal Actions: A Test of Their Relative Importance in Economic Stabilization,"
Federal Reserve Bank of St. Louis Review, November 1968, pp. 11-24; and Karl Brunner, "The Role
of Money and Monetary Policy," Federal Reserve Bank of St. Louis Review, September-October 1989,
pp. 4-22 (reprinted from the July 1968 Review).
80
regression. The sum of the coefficients and the intercept term were recorded for each time period and
plotted over time so that it would be possible to see whether these coefficients have tended to be
stable. Stable coefficients, of course, would suggest that these reduced form equations would be
useful guides for policy, whereas coefficients varying over wide ranges would suggest that these
equations would be of only limited use for policy purposes. In addition, the equations were estimated
over the entire time period and the residuals were plotted to examine whether the results would
suggest the same deterioration in stability of the Ml relationships since the 1970s as the tables in the
first section had indicated.
Table 6 contains the estimated equations for the money supply-GNP relationships and for the
base-money supply relationships over the entire 1952 to 1988 period. In all cases, significant
relationships emerge between money growth and GNP growth and between base growth and money
growth. The relationships between base growth and money growth appear to be tighter than the
relationships between GNP growth and money growth. In particular, the coefficient on Ml growth in
the GNP relationship, while significantly different from zero, is also significantly below one. A value
of one is generally the value expected, that is, a 1 percentage point increase in money growth should
lead to a 1 percentage point increase in nominal spending if velocity growth is constant. The lower
value no doubt reflects the sharp decline in velocity during the 1980s.15 For the relationship between
base growth and Ml growth, however, a coefficient not significantly different from one is obtained,
suggesting that Ml and base growth have about equal trends in the long run. This result does not
appear to hold for M2 and M3, however.
To explore how stable the coefficients have been over time, the six equations in Table 6 were
run for successive ten-year periods, with one quarter added and one quarter dropped each time. The
constant terms and sum of coefficients on current and lagged values were recorded for each regression
and plotted in Charts 1 and 2 for the money-GNP relationships and in Charts 3 and 4 for the basemoney relationships.
The results in Charts 1 and 2 suggest considerable variation in the coefficients in the
15
In these regressions, the constant term can be interpreted as the average growth rate of velocity
during the sample period, while the coefficient can be interpreted as the increase in GNP growth per
percentage point increase in Ml growth. In the regressions for Ml and the monetary base, the
constant term can be interpreted as the average growth rate in the money multiplier, while the
coefficient represents the increase in Ml growth per percentage point increase in the base.
81
Table 6
Long-Run Relationships
Money and GNP
Base and Money
(1952-1 to 1988-IV)
Growth of GNP as a Function of Current and Lagged Moneir Growth
Constant
Term
Sum of
Coefficients
£2
Standard
Error
Ml
4.7
(6.4)
0.57
(4.7)
0.14
4.5
M2
0.4
(1.2)
0.95
(6.2)
0.20
4.3
M3
1.1
(0.9)
0.80
(5.9)
0.19
4.4
Money Growth as a Function of Current and Lagged Base Growth
Constant
Term
Sum of
Coefficients
R2
Standard
Error
Ml
0.19
(0.3)
0.92
(10.4)
0.50
2.9
M2
4.8
(10.1)
0.51
(6.4)
0.31
2.6
M3
4.7
(10.5)
0.65
(8.7)
0.42
2.4
82
regressions for money and GNP over time. In particular, the substantial change in the M2 and M3
constant terms from the 1960s to the 1970s and the large increase in the Ml constant term in the
1980s suggest underlying instability at times (Chart 1). There were gradual declines in all coefficients
on money growth (Ml, M2, and M3) from the early 1960s to the early 1970s. During the 1970s, the
coefficient on Ml remained close to one, then it declined to zero in the 1980s. The M2 coefficient
has remained at around 0.5 since the early 1970s, while the M3 coefficient has moved close to zero in
the late 1980s (Chart 2). In general, for the money-income relationships, the total period results in
Table 6 tend to mask considerably variation in the coefficients in subperiods.16
Next we turn to the subperiod coefficients in the base-money relationships shown in Charts 3
and 4. The constant term for the Ml equation has remained near zero for the most part, and the sum
of the coefficients on base growth, except for a few brief periods, has remained fairly close to one.
The most striking feature of the two charts is the extreme fluctuation in the M2 and M3 constant terms
and coefficients in the late 1970s and early 1980s. Indeed, the M2 and M3 regressions suggest that
there has not been a significant relationship between base growth and M2 or M3 growth since the late
1970s. Earlier studies, using reduced form equations of this type, also concluded that the relationship
between Ml and the base was superior to the relationships incorporating M2 and M3.17
The next step in the evaluation of the monetary aggregates as intermediate variables was to plot
the in-sample errors for the reduced form equations estimated over the entire period (Charts 5 and 6).
Chart 5 contains the errors (on a four-quarter moving average basis) in estimating GNP growth with
money growth. The results from the chart suggest rather large errors in the late 1950s and early 1960s
and a high degree of correlation during that period in the errors from the Ml, M2, and M3 equations.
From the mid-1960s to the mid-1970s, the errors tend to diminish somewhat and to become less
16
There has been some investigation of the reasons for the instability in the Ml-GNP reduced form
relationship. Explanations advanced include the introduction of nationwide NOW accounts and the
large international capital flows that occurred in the 1980s. For more detail, see John Wenninger,
"The Ml-GNP Relationship: A Component Approach," Federal Reserve Bank of New York Quarterly
Review. Autumn 1984, pp. 6-15; and John Wenninger and Thomas Klitgaard, "Exploring the Effects
of Capital Movements on Ml and the Economy," Federal Reserve Bank of New York Quarterly
Review. Summer 1987, pp. 21-31.
17
R.W. Hafer, "Much Ado About M2," Federal Reserve Bank of St. Louis Review. October 1981,
pp. 13-18; John Tatom, "Money Stock Control Under Alternative Definitions of Money," Federal
Reserve Bank of St. Louis Review, November 1979, pp. 3-9.
83
Chart 1
Constant Terms for Money-GNP Reduced Form Equations
Estimated over Successive fen-Year Periods from 1961 to 1988
Percentage points
10
1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988
Chart 2
Sum of Coefficients for Money-GNP Reduced Form Equations
Estimated over Successive Ten-Year Periods from 1961 to 1988
Percentage points
3
84
1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988
Note: Currant and lagged four quarters.
Chart 3
Constant Terms for Monetary Base-Money Equations
Estimated over Successive Ten-Year Periods from 196i to 1988
Percentage points
40
— i n
11111111111111111111111111111111111111111111111111111111111111111111111111111
1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988
Chart 4
Sum of Coefficients for Monetary Base-Money Equations
Estimated over Successive Ten-Year Periods from 1961 to 1988
Percentage points
2
85
1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988
http://fraser.stlouisfed.org/Note: Currant and lagged four quarters.
Federal Reserve Bank of St. Louis
Chart 5
Money-GNP Relationships
Four-Quarter Moving Average of In—Sample Errors*
Percentage points
10
-5 4
—
1 Q
11 • i • 1 1 1 1 1 • i • 1 1 1 • l • • • I • . • I . .
• I . . .
I ••• I ••. I . . .
I . . .
I . . .
i . . .
i . . .
t . . .
i . . .
i . . .
i . . .
i . . .
i . . . . . . .
i . . .
i
i • ... i . .
.
1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988
• Equation estimated 1952 to 1988
Chart 6
Monetary Base—Money Relationships
Four-Quarter Moving Average of In-Sample Errors*
Percentage points
8
86
_C
*•"'
1Q*9
ilniliiilinliiiliiiliiilinlinlinlmlinlnilmlinlmliiiliiilnilinliiiliiiliiiliiiliiilinltiiliiiliiiliiiliiiliiiliiilllili
1CW5
1Q5R
1QR1
1QfiA
1Qfi7
1Q70
1Q7*
1Q7fi
1Q7Q
1QR?
1QR5
1QRK
correlated across equations. From the late 1970s to the late 1980s, the errors become quite large
again, especially for Ml. All three equations show large negative errors in 1982, and the Ml equation
shows another large negative error in 1986. Overall, the 1980s have been dominated by negative
errors on average, while from the mid-1970s to the late 1970s the errors tended to be positive. These
results are generally consistent with those shown in Table 2.
Chart 6 contains the comparable errors for the reduced form equations relating the growth of
Ml, M2, or M3 to current and lagged values of base growth. Until the late 1960s, the errors tended to
be rather small. Since that time, however, the errors have been quite large-especially for M2 and M3
in the 1970s and for Ml in 1985 and 1986. By and large, these results confirm the results shown
earlier in Table 4 that indicated some slippage in recent years in the relationships between the
monetary base and the monetary aggregates.
V. Recent Problems with Money as an Intermediate Target
The Federal Reserve's difficulties with the monetary aggregates, particularly Ml, since the mid1970s illustrate in general the problems of using intermediate targets when the financial structure is
rapidly changing as a result of innovation and deregulation. These difficulties have been well
documented in other studies and will be only briefly discussed here.18
In the mid-1970s, when the Federal Reserve was beginning to set money targets, Ml growth
became considerably weaker than would have been expected from most econometric relationships.
Considerable research was undertaken to explain this unusually weak growth in Ml. By and large, it
was concluded that the high interest rate levels of the early 1970s had spurred increased emphasis on
cash management by the corporate sector.19
18
For detailed discussions of this subject, see Stephen Axilrod, "Monetary Policy, Money Supply
and the Federal Reserve's Operating Procedures," Federal Reserve Bulletin, January 1982, pp. 13-24;
Stephen Axilrod, "U.S. Monetary Policy in Recent Years: An Overview," Federal Reserve Bulletin,
January 1985, pp. 14-24; and Donald Kohn, "Monetary Policy in an Era of Change," Federal Reserve
Bulletin, February 1989, pp. 53-7.
19
For a detailed discussion of this downward shift in money demand and possible explanations, see
Thomas Simpson and Richard Porter, "Some Issues Involving the Definition and Interpretation of the
Monetary Aggregates," Controlling Monetary Aggregates HI (Proceedings of a conference sponsored
by the Federal Reserve Bank of Boston), October 1980, pp. 161-234; and Richard Porter, Thomas
Simpson, and Eileen Mauskopf, "Financial Innovation and the Monetary Aggregates," Brookings
Papers on Economic Activity. 1:1979, pp. 213-29.
87
By the late 1970s, it appeared that the next source of distortion for the money supply was
likely to come from the consumer sector. Some individual states were allowing interest-earning
checking accounts (NOW accounts), and it appeared that NOW accounts eventually would be offered
nationwide. Since consumers would most likely combine savings and transactions balances in their
NOW accounts because NOW accounts paid the same rate as savings accounts, it was not clear
whether these accounts should be part of Ml or M2. This uncertainty raised the general question
whether the monetary aggregates needed to be redefined in light of past and pending changes in the
banking system.20
Ultimately, NOW accounts were defined as part of Ml. However, during the period when
consumers were shifting funds into their NOW accounts from sources other than demand deposits
(1981 and early 1982), the Federal Reserve used a "shifted adjusted" version of Ml that assumed that
something on the order of 25 percent of the balances in NOW accounts were transferred from savings
balances, and hence tended to overstate the growth of narrowly defined money.
In 1983, so-called Super NOWs—accounts not subject to the 5.25 percent ceiling rate affecting
conventional NOW accounts-were authorized. And by 1986, all NOW accounts were completely
deregulated. Many observers at first argued that the demand for Ml would become less interestsensitive with deregulated NOW accounts. They reasoned that the rate on Ml balances would tend to
move with market rates, thereby providing no varying incentives to economize on cash balances. As it
turned out, however, banks were very slow in adjusting the rates on NOW accounts when market rates
changed, and as a result the demand for Ml retained a significant interest elasticity. Nonetheless,
considerable uncertainty about the interest responsiveness of the demand for Ml made it necessary for
the Federal Reserve to exercise flexibility and judgment in using Ml as an intermediate target. In
addition, the combining of savings and transactions balances in Ml also raised questions about the
stability of the income elasticity in the demand for Ml function.21 Thus, Ml appeared to be an
unsatisfactory intermediate target for both reasons cited in the theoretical discussion in Section I. The
demand for Ml seemed likely to be unstable and unpredictable because it depended on the pricing
policies of the banks that set the rates for Ml deposits and because the continued high interest rate
^ o r more detail, see Thomas Simpson, "The Redefined Monetary Aggregates," Federal Reserve
Bulletin. February 1980, pp. 97-114.
21
For more detail, see John Wenninger, "Financial Innovation, A Complex Problem Even in a
Simple Framework," Federal Reserve Bank of New York Quarterly Review, Summer 1984, pp. 1-8.
88
elasticity made it difficult to determine an appropriate target value in the face of possible shifts in
aggregate demand.
Overall, the increased emphasis on cash management by the corporate sector and the additional
savings balances held in Ml by the consumer sector have greatly altered the sectoral composition of
Ml. In the early 1970s, about 33 percent of transactions balances were held by the consumer sector,
and 50 percent by the corporate sector. Now the reverse is true (see Chart 7). With such dramatic
shifts in the sectoral composition of Ml holdings, it is not surprising that Mi's relationship to
economic activity has not been stable.
For M2 and M3, the shifting of funds from time deposits and savings accounts into and out of
NOW accounts as rate spreads changed did not create as much of a problem because the shifts, for the
most part, were internalized. Only in 1983 did the Federal Reserve need to make some adjustments to
its M2 target to allow for possibly large inflows from outside M2 into the newly created money
market deposit accounts (MMDAs).
Even though the flows among accounts were internalized, the interest elasticities of the broad
aggregates seemed likely to decline as more and more deposits became deregulated.22 Again, some
analysts expected that this elasticity would eventually approach zero. As it turned out, however, the
demand for M2 retained a significant elasticity over the short and intermediate terms (see Chart 8)
because banks only partially adjusted the rates on NOW accounts, savings accounts, and MMDAs to
increases in market rates.23 Apparently, banks preferred to see a fraction of these deposits flow out
as market rates increased rather than pay higher rates on the entire stock of these liquid instruments.
They seemed to believe it would be less costly in the short run to replace the outflow of consumer
For a detailed discussion of the phaseout of Regulation Q, see R. Alton Gilbert, "Requiem for
Regulation Q: What It Did and Why It Passed Away," Federal Reserve Bank of St. Louis Review,
February 1986, pp. 22-37.
23
The opportunity cost in Chart 9 is the three-month bill rate less the weighted average of interest
rates (by dollar volume in each component of M2) on the various components of M2. In estimating
how sensitive M2 is to changes in interest rates, one should keep in mind that a 1 percentage point
increase in the bill rate will not necessarily result in a 1 percentage point increase in this spread. How
much the spread increases depends on how bankers adjust the interest rates on the various components
of M2 in response to the increase in market rates and what shares of M2 these various components
happen to represent when the Treasury bill rate increases.
89
Chart 7
Changing Composition of Checkable Deposits
As Percent of Total
Percent
2nd
quarttr
m^jT
1 Q
* ' ' • * ' ' ' 1 ' ' ' ^ • ' ' 1 ' T ' * ' ' ' ^ ' * * 1 ' ' * ^ ' ' ' i ' ' ' ' ' ' ' ^ ' ' ' ' ' ' ' ^ ' ' '
1971
1973
1975
1977
1979
1981
f
1983
' ' ' ^ ' ' '
1985
Chart 8
M2 Velocity and Average M2 Opportunity Cost
Ratio scale
Ratio seals
1.9-1
—
—r18
M4
Percentage points
Ml
1.8-J
,\
U7,
/K"\
Three-month Treasury bill rate*
Right seal*
M
1,7H
M2 Velocity
1.6H
90
Ltft icalt
i
1.5
1979
1980
I
i
i
i
1981
I
i
* Two-quarter moving average.
i
i
1982
I
i
i
i
1983
I
i
i
i
1984
I
i
i
i
1985
I
i
i
i
1986
l
-j
1987
L-J
u.
1988
deposits with wholesale managed liabilities such as large CDs.24 In any case, the uncertainty about
how responsive M2 and M3 would be to changes in market rates limited the usefulness of the broad
monetary aggregates for policy purposes during much of the 1980s.
Overall, for both the narrow and broad aggregates, it appears that much of the instability during
the 1980s resulted from the fact that in a deregulated banking system, bank pricing policies can affect
the monetary aggregates. If banks quickly change deposit rates on all deposit categories to match
movements in market rates, the monetary aggregates should not be affected by the movements in
market rates. If the banks fully respond on some deposit categories but not others, however, there can
be deposit flows that affect both Ml and M2. And if banks tend to behave differently when rates are
falling than when rates are rising, it could be very difficult to estimate the interest responsiveness of
the monetary aggregates. This last outcome seems to have characterized much of the 1980s.
With Chart 9, the effects of banks' pricing policies on Ml can be illustrated. As rates fell in
1984, 1985, and 1986, banks moved the time deposit rate in step with market rates but only changed
the NOW account rate very slowly. As a result, the spreads between the market rate and the NOW
account rate and between the time deposit rate and the NOW account rate narrowed significantly.
Consumers responded to the narrower spreads by increasing more rapidly their holdings of liquid
NOW accounts and reducing the rate at which they increased their holdings of time deposits. As a
result, Ml grew very rapidly.
In contrast, from early 1987 to early 1989, market rates and the rate paid on time deposits
generally increased, while the NOW account rate remained virtually unchanged. Wider spreads
between market rates and the NOW account rate and between the time deposit rate and the NOW
account rate opened up. In response, consumers rapidly increased their holdings of time deposits and
sharply reduced the rate at which their NOW account balances increased. As would be expected, Ml
growth declined sharply.
Hence, in a deregulated system the pricing decisions of commercial banks can have a strong
influence on monetary growth as consumers respond to changing interest rate spreads. And since
^For more detail, see Richard Davis, Leon Korobow, and John Wenninger, "Bankers on Pricing
Consumer Deposits," Federal Reserve Bank of New York Quarterly Review. Winter 1986, pp. 6-13;
reprinted in the American Banker. September 28, 1987. In the long run, the interest elasticity of the
demand for M2 could be close to zero if banks even very gradually adjust deposit rates to match
changes in market rates.
91
Chart 9A
Bank Offering Rates and Market Interest Rates
Percent
1984
1985
1986
1987
1988
1989
1988
1989
Note: As of January 1986, Super NOW rate reflects all NOW accounts.
Chart 9B
M1 Growth
Change from Twelve Months Earlier
92
1984
1985
1986
1987
Chart 9C
Small Time Deposits Growth
Change from Twelve Months Earlier
1984
1985
1986
1987
1988
1989
Chart 9D
Other Checkable Deposits Growth
Change from Twelve Months Earlier
93
1984
1985
1986
1987
1988
1989
banks have left the NOW account rate virtually unchanged during the more recent period of rising
rates, while they reduced it somewhat in the earlier period when rates were falling, it could turn out
that the interest rate elasticity of Ml demand will vary with the rise or fall of market rates.
In sum, it appears that the recent deregulation and innovation in the financial system have
reduced the usefulness of monetary aggregates as intermediate targets for policy. Ml, in particular,
has suffered from these changes. Nevertheless, it still appears that the broader aggregates can be used
to some extent as guides for policy.25
Appendix 1: Money and Inflation
Although most of the empirical work has addressed the relationship between Ml and nominal
income, a few studies have focused on the relationship between Ml growth and inflation.26
Generally, these studies have found rather long lags from Ml growth to prices, with a mean lag of
about 2 1/2 years in the earliest studies and 1 1/2 years in the later studies that include price control
dummies and energy price shocks in the reduced form equations.
As the first chart in this appendix shows, the relationship between Ml growth and inflation has
completely broken down during the 1980s (Chart 10). On the basis of past relationships, inflation
should have averaged around 10 percent during the 1980s, about S percentage points higher than the
actual outcome thus far. Food and energy and other price shocks cannot account for the large errors,
and the breakdown may be due to several factors, including international price competition.27
Recent work at the Board of Governors of the Federal Reserve System suggests that in
forecasting inflation, analysts should use M2 growth rather than Ml growth (because of M2's more
^Other analysts have reached similar conclusions. For example, see W. Michael Cox and Harvey
Rosenblum, "Money and Inflation in a Deregulated Financial Environment: An Overview," Federal
Reserve Bank of Dallas Economic Review. May 1989, pp. 1-19.
^ e e , for example, Keith Carlson, "The Lag from Money to Prices," Federal Reserve Bank of St.
Louis Review. October 1980, pp. 3-10; and Denis Karnosky, "The Link Between Money and Prices,"
Federal Reserve Bank of St. Louis Review, June 1976, pp. 17-23.
27
For more detail, see John Wenninger and Thomas Klitgaard, "Exploring the Effects of Capital
Movements in Ml and the Economy," Federal Reserve Bank of New York Quarterly Review. Summer
1987, pp. 21-31.
94
Chart 10
Money and Price Changes in the Long Run
Change at Annual Rates, Measured from Eight Quarters Earlier
Percent
16
4th
quarter
14 4
12-1
Change in M1
two years earlier
10
8
6
4
Change in
GNP deflator
2
li n l i
1965
4th
quarter
i I i i i Li t I I t I I 1 I I I I I I I l,i i 11 111 i 111 I1, i i 1 i i i I i i i I i i i I i i i I i 11 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
1967
1969
1971 1973
1975
1977
1979
1981 1983 1985 1987
stable velocity) and also allow for the amount of unused capacity in the economy.28 In other words,
if unused capacity was high, a given growth rate of money would have different implications for
inflation than if resources were fully employed.
Since this approach to using money as an intermediate variable is rather recent, the remainder
of Appendix 1 contains a brief discussion of the theory behind the approach. The Board staff begins
with the basic equation of exchange:
28
Jeffrey Hallman, Richard Porter, and David Small, M2 Per Unit of Potential GNP as a Price
Level Anchor. Board of Governors of the Federal Reserve System, Staff Studies, forthcoming.
95
(1) P =
where
M*V,
Y
P = the price level
M = M2
V = M2 velocity
Y = real GNP.
They then use equation 1 to calculate a measure of potential inflationary pressures (FN) by
setting V equal to its long-run average (AV), and real GNP equal to its potential level (Yp).
(2) PN = M*AV.
The ratio of (or the difference between) PN (equation 2) and P (equation 1) is then used as a
measure of inflationary pressures. When PN exceeds P, for example, inflation can be expected to
accelerate. If we divide equation 2 by equation 1 to solve for this measure of inflationary pressure
(PN/P), we can isolate the factors that cause PN to deviate from P.
(3)
PN = M*AV * Y =
P
Yp
M*V
Y * AV.
Yp
V
Equation 3 shows that M drops out of this measure of inflationary pressure. Hence, two factors
remain that could contribute to PN's rising above P and therefore to a forecast of greater inflation.
One would be an increase in Y relative to Yp; the other would be a reduction in V relative to AV. In
addition, a combination of these two factors could result in a forecast of higher inflation. If velocity
was stable (AV/V = 1), of course, then forecasts of inflation would depend only on where actual real
GNP stood relative to potential GNP, and this approach would not be conceptually different from other
approaches that look at the amount of slack in the economy as an indicator of future inflation trends.
Hence, weighting the Y/Yp ratio by the AV/V ratio is the feature of the Board staffs approach that
differentiates it from more conventional approaches.
How should this measure of inflationary pressure be interpreted when V is not equal to AV?
For a given ratio of Y to Yp, if the Federal Reserve was easing policy by increasing the money supply
relative to the average need for money as GNP grew, then V would be falling relative to AV and the
ratio of AV to V would be increasing. Hence, the given ratio of Y to Yp would be associated with
more than normal potential inflationary pressures because the money supply had been increased.
Alternatively, if the Federal Reserve was reducing the money supply relative to the average need for
money as GNP grew, the ratio of AV to V would be falling because V would be increasing. This
would indicate less inflation for the given Y to Yp ratio. Hence, at a theoretical level a case could be
96
made for also looking at the ratio of AV to V in predicting the inflation that is likely to be associated
with a given potential GNP gap.
However, if V shifts for reasons not associated with policy, then using this ratio could be
misleading. This would have been particularly true during the 1980s, when a sharp fall in inflation
and inflationary expectations caused interest rates to fall sharply and M2's velocity to decline. It
would seem that under those circumstances the ratio of AV to V would not add anything to our ability
to track inflation because it would be increasing in response to a decline in inflation, not to a change
in policy per se.
It is also possible to look at this indicator from a slightly different perspective. If PY/M is
substituted for V on the right side of equation 3, the following relationship results:
(4)
PN = AV*M.
P
Yp*P
Equation 4 suggests that this inflation index is a nominal approach to using the potential GNP
gap to forecast inflation, with a monetarist twist. AV*M in the numerator is the basic monetarist
approach to forecasting nominal GNP (or nominal demand) as the product of average long-run velocity
and M2. Yp*P in the denominator represents potential real GNP converted into nominal terms by
using the current price level. Hence, equation 4 suggests that when nominal GNP, calculated by
AV*M, exceeds nominal potential GNP as calculated by Yp*P, prices will increase over time to make
up the difference because PN exceeds P. This approach is not greatly different from conventional
monetarist equations that simply relate P to M. The only difference is that for P to increase, M must
increase by enough so that AV*M exceeds the critical value of Yp*P. In other words, nominal
demand generated by the increase in M2 must exceed potential output of the economy evaluated in
nominal terms by using the current price level. If the economy was coming out of a recession, such
that AV*M was considerably less than Yp*P, P would not be expected to rise, even though nominal
demand might be increasing. This result occurs because nominal demand would still be below
"nominal capacity." Hence, the inflationary consequence of an increase in M depends on where the
current level of economic activity stands relative to its long-run potential.
Chart 11 shows the recent track record of this inflation indicator. By and large, it appears that
this indicator would be useful in projecting longer run trends in inflation but might not be very useful
97
Chart 11A
Inflation Indicator Based on M2
Ratio scale
140-
iiiiitiiiii.il
l»>.U«tli>llli.l.li.tll.>lJLlHlllMl
1956 1958 1960 1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988
Chart 11B
Inflation
Change in the Implicit GNP Deflator from Four Quarters Earlier
Percentage Points
12-
1956 1958 1960 1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988
The current price level (P, the solid line in the top panel) is the implicit GNP deflator, which is set to
100 in 1982.
The long-run equilibrium price level given current M2 (P*, the dashed line in the top panel) is
calculated as P* =(M2 x V*)/Q*, where V* is an estimate of the long-run value of the GNP velocity of M2-the
mean of V2 from 1955-1 to 1988-IV-and Q* is a Federal Reserve Boarcl staff measure of potential real GNP.
The vertical lines mark the quarters when the difference between the current price level (P) and the
long-run equilibrium price level (P*) switches sign, and thus when inflation, with a lag, tends to begin
accelerating or decelerating.
Inflation (bottom panel) is the percentage change in the implicit GNP deflator from four quarters earlier.
For more details, see Jeffrey Hallman, Richard D. Porter, and David H. Small, M2 Per Unit of Potential
GNP as a Price-Level Anchor. Board of Governors of the Federal Reserve System, Staff Studies, forthcoming.
The model for this chart and its accompanying commentary was taken from Federal Reserve Chairman
Alan Greenspan's statement before the Committee on Banking, Housing and Urban Affairs of the U.S. Senate on
February 21, 1989.
98
in the shorter run:
Appendix 2: Federal Reserve Operating Procedures
The Federal Reserve has attempted to control the money stock in different ways over the last
twenty years. Several recent articles provide excellent overviews of the Federal Reserve's techniques
since the early 1970s.30 More detailed and technical analysis can be found in the Federal Reserve's
two-volume study, New Monetary Control Procedures, published in 1981.31 Because a great deal of
material is already available on this subject from both general and technical perspectives, this appendix
will simply highlight some of the more important changes in the Federal Reserve's operating
procedures.
Analysts of U.S.financialmarkets divide the Federal Reserve's operating procedures into three
distinct periods. The first, prior to October 1979, was characterized by relatively stable interest rates
and low real rates of interest. The second period, from October 1979 to about October 1982, showed
high real rates and extreme volatility of interest rates. The last period, October 1982 to the present,
has had continued high real interest rates but with much less volatility of interest rates than the
previous period (about as much volatility as in the first period).
Before October 1979, the Federal Reserve attempted to keep Ml within its target ranges by
29
For an econometric analysis that supports this conclusion, see Kenneth N. Kuttner, "Inflation and
the Growth Rate of Money," Federal Reserve Bank of Chicago Economic Perspectives, JanuaryFebruary 1990, pp. 2-11.
^Stephen Axilrod, "U.S. Monetary Policy in Recent Years," Federal Reserve Bulletin. January
1985; Henry C. Wallich, "Recent Techniques of Monetary Policy," Federal Reserve Bank of Kansas
City Economic Review. May 1984. See also Stephen Axilrod, "Monetary Policy, Money Supply and
the Federal Reserve's Operating Procedures," Federal Reserve Bulletin. January 1982; and Stephen
Axilrod and David Lindsey, "Federal Reserve System Implementation of Monetary Policy: Analytical
Foundations of the New Approach," American Economic Review, vol. 71, no. 2 (May 1981). For an
international perspective, see Changes in Money Market Instruments and Procedures: Objectives and
Implications. Bank for International Settlements, March 1986. This appendix is adapted from my
contribution to this volume. For a historical perspective, see Ann-Marie Meulendyke, "A Review of
Federal Reserve Policy Targets and Operating Guides in Recent Decades," Federal Reserve Bank of
New York Quarterly Review. Autumn 1988, reprinted in this volume.
3I
Board of Governors of the Federal Reserve System, New Monetary Control Procedures. Staff
study, February 1981.
99
gradually changing short-term interest rates. In essence, the Federal Reserve was operating through
the demand-for-money function, supplying whatever quantity of reserves was demanded at the interest
rate level estimated as most likely to hit the Ml target. The obvious problem with such an approach
was that changes in the demand for reserves, until they were recognized, were fully accommodated in
the short run by changes in the supply of reserves, with no automatic response in interest rates (see
Figure 1).
In October 1979, the Federal Reserve began to place less emphasis on short-term interest rates
and more emphasis on the supply of reserves to control Ml. This shift in emphasis meant, in effect,
that interest rates would change automatically to help balance the supply and the demand for money.
Because the Federal Reserve controlled only nonborrowed, not total, reserves in the short run, total
reserves still responded to changes in the demand for reserves, although not as much as in the earlier
period (see Figure 2). Interest rates, of course, were not completely ignored. The Federal Reserve
still set consulting ranges for the federal funds rate. If the funds rate appeared likely to go beyond
these ranges, the FOMC would discuss adjusting the range in the light of economic developments.
Since the autumn of 1982, the Federal Reserve has focused less on the supply of reserves and
more on the amount of pressure on banks' reserve positions, as measured by the quantity of borrowed
reserves. If the Federal Reserve maintains the same borrowing target when the demand for reserves
shifts, then the procedure is similar to the pre-October 1979 approach because the level of borrowing
is related to the spread between the funds rate and the discount rate. (Compare Figure 3 with Figures
1 and 2. Figure 4 shows the determinants of open market operations under current procedures.)
Under current procedures, however, the funds rate could fluctuate more in the very short run
(compared with the period before October 1979) because the relationship between the funds rate and
the level of borrowing is not exact and because the Federal Reserve attempts to attain its objective for
borrowing on average for a two-week period rather than on a day-to-day basis.
100
Figure 1
Example of Operating Procedures before October 1979
The demand for total reserves is determined by: (1) the demand for required reserves that banks
must hold against transactions deposits and certain other liabilities, and (2) the demand for excess
reserves, which is largely seasonal in nature. Excess reserves are a very small percentage of total
reserves. The quantity of total reserves demanded increases as interest rates decline* because the
demand for transactions deposits (and hence for the reservesrequiredfor them) increases as interest
rates decline. The supply of reserves consists of two components, borrowed reserves and nonborrowed
reserves. Nonborrowed reserves are equivalent to the sum of several technical factors and open market
operations. Borrowed reserves tend to increase as the spread between the federal funds rate and the
discount rate widens. Hence the overall supply of reserves increases as the funds rate increases. (For
a more formal model of the supply and demand for reserves, see the paper by Brian F. Madigan and
Warren T. Trepeta in Bank for International Settlements, Changes in Money Maiket Instruments and
Procedures.)
Funds rate
Reserves
Before October 1979, the Federal Reserve would tend to stabilize the funds rate in the very short
run. This, in effect, made the supply of reserves (S) a horizontal line at the desired funds rate level.
If the demand for total reserves increased from Dx to D2, there would be no automatic upward pressure
on the funds rate (r^. If the shift from Dx to D2 was ascribed to excessive money growth, the Federal
Reserve would undertake operations to raise the funds rate from rt to some higher level (r2).
10 i
Figure 2
Example of Operating Procedures from October 1979 to Autumn 1982
Funds rate
Supply of
total reserves
r2
.
Reserves
During this period, the Federal Reserve would set the level of nonborrowed reserves (NB) in line
with its desired growth rate for Ml and associated required reserves. If the demand for reserves (D^
was greater than the supply of nonborrowed reserves, the fund rate fo) would exceed the discount
rate (rd) and banks would borrow the difference between total reserves (TRX) and nonborrowed
reserves. If the demand for reserves increased because of growth in Ml (Dx to D2) that was more
rapid than desired, the funds rate would increase automatically (rx to r2), and borrowed reserves would
increase (TI^ - NB to TR2 - NB) because the spread between the funds rate and the discount rate
would increase (rrrd to r2-rd). Over a longer period, the rise in rates could be expected to exert
restraint on the excessively rapid growth in Ml.
102
Figure 3
Example of Operating Procedures since Autumn 1982
Funds rate
Reserves
During this period, the Federal Reserve has set the supply of nonborrowed reserves (NB^ to
achieve a certain degree of pressure on the banks' reserve positions (TRj-NBi). Since TRpNBj =
borrowings (B) and B responds in a loose way to the spread between the funds rate (rx) and the
discount rate (rd), increases in the desired level of pressure on the banks' reserve positions put upward
pressure on the funds rate. If the demand for reserves increases from Dx to D2, the funds rate will not
necessarily increase from rx to r2 (as in the previous example) if the Federal Reserve has kept the
desired level of B the same. If the Federal Reserve wants to keep the level of borrowing constant, it
increases the supply of reserves from Sx to S2 by increasing nonborrowed reserves from NBi to NB2.
The degree of pressure on the banks' reserve positions will be unchanged because TR3 - NB2 = TRj NBX. Hence the funds rate will not automatically increase when the demand for reserves increases
from Dx to D2.
103
Figure 4
The Determinants of Open Market Operations
Money supply
44
t
M1
Currency
Demand deposits*
\
Other checkable*
\
\
,
MMDAs*
\
M2
Savings*
Small time*
M3
Money funds
Total
reserves
Required
on
transactions
deposits
Total
reserves
Borrowed
Required
on nontransactions
deposits
Nonborrowed
Overnight RPs
and Euros
Excess
Large time*
Money funds
Term RPs & Euros
Other market
operations
*These eomnonfints
as well as some
other liabilities
have reserve
requirements.
Reserve factors
At the start of each two-week maintenance period, a forecast of required reserves is made on the
basis'of expected movements (largely seasonal) in the components of the monetary aggregates and
other liabilities that have reserve requirements. This forecast forrequiredreservesis then added to the
expected demand for excess reserves (also largely seasonal) to obtain a forecast for total reserves.
Given the FOMCs objective for borrowed reserves, this forecast of total reserves implies a target for
nonborrowed reserves. Nonborrowed reserves, in turn, consist of two components: open market
operations and the net reserve impact of several factors such as currency, float, and the Treasury's
balance at the Federal Reserve. The Open Market Desk undertakes open market operations equal to
the difference between the target for nonborrowed reserves and the net reserve impact of these other
reserve factors. These estimates are revised daily, and at times the volume of open market operations
called for by the forecasts can change appreciably over the course of a two-week period.*
* For more detail, see John Partlan, Kausar Hamdani, and Kathleen M. Camilli, "Reserves
Forecasting for Open Market Operations," Federal Reserve Bank of New York Quarterly Review,
Spring 1986, pp. 19-33.
104
Bibliography
Andersen, Leonall, and Jerry L. Jordan. "Monetary and Fiscal Actions: A Test of Their Relative
Importance in Economic Stabilization." Federal Reserve Bank of St. Louis Review, November
1968.
Axilrod, Stephen. "Monetary Policy, Money Supply and the Federal Reserve's Operating Procedures."
Federal Reserve Bulletin, January 1982.
. "U.S. Monetary Policy in Recent Years: An Overview." Federal Reserve Bulletin, January
1985.
Axilrod, Stephen, and David Lindsey. "Federal Reserve System Implementation of Monetary Policy:
Analytical Foundations of the New Approach." American Economic Review, vol. 71, no. 2 (May
1981).
Rpfthdl9£Anton. "How Controllable is Money Growth." Federal Reserve Bank of St. Louis Review,
Bank for International Settlements. Changes in Money Market Instruments and Procedures:
Objectives and Implications. Proceedings of a meeting, March 1986.
Board of Governors of the Federal Reserve System. New Monetary Control Procedures. Staff study,
February 1981.
Brunner, Karl. "The Role of Money and Monetary Policy." Federal Reserve Bank of St. Louis
Review, September-October 1989.
Burger, Albert E. "Money Stock Control." Controlling Monetary Aggregates II: The Implementation.
Proceedings of a conference sponsored by the Federal Reserve Bank of Boston, September 1972.
. "The Puzzling Growth of the Monetary Aggregates in the 1980s." Federal Reserve
Bank of St. Louis Review, September-October 1988.
Carlson, Keith. "The Lag from Money to Prices." Federal Reserve Bank of St. Louis Review,
October 1980.
Cox, W. Michael, and Harvey Rosenblum. "Money and Inflation in a Deregulated Financial
Environment: An Overview." Federal Reserve Bank of Dallas Economic Review, May 1989.
Darby, Michael, Angelo Masarco, and Michael Marlow. "The Empirical Reliability of Monetary
Aggregates as Indicators." Economic Inquiry, October 1989.
105
Darby, Michael, William Poole, David Lindsey, and Milton Friedman. "Recent Behavior of the
Velocity of Money." Contemporary Policy Issues. January 1987.
Davis, Richard G. "Monetary Aggregates and the Use of 'Intermediate Targets' in Monetary Policy."
New Monetary Control Procedures, vol. 1, Board of Governors of the Federal Reserve System Staff
Study, February 1981.
Davis, Richard G., and Frederick C. Schadrack. "Forecasting the Monetary Aggregates with Reduced
Form Equations." Monetary Aggregates and Monetary Policy, Federal Reserve Bank of New York,
October 1974.
Davis, Richard G., Leon Korobow, and John Wenninger. "Bankers on Pricing Consumer Deposits."
Federal Reserve Bank of New York Quarterly Review, Winter 1986.
Friedman, Benjamin. "Lessons on Monetary Policy in the 1980s." The Journal of Economic
Perspectives, Summer 1988.
. "The Value of Intermediate Targets in Implementing Monetary Policy." Price Stability and
Public Policy. Proceedings of a symposium sponsored by the Federal Reserve Bank of Kansas City,
August 1984.
. "Conducting Monetary Policy by Controlling Currency Plus Noise." Carnegie-Rochester
Conference Series on Public Policy, no. 29, 1988.
Garfinkel, Michelle R., and Daniel Thornton. "The Link Between Ml and the Monetary Base in the
1980s." Federal Reserve Bank of St. Louis Review, September-October 1989.
Gilbert, R. Alton. "Requiem for Regulation Q: What It Did and Why It Passed Away." Federal
Reserve Bank of St. Louis Review, February 1986.
Goldfeld, Stephen. "The Case of the Missing Money." Brookings Papers on Economic Activity,
3:1976.
Hafer, R.W. "Much Ado About M2." Federal Reserve Bank of St. Louis Review, October 1981.
Hallman, Jeffrey, Richard Porter, and David Small. "M2 Per Unit of Potential GNP as a Price Level
Anchor." Board of Governors of the Federal Reserve System, Staff Studies, forthcoming.
Hamburger, Michael. "Behavior of the Money Stock: Is There a Puzzle?" Journal of Monetary
Economics, July 1977.
Johannes, James, and Robert Rasche. "Predicting the Money Multiplier." Journal of Monetary
Economics, July 1979.
Kane, Edward. "Selecting Monetary Targets in a Changing Financial Environment." Money Policy
Issues in the 1980s. Proceedings of a symposium sponsored by the Federal Reserve Bank of Kansas
City, August 1982.
106
Karnosky, Denis. "The Link Between Money and Prices." Federal Reserve Bank of St. Louis
Review, June 1976.
Kohn, Donald. "Monetary Policy in an Era of Change." Federal Reserve Bulletin, February 1989.
Kuttner, Kenneth N. "Inflation and the Growth Rate of Money." Federal Reserve Bank of Chicago
Economic Perspectives, January-February 1990.
Mayer, Thomas, James Duesenberry, and Robert Aliber. Money, Banking and the Economy. New
York: W.W. Norton and Company, 1987.
Meulendyke, Ann-Marie. "A Review of Federal Reserve Policy Targets and Operating Guides in
Recent Decades." Federal Reserve Bank of New York Quarterly Review, Autumn 1988. Reprinted
I in this volume.
Partlan, John, Kausar Hamdani, and Kathleen M. Camilli. "Reserves Forecasting for Open Market
Operations." Federal Reserve Bank of New York Quarterly Review, Spring 1986.
Pierce, James L., and Thomas D. Thomson. "Some Issues in Controlling the Stock of Money."
Controlling Monetary Aggregates II: The Implementation. Proceedings of a conference sponsored
by the Federal Reserve Bank of Boston, September 1972.
Poole, William. "Monetary Policy Lessons of Recent Inflation and Disinflation." The Journal of
Economic Perspectives, Summer 1988.
. "Optimal Choice of Monetary Policy Instruments in a Simple Stochastic Model."
Quarterly Journal of Economics, May 1970.
Porter, Richard, Thomas Simpson, and Eileen Mauskopf. "Financial Innovation and the Monetary
Aggregates." Brookings Papers on Economic Activity, 1:1979.
Ritter, Lawrence, and William Silber. Principles of Money, Banking and Financial Markets. New
York: Basic Books, Inc., 1989.
Simpson, Thomas. "The Redefined Monetary Aggregates." Federal Reserve Bulletin, February 1980.
Simpson, Thomas, and Richard Porter. "Some Issues Involving the Definition and Interpretation of the
Monetary Aggregates." Controlling Money Stock in. Federal Reserve Bank of Boston, October
1980.
Stone, Courtney, and Daniel Thornton. "Solving the 1980s* Velocity Puzzle: A Progress Report."
Federal Reserve Bank of St. Louis Review, August-September 1987.
Tatom, John. "Money Stock Control Under Alternative Definitions of Money." Federal Reserve Bank
of St. Louis Review, November 1979.
Tobin, James. "Monetary Policy: Rules, Targets, and Shocks." Journal of Money, Credit, and
Banking, November 1983.
107
Wallich, Henry C. "Recent Techniques of Monetary Policy." Federal Reserve Bank of Kansas City
Economic Review, May 1984.
Wenninger, John. "Financial Innovation, A Complex Problem Even in a Simple Framework." Federal
Reserve Bank of New York Quarterly Review, Summer 1984.
. "Responsiveness of Interest Rate Spreads and Deposit Flows to Changes in Market Rates."
Federal Reserve Bank of New York Quarterly Review, Autumn 1986.
. "The Ml-GNP Relationship: A Component Approach." Federal Reserve Bank of New
York Quarterly Review, Autumn 1984.
. "Money Demand: Some Long-Run Properties." Federal Reserve Bank of New York
\ Quarterly Review, Spring 1988.
Wenninger, John, and Thomas Klitgaard. "Exploring the Effects of Capital Movements on Ml and the
Economy." Federal Reserve Bank of New York Quarterly Review, Summer 1987.
Volcker, Paul. "The Role of Monetary Targets in an Age of Inflation." Journal of Monetary
Economics, vol. 4, no. 2 (April 1978).
108
LIQUID ASSET MEASURES AS INTERMEDIATE TARGETS AND INDICATORS
FOR MONETARY POLICY
Gabriel S. P. de Kock and Lawrence J. Radecki
This paper reviews the potential use of aggregate measures of liquid assets in formulating
monetary policy. It addresses the rationale for using aggregate measures of liquid assets as targets or
indicators of monetary policy, the appropriate definition of a liquid asset aggregate, and the empirical
evidence in favor of the Federal Reserve's monetary aggregate L as an informative indicator or reliable
target for monetary policy. Finally, practical issues relating to implementing a liquid asset target,
including controllability and the timely availability of data, are discussed.
Three main rationales can be advanced to justify the use of liquid asset measures as indicators or
intermediate targets for monetary policy:
• A broad monetary aggregate includes all assets yielding monetary services and
internalizes possible changes in their relative attractiveness and functions;
thus it should have a more stable relationship with policy objectives.
• The existence of capital market imperfections implies that firms' and
households' portfolio and spending decisions cannot be separated and that
they may be liquidity constrained. Thus, liquid asset aggregates may be
stably related to private sector spending decisions and plans as well as to
the ability of firms and households to finance spending. A related
justification is that liquid asset measures provide timely information about
lending to households and firms byfinancialinstitutions.
• It has been argued that there is no a priori reason to presume that a single
monetary aggregate or asset price would capture all the information pertinent
to monetary policy decisions. Policy makers can minimize the risk of policy
errors by diversifying over a number of policy indicators/intermediate
targets, including a broad liquid assets measure.
We review a number of recent studies that have provided empirical evidence on capital market
imperfections and thus indirectly favor the use of liquid asset aggregates. We point out, however, that
the relationship between a liquid asset aggregate and spending need not be stable if the proportions of
households and firms that are liquidity constrained change over time. Furthermore, we note that there
109
is considerable ambiguity whether capital market imperfections constitute a rationale for the use of
liquid asset aggregates or credit aggregates in the formulation of monetary policy.
Two measurement problems complicate the practical definition of a liquid asset measure:
identifying (1) the assets that should be included in a liquid asset measure and (2) the aggregation
principle that should be used in compiling the aggregate. We discuss three potential shortcomings of
the Federal Reserve System's liquid asset aggregate, L:
• L overlaps considerably and is highly correlated with other high-level financial
aggregates and is unlikely to convey much information that is not provided by
these aggregates.
• If a liquid asset aggregate is to measure readily available spending power in
the economy, a potentially serious measurement error arises from the
exclusion of prearranged lines of credit from L.
• L is calculated as a simple-sum aggregate. If the assets included in an
aggregate are imperfect substitutes, both the demand for the aggregate and
its relationship with economic activity will not be independent of its
composition; consequently, it may be of limited value to policy makers.
We also review the conceptual case for and experience with monetary-services, or Divisia, indexes that
have been proposed as alternatives to the conventional simple-sum aggregates, and we conclude that
Divisia aggregates are subject to different, but equally serious, measurement problems.
A liquid asset aggregate can play a role in the formulation of monetary policy in three
conceptually different roles, as an intermediate target, an indicator conveying information about future
economic events or the current stance of policy, and a constraint. A successful intermediate target is
related in a stable and predictable manner to both policy objectives and policy instruments. To assess
whether a liquid asset measure would satisfy these requirements, we review earlier empirical research
and present the results of our own empirical analysis using the liquid asset measure, L. Our findings
are as follows:
• There is a stable long-run demand for real L.
• L adjusts very slowly to changes in its determinants.
• The relationship that describes the dynamic adjustment of L in response to
changes in its determinants is difficult to capture statistically, as evidenced by
poor out-of-sample simulation results, and it has shifted in 1980.
110
• The monetary base L multiplier is less volatile and marginally easier to
forecast than the M2 multiplier. However, it is not clear that this advantage
will be of practical consequence or can be exploited by a policy regime in
which the base is adjusted to hit an intermediate target for L.
The evidence on the relationship between L and policy goal variables does not suggest the use of L as
an intermediate target. Since L responds slowly to changes in its determinants, policy makers may
have to accept long-lasting deviations of L from its intermediate target path. Moreover, control over L
may offer only imprecise control over final objectives of policy, because the short-run relationship
between L and goal variables is difficult to predict and subject to unpredictable shifts. Finally, the
long time delay in the availability of data on L compared with that for other monetary aggregates
would make an L intermediate target difficult to implement.1
A liquid asset measure can be used as a monetary policy indicator in two conceptually distinct
roles: first, as a predictor of future economic developments or an "information variable" that is used
to signal the need to change instrument settings; second, as a measure of policy stance that is used to
indicate the need to change an intermediate target path that has become inappropriate. L holds little
promise as an information variable, although it fares somewhat better than M2. A vector
autoregression analysis of data from 1959 to 1989 shows that once the commercial paper rate is
included in the information set, L has only marginal predictive value for nominal GNP, none for real
GNP, and more surprisingly, none for the GNP deflator.
Finally, we do not foresee a role for L as a constraint in the policy process. It seems implausible
that one could specify ex ante a level or rate of change for L that, if breached in one direction or the
other, will signify an unambiguous policy error.
I. Rationale for Liquid Asset Measures
There are three possible motivations for using a broad measure of liquid assets in the formulation
of monetary policy. First, there may be no well-defined set of financial assets that corresponds to the
theoretical concept of money. Different assets with varying degrees of moneyness exist, and it can be
argued that a "money" aggregate used as an intermediate target or indicator for monetary policy should
include all these assets. Secondly, capital market imperfections imply that private agents may often be
1
Final data for L are available one month after the monetary aggregates for the same date.
ill
liquidity constrained2 and that their spending and portfolio decisions cannot be separated. Thus,
liquid asset holdings may be stably related to private sector spending decisions and plans as well as to
the ability of firms and households to finance spending. Finally, there is no a priori reason to presume
that any single financial aggregate accurately captures the influence of financial decisions on real
economic activity; thus it can be argued that policy makers should use a liquid asset measure as one of
the targets/indicators of monetary policy.
In this section we review the conceptual case for using a liquid asset measure in the monetary
policy process. We also survey the empirical evidence on the existence of liquidity constraints and
discuss the ambiguities arising from the fact that the rationale for a liquid asset aggregate proves, on
examination, to be fundamentally the same as the rationale for credit measures.
A. The Conceptual Case for Liquid Asset Measures
An argument for specifically using broad financial aggregates or measures of liquid assets to guide
monetary policy decisions rests on the close substitutability of different financial assets. The
theoretical distinction between transactions assets and investment media has no real-world counterpart.
Private agents economize on transactions balances by substituting higher yielding liquid assets for
them. Hence, failing to consider substitution in and out of transactions money in a world where
financial intermediaries create assets that increasingly blur the distinction between transactions and
investment balances will, in general, lead to failures in judging the stance of policy. Thus, a reliable
measure of "the quantity of money" or spending power should include the close substitutes for
checkable deposits offered by both banks and nonbank financial intermediaries. However, there exists
a virtual continuum of assets of varying degrees of moneyness or substitutability for transactions
balances, and consequently the set of assets that should be included in the measures intended to guide
monetary policy has no clearly defined boundary.
One solution is to choose a broad liquid asset aggregate and thereby to ensure that all relevant
assets are included. But this introduces a new problem: how to aggregate the relevant assets into a
single quantity. A proposed solution to this problem, discussed in greater detail in the section dealing
with measurement issues, is to use an aggregate with components weighted to reflect their
"moneyness" or the monetary services that they provide and thus to capture the implications of both
2
The terms "liquidity constrained" and "credit rationed" are used interchangeably in this paper. In
the literature, the term "liquidity constraint" is often used when referring to consumers and "credit
ration" when referring to firms.
112
shifts in quantities and composition. The use of broad aggregates weighted to measure the quantity of
"money" accurately can also be justified as a useful simplification—a substitute for the complex
disaggregated models of the financial industry favored by, for example, James Tobin and his
associates.3
A related argument in favor of broad liquid asset aggregates relies on the recent experience with
financial innovation and its implications for a long-run strategy of monetary targeting.4 After the
"missing money" episode of the mid-seventies, a long-run intermediate targets strategy still seemed
feasible to many.5 It was argued that although the demand for a particular monetary aggregate may
be unstable in the short run, such instabilities are bound to be reversed and therefore less important in
the longer run. Thus, economists who held this view concluded that an intermediate targets strategy
could be used to foster long-run objectives, even if not to fine-tune the economy. However, the
financial innovations of the early eighties have cast considerable doubt on this presumption, at least as
far as an intermediate target strategy based on narrow financial aggregates is concerned. Financial
innovations have rendered the demand for narrow aggregates even more unstable in the long run than
in the short run by changing the economic roles of the assets encompassed by these aggregates. Such
changes largely involve changes in the rates of return of different liquid assets and therefore affect
narrow monetary aggregates much more strongly than broad aggregates.6 On these grounds it might
3
See, for example, James Tobin, "A General Equilibrium Approach to Monetary Theory," Journal
of Money. Credit and Banking, vol. 1, no. 1 (1970); and James Tobin and Donald Hester, Financial
Markets and Economic Activity (New York: Wiley, 1967).
4
Benjamin Friedman summarizes this line of reasoning in his discussion of the paper presented by
Simpson and Porter at the 1980 Boston Fed conference on the control of monetary aggregates. See
Thomas D. Simpson and Richard D. Porter, "Some Issues Involving the Definition and Interpretation
of the Monetary Aggregates," and Benjamin M. Friedman, "Discussion," in Controlling Monetary
Aggregates HI. Federal Reserve Bank of Boston, Conference Series no. 23, 1981.
5
Friedman, in his discussion of Simpson and Porter, characterizes this view as held by the
majority of macroeconomists. An example is the paper by Karl Brunner, "The Control of Monetary
Aggregates," in the same volume.
6
A recent example of this phenomenon is the slowing of Ml and M2 growth from the third
quarter of 1988 to the second quarter of 1989, which occurred partly in reponse to changes in bank
deposit pricing policies. Banks allowed the differential between market rates such as the three-month
Treasury bill yield and the rates on their NOW accounts, savings accounts, and MMDAs to increase
sharply. For example, in the first quarter of 1989, Treasury bills yielded 398 basis points more than
NOW accounts, 360 basis points more than savings accounts, and 62 basis points more than small sixcontinued...)
113
seem advisable to base a long-run intermediate target strategy on a broad concept of total liquid assets
rather than a narrow concept of mohey.
Many observers point to the existence of capital market imperfections as diverse as transactions
costs and information asymmetries and emphasize that households' and firms' responses to income
shocks and financial incentives are significantly different from those predicted by theories predicated
on the existence of perfect capital markets.7 The existence of capital market imperfections modifies
standard theory in two ways; on the one hand, households' and firms' anticipated spending will be
reflected in the maturity composition of their portfolios while, on the other hand, economic agents'
responses to income and interest rate changes will depend on their holdings of liquid assets. Thus,
liquid asset aggregates may provide policy makers with information about planned spending and may
have a stable relationship to aggregate expenditures.
Pissarides provides a theoretical analysis of the interrelationship between a consumer's portfolio
decisions and consumption spending that arises from transactions costs and differences between
borrowing and lending rates.8 He shows that precautionary balances and funds earmarked for
spending in the near future are held in the form of liquid assets because liquid assets have investment
characteristics but also allow resources to be transferred to transactions media easily and at low cost.
Furthermore, consumption is more sensitive to current income than predicted by the permanent income
hypothesis because transitory decreases in current income lower consumption if the household has
insufficient liquid assets.9 Consequently, the information in current income flows and portfolio
6
(...continued)
month time deposits, compared with differentials of 68, 50, and -8 basis points, respectively, in the
first quarter of 1987.
7
For examples of this line of thought, see James Tobin and Walter Dolde, "Wealth, Liquidity and
Consumption," in Consumer Spending and Monetary Policy: The Linkages, Federal Reserve Bank of
Boston, Conference Series no. 5, 1971; and Richard Jackman and John Sutton, "Imperfect Capital
Markets and the Monetarist Black Box," The Economic Journal, March 1982, pp. 108-28.
8
Christopher A. Pissarides, "Liquidity Considerations in the Theory of Consumption," Quarterly
Journal of Economics, May 1978, pp. 279-96. In particular, Pissarides assumes that the costs of
transactions in long-term assets exceed those of transactions in short-term assets and that the interest
rate at which consumers can borrow exceeds the rates on both the short- and long-term investment
media available to them.
9
Empirical evidence on the "excess sensitivity of consumption to current income" is provided by
Marjorie Flavin, "The Adjustment of Consumption to Changing Expectations About Future Income,"
(continued...)
114
composition (for example, the ratio of liquid to illiquid assets) may predict consumption better than
measures of human and nonhuman wealth calculated on the basis of market interest rates.10
Credit rationing can occur because financial intermediaries cannot monitor perfectly borrowers'
income prospects and spending while borrowers' potential liability is limited by the possibility of
bankruptcy.11 The existence of a limit to the amount that a firm or household can borrow will, in
general, raise the true cost of borrowing above the interest rate at which financial institutions lend and,
by implication, above the opportunity cost of the firm's or household's own funds. This may occur
even if such a liquidity constraint is not strictly binding, simply because current borrowing reduces the
amount of funds available for future borrowing. Thus, households' consumption spending will be
more sensitive to current income than predicted by the life-cycle hypothesis. Similarly, firms'
investment spending will be more sensitive to current cash flows than predicted by neoclassical models
that assume perfect capital markets. To assess the influence of liquidity constraints on consumption
behavior, Tobin and Dolde used simulations of a simple macroeconomic model.12 They assumed that
liquidity constraints bind for younger and poorer segments of the population and that these economic
agents cannot borrow at all against future labor income. They found that, at realistic parameter values,
liquidity constraints raise the marginal propensity to consume out of resources (nonhuman wealth plus
the present value of future labor income) from 0.055 to between 0.09 and 0.12.
9
(...continued)
Journal of Political Economy, vol. 89 (1981), pp. 974-1009; "Excess Sensitivity of Consumption to
Current Income: Liquidity Constraints or Myopia?" Canadian Journal of Economics. February 1985,
pp. 117-36; and Robert E. Hall and Frederic S. Mishkin, "The Sensitivity of Consumption to
Transitory Income: Evidence from Panel Data," Econometrica, vol. 50 (1982), pp. 461-81.
10
Scheinkman and Weiss obtain similar results with a general equilibrium model in which
individuals face borrowing constraints and incomplete insurance markets. The model reproduces some
of the qualitative features of real-world business cycles that cannot be generated by models with
complete markets. See Jose A. Scheinkman and Laurence Weiss, "Borrowing Constraints and
Aggregate Economic Activity," Econometrica, January 1986, pp. 23-45.
11
The standard reference on credit rationing is Joseph E. Stiglitz and Andrew Weiss, "Credit
Rationing in Markets with Imperfect Information," American Economic Review, vol. 71, no. 3 (June
1981), pp. 393-410. For a readable discussion of the macroeconomic implications of credit rationing,
see Mark L. Gertler, "Financial Structure and Aggregate Economic Activity: An Overview," Journal of
Money, Credit, and Banking, vol. 20, no. 3 (August 1988), part 2, pp. 559-88.
12
Tobin and Dolde, "Wealth, Liquidity and Consumption," in Consumer Spending and Monetary
Policy: The Linkages.
115
In recent years, a number of authors have documented the existence of liquidity effects in
consumption and investment behavior. In particular, liquidity constraints have been advanced as a
convincing explanation for the empirical failure of the life cycle/permanent income theory of
consumption and the neoclassical theory of investment. Regression results from quarterly data on
aggregate consumption of nondurables and services reported by Campbell and Mankiw indicate that
consumers earning about one-half of disposable income are liquidity constrained.13 Auerbach and
Hasset extend Campbell and Mankiw's results to provide further support for the liquidity constraints
hypothesis.14 Their model, estimated with annual data, indicates that liquidity constraints are
particularly significant in explaining aggregate consumption in periods characterized as credit crunches.
Moreover, Auerbach and Hasset also find that about one-half of households are liquidity constrained
when the economy is in a credit crunch. Zeldes uses panel data on food consumption to compare the
behavior of households holding low levels of liquid assets with those holding high levels of liquid
assets.15 His results show that households with high levels have spending patterns consistent with
unconstrained consumption choices. The consumption behavior of households with low levels of
liquid assets, by contrast, is inconsistent with unconstrained choice but consistent with the existence of
liquidity constraints. Researchers have also had success using proxies such as the unemployment rate
and the nominal interest rate on auto loans to measure the effects of liquidity constraints.16 Similar
proxies also play an important role in the consumption functions of Ray Fair's midsize
macroeconometric model.17 Fair's work establishes that taking account of liquidity effects improves
the predictive power of econometric models. It is not clear, however, that the predictive value of his
13
See John Y. Campbell and N. Gregory Mankiw, "Consumption, Income, and Interest Rates:
Reinterpreting the Time Series Evidence," Princeton University, April 1989, mimeo.
14
See Alan J. Auerbach and Kevin Hasset, "Corporate Saving and Shareholder Consumption,"
University of Pennsylvania, February 1989, mimeo.
15
See Steven P. Zeldes, "Consumption and Liquidity Constraints: An Empirical Investigation," Journal
of Political Economy, vol. 97, no. 2 (April 1989), pp. 305-46.
16
See James A. Wilcox, "Consumption Constraints: The Real Effects of 'Real* Lending Policies"
(Paper presented at the spring 1989 meeting of the Federal Reserve System Committee on Business
Conditions); and Flavin, "Excess Sensitivity of Consumption to Current Income: Liquidity Constraints
or Myopia?"
17
Ray C. Fair, The Specification. Estimation, and Analysis of Macroeconometric Models (Cambridge:
Harvard University Press, 1984).
116
proxy for liquidity constraints is due solely to the existence of such constraints.
Recent work on investment spending parallels the research on liquidity effects and consumption.
The most extensive paper, by Fazzari, Hubbard, and Petersen, explores several financial factors and
their impact on investment spending.18 The authors find that for large mature firms liquidity
variables (cash flow and initial stocks of cash on hand) do not help explain investment spending. But
liquidity does have a significant role in explaining investment spending in the case of medium-size
firms and especially small firms in the same industry. The authors interpret this result as indicating
that financial constraints are binding for a significant segment of the nonfinancial business sector and
thus may have an important role in macroeconomic fluctuations. Similarly, Gertler and Hubbard,
using cross-sectional time-series data, show that cash flow is an important factor determining
investment spending by manufacturing firms.19 In general, the finding that financial factors such as
liquidity or cash flow are important for investment spending by the individual firm seems to be in
accord with an inability to find an important interest rate effect on aggregate investment spending.
The empirical research on liquidity constraints reviewed here has used diverse techniques to
establish the relevance of liquidity effects in a variety of settings. Thus, the existence of liquidity
constraints appears robust, a result which lends credence to the presumption that spending should be
closely related to agents' holdings of liquid assets. But research to date has focused on testing
whether liquidity effects influence spending; thus, it is still too early to judge whether this strand of
research would significantly improve the predictive power of macroeconometric models. Moreover, it
should be emphasized that the existence of liquidity effects on consumption and investment spending
does not imply a stable relationship between broad liquid asset aggregates, however defined, and
aggregate spending in the economy. First, the pool of liquidity-constrained households and firms may
change over time and in response to changes in economic conditions, rendering simple relationships
between aggregate economic activity and any particular liquid asset measure unstable. Second, and
potentially more damaging to the the use of a broad measure of liquid assets in the formulation of
monetary policy, financial innovations that change borrowers' access to credit will bring about
permanent shifts in the relationship between aggregate spending and the liquid asset measure.
18
Steven M. Fazzari, R. Glen Hubbard, and Bruce C. Petersen, "Financing Constraints and Corporate
Investment," Brookings Papers on Economic Activity, 1:1988, pp. 141-206.
19
Mark L. Gertler and R. Glen Hubbard, "Financial Factors in Business Fluctuations," in Financial
Market Volatility, a symposium sponsored by the Federal Reserve Bank of Kansas City, August 1988;
also, First Boston Working Paper Series, no. FB-88-37, September 1988.
117
However, in principle, it may be possible to design a measure with more restricted scope, such as the
liquid assets of low- and middle-income households and small and medium-size firms, that performs
better than economy-wide totals.
The final argument for the use of liquid assets in judging monetary policy actions simply states
that policy makers should diversify their use of policy indicators. Financial variables-the structure of
rates of return and the quantities of assets outstanding—are determined in general equilibrium by the
interaction of supplies of and demands for financial assets. Specific financial assets have special
characteristics tailored to particular financing requirements, but no asset or subset of assets is special
for the interaction of real and nominal quantities.20 In a world of certainty the monetary authorities
can control nominal income by controlling any one financial aggregate and one rate of return.21
However, in an uncertain world there is no reason to presume that any specific financial aggregate or
interest rate would fully capture the implications of financial decisions for real economic activity, or
provide all the information pertinent to monetary policy decisions. On this view, an exclusive focus
on any particular monetary aggregate is inappropriate. Instead, in making decisions on monetary
policy, the authorities could use a very broad measure of the money supply, an aggregate of liquid
assets, along with other financial aggregates and rates of return.
Measures of liquid assets may also have another, very different, value for policy makers. The
holdings of liquid assets by households and firms calculated relative to their income or indebtedness
may give an indication of the economy's ability to withstand economic shocks. Benjamin Friedman,
for example, has interpreted the decline of the household and nonfinancial business sectors9 holdings
of liquid assets relative to total financial assets and net worth as increasing appreciably the threat of
financial instability.22
20
For statements of this viewpoint, see Tobin, "A General Equilibrium Approach to Monetary
Theory"; and James Tobin and Donald Hester, Financial Markets and Economic Activity (New York:
Wiley, 1967).
21
For example, control of any monetary aggregate and the zero rate of return on currency is sufficient.
See James Tobin, "Money and Finance in the Macroeconomic Process," Journal of Money, Credit and
Banking, vol. 14, no. 2 (May 1982), pp. 171-204; and Franco Modigliani and Lucas Papademos, "The
Structure of Financial Markets and the Monetary Mechanism," in Controlling Monetary Aggregates III,
Federal Reserve Bank of Boston, Conference Series no. 23, 1980.
22
Benjamin M. Friedman, "Increasing Indebtedness and Financial Stability in the United States,"
in Debt Financial Stability, and Public Policy. Federal Reserve Bank of Kansas City, 1986, pp. 27-53.
118
B. Liquid Asset Measures Compared with Credit Measures
The main arguments for using liquid asset measures in the formulation of central bank
policy-the instability in the money-income relationship resulting from financial innovation and
deregulation, the existence of liquidity or borrowing constraints on the budgets of individual
households and business firms, and the diversification of indicators from the financial side of the
economy-apply equally well to the use of credit aggregates.23 The ambiguity between liquidity and
credit can be traced back to the earliest thinking on measures of credit and liquid assets, as
exemplified by the "Radcliffe view."24 The Radcliffe Report obscures the distinction between
liquidity and credit and offers an ambiguous account of the channels through which monetary policy
affects the economy. It downplays the role of interest rates in the transmission of monetary policy and
emphasizes the availability of credit-to the point of recommending the use of selective credit controls.
In the appendix we argue that although the Radcliffe Committee used the term "liquidity" to indicate
the broadest possible measure of the availability of financing, committee members may primarily have
been concerned with credit.25
More recently, Modigliani and Papademos have argued that changes in broad monetary aggregates
are correlated with changes in real economic activity not because they change the "liquidity" of the
economy or because some of the assets in the aggregate are endowed with "moneyness" properties
(and hence may be a better measure of true "money" than narrow aggregates) but because they are the
unavoidable accompaniment of credit expansion.26
23
For more details on the case for using credit aggregates in the formulation of monetary policy,
see Lawrence Radecki, "Credit Measures as a Policy Variable," in this volume.
24
See the Report of the Committee on the Working of the Monetary System, Cmnd. #827, Her
Majesty's Stationery Office, August 1959; R S. Sayers, "Monetary Thought and Monetary Policy in
England," Economic Journal. December 1960, pp. 710-24; and Nicholas Kaldor, The Scourge of
Monetarism (New York: Oxford University Press, 1982).
25
The Radcliffe Committee and commentators such as Gurley and Shaw were also concerned with
the monetary policy implications of the substitutability of the liquid liabilities of nonbank financial
intermediaries for those of commercial banks. The current definitions of the monetary aggregates,
which treat similar liabilities of nonbank intermediaries and banks as perfect substitutes, reflect their
thinking.
26
Modigliani and Papademos, "The Structure of Financial Markets and the Monetary Mechanism,"
in Controlling Monetary Aggregates HI. Recently, Joseph Stiglitz has taken a more extreme position,
arguing that money affects real economic activity not because it is used as a transactions medium but
(continued...)
119
In a pair of articles that reflected the conceptual and empirical overlap of the credit and liquid
asset aggregates, Frank Morris, the former president of the Federal Reserve Bank of Boston, suggested
in 1982 and 1983 that the Federal Open Market Committee (FOMC) should move away from money
supply targets and adopt either a liquid asset aggregate or a total credit aggregate.27 He had a weak
preference for a liquid asset measure because the collection of credit data presented more of a
problem.
The fact that capital market imperfections underlie both liquidity and credit effects on aggregate
demand does not preclude liquid asset measures and credit aggregates from having different
information. The investment properties of liquid assets do provide some independent rationale for the
use of liquid asset measures for the formulation of policy. In fact, Pissarides' analysis shows why
consumers' and firms' portfolios should provide information independent of that contained in, for
example, credit measures.28
II. The Appropriate Definition of a Liquid Asset Measure
The appropriate definition of any monetary aggregate involves two potentially controversial issues:
Which assets should be included in the aggregate? What weights should be given to the various assets
in compiling the aggregate? The choice of assets determines whether the aggregate corresponds to the
theoretical rationale for its use in the formulation of policy. The weights given to the constituent
assets in an aggregate should depend on their degree of substitutability; for example, giving equal
weights to the components of an aggregate requires in principle that the assets be perfect substitutes, a
condition that is particularly stringent in the case of high-level aggregates such as L or M3.
In this section we discuss alternative definitions of liquid asset aggregates that have been proposed,
especially the Federal Reserve System's current measure L and the Divisia (or monetary-services)
index of total liquid assets. This second measure was constructed by the staff of the Board of
^(...continued)
because the creation of money is the counterpart of the creation of bank credit. See Joseph E. Stiglitz,
"Money, Credit and Business Fluctuations," National Bureau of Economic Research, Working Paper
no. 2823, January 1989.
27
Frank E. Morris, "Do the Monetary Aggregates Have a Future as Targets of Federal Reserve
Policy?" New England Economic Review, March-April 1982, pp. 5-14; and "Monetarism without
Money," New England Economic Review, March-April 1983, pp. 5-9.
28
Pissarides, "Liquidity Considerations in the Theory of Consumption."
120
Governors as part of its effort to apply well-founded aggregation procedures to broad measures of the
money supply.
An issue closely related to the selection of assets for a liquid asset aggregate is the proper
treatment of prearranged lines of credit. If a liquid asset aggregate is a measure of the dollar volume
of assets that serve as a "temporary abode of purchasing power," credit lines are not relevant. But if
an aggregate of liquid assets is intended to be a measure of readily available spending power in the
economy, then established lines of credit might properly be included. One quickly sees the logic of
including them, but it is unclear how they should be incorporated.
A. The Composition of a Liquid Asset Measure
The definitions and magnitudes of the major financial aggregates reported by the Federal Reserve
System are shown in Table 1. The liquid asset aggregate, L, is defined to include the assets in the M3
measure of the money supply plus the nonbank public's holdings of U.S. savings bonds, short-term
Treasury securities,29 commercial paper, and bankers' acceptances (net of money market mutual fund
holdings of these assets). Several other assets might be included, but for various reasons they are not.
For instance, some assets can be sold quickly but at a relatively high transactions cost or significant
price risk.30 A no-load mutual fund holding long-term bonds reduces the transactions costs, but its
combined market value still fluctuates. Savings held in an employer-sponsored thrift plan may have
stable value and could be liquidated with no transactions costs, but the number and timing of
withdrawals are typically quite limited and one's funds are only available after a substantial delay.
A liquid asset aggregate, however defined, must have considerable overlap with two other highlevel financial aggregates, M3 and Total Nonfinancial Sector Debt, as is evident in Table 1. The
dollar value of L, as it is currently defined, is six times as large as that of Ml, one and a half times as
large as that of M2, and exceeds that of M3 only by about 20 percent. In contrast, Total Nonfinancial
Sector Debt is about twice as large as L. One cannot help but question whether augmenting M3 by
this modest amount provides any substantial additional information about developments on the
financial side of the economy. It is the case, however, that the growth rates of M3 and L do diverge
29
This category comprises marketable Treasury obligations with less than eighteen months to
maturity.
30
The price of a bond with a long term to maturity fluctuates markedly.
121
Table 1
Money Stock, Liquid Assets, and Debt Measures
(Billions of Dollars)
Component
Currency
Travelers' checks
Demand deposits
Other checkable deposits
Ml
August 1989 Total
218.4
7.1
277.5
274.4
777.4
Overnight RPs
Overnight Eurodollar deposits
Money market deposit accounts
Savings and small time deposits
Money market mutual funds
M2
59.7
15.2
465.4
1534.2
285.5
3136.5
Large denomination time deposits
Term RPs
Term Eurodollar deposits
Institutional MMMFs
M3
569.0
105.3
97.6
100.6
4009.0
Public's holdings of U.S. savings bonds
Short-term Treasury securities
Commercial paper
Bankers' acceptances
L
114.7
286.4
353.1
42.7
4812.7
U.S. government debt
Private nonfinancial sector debt
Total Nonfinancial Sector Debt
122
2199.9
7395.0
9558.9
at times; L may grow more rapidly than M3 when the public is holding a larger proportion of its
liquid assets as directly owned market instruments. But what is gained by knowing when and the
extent to which this is happening is unclear.
Before adopting the current definitions of the monetary aggregates in February 1980, the Board of
Governors considered greater differentiation between M3 and L. According to a September 1979
memorandum that amended the definitions proposed in January of that year, M3 was to include small
time deposits but not large time deposits, term repurchase agreements, or Eurodollar deposits.31
Under this scheme L would have been considerably larger than M3. Subsequent revisions to the
proposed definition of M3 reduced the degree of differentiation.
Although the quantitative difference between L and Total Nonfinancial Sector Debt is larger, the
two aggregates still overlap in that they both include savings bonds, Treasury securities, commercial
paper, and bankers' acceptances. These assets represent 20 percent of L and 9 percent of Total Debt.
If the long-term component of Total Debt (mortgages and Treasury, corporate, state, and municipal
bonds) were removed, the overlap between the remainder, total short-term debt, and L would
obviously be much greater and the aggregates would be considerably closer in size.
Over the past thirty years, L has grown at an average rate of 9.0 percent, not quite as rapidly as
M3 and Total Debt, which averaged growth rates of 9.3 percent and 9.4 percent, respectively. (The
annual growth rates of the monetary and credit aggregates are plotted in Chart 1.) L's annual growth
rates are a little more stable than the growth rates of the other monetary aggregates and a little less
stable than those of Total Debt when variability is measured by the standard deviation. But L's
growth rate is noticeably more stable than the other monetary aggregates', but still less stable than that
of Total Debt when variability is measured by the absolute percentage change from one year to the
next. The growth rate of L is highly correlated with that of M3 (the simple correlation coefficient is
0.87); indeed, no other pair of financial aggregates has a higher correlation over the 1960-88 period.
(See Charts 2 through 5 and Tables 2 and 3.) Nevertheless, the data reject at the 1 percent level the
hypothesis that the correlation between M3 and L growth rates is perfect.32 The correlation between
the growth rates of L and Total Debt over the past thirty years, when quarterly data are used, has been
31
The contents of the memorandum are discussed in Neil G. Berkman, "Abandoning Monetary
Aggregates," in Controlling Monetary Aggregates in, Federal Reserve Bank of Boston, Conference
Series no. 23, pp. 76-100.
32
However, as noted above, it is not clear whether differences in the growth rates of L and M3
convey information that is of potential use to policy makers.
123
Chart 1
Money Stock, Liquid Assets, and Debt
Change from Four Quarters Earlier
I t •• I •! I I I I I I I I I I I •• I 1 I I I 1 I • I •• I I I 1 I I I I • I 1 1 I I I I I I I I I I I H I I I I I I I I I I I 1 I I I I I I I 1 I 1 I I I I I I I I I I I I I I I I I I I I I I I I t I I I • I I • I I I I I I I
1960 1962 1964 1966 1968 1970 1972 1974 1976 1978 1980 1982 1984 1986 1988
0.62--rather high compared with the other correlations found.
B. The Application of Aggregation Theory
The money supply has conventionally been measured by calculating the simple sum of the dollar
volume in each component (currency, demand deposits, savings account deposits, and so forth).
Economic theory points out, however, that a necessary and sufficient condition for justifying the use of
a simple sum as the method of aggregation is that all the components be perfect substitutes for one
another.33 The broader an aggregate, the less substitutable are its components and hence the less
suitable is the simple sum method.
The current definition of L, as well as virtually any conceivable liquid assets aggregate, can be
33
See William A. Barnett, "Economic Monetary Aggregation: An Application of Aggregation and
Index Number Theory," Journal of Econometrics, 1980, pp. 11-48, for a formal discussion of
assumptions required for the validity of different aggregation schemes.
124
Chart 2
Land Ml
Change from Four Quarters Earlier
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
1960 1962
• 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
1964 1966 1968
• 1 1 1 1 1 1 1 1
• 1 1 1 1 1 1 1 1 1
•
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
1970 1972 1974 1976 1978 1980 1982 1984 1986 1988
Chart 3
L and M2
Change from Four Quarters Earlier
_5
—J
I I H I M I I I M I I I I I n I I I It I I I I I I I i l M l I I it I I I I l l I I I I I I l l I I l i i i l t i i i i n l . i . I i i . I I I I I I I I I I I i l i i i l i i i i . n l
1960 1962 1964 1966 1968 1970 1972 1974 1976 1978 1980
t.il,.il.,il,..l
1982 1984 1986 1988
Chart 4
L and M3
Change from Four Quarters Earlier
Percent
20
15 H
10 H
5H
.5
1 1111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111 M 1111111M
1960
1962
1964
1966
1968
1970
1972
1974
1976
1978
1980
1982
1984
1986
1988
Chart 5
L and Nonfinancial Sector Debt
Change from Four Quarters Earlier
Percent
20
Nonfinancial sector debt
. \ :
15 H
:
* t :
:/••-.«»
;..»ij»*;.
10 -\
l
:'» ^
.* :
:
- •* :
71
126
I I • I I• ^ • I • I I I 1 I I I I • I I • I t I 1 I 1 I I I I I I I • I I I • I I I I I t I I I I I I I I I I I I I I I I I I I I 1 I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I H I I • I I I I • I I I I I I I I I
1960
1962
1964
1966
1968
1970
1972
1974
1976
1978
1980
1982
1984
1986
1988
Table 2
Major Financial Aggregates: Descriptive Statistics
Relative Sizes of Financial Aggregates (As of June 1989)
Dollar Volumes
On Billions)
Size in Relation to L
(In Percent)
Ml
M2
M3
L
770.8
3090.0
3975.0
4739.3
16.3%
65.2
83.9
100.0
Total Nonfinancial
Sector Debt
9339.5
197.1
Annual Growth Rates (In Percent; 1960-88)
Mean
Standard
Deviation
Minimum
Mean
Absolute
Change^
Maximum
Ml
M2
M3
L
6.2
8.4
9.3
9.0
3.1
2.6
2.8
2.5
0.4
4.2
1.9
3.6
15.6
13.5
14.7
12.8
2.2
2.3
2.2
1.8
Total Nonfinancial
Sector Debt
9.4
2.4
5.3
13.9
1.2
Simple Correlations among the Annual Growth Rates of the Financial Aggregates (1960-88)
Ml
M2
M3
Ml
M2
M3
L
1.00
0.44
0.35
0.44
1.00
0.78
0.65
1.00
0.87
1.00
Total Nonfinancial
Sector Debt
0.71
0.34
0.43
0.69
_L_
Total Nonfinancial
Rector Debt
1.00
Note: Annual growth rates are measured from fourth quarter to fourth quarter.
- This is the average change in the annual growth rate from one year to the next regardless of sign.
127
Table 3
Correlations among the Monetary and Credit Aggregates
1960-H to
1989-1
Monetary base
Ml
M2
M3
L
Total Nonfinancial
Sector Debt
1960-nto
1974-m
Monetary
Base
Total Nonfinancial
Sector Debt
Ml
M2
M3
1.00
0.67
0.36
0.41
0.57
1.00
0.54
0.40
0.43
1.00
0.75
0.61
1.00
0.83
1.00
0.70
0.48
0.20
0.35
0.62
1.00
Ml
M2
M3
L
Total Nonfinancial
Sector Debt
Monetary
Base
Monetary base
Ml
M2
M3
L
Total Nonfinancial
Sector Debt
1.00
0.74
0.38
0.52
0.74
1.00
0.69
0.63
0.74
1.00
0.87
0.71
1.00
0.84
1.00
0.79
0.56
0.37
0.49
0.78
1974-IV to
1989-1
Monetary
Base
Ml
M2
M3
L
1.00
0.63
0.29
0.20
0.15
1.00
0.46
0.30
0.17
0.68
0.49
1.00
0.88
1.00
0.35
0.24
-0.06
0.27
0.41
Monetary base
Ml
M2
M3
L
Total Nonfinancial
Sector Debt
128
1.00
Total Nonfinancial
Sector Debt
1.00
criticized on the grounds that it violates sound aggregation principles.34 It has been argued that the
components of L are not uniformly close substitutes because they serve different purposes for different
groups in the economy. For example, repurchase agreements and overnight Eurodollar deposits may
be close substitutes for large firms but not for households or small firms. Thus, shifts in the
composition of L will tend to change its relationships with important macroeconomic variables. In this
regard, economic research on the elasticities of substitution between monetary assets heavily supports
the conclusion that substitutability between many of the components of the monetary aggregates is
low.35 In sum, it has been argued, the standard method of measuring may be seriously flawed when
applied to broad measures of the money supply such as L. This criticism applies equally well to M3
and M2, although it is potentially more damaging for L.
In response, economists on the staff of the Board of Governors have tried to apply index number
and aggregation theory rigorously to the measurement of the money supply. One specific approach
they have taken has been to construct a monetary index based on money defined as the medium of
exchange. This transactions-money-stock index is an alternative to the conventionally measured Ml
and M2, the narrower definitions of the money supply.36 A second approach has been to construct an
index based on money defined as providing a broad range of services, including liquidity and store of
value as well as means of payment.37 This monetary-services (or Divisia) index for a particular
aggregate is meant to capture the value of all services provided by the monetary assets in the
aggregate. The Divisia aggregates are intended primarily as alternatives to M3 and L, because the
assumptions required for simple-sum aggregation of these broad measures of the money supply are
34
Berkman, "Abandoning Monetary Aggregates."
35
Barnett formally tested the suitability of the simple-sum approach and rejected it; see Barnett,
"Economic Monetary Aggregation." An alternative, nonparametric test is reported by Michael T.
Belongia and James A. Chalfant, "The Changing Empirical Definition of Money: Some Estimates from
a Model of the Demand for Money Substitutes," Journal of Political Economy, vol. 97, no.2, pp. 38797.
36
Paul A. Spindt, "Money Is What Money Does: Monetary Aggregation and the Equation of
Exchange," Journal of Political Economy, February 1985, pp. 175-204.
37
The monetary services index is based on Friedman's restatement of the quantity theory of
money, which defines money as providing a broad range of services rather than as transactions
medium only. See Milton Friedman, "The Quantity Theory of Money: A Restatement," in Friedman,
ed., Studies in the Quantity Theory of Money (Chicago: University of Chicago Press, 1956).
129
unlikely to be satisfied in practice.38
A monetary services index of L is constructed by giving a weight to each component of L equal to
the value of monetary services provided per dollar by that component.39 In turn, the monetary
services provided per dollar by each component should be equal to the opportunity cost of holding the
monetary asset instead of a higher yielding asset that provides absolutely no monetary services, such
as a corporate bond. That is, the monetary services gained per dollar equals the difference between
the expected total return from nonmonetary assets and the expected total return on that particular
monetary asset, both calculated over the same holding period. In practice, opportunity cost has been
measured by the difference between the market yield of a benchmark asset, such as the Baa rated
corporate bond rate, and the actual or realized yield of the monetary asset.
In general, the growth rate of a monetary aggregate, to a first approximation, is equal to the
weighted average of the growth rates of its components. For a conventionally computed measure of
the money supply, the weight given to the growth rate of each component equals its dollar share of the
total. For a monetary services index, the weight given to the growth rate of each component is
approximately equal to its share of total monetary services provided, which in turn is proportional to
the product of its dollar volume and its opportunity cost.40 Hence, the growth rate of NOW account
deposits is given less weight than money market mutual fund shares in calculating the growth of the
simple sum measure of L since the dollar volume of deposits in NOW accounts is smaller ($273
billion for NOW deposits and $374 billion for money fund shares, as of July 1989). But NOW
account deposits currently have a significantly higher opportunity cost than do money market mutual
fund shares, and hence the growth rate of NOW account deposits would receive more weight in the
monetary services index of L than would the growth rate of money market mutual fund shares.
A monetary services index of L (as for the other aggregates) has proven tricky to develop in
practice. The basic problem is the correct measurement of the opportunity cost of the components of
L. As mentioned above, the opportunity cost of holding a particular monetary asset, as well as the
monetary services gained, is the difference between the expected total return from nonmonetary assets
38
William A. Barnett, Edward A. Offenbacher, and Paul A. Spindt, "New Concepts of Aggregated
Money," Journal of Finance, May 1981, pp. 497-506.
39
The practice has been to use chain index numbers, not weights fixed at a particular year.
40
Strictly speaking, the weight of asset i is given by s{ = (R-r^V/tXCR-rpVj}, where R is the
expected return on the benchmark asset, r{ is the expected return on asset i, and W{ is the dollar value
of the i-th asset in circulation.
130
(capital gains plus interest payments) and the expected total return of the monetary asset, both
measured over the same holding period. In practice, it has been measured by the difference between
the actual yields. Hence, a serious problem develops when the yield curve inverts such that the bond
rate is below money market rates and the measured opportunity cost of components of L is negative.
As a quick fix when this occurs, the highest rate posted by any of the components of L is used instead
of the corporate bond rate so that the opportunity cost of the other components and the value of
monetary services they provide are measured as positive.41 But this is unsatisfactory and merely
highlights the fundamental problem in trying to compute monetary-services indexes when reliable data
on expected yields on nonmonetary assets are not available.
The experience with the monetary-services measure ofL
From the first quarter of 1970 to the first quarter of 1988, the last full quarter for which Divisia L
was calculated, the average annual growth rate of conventional L was 10.2 percent, compared with 7.3
percent for the monetary-services measure of L. Consequently, while the velocity of conventional L
has declined about 1 percent per year, the velocity of monetary-services L has risen at an average rate
of between 1.5 and 2 percent. Almost all of the velocity growth of the monetary-services index
occurred in the four years 1978 to 1981.
As discussed in greater detail below, the Divisia index measures for L and the lower level
aggregates did not live up to their theoretical promise, presumably because of the measurement
problems cited. As a result, the staff of the Board of Governors abandoned this theoretically attractive
approach to the compiling of monetary aggregates in May 1988.
C. Measurement Issues: Lines of Credit
We observed earlier that at least part of the rationale for focusing on a measure of liquid assets is
the notion that the amount of liquidity, or readily accessible purchasing power, held by households and
firms at least indirectly limits their spending. Households must hold liquid assets in reserve against
the possibility of unemployment or an unexpected expense; business firms against the possibility of an
unexpected expense or disappointing sales. If liquid reserves are not held in desired amounts,
households and firms will presumably cut back their spending on goods, services, and inventories to
41
The opportunity cost and hence the monetary services provided by the component of L serving
as the benchmark asset will be zero in this situation—a result which is wrong but cannot be helped.
131
build up their liquidity. The dollar volume of liquid assets held by households and firms would seem,
however, to be an inadequate measure of the amount of liquidity at their disposal. Many households
and firms add to the readily available purchasing power at their disposal by establishing a line of
credit, usually from a commercial bank,. The widespread use of prearranged lines of credit
documented below not only indicates that liquid asset aggregates mismeasure the amount of liquidity
in the economy, but also casts some doubt on the importance of liquidity constraints as a justification
for using liquid asset aggregates in the formulation of monetary policy.
For business firms, the line of credit is usually provided by a commercial bank to finance
inventories of materials and goods at the various stages of production. For households, the line of
credit may be established to finance a major expense, such as college tuition or an addition to a house,
or to consolidate small debts. Such credit lines can take several forms:
(i) unsecured, provided by a bank, and intended to give overdraft protection for the customer's
checking account;
(ii) unsecured and accessible through a credit card or other means; or
(iii) secured by home equity.
The unused portion of a household's or firm's line of credit would seem to be a near-perfect substitute
for holdings of liquid assets. That is, if a household's credit line were increased by $1000 and
simultaneously its liquid assets were reduced by the same amount (and its total net wealth held
constant), one would not expect the household's consumption behavior to be affected very much. In
this case, the household's available liquidity is the sum of liquid assets it does hold and unused
amounts of its credit lines.42 Similarly, a firm with relatively little cash on hand is not truly
constrained if it also has large unused amounts on its prearranged lines of credit.
Little can be done about this measurement problem at present. The data available on credit lines
is limited, and certainly far too modest to establish whether and how aggregate measures of liquid
assets should be adjusted for unused credit lines. Nevertheless, some information is available which
gives a rough indication of the size of the measurement problem.
The Survey of Currency and Transactions Account Usage (1986) provides information on the
42
The individual's available liquidity would be close to, if not exactly equal to, the sum of liquid
assets held and unused lines of credit, because unused lines of credit and liquid assets are unlikely to
be perfect substitutes.
132
amount of credit available to households through their credit cards.43 From the survey results, we
would estimate that at the time of the survey each American family held an average of four credit
cards.44 If each card has an average unused credit line of $500, then there is $2000 of credit
available from credit cards per family. Using an estimate of 125 million American families, we
calculate that there is in the aggregate about $250 billion of unused credit card lines outstanding. This
figure compares with $128 billion of consumer installment debt outstanding at the end of 1986, and
with L measured at $4112 billion as of November 1986.
Open-end revolving lines of credit secured by home equity are also held by a significant number
of American households. Another recent survey (1988) indicated that 5.6 percent of homeowners held
(or had applied for at the time of the survey) a home equity line of credit.45 Since about 60 percent
of American families owned their own homes, about 3 percent of all families held (or had applied for)
a home equity line of credit. The median amount of the line is around $31,250; the median proportion
used is only 32 percent. From reports obtained from banks and other financial intermediaries, Canner,
Luckett, and Durkin estimate the total debt outstanding on home equity lines of credit at $74 billion as
of the end of December 1988. Taking the survey's finding that typically about one-third of the
available credit line was used, we calculate that the total amount of credit still available from unused
home equity lines would be around $150 billion-and growing rapidly.
In comparison with the estimated $400 billion of unused credit lines available to the household
sector ($250 billion from credit cards and $150 billion from home equity lines), the business sector has
about $625 billion of unused formalized loan commitments from insured commercial banks (as of
December 1988). The data, collected from banks on the quarterly Reports of Conditions and Income,
show that unused formalized commitments grew at a 9.2 percent annual rate from 1983 to 1988.46 L
43
R B. Avery, George E. Elliehausen, Arthur B. Kennickell, and Paul A. Spindt, "Changes in the
Use of Transactions Accounts and Cash from 1984 to 1986," Federal Reserve Bulletin, March 1987,
pp. 179-96.
44
The survey found that 71 percent of all families held at least one credit card, and among these
families the mean number of cards per family was six.
45
George C. Canner, Charles A. Luckett, and Thomas A. Durkin, "Home Equity Lending," Federal
Reserve Bulletin, May 1989, pp. 333-44; and George C. Canner, Thomas A. Durkin, and Charles A.
Luckett, "Recent Developments in the Home Equity Loan Market," Journal of Retail Banking, Summer
1989, pp. 35-47.
46
The Federal Reserve System also collected data on unused formalized loan commitments from
112 large commercial banks beginning in July 1973, but the survey was discontinued in June 1987.
133
grew at a similar annual rate, 8.6 percent over the same period. The data also show that unused loan
commitments increased more rapidly than commercial and industrial loans outstanding, which grew at
an annual rate of 3.4 percent over the same period.
Combining the household and business sectors yields roughly $1 trillion of unused prearranged
credit, compared with a volume of L on the order of $4 trillion. It is clear that the amount of
available but unused credit is large enough to affect the relationship between the dollar volume of L
and economic activity. Furthermore, unused credit lines may also alter the length of the lag between
changes in Federal Reserve policy instruments and the effect on the volume of credit outstanding. If
households and firms have prearranged credit lines, a tightening of Federal Reserve policy may induce
firms and households to make initial use of their lines of credit as sales or personal income falter, and
only sometime later would the demand for credit slow or contract.
The widespread use of lines of credit shows that a significant fraction of households and firms are
not liquidity-constrained. This raises the question whether liquidity effects are sufficiently pervasive
for measures of liquid assets to be of practical use in the formulation of policy. The existence of
unused lines of credit will obviously weaken liquidity effects, but it is not sufficient to rule them out.
As noted earlier, liquidity effects in consumption and investment may arise even if households and
firms are not up against their liquidity constraints. The possibility of future rationing can be sufficient
to induce current liquidity effects. The data on unused lines of credit also imply that a large share of
firms and households, presumably those without collateral, do not necessarily have access to lines of
credit. Thus, we can conclude that very broad liquid asset aggregates may not be very good measures
of liquidity effects because they include liquid asset holdings by households and firms that are unlikely
to be constrained in financial markets. It may be possible to define narrower liquid asset measures
that correspond more closely to the theoretical rationale for their use, or to combine the amount of
unused credit lines with the dollar volume of liquid assets to form a more accurate measure of
liquidity. But how such measures could be constructed is an insoluble problem given the limited data
we have at present. In the final analysis, whether broad liquid asset measures as currently defined are
useful for the formulation of monetary policy can only be settled on empirical grounds. We review
the empirical evidence in a later section.
D. Summary
In this section we have reviewed issues related to the measurement of a liquid asset aggregate.
The Federal Reserve System's broadest liquid asset measure, L, is only 20 percent larger than and so
134
highly correlated with M3 that it is not clear that it would provide significant additional information to
policy makers. Simple sum monetary aggregates from Ml to L are subject to the criticism that their
constituent assets are imperfect substitutes. Thus, on a priori grounds, one would not expect them to
have stable relationships with macroeconomic variables of interest to policy makers. Nevertheless,
efforts to apply aggregation techniques based on index number theory to the monetary aggregates were
abandoned in 1988 because the Divisia aggregates, after considerable initial promise, proved to be no
more reliable than the simple sum aggregates in the 1980s.
We estimated that unused prearranged credit lines of households and nonfinancial firms represent
about one-quarter of L. The widespread availability of lines of credit indicates that L, or for that
matter M3 or M2, may not capture liquidity effects in consumption and investment spending very
well.
ID. An Empirical Assessment of Liquid Asset Measures as Intermediate Targets
There are two crucial requirements for the use of a financial aggregate as an intermediate target:
(i) stability and predictability of the relationship between the intermediate target and
macroeconomic goal variables such as the inflation rate or unemployment, and
(ii) stability and predictability of the relationship between policy instruments and the intermediate
target.
In addition, data needed to construct the aggregate should be available promptly so as to allow timely
adjustment of policy instruments in light of macroeconomic developments. The longer the delay in the
availability of data, the more important the predictability and stability of these relationships become.
In this section we review the empirical evidence relevant to the use of a liquid asset aggregate as
an intermediate target. We summarize earlier studies and also report the results of empirical work that
we undertook to update the earlier studies. We find that the relationship between L and the monetary
base is marginally more stable and predictable than that between M2 and the monetary base. Our
results indicate that there is a stable long-run demand for L, suggesting that L could be used as a longrun nominal anchor for policy. The dynamic process whereby L adjusts to its long-run equilibrium
value is slow, however, and is also subject to considerable uncertainty, as evidenced by poor out-ofsample simulation results and a structural break associated with the financial innovations and
deregulation of the early eighties. Thus, we conclude that targeting L would not be a viable strategy.
The slow adjustment of actual to desired holdings of liquid assets implies that policy makers may have
to tolerate large and sustained deviations of L from its target path. Furthermore, setting target paths
135
for L over short- and medium-term horizons would be difficult given the risks in forecasting the
demand for L and the delays involved in collecting data on some components of L.47
A. L and Policy Objectives
It has become standard practice to evaluate the stability of the relationships between monetary
aggregates and goal variables in one or more of three ways:
(i) the analysis of the behavior of the aggregate's income velocity using summary
statistics such as growth rates and deviations from trend,
(ii) the statistical stability of estimated demand equations for the aggregates, and
(iii) reduced-form regressions, either of the form popularized by Anderson and
Jordan48 or vector autoregressions, a simple generalization of the Anderson-Jordan
reduced-form methodology introduced into monetary analysis by Sims.49
In our analysis of the potential of L as an intermediate target we concentrate on velocity analysis and
the stability of an estimated demand equation. The use of reduced-form regression results to draw
policy conclusions—for example, conclusions about the potential value of a particular monetary
aggregate as an intermediate target—is subject to two criticisms. First, the results can be interpreted in
conflicting ways on the basis of different structural models and, second, the estimated relationships
cannot be exploited by policy because the parameters are not independent of the policies in place.50
But vector autoregressions are a convenient means of summarizing the predictive content of economic
47
The data lag is about one month, primarily because of the delay in obtaining data on offshore
deposits.
48
See Leonall C. Anderson and James L. Jordan, "Monetary and Fiscal Actions: A Test of Their
Relative Importance in Economic Stabilization," Federal Reserve Bank of St. Louis Review, vol. 50
(1968), pp. 1-24.
49
Christopher A. Sims, "Money, Income and Causality," American Economic Review, vol. 62
(1972), pp. 540-52.
50
See Albert Ando and Franco Modigliani, "The Relative Stability of Monetary Velocity and the
Investment Multiplier," American Economic Review, September 1965, pp. 693-728, for a statement of
the first point; and Robert E. Lucas, Jr., "Econometric Policy Evaluation: A Critique," in Karl Brunner
and Alan Meltzer, eds., The Phillips Curve and Labor Markets, for the classic statement of the second.
Thomas F. Cooley and Stephen Leroy, "Atheoretical Macroeconometrics: A Critique," Journal of
Monetary Economics, vol. 16, pp. 283-308, discuss the limitations of vector autoregression
methodology.
136
variables. Therefore, vector autoregression results are presented later when we discuss the use of L as
an indicator. We should note, however, that the use of estimated demand equations to derive
conclusions about intermediate targeting is subject to the Lucas critique-that the demand equation is
not a true structural form and is therefore unlikely to be invariant to changes in policy regime.51
Thus, our results are merely suggestive and do not represent firm conclusions.
The behavior of velocity and the demand for L
The relationship between monetary aggregates and nominal GNP over the medium-term horizon
remained relatively stable from the end of World War II until 1980, with the GNP velocity of Ml
rising steadily at a rate of about 3 percent per year and that of M2 declining at about 0.1 percent. The
velocity of L showed a tendency to contract slowly (less than 1 percent per year), similar to the trend
of the other broad financial aggregates (see Table 4). The velocities of L and Total Debt tended to be
the most stable.
During the 1970s, debates over the advisability of monetary targeting were largely concerned with
the behavior of velocity (or the demands for different monetary aggregates) over higher-than-businesscycle frequencies. Short-run instabilities in the money demand function became more pronounced
during the mid-1970s, prompting the redefinition of the monetary aggregates in 1980. However, in
retrospect, these instabilities appear minor compared with the dramatic breakdown, from 1980
onwards, of the long-run relationship between monetary aggregates and income. Structural change in
financial markets, brought about by financial innovation and deregulation in the early eighties, has
reversed the previous steady rise in the velocities of the financial aggregates (Charts 6 and 7). The
most dramatic decline in the velocities of the major financial aggregates occurred from 1982 to 1987.
During this period, Ml velocity declined at an annual rate exceeding 2.4 percent, while the velocities
of M2 and L declined by about 1.7 percent and 1.4 percent, respectively. The trend towards lower
velocities was reversed in 1988; all the aggregates posted modest increases thereafter.52
Table 5 reports velocity trends and deviations from trend for L and M2 from the first quarter
51
Lucas, "Econometric Policy Evaluation."
52
The annual rates of change of velocity for the major aggregates from 1981-IV to 1987-IV were
as follows: -2.44 percent for Ml, -1.44 percent for M2, -1.67 percent for M3, and -1.91 percent for L.
From 1987-IV to 1989-IH the annual percentage changes were 4.43 percent for Ml, 1.99 percent for
M2, 2.09 percent for M3, and 1.11 percent for L.
137
Table 4
Velocity Growth Rates: Major Financial Aggregates
Annual Growth Rates: 1960-88
(In Percent)
Standard
Deviation
Mean Ab
Change*'
Aggregate
Mean
Ml
2.0
3.5
-9.4
6.2
3.2
M2
-0.1
2.8
-5.4
6.3
3.1
M3
-1.0
2.6
-6.1
5.3
3.0
L
-0.6
2.1
-6.3
2.7
2.6
Total Nonfinancial Sector Debt
-1.0
2.4
-6.8
3.1
2.3
Minimum Maximum
Note: Annual growth rates are measured from fourth quarter to fourth quarter.
- This is the average change in the annual growth rate from one year to the next regardless of sign.
of 1959 to the first quarter of 1989. The results were obtained by fitting time trends, both with and
without correction for serial correlation, to quarterly data. Average deviations from trend were
calculated by expressing the standard error of the regression as an annualized percentage growth
rate.53 The table shows that the velocity of L declined at an annual rate of about 1.5 percent since
1979. Although this rate of decline is substantially higher than the average annual rate of decline of
0.6 percent over the whole period, the difference is not statistically significant. M2 velocity, by
contrast, has not exhibited a statistically significant trend over the 1959-89 period or, more
surprisingly, from 1979 to 1989. When serial correlation is taken into account, deviations from trend
velocity are smaller for L than for M2 over the sample, despite the rapid contraction of L velocity
during the eighties and the absence of such a tendency in M2 velocity. But we should not give too
much weight to this result, because the available evidence suggests that deviations from M2 trend
velocity are eliminated over time while those for L are not. The Said-Dickey test statistics (S-D) in
53
Thus, the deviation calculated from a regression with correction for serial correlation is a
measure of the variability of the uncertain component of velocity-the component that cannot be
forecast by a time trend and serial correlation. It is naturally much smaller than the deviation
calculated from a trend regression without serial correlation correction.
138
Chart 6
L Velocity
1.35
1.3
h*
I
1.25
v
H
h
1.2
H
k
1.15
h\ J \
L
1.1
''• GNP
L
\
H-
L
1.05
I
I
I
I
1959
I
I
1962
I
I
I
1965
I
I
I
1968
1 __l
I I
I
1971
1974
L_J
1 L J
1977
1
1980
1
1
L
1983
L
J
1
1986
Note: Quarterly observations for L are averages of monthly data.
Chart7
M2 Velocity
1.85
h
1.8
h1.75
i
1.7
h-
i-
I
1.65
1.6
h\
\
1.55
/ GNP
/ M2
h-
I
1.5
1959
I
I
Ii
1962
i
i
i
1965
i
i
i
1968
i
i
i
1971
i
i
i— . 1 — 1
1974
Note: Quarterly observations for M2 are averages of monthly data.
1—J
1977
L
._!,., J
1980
1 1 1
1983
1986
i
i
1
139
Table 5
L and M2 Velocities: Descriptive Statistics
Period
Deviations from Trend
Maximum
Minimum
Trend
Average
1959-1 to
1989-1
-0.60**
-0.58s**
14.45
3.87s
31.88
13.19s
-26.54
-9.32s
1959-1 to
1969-1
-0.19*
-0.16s
6.25
3.56s
12.31
9.63s
-12.76
-6.63s
1969-H to
1979-1
-0.54**
-0.44s*
6.06
3.73s
11.34
11.34s
-11.11
-7.07s
1979-H to
1989-1
-1.85**
-1.44s**
9.12
4.08s
22.29
8.47s
-12.15
-8.59s
SD
-1.44
L Velocity
M2 Velocity
1959-1 to
1989-1
0.01
-0.01s
12.86
4.61s
41.36
15.31s
-23.49
-12.46s
1959-1 to
1969-1
-0.39**
-0.30s
8.17
3.86s
16.97
10.21s
-15.18
-8.07s
1969-H to
1979-1
0.01
0.26SS
9.27
4.71s
20.54
13.93s
-13.80
-7.98s
1979-H to
1989-1
-1.14**
-0.65s
10.52
5.33s
26.49
14.04s
-17.51
-12.66s
SD
-3.11*
Notes: Measures are expressed as annual percentage changes.
Statistics are based on trend regressions.
A superscript "s" indicates thatfirst-orderserial correlation correction was used.
SD is the Said-Dickey statistic to test for a unit root.
** indicates significance at the 1 percent level.
140
Table 5 confirm this pattern. The hypothesis that velocity is a random walk with drift can be rejected
at the 5 percent level for M2 but not for L.54 However, as shown in Charts 8 and 9, movements in
the velocities of L and M2 relative to trend correlate closely with movements in interest rates on
alternative assets, so that this conclusion may be reversed when the demands for L and M2 are
modeled more explicitly.55
The demand for liquid assets
Earlier studies. When the monetary aggregates were redefined in 1981, the Board staff economists
did a comprehensive study of the econometric properties of the new and old aggregates.56 To
analyze the broad monetary aggregates, a portfolio model was used to specify the public's demands for
M2, M3, and L. (An inventory model was used to specify the demand for Ml.) In the demand
equation for L, the share of nominal household net worth held in L was determined by the three-month
Treasury bill rate, the commercial bank passbook rate, the maximum rate on commercial bank time
deposits, the ratio of nominal GNP to household net worth, and the share of household net worth held
in L the previous quarter. The results obtained from estimating this model were somewhat
unsatisfactory, as were the results from the estimation of comparable models for M2 and M3. The
coefficient estimates were very sensitive to the sample period, they sometimes had the "wrong" sign,
and the t-statistics often indicated coefficients insignificantly different from zero. Nevertheless, in a
postsample simulation, the demand equation for L did relatively well. These results quickly became
outdated, however, when the ceilings on the interest rates banks could pay depositors were removed.
54
The Said-Dickey statistic is used to test whether a general autoregressive moving average
process has unit roots. See Said E. Said and David A. Dickey, "Testing for Unit Roots in
Autoregressive Moving Average Models of Unknown Order," Biometrika, vol. 71 (1984), pp. 599-607.
This statistic is more commonly used in tests for cointegration and known as the augmented DickeyFuller statistic after the terminology used by Robert F. Engle and Clive W. J. Granger, "Cointegration
and Error Correction: Representation, Estimation and Testing," Econometrica, vol. 55 (1987), pp. 25176. The result that L velocity is not trend-stationary should be interpreted with some caution because
of the poor power of the Said-Dickey test.
55
The relevant interest rate for L is taken to be Moody's AAA corporate bond yield, and for M2
we use the six-month commercial paper rate.
56
David C. Bennett, Flint Brayton, Eileen Mauskopf, Edward K. Offenbacher, and Richard D.
Porter, "Econometric Properties of the Redefined Monetary Aggregates," Board of Governors of the
Federal Reserve System memorandum, February 1980. Some of the estimation results are also
reported by Berkman, "Abandoning Monetary Aggregates."
141
Chart 8
L Velocity Deviations from Trend
and Long-Term Bond Yield
Percent
20
r
15
H
:
:
::
:
:
:
i
:
^
/ \i
/
,'
J*
i.
:
H
i
10
•'
h
j
y
V*
**\
""
/ •
H
^^v
A/V
AA/
-5
-10
H
1
J
L._L.-„ - . . l
I-..L.J
1962
1965
1959
J
1\
1
i._ i
J _i
1968
1..1....LJ
1971
1974
Long-term
bond yield*
i l l
^t/
W:
)
j
j As
^ y
5
\
t!
i i
l^brl
\1 \ //
i
i
1977
i
i
i
1980
i
i
i
i
i
i
i
\j
1 1 1
1986
1983
Moody's AAA corporate bond yield.
Chart 9
M2 Velocity Deviations from Trend
and Six-Month Commercial Paper Rate
Percent
20
-
Commercial
paper rate
:
15
j
j
:
: .' »
A
h-
'A
#\ J \
%
»'
:
M/
\
:
.4 : I*.
I
,
':1: '. •'• :
h
10
o
•
f
L
«* **
h
^A J
A
/
-5
M2 deviations j V
142
-10
I
i
1959
i
i
i
1962
i t
i -i
1965
i
1 i—l
1968
i—l
1971
i _ . l _ i _ i _ . l _i—l
1974
1977
i
1980
i
i
i
1983
i
i
i
1986
i
i
The passbook savings rate became essentially irrelevant, and the rates paid on small time deposits
adjusted to Treasury bill rates with only minor time lags.
Reduced-form equations for the monetary aggregates were also estimated. The explanatory power
of L for nominal GNP was greater than that of Ml or M3, but less than that of M2. In out-of-sample
simulations, L again did well relative to the other aggregates, providing slightly more accurate
forecasts of nominal GNP growth than Ml and M3 and slightly less accurate forecasts than M2.57
Besides the work of the Board staff economists, Frank Morris reports the results of estimating the
public's demand for Ml, L, and total debt.58 All three econometric equations have the same
inventory-theory specification and use the same explanatory variables. The equations are estimated
over the period 1959-73 and are then used to predict the aggregates' income velocity from 1974 to
1981. The annual forecasting errors of L are not quite as good as those of total debt, but both L and
total debt perform far better than Ml.
Most other econometric studies of liquid asset measures have attempted to apply aggregation
methods more refined than those used in the construction of L. The early findings of Bamett, Spindt,
and Offenbacher on a monetary-services index of L were encouraging. They used data covering the
1960s and 1970s and applied several methodologies: Granger causality, stability of velocity, simulation
performance of demand equations, and stability and simulation performance of both portfolio
allocation equations and reduced-form equations. In nearly all cases, a monetary aggregate measured
by a monetary-services index was superior to a simple sum aggregate; the advantages of Divisia
aggregation appeared more evident the more broadly money was defined. These results promised that
more rigorous aggregation methods would in fact provide more informative measures of the money
supply.59
Unfortunately, later results for the Divisia aggregates, obtained by adding data from the years
57
Berkman also ran reduced-form equations explaining the growth of real GNP and the inflation
rate. L performed relatively poorly in forecasting real growth and relatively well in predicting prices,
although Berkman felt the differences in performance of different aggregates were not economically
meaningful. See Berkman, "Abandoning Monetary Aggregates."
58
See the Appendix to Morris, "Do the Monetary Aggregates Have a Future as Targets of Federal
Reserve Policy?"
59
William A. Barnett, Paul A. Spindt, and Edward K. Offenbacher, "Empirical Comparisons of
Divisia and Simple Sum Monetary Aggregates," Board of Governors of the Federal Reserve System,
Special Studies Paper no. 158, July 1981.
143
1980-85, were judged by Lindsey and Spindt to be quite disappointing.60 Standard demand functions
were estimated for the conventionally measured aggregates and for monetary services L by using, for
each aggregate, an average opportunity cost.61 For each conventionally measured aggregate, the
opportunity cost was the difference between the three-month Treasury bill rate and its weightedaverage offering rate; for monetary-services L, the opportunity cost was the index of "user cost" used
to construct the monetary-services index. The estimated income elasticity of the demand for Divisia L
was found to be negative, although insignificant. (The regression equations were estimated from 1971
to 1979.) In simulations over the period 1980-85, the prediction errors of monetary-services L were
Unacceptable, larger than the errors of each conventionally defined monetary aggregate. The poor
performance of monetary-services L was attributed to mismeasurement of opportunity cost. In simple
reduced-form equations relating the growth of nominal GNP to fiscal policy and growth of monetary
services L, the R-bar squared was actually negative.
Recent experience with the demand for liquid assets. To explore the stability of the demand for
liquid assets, we specify a quasi-reduced form demand equation for L.62 In principle, the demand for
L depends on factors such as real income, real wealth, and the opportunity cost of holding each of its
components. However, the own rates of return on assets included in L are not exogenous and respond
to changes in the yields on alternative assets. Similarly, real wealth holdings depend on income and
real interest rates. Thus, a model of the demand for L that uses opportunity cost measures and real
wealth as explanatory variables is incomplete without models explaining the interrelationships among
the own rates of return on assets in L, the interrelationships between own rates and yields on
alternative assets, and the relationship between real wealth, real GNP, and interest rates. We do not
model these relationships explicitly. Rather, to simplify the analysis we substitute them into the
demand for L to obtain a conventional specification of the demand as a function of rates of return on
60
David E. Lindsey and Paul A. Spindt, "An Evaluation of Monetary Indexes." Earlier results
were reported in Paul A. Spindt, "Money Is What Money Does: Monetary Aggregation and the
Equation of Exchange."
61
See Appendix D of the Lindsey and Spindt paper.
62
For an example of this approach applied to the demand for M2, see John P. Judd, Brian Motley,
and Bharat Trehan, "The Demand for Money: Where Do We Stand?" Federal Reserve Bank of San
Francisco, Working Paper no. 88-02, July 1988.
144
alternative assets, real income, and the inflation rate.
This strategy naturally implies shifts in the
demand for liquid assets, as we specify it, when the relationships among own rates and between own
rates and the rates on alternative assets change, as they did in the wake of financial deregulation and
innovation in the early eighties.
We model the demand for liquid assets in a cointegration-error correction framework.64 That is,
we use two equations to represent demand. The first, or cointegrating equation, captures the long-run
relationship between the real value of L demanded and its determinants. The dynamic process
whereby the real quantity of L outstanding adjusts to its long-run equilibrium value is captured by the
second equation, or error correction model. The cointegration-error correction framework is
appropriate because the variables used to explain the demands for financial asset aggregates are
generally nonstationary. This framework is also intuitively appealing. It is not difficult to imagine a
stable long-run demand for liquid assets, despite a changing institutional and regulatory environment.
Financial institutions respond to regulations governing interest payments on deposits, for example, by
devising new instruments or providing services that provide asset holders with nonpecuniary interest
payments. Thus, the removal of interest rate regulations may change the form but not the value of
interest payments to deposit holders, and therefore may leave the long-run demand for liquid assets
unaffected. At the same time, we would expect the removal of interest rate regulations to affect the
dynamic adjustment of the value of liquid assets outstanding, because it allows financial institutions to
change interest rates on their deposits quickly in response to changes in the yields on alternative
assets, rather than changing over time the services that provide implicit interest payments to deposit
holders.
The long-run demand for, or desired, real L is based on an equation of the form
63
Formally, we assume a demand of the form
L/P = F{Y,W/P,RA,rL,TC},
where L/P is real L, Y is real income (GNP), W/P is real wealth, and % is a measure of inflation.
Interest rates on assets included in L are denoted by the vector rL, while RA represents nominal interest
rates on alternative assets L. We then substitute the equilibrium relationships between the own rate of
return on L and the return on alternative assets,
rL = H(RA),
and between wealth, real GNP, nominal interest rates and inflation,
W/P = G(Y,RA,rL,7c),
into this demand equation to get the semireduced form
L/P = M(Y,RA,TC).
64
See Engle and Granger, "Cointegration and Error Correction."
145
(1) logeOW = Yo + YiTt + 'felqg.aW + y3LBRt + y4nt + et,
where P is the GNP deflator, T is a time trend, Y/P is real GNP, LBR is a long-term bond rate,
n is the inflation rate,65 and 8 an error. The dynamic adjustment of real L to its desired value is
modeled as an error-correction equation of the form
(2) Alog^L/Pt) = 80 + 8 ^ + £ 82JAlogc(Yt./Pt.j) + 1
n
b^ALR^
n
+ I84i|AKt.j + I85JAlogc(Lt.j/Pt.j),
where e denotes the residual from the long-run demand equation, equation (1), and A is the difference
operator. That is, the rate of change of real L depends on its own past, the lagged deviation of the
actual from the desired level of real L, lagged rates of change of real GNP, and lagged changes in the
levels of long-term interest rates and inflation.
Estimation results for the demand for liquid assets are given in Table 6. In the table we present
three regressions:
(i) an estimated long-run demand for L (equation 1),
(ii) the estimated error-correction model (equation 2) governing the adjustment of liquid assets to
their desired level implied by the long-run demand equation, and
(iii) an augmented Dickey-Fuller regression for testing the hypothesis that the long-run demand
equation is a cointegrating regression.
After some experimentation with different long-term bond yields and holding-period returns, Moody's
AAA corporate bond yield was chosen as the relevant long rate. Inflation is measured by the
annualized quarterly rate of change of the GNP deflator.
The estimated long-run income elasticity of the demand for L, 0.57, is surprisingly low. But this
result is not inconsistent with the postwar decline in L shown in Table 4 and Chart 6; both GNP and
65
The inflation rate enters the demand equation for two reasons: to capture substitution between
money and commodity holdings, and because we have written the relationship between real wealth and
its determinants in terms of nominal interest rates.
146
Table 6
Demand for L
Cointegrating Regression
Dependent Variable: log(L/P)
Regressors
Trend
Constant
0.006
(0.0004)
-1.671
(0.386)
LR
it
-1.059
(0.122)
-0.453
(0.074)
log(Y/P)
0.573
(0.052)
R2 = 0.997
DW = 0.37
Error-Correction Model
Dependent Variable: Alog(L/P)
Regressors
Constant
AIog(L/P).,
0.004**
(0.001)
(0.001)*
0.616**
(0.073)
(0.078)*
ii
-0.059*
(0.031)
(0.029)*
ALR.!
An.,
-0.336**
(0.116)
(0.096)*
0.062"
(0.027)
(0.024)*
AHogWh
0.078*
(0.041)
(0.035)*
R2 = 0.49
DW = 1.95
Durbin t = 0.138
Augmented Dickey-Fuller Regression
Dependent Variable: Ae
Regressors
Constant
-0.00003
(0.0009)
e..
-0.218**
(0.067)
AS.,
0.006
(0.101)
Ae\2
0.028
(0.101)
Aej
0.055
(0.099)
Afi.4
Ae\,
0.142
(0.099)
0.137
(0.097)
R2 = 0.12
DW = 1.97
Durbin t = 0.044
Notes: Sample period: 1959-1 to 1989-L
6 denotes the residuals from the cointegrating regression.
Standard errors are given in parentheses; those superscripted by a "W" are based on White's heteroskedasticityconsistent estimate of the covariance matrix.
T-statistics for the cointegrating regression do not have the standard distribution.
* indicates significance at the 5 percent level.
** indicates significance at the 1 percent level.
147
the demand for L have strong trend components. A 1 percentage point increase in the corporate bond
yield lowers the demand for real liquid assets by about 1 percent in the long run, and a 1 percentage
point increase in inflation lowers it by about 1/2 of 1 percent. Including rates of return on assets in L
(for example, the three-month Treasury bill rate and the six-month commercial paper rate) does not
add significantly to the explanatory power of the long-run demand equation.66
The cointegration tests discussed below indicate that the estimated equation is in fact a stable longrun relationship. This relationship may be stable because, as we speculated earlier, the supply of
liquid assets is demand determined in the long run. Alternatively, we can interpret its stability as an
ex post property generated by the redefinition of financial asset aggregates. The estimated coefficients
in the long-run demand for real L are consistent estimates of the long-run responses of real liquid asset
holdings to changes in right-hand side variables despite the endogeneity of the regressors. The usual
simultaneity problems do not arise because the variables in the regression are not stationary. However,
for the same reason, the ratios of estimated coefficients to their standard errors do not have the usual tdistribution.67 The explanatory variables are nevertheless clearly significant.
The estimated error correction model indicates that 6 percent of the discrepancy between actual
and desired real liquid asset holdings is eliminated each quarter if lagged values of real income,
nominal interest rates, inflation, and real liquid assets are held constant. Adjustment is more rapid if
the deviation between actual and desired liquid asset holdings is attributable to changes in long-term
interest rates or real GNP; adjustment is slower if the deviation is due to a change in inflation. Note,
however, that real L responds much more rapidly to interest rate changes than to real GNP or inflation
rate changes. For example, 2/5 of the 1 percent decrease in the long-run demand for real L in
response to a 1 percentage point increase in the nominal long-term corporate bond yield occurs within
one quarter. By contrast, only 6 percent of the change in real L in response to a real GNP change
occurs in the first quarter; another 14 percent occurs in the second quarter after the shock.68 As in
the case of the long-run demand for L, yields on assets included in L do not add significantly to the
66
For space considerations these results are not shown.
67
See Clive W. J. Granger and Paul Newbold, "Spurious Regressions in Econometrics," Journal of
Econometrics, vol. 2 (1974), pp. 111-20, for a discussion of the "spurious regressions" problem that
arises when a relationship between nonstationary variables is estimated by ordinary least squares.
68
These calculations are only first approximations; we can obtain more accurate responses by
calculating total multipliers taking both the long-run demand equation and the error-correction models
for all four variables into account.
148
explanation of the adjustment dynamics of holdings of L.
Real L, as well as the variables in the long-run demand for L, is nonstationary. Thus, to test
whether the estimated long-run demand for L does in fact represent a stable long-run relationship—that
is, whether the variables in the equation are cointegrated-we test whether the residuals from the
cointegrating regression are stationary. The null hypothesis of nonstationarity is tested by the
augmented Dickey-Fuller test, which establishes whether the lagged level of the residual enters with a
significantly negative coefficient in a regression of the change of the residual on lagged changes and
the lagged level of the residual. Two difficulties are associated with the augmented Dickey-Fuller test:
the critical values for the test are very sensitive to the number of parameters estimated in the
cointegrating regression, and the test has very low power.70 The test rejects at the 1 percent level the
null hypothesis that the residuals from the long-run demand equation are nonstationary if Fuller's
critical values are used and at the 5 percent level with the Granger and Engle critical values.
However, when Engle and Yoo's critical values are used, we can only reject the null hypothesis in a
one-tailed test at a confidence level between 5 percent and 10 percent. Thus, the evidence in favor of
the existence of a stable long-run relationship is not unambiguous. An alternative test, the
cointegrating regression Durbin-Watson (CRDW) test, rejects the null hypothesis of nonstationarity if
the Durbin-Watson statistic is large enough. In this case, it rejects nonstationarity at a significance
level greater than 5 percent but less than 10 percent.71 However, this test is known to be sensitive to
the parameters under the null. Whether a cointegrating relationship is found is asymptotically
independent of the normalization of the long-run relationship. However, nonstationarity of the
69
Results are not shown.
70
Critical values are obtained by Monte Carlo experiments. The most commonly used critical
values are those given in Wayne A. Fuller, Introduction to Statistical Time Series, p. 373 (values
which, strictly speaking, should only be used in testing for unit roots) and those given in Granger and
Engle, "Cointegration and Error Correction" (which assume that there are only two variables in the
cointegrating regression). Robert F. Engle and Byung Sam Yoo, "Forecasting and Testing in
Cointegrated Systems," Journal of Econometrics, vol. 35 (1987), pp. 143-59, provide critical values for
some cases where more than two variables are included in the cointegrating regression. A quick
perusal of Engle and Yoo's Table 3 reveals, surprisingly, that the critical values do not decrease
monotonically in the number of observations in the sample. For this reason, we decided not to base
our inferences on this table of critical values alone.
71
To reject at the 5 percent level the null hypothesis that the residuals from the cointegrating
regression have a unit root, the Durbin-Watson statistic for the cointegrating regression must exceed a
critical value of 0.386. The critical value at the 10 percent level of significance is 0.322. Thus the
true P-value for this test is much closer to 5 percent than to 10 percent.
149
residuals cannot be rejected when the corporate bond yield is chosen as the left-hand-side variable, but
this result may be attributable to the size of the sample. Thus, on this evidence, and taking into
account the power properties of tests for nonstationarity, we tentatively conclude that the estimated
long-run demand equation captures a stable long-run relationship.
Error correction equations for the other variables in the long-run demand for liquid assets are
shown in Table 7. For each of these equations we derived the residuals for the estimation of the error
correction term, which indicates the speed of adjustment of the dependent variable towards the longrun equilibrium relationship captured by the cointegrating regression, from a renormalized long-run
relationship with the same left-hand-side variable.72 Three results are worth noting: Real GNP
adjusts very slowly-the coefficient on the error correction term is not significantly different from zero.
The effects on the adjustment process of changes in real L are largely reversed after four quarters.
Finally, the inflation rate adjusts surprisingly rapidly, a quarter of the deviation between the actual
inflation rate and the equilibrium inflation rate is eliminated each quarter, and adjustment in response
to changes in real liquid assets is particularly rapid.
Stability. It is not possible to test directly whether the parameters in the long-run demand for L
were stable over time, but cointegration is in itself evidence of long-run stability. It is reasonable,
however, to expect the dynamic adjustment of the demand for L to have changed as a result of the
financial innovations and deregulation of the early eighties and to test for such a change. Liquid
financial assets introduced in the early 1980s, such as money market mutual funds, money market
deposit accounts, and time deposits with resettable or guaranteed minimum interest rates, yield returns
that respond quickly and flexibly to movements in market-determined rates on alternative assets.
These assets serve to tie the own return on L more closely to long-term free market rates. Because the
own rate of return on L can respond more flexibly to movements in long-term free market interest
rates, we would expect a greater share of the response to such movements to take the form of changes
in the rates of return on liquid assets rather than expansion or contraction of the volume of liquid
72
The cointegrating equation only captures a statistical long-run equilibrium relationship. We
chose to interpret this relationship as a demand for real liquid assets. However, it could as well be
interpreted as a model of the equilibrium inflation rate. Associated with the cointegrating relationship
are error correction equations for each of the included variables. Together, these equations capture
the adjustment process whereby the long-run equilibrium is attained.
150
Table 7
Estimated Coefficient Values of the
Error Correction Models
Dependent Variable
Constant
Arc
ARL
AloR(Y/P)
-0.007*
(0.003)
(0.003)*
0.0003
(0.0004)
(0.0004)*
0.006**
(0.002)
(0.002)*
-0.268*
(0.115)
(0.122)*
-0.137**
(0.039)
(0.043)*
-0.039
(0.040)
(0.045)
Alog(Y/P).,
0.193*
(0.091)
(0.085)*
A^ogOT/P).;
0.154*
(0.066)
(0.062)*
ALR.,
-0.472*
(0.185)
(0.178)*
ALR.j
-0.642**
(0.196)
(0.154)*
Alog(L/P).:
0.684*
(0.289)
(0.308)*
0.578**
(0.173)
(0.197)*
Alog(L/P).3
+
Alog(L/P).4
-0.370**
(0.111)
(0.117)*
Alog(L/P).5
0.271*
(0.136)
(0.122)*
An,
-0.355"
(0.117)
(0.105)*
Arc,
ALR.,
+
ALR.,
0.151"
(0.051)
(0.051)*
0.276**
(0.059)
(0.104)*
151
Table 7
Estimated Coefficient Values of the
Error Correction Models (Continued)
Regressors
Dependent Variable
Arc
ALRj
+
ALR.4
ARL
Alog(Y/P)
0.541*
(0.246)
(0.308)w
R2
0.29
0.20
0.35
DW
2.05
1.94
1.88
Durbin t
-0.98
0.89
Notes: Sample period: 1959-1 to 1989-1.
6 denotes the residuals from the cointegrating regression.
Standard errors are given in parentheses; those superscripted by a "W" are based on White's heteroskedasticityconsistent estimate of the covariance matrix.
T-statistics for the cointegrating regression do not have the standard distribution.
* indicates significance at the 5 percent level.
** indicates significance at the 1 percent level.
assets held by the public. Consequently, we would expect the demand for L to respond less strongly
and less rapidly to movements in long rates as a result of the financial changes of the early eighties.
Table 8 provides empirical support for this view.
The second column of Table 8 shows estimates of the shifts in the parameters of the error
correction model after 1979. Two sets of standard errors are shown; the second set is based on
White's heteroskedasticity-consistent estimate of the residual covariance matrix.73 The hypothesis
that all the parameters have remained constant is rejected at the 5 percent level by the likelihood ratio
test and the Wald tests based on ordinary least squares (OLS) residuals and by the Wald and F tests
73
See Halbert White, "A Heteroskedasticity-Consistent Covariance Matrix Estimator and a Direct
Test for Heteroskedasticity," Econometrica, vol. 48, pp. 817-38.
152
Table 8
Stability Tests: Error Correction Model
Dependent Variable: Alog(L/P)
Regressors
Coefficients
1980-89 Dummies
Constant
0.005**
(0.001)
(0.001)w
-0.003
(0.002)
(0.002)w
Alog(L/P).,
0.490**
(0.086)
(0.086)*
0.307*
(0.180)
(0.146)w
-0.051
(0.036)
(0.032)w
0.018
(0.098)
(0.080)w
A^ogCr/P)^
0.096*
(0.048)
(0.045)w
-0.049
(0.089)
(0.062)w
ALR.,
-1.201**
(0.313)
(0.375)w
1.053**
(0.337)
(0.385)w
Are.,
0.058*
(0.029)
(0.026)w
0.007
(0.070)
(0.053)w
R2 = 0.546
DW = 2.04
Durbin t = -0.559
LR = 13.61*
W = 12.96*
F = 2.16
W w = 13.98*
F w = 2.33*
Notes: Sample period: 1959-1 to 1989-1.
Standard errors are given in parentheses.
F, W, and LR are F, Wald, and likelihood ratio statistics for the hypothesis that parameters are die same before
and after 1980.
A "W" superscript indicates a statistic based on White's heteroskedasticity-consistent covariance matrix
estimator.
* indicates significance at 5 percent level.
** indicates significance at 1 percent level.
153
based on the White standard errors.74 The regression results suggest that this rejection is due to less
marked adjustment to changes in long-term bond yields from 1980 onwards, and to slower adjustment
in general. The coefficient on the long-term bond yield declines in absolute value by more than
seven-eighths, and the decline is significant at the 1 percent level regardless of which standard error
was used. The coefficient on the lagged dependent variable increases by 60 percent after 1979,
indicating a generally slower adjustment of L.75 However, this increase is only significant if judged
by the White standard error.
Forecasting performance. Despite a reasonable in-sample performance, the cointegration-error
correction model in Table 6 performs disappointingly in out-of-sample forecasting. Table 9 documents
the results of some relatively undemanding forecasting experiments. These experiments were
performed as follows: The cointegration model was estimated over the whole sample (1959-1 to 1989I) to generate a vector of residuals. The error correction model, estimated with data up to and
including the first quarter of 1987, was then used with actual data on its right-hand-side variables and
the residuals from the cointegrating regression to generate static and dynamic forecasts of the
annualized percentage rate of growth of real liquid assets for the period from the second quarter of
1987 to the first quarter of 1989.
The model overpredicts the rate of real L growth over the simulation period by about 40 percent in
the static simulation and more than one-half in the dynamic simulation. Overprediction occurs in
periods of both low and high liquid asset growth, as the maximum and minimum values of actual and
predicted growth rates suggest. The root mean squared prediction error exceeds 2.4 percent for the
static forecast and 2.7 percent for the dynamic forecast compared with a standard error of real L
growth of 1.85 percent. The mean absolute forecast errors for the both the static and dynamic
forecasts are large enough to exceed the actual annual growth rate of real L of 2.2 percent. Finally, it
is clear that the model does poorly at predicting turning points in L growth. The static predictions are
74
It is not possible to calculate a straightforward likelihood-ratio test of parameter stability for the
model with "White standard errors.1' Such a test would necessitate the specification of a model of the
heteroskedasticity of the errors.
75
Further evidence of slower adjustment after 1980 is provided by the increase in the coefficient
on the error correction term (e), but this increase is not statistically significant. Note also that no
specific conclusions can be drawn from the fact that neither of the coefficients on the error correction
term is individually significant.
154
Table 9
Forecasting Performance: Error Correction Model
Growth of Real Liquid Assets: 1987-11 to 1989-1 (Percent per Annum)
Parameter
Actual
2.21
1.85
4.93
-0.08
Average
Standard deviation
Maximum
Minimum
Dynamic Forecast
Static Forecast
3.05
1.56
5.85
0.51
3.38
1.02
4.74
1.90
Measures of Forecast Accuracy
Measure
Static Forecast
Dynamic Forecast
2.41
2.25
-0.82
-0.002
Root mean squared error
Mean absolute error
Mean error
Correlation coefficient
(actual/forecast)
2.77
2.43
-1.14
-0.73
Notes: Estimation period: 1959-1 to 1987-1.
Forecast error statistics are calculated as annualized values of quarterly rates of growth.
Chart 10
L and M2 Multipliers
18
L
Base
, * * -
|16
I
h-
:
•
:
•
;
:
;
;
* \
/
•
* — **
*
/
/
j
*
•
14
h-
:
\-
—*
*
\
•
*
12
—
M2^"
Base
'.
10
r~-""'"
h-
8
6
!
i
*
*'*
h-
I
1959
I
I
\
1962
I
I
I
1965
L_J
II — I — i
1968
i
1971
i
i
i--\
1974
i-J
1977
i
i
i
1980
i
i
i
1983
i
i
'
1986
I
I
155
virtually uncorrected with growth of real L, and the dynamic predictions are strongly negatively
correlated with real L growth. Thus, one could make better forecasts by betting against the model!
Moreover, these results are little affected by changing the period over which the error correction model
is estimated or by allowing for shifts in the parameter values after 1980.
B. The Relationship between L and Policy Instruments: The L Multiplier
The L and M2 multipliers (ratios of L to the monetary base and M2 to the base) are shown in
Chart 10. Descriptive statistics obtained by fitting time trends, with and without correction for serial
correlation, to quarterly data are given in Table 10. The L multiplier grew at an average annual rate
of 2.4 percent from the first quarter of 1959 to the first quarter of 1989; the M2 multiplier grew at
about 1.8 percent over the same period. These trends consist of growth, at 2.5 percent for L and 2
percent for M2, until the end of 1985 and declines thereafter at rates of 1.4% and 2.2% per year.
To explore the stability of the L and M2 multipliers, we calculate deviations from trend by
expressing the standard errors of the regressions as annual percentage growth rates. The deviations
from trend indicate that both the L and M2 base multipliers have been more volatile in the last decade
than in the two preceding ones. The deviations of the L multiplier are notably smaller than those of
M2, and the estimated trend growth rate of the L multiplier varies less over time than that of the M2
multiplier.
The Said-Dickey statistics reported in Table 10 indicate that we cannot reject the hypothesis of
nonstationarity for either the L multiplier or the M2 multiplier. However, we were unable to find
simple long-run relationships that linked the L and M2 multipliers with potential determinants such as
the federal funds rate, own rates of return on assets included in L and M2, GNP, and the inflation rate.
To evaluate the predictability of the L and M2 multipliers, we performed the following experiment:
We used data from the first quarter of 1959 to the first quarter of 1987 to estimate simple
autoregressive models for the rates of change (log differences) of the L and M2 multipliers and then
used these models to forecast the rates of change of the L and M2 multipliers for the period from the
second quarter of 1987 to the first quarter of 1989.76 Statistics summarizing static and dynamic
forecasts are given in Tables 11 and 12. The simple models tended to underpredict the rates at which
the L and M2 multipliers declined over the forecast period. Moreover, in the dynamic simulations the
76
The model for the L multiplier included a constant trend and ten of its own lagged values; the
model for the M2 multiplier, a constant and nine lagged values.
156
Table 10
The L and M2 Multipliers: Descriptive Statistics
Period
Trend
Average
Deviations from Trend
Maximum
Minimum
L Multiplier
1959-1 to
1989-1
2.4"
1.7*
8.9
6.1*
19.2
0.01*
0.33
1959-1 to
1969-1
3.0"
3.0"*
5.9
1.1*
10.6
3.5*
0.30
0.03*
1969-H to
1979-1
2.9"
2.7"*
3.1
1.5*
12.3
3.9*
0.02
0.14*
1979-H to
1989-1
1.5"
1.2"*
10.6
2.2*
20.2
7.2*
1.91
0.06s
SD
2.17
M2 Multiplier
1959-1 to
1989-1
1.8"
1.8"*
12.3
2.5*
40.4
7.9*
0.02
0.01*
1959-1 to
1969-1
3.2"
3.1"*
7.7
1.5*
14.6
4.0*
1.63
0.49*
1969-H to
1979-1
2.3"
2.0"*
7.2
3.0*
16.7
6.8*
0.04
0.04*
1979-H to
1989-1
0.7"
0.3"*
9.6
2.3*
20.1
9.7*
1.48
0.14*
SD
2.80
Notes: Measures are expressed as annual percentage changes.
Statistics are based on trend regressions. An "S" superscript indicates thatfirst-orderserial correlation
correction was used.
SD is the Said-Dickey statistic to test for a unit root.
** indicates significance at the 1 percent level.
157
Table 11
Forecasting Performance: L Multiplier
Changes in L Multiplier: 1987-n to 1989-1 (Percent per Annum)
Parameter
Average
Standard deviation
Maximum
Minimum
Actual
-0.40
1.46
1.21
-2.89
Static Forecast
0.31
1.20
2.13
-1.49
Dynamic Forecast
0.92
1.16
1.79
-1.49
Measures of Forecast Accuracy
Measure
Static Forecast
Root mean squared error
Mean absolute error
Mean error
Correlation coefficient
(actual/forecast)
1.67
1.47
-0.72
0.28
Dynamic Forecast
1.75
1.42
-1.31
0.60
Notes: Estimation period: 1959-1 to 1987-1.
Forecasts are based on a tenth-order autoregression.
Forecast error statistics are calculated as annualized values of quarterly percentage changes.
The rate of change of the multiplier is forecast using afifth-orderautoregression, with trend, for its level.
models predicted increases in the multipliers, rather than the declines that occurred in fact. When
evaluated in terms of standard measures of forecast accuracy-root mean squared error, mean absolute
error, and mean error, relative to standard error-the rate of change of the L multiplier appeared more
predictable than that of the M2 multiplier. On the other hand, the L multiplier was somewhat more
volatile than the M2 multiplier over the forecast period, but the errors that occurred in forecasting it
appeared smaller in absolute magnitude. It was not clear whether the turning points of the L
multiplier were easier to forecast: static forecasts for the M2 multiplier were more highly correlated
with actual values than were forecasts for the L multiplier, but the reverse was true for the dynamic
forecasts. Thus, we tentatively concluded that the L multiplier could be forecast more accurately than
the M2 multiplier, but that this advantage was probably too small to be of practical significance.
Furthermore, neither the L or M2 multiplier was predicted accurately on the basis of its past values;
for both, the root mean squared errors exceeded their standard deviations over the forecast period.
158
Table 12
Forecasting Performance: M2 Multiplier
Change of M2 Multiplier: 1987-11 to 1989-1 (Percent per Annum)
Parameter
Average
Standard deviation
Maximum
Minimum
Actual
-2.19
1.27
-0.36
-4.37
Static Forecast
Dynamic Forecast
-0.34
0.80
0.25
-1.93
1.32
1.56
2.66
-1.93
Measures of Forecast Accuracy
Measure
Static Forecast
Root mean squared error
Mean absolute error
Mean error
Correlation coefficient
(actual/forecast)
2.17
1.88
-1.85
3.79
3.56
-3.46
0.44
0.56
Dynamic Forecast
Notes: Estimation period: 1959-1 to 1987-1.
Forecasts are based on a ninth-order autoregression.
Measures other than the correlation coefficient between actual and predicted changes in the M2 multiplier are
measured as annualized values of quarterly percentage changes.
The rate of change of the multiplier is forecast using a ninth-order autoregression, with trend, for its level.
C. Implications for Intermediate Targeting
Our comparison of the velocity behavior of L and M2 proved inconclusive, because judgments on
the stability of velocity depend on the volatility measure used. Thus a more in-depth study of the
demand for liquid assets is needed to decide whether L can be used as an intermediate target. Recent
studies on monetary targeting have concentrated on lower level aggregates; earlier studies of L
undertaken by the staff of the Board of Governors and by Neil Berkman and Frank Morris of the
Federal Reserve Bank of Boston produced mixed results and, in light of the financial changes of the
eighties, are probably too out of date to be used with confidence. Our econometric work, undertaken
to update the earlier work, yielded four main findings:
(i)
There is a stable long-run demand for L.
(ii)
The adjustment of L to its long-run level is slow, with only 6 percent of the deviation of
159
L from its long-run equilibrium value eliminated each quarter.
(iii)
The adjustment process underwent significant structural change as a result of the financial
innovations and deregulation of the early 1980s.
(iv)
The short-run behavior of L is difficult to predict out of sample; it may in fact pay to bet
against a simple model of this adjustment process.
We also analyzed the behavior of the L and M2 multipliers to evaluate the feasibility of a strategy of
targeting L. This work led to two conclusions:
(i)
The estimated trend growth rate of the L multiplier is less variable than that of the M2
multiplier, and deviations from trend of the L multiplier are smaller than those of the M2
multiplier.
(ii)
Time series forecasting experiments favor, albeit weakly, the conclusion that the L
multiplier is more predictable than the M2 multiplier.
These results suggest that control over L need not be more difficult to achieve than control over M2,
but that any advantage of L is probably too small to be of practical significance.
The evidence on the relationship between L and variables of policy interest leads us to conclude
that targeting L is not a viable strategy. The existence of a stable long-run demand for L suggests that
L may be useful as a long-run nominal anchor for monetary policy. However, the slow adjustment of
actual to desired holdings of liquid assets implies that policy makers may have to tolerate large and
sustained (up to three quarters or longer) departures of L from its predetermined intermediate target
path.77 Furthermore, even if L is kept on a track, goal variables may deviate from desired paths for
even longer periods when L is used in this role. This appears to be the case for real GNP, which was
shown to adjust very slowly to its long-run equilibrium value, but not necessarily for the inflation rate,
which adjusts surprisingly rapidly. Finally, the structural break in, and poor forecasting performance
of, the error correction model for L does not augur well at a time when policy involves more active
adjustments of intermediate targets. The poor forecasting performance is all the more damaging
because some elements of L (primarily offshore deposits) are subject to long reporting lags.
77
If the Fed could control long-term nominal interest rates precisely, L could be brought back on
track in three quarters. Imperfect control would probably lengthen the interval between policy actions
and the desired result.
160
IV. Liquid Asset Measures as Indicators
An indicator can be useful in the formulation of monetary policy in two conceptually distinct
roles: It can be used as an indicator of the state of the economy as a whole, that is, as a source of
information on current and potential future developments in the economy that may require changes in
policy. Alternatively, it can be used as a measure of the current stance of monetary policy, that is, as
an independent indicator that tells policy makers whether policy is, or has been, loose, tight, or neutral.
The authorities will not rely solely on indicators of policy stance in deciding whether policy should be
changed. Rather, policy decisions are made on the basis of both the stance of monetary policy and the
state of the economy. Unlike an intermediate target, an indicator variable need not be under the
control of the central bank. This section explores the potential of a liquid asset aggregate as an
indicator in these two roles.
An information variable must satisfy two requirements. It should bear a stable relationship to
target or goal variables and it should provide information about the state of the economy early enough
for policy makers to respond to ongoing or prospective developments. To evaluate the potential of L
as an information variable, we use simple vector autoregressions (VARs) to determine whether L has
marginal predictive power for movements in goal variables once other readily available sources of
information have been taken into account.
There is no conceptually unambiguous measure of the stance of monetary policy comparable to the
structural deficit for fiscal policy. Thus policy makers are faced with a choice between financial prices
and quantities, both of which tend to be procyclical. However, a variable will still be useful as a
measure of the stance of policy if it provides reliable information about the impact of monetary policy
on the economy. In this case, the variable need not lead economic activity, but it is of the utmost
importance that it be reliable and timely. The value of a particular aggregate or price as a measure of
the stance of policy will also depend on the specifics of the policy-making setup: the intermediate
target(s), information variables, and other indicator variables used, and the precise rules chosen to
signal the need to reset the target path for the intermediate target. Given such a set of specifics, the
usefulness of L as an indicator of the stance of monetary policy can be evaluated in two ways. First,
one can study episodes in the past when policy errors may have occurred to determine whether the use
of a liquid asset indicator would have enabled policy makers to avoid mistakes. Alternatively, one
could evaluate the usefulness of L by simulating the effects of combining an intermediate target with
an indicator and a decision rule in a simple macroeconomic model. We shall not follow either of
these approaches since both require more information about the policy framework than we can assume
161
here.
In principle, broad financial aggregates are less likely to provide misleading signals about the
stance of policy because they are less susceptible to financial innovations and interest rate changes, but
in practice, long reporting lags may limit the aggregates' usefulness. The experience with financial
aggregates in the early eighties suggests that neither broad nor narrow aggregates would be reliable
indicators of the stance of policy in times of uncertainty. For example, L growth measured on a fourquarter basis accelerated from the fourth quarter of 1980 to the third quarter of 1981, a period during
which an inverted term structure of interest rates, as measured by the difference between the nominal
yields on ten-year government bonds and three-month Treasury bills, indicated a monetary
tightening.78 Thus, the evidence suggests that L need not, in fact, be a more reliable indicator of the
stance of monetary policy than narrower aggregates.
A. The Information Content of L: A VAR Analysis
VAR models provide a simple way of capturing the contribution of an indicator to the policy
maker's information about future developments in the economy. Interpreted in this way, rather than as
a means to capture causation, VAR models are uncontroversial. To evaluate the information content
of L, we regressed the quarterly rate of change (log difference) of each of three variables of policy
concern—nominal GNP, real GNP, and the GNP deflator—on four lags of itself and four lags of the
rate of change of L. To test whether L would have predictive power in a broader information set, we
successively added four lags of changes of the six-month commercial paper rate and four lags of
Cohen's79 structural deficit measure to the regression. M2 is used as the benchmark against which L
is evaluated.
Tables 13 and 14 show the regression results when nominal GNP is the dependent variable. Table
13A shows that four lags of the growth rate of L jointly have significant predictive power (at the 5
percent level) for nominal GNP growth, although no single lagged value does. When the commercial
paper rate is added to the regression, L has predictive power for nominal GNP two periods ahead at
the 5 percent level, as does the commercial paper rate (at the 1 percent level). However, as Table 13B
78
The L growth rate measured on a quarterly basis declined in the fourth quarter of 1980 and
increased thereafter.
79
See Darrel Cohen, "A Comparison of Fiscal Measures Using Reduced-Form Techniques," Board
of Governors of the Federal Reserve System, Division of Research and Statistics, 1989, mimeo.
162
Table 13A
"Causality" Tests: Nominal GNP
Lags
Nominal GNP
0.16
(0.10)
0.20
(0.26)
0.05
(0.10)
0.60
(0.32)
-0.06
(0.10)
-0.32
(0.31)
-0.01
(0.10)
-0.02
(0.24)
F
0.87
2.81*
LR
3.69
11.59*
Table 13B
"Causality" Tests: Nominal GNP
Lags
Nominal GNP
L
CPR
0.19
(0.10)
-0.03
(0.26)
0.001
(0.0010)
0.12
(0.11)
0.64*
(0.31)
-0.004**
(0.001)
0.003
(0.11)
-0.32
(0.31)
0.0003
(0.001)
0.07
(0.10)
0.02
(0.24)
-0.001
(0.001)
F
1.28
1.67
3.57**
LR
5.63
7.28
15.08**
Notes: Sample period: 1959-1 to 1989-1.
Variables other than the commercial paper rate are measured as quarterly rates of change.
The figures in brackets are standard errors.
F is the F-statistic and LR the likelihood ratio statistic for the hypothesis that four lags of a variable are
insignificant in explaining changes of nominal GNP.
* indicates significance at the 5 percent level.
** indicates significance at the 1 percent level.
163
Table 13C
"Causality" Tests: Nominal GNP
Lags
Nominal GNP
CPR
0.19
(0.11)
-0.02
(0.27)
0.001
(0.001)
0.001
(0.001)
0.12
(0.11)
0.65*
(0.32)
-0.003
(0.003)
-0.004"
(0.001)
0.005
(0.11)
-0.36
(0.31)
0.003
(0.003)
0.0003
(0.001)
0.05
(0.11)
0.05
(0.25)
0.001
(0.003)
-0.001
(0.001)
F
1.22
1.69
0.64
3.41*
LR
5.69
7.65
2.96
14.97"
Table 14A
"Causality" Tests: Nominal GNP
Lags
Nominal GNP
M2
0.12
(0.10)
0.33*
(0.14)
0.13
(0.10)
0.19
(0.17)
-0.02
(0.94)
-0.04
(0.17)
0.03
(0.09)
0.11
(0.15)
F
0.98
4.23**
LR
4.16
17.06"
Notes: Sample period: 1959-1 to 1989-1.
Variables other than the commercial paper rate are measured as quarterly rates of change.
The figures in brackets are standard errors.
F is the F-statistic and LR the likelihood ratio statistic for the hypothesis that four lags of a variable are
insignificant in explaining changes of nominal GNP.
* indicates significance at the 5 percent level.
** indicates significance at the 1 percent level.
164
Table 14B
"Causality" Tests: Nominal GNP
Nominal GNP
Lags
M2
CPR
1
0.11
(0.10)
0.23
(0.19)
0.002
(0.001)
2
0.14
(0.11)
0.16
(0.19)
-0.002
(0.001)
3
-0.03
(0.11)
0.05
(0.19)
0.0009
(0.001)
4
0.06
(0.11)
0.05
(0.16)
-0.0005
(0.001)
F
0.84
1.16
1.70
LR
3.71
5.10
7.42
Table 14C
"Causality" Tests: Nominal GNP
Lags
Real GNP
M2
G
CPR
1
0.11
(0.11)
0.23
(0.19)
0.001
(0.003)
0.0017
(0.0011)
2
0.14
(0.11)
0.18
(0.20)
-0.003
(0.003)
-0.0022
(0.0014)
3
-0.03
(0.11)
0.04
(0.20)
0.003
(0.003)
0.0009
(0.0014)
4
0.04
(0.11)
0.06
(0.16)
0.001
(0.003)
-0.0002
(0.0013)
F
0.78
1.22
0.66
1.70
LR
3.61
5.56
3.06
7.73
Notes: Sample period: 1959-1 to 1989-L
Variables other than the commercial paper rate are measured as quarterly rates of change.
The figures in brackets are standard errors.
F is the F-statistic and LR the likelihood ratio statistic for the hypothesis that four lags of a variable are
insignificant in explaining the rate of change of nominal GNP.
* indicates significance at the 5 percent level.
** indicates significance at the 1 percent level.
165
shows, the other lags of the growth rate of L fare so poorly that, in contrast with the commercial paper
rate, we can reject the hypothesis that four lags of L jointly have significant predictive power for
nominal GNP growth. These results do not change when the structural budget deficit measure is
added to the regression in Table 13C. Table 14A shows that M2 growth does somewhat better than L
growth in a bivariate model; it helps to predict nominal GNP one period ahead at the 1 percent level.
However, M2 loses its predictive value completely when lagged values of the commercial paper rate
are added to the regression, as Tables 14B and 14C show.
Tables IS and 16 show the regression results when real GNP is the dependent variable. Both L
and M2 are useful in predicting real GNP growth one period ahead. However, neither of the monetary
aggregates nor the structural budget deficit measure significantly improves predictions based on an
information set restricted to real GNP and interest rates. The commercial paper rate is significant at the
1 percent level.
Finally, Tables 17 and 18 summarize the regression results when the inflation rate, measured by
the GNP deflator, is the dependent variable. Surprisingly, neither L nor M2 has any predictive power
for future inflation. Only lagged inflation and the commercial paper rate are significant in predicting
inflation, both at the 1 percent level of significance. This result could be explained by the fact that the
inflation rate is nonstationary in the sample.80 One would not expect the growth rates of L and M2,
both of which are stationary, to predict the inflation rate, or for that matter, the growth rate, of
nominal GNP. But as Table 7 illustrates, the growth rate of L does predict changes in the inflation
rate two quarters ahead, even when lagged values of the change in inflation are taken into account.
The econometric results of this section, based on data spanning the period from the first quarter of
1959 to the first quarter of 1989, cast doubt on the value of monetary aggregates as information
variables. Neither L nor M2 has significant predictive power for quarterly rates of change of nominal
or real GNP and of the GNP deflator once lagged values of the relevant variable and short-term
nominal interest rates are taken into account.81 But L does seem to contain information about future
80
Once again we should caution that this result may simply reflect the low power of tests for unit
roots. Tests based on larger data samples than are used here may indicate that the (log of the) price
level is difference-stationary.
81
Different results may be obtained when variables are measured as four-quarter rates of change,
because measuring variables as four-quarter rates of change may induce correlation between lagged
regressors and the residual of a time-series model. If this is the case, regressions based on four-quarter
rates of change will tend to exaggerate the information content of the right-hand-side variables in the
regression.
166
Table 15A
"Causality" Tests: Real GNP
Lags
Real GNP
1
0.22**
(0.10)
0.01*
(0.24)
2
0.15
(0.10)
0.46
(0.30)
3
-0.05
(0.10)
-0.42
(0.29)
4
0.01
-0.19
(0.09)
(0.23)
2.48
1.80
10.29*
7.54
F
LR
Table 15B
"Causality" Tests: Real GNP
Lags
Real GNP
L
1
0.17
(0.10)
-0.13
(0.23)
-0.00002
(0.0009)
2
0.19
(0.10)
0.47
(0.29)
-0.004*'
(0.0009)
3
0.06
(0.10)
-0.36
(0.28)
-0.0004
(0.0010)
4
0.11
-0.06
-0.002*
(0.09)
(0.22)
(0.0010)
3.67"
1.02
5.89**
15.46**
4.51
23.89**
F
LR
CPR
Notes: Sample period: 1959-1 to 1989-1.
Variables other than the commercial paper rate are measured as quarterly rates of change.
The figures in brackets are standard errors.
F is the F-statistic and LR the likelihood ratio statistic for the hypothesis that four lags of a variable are
insignificant in explaining changes of real GNP.
* indicates significance at the 5 percent level.
** indicates significance at the 1 percent level.
167
Table 15C
"Causality" Tests: Real GNP
Lags
Real GNP
L
G
CPR
1
0.16
(0.10)
-0.12
(0.23)
0.002
(0.003)
0.0001
(0.0001)
2
0.18
(0.10)
0.47
(0.29)
-0.002
(0.003)
-0.004**
(0.001)
3
0.05
(0.10)
-0.38
(0.28)
0.003
(0.003)
-0.0003
(0.001)
4
0.09
(0.10)
0.06
(0.23)
0.002
(0.003)
-0.0019
(0.0010)
F
2.79*
1.02
0.80
4.57**
12.37*
4.69
3.68
19.65**
LR
Table 16A
"Causality" Tests: Real GNP
Lags
Real GNP
M2
1
0.17
(0.10)
0.28*
(0.14)
2
0.21*
(0.10)
0.12
(0.16)
3
-0.04
(0.10)
-0.11
(0.16)
4
0.006
0.06
(0.09)
(0.14)
2.46
2.36
10.20*
9.81*
F
LR
Notes: Sample period: 1959-1 to 1989-1.
Variables other than the commercial paper rate are measured as quarterly rates of change.
The figures in brackets are standard errors.
F is the F-statistic and LR the likelihood ratio statistic for the hypothesis that four lags of a variable are
insignificant in explaining changes of real GNP.
* indicates significance at the 5 percent level.
** indicates significance at the 1 percent level.
168
Table 16B
"Causality" Tests: Real GNP
Real GNP
M2
CPR
0.14
(0.10)
-0.01
(0.16)
-0.0002
(0.0009)
0.21*
(0.10)
0.08
(0.18)
-0.004"
(0.001)
0.06
(0.10)
-0.12
(0.18)
-0.0009
(0.001)
0.11
(0.09)
0.14
(0.15)
-0.003*
(0.001)
3.24*
0.33
4.48"
13.74"
1.50
18.61"
Table 16C
"Causality" Tests: Real GNP
Lags
Real GNP
M2
G
CPR
0.14
(0.10)
-0.02
(0.16)
0.0015
(0.0027)
0.0001
(0.001)
0.20
(0.10)
0.01
(0.18)
-0.0019
(0.0027)
-0.0038"
(0.0011)
0.04
(0.10)
-0.12
(0.18)
0.0037
(0.0027)
-0.0008
(0.0012)
0.08
(0.10)
0.15
(0.15)
0.0021
(0.0028)
-0.0022*
(0.0011)
2.36
0.40
0.84
3.32*
10.56*
1.88
3.88
14.62"
Notes: Sample period: 1959-1 to 1989-1.
Variables other than the commercial paper rate are measured as quarterly rates of change.
The figures in brackets are standard errors.
F is the F-statistic and LR the likelihood ratio statistic for the hypothesis that four lags of a variable are
insignificant in explaining changes of real GNP.
* indicates significance at the 5 percent level.
** indicates significance at the 1 percent level.
169
Table 17A
"Causality" Tests: GNP Deflator
Lags
GNP Deflator
1
2
3
0.38*
0.06
(0.10)
(0.11)
0.19
(0.10)
0.07
(0.13)
0.12
0.06
(0.10)
(0.13)
0.11
-0.02
(0.09)
(0.10)
F
28.91**
1.29
LR
85.00**
5.46
4
Table 17B
"Causality" Tests: GNP Deflator
Lags
GNP Deflator
L
CPR
0.35"
(0.10)
0.03
(0.11)
0.001**
(0.0004)
0.22*
(0.10)
0.10
(0.13)
-0.0002
(0.0004)
0.06
(0.10)
0.05
(0.13)
0.0004
(0.0004)
0.16
(0.09)
-0.04
(0.10)
0.0006
(0.0004)
F
28.59**
1.13
2.66*
LR
86.63**
4.98
11.42*
Notes: Sample period: 1959-1 to 1989-1.
Variables other than the commercial paper rate are measured as quarterly rates of change.
The figures in brackets are standard errors.
F is the F-statistic and LR the likelihood ratio statistic for the hypothesis that four lags of a variable are
insignificant in explaining changes of the GNP deflator.
* indicates significance at the 5 percent level.
** indicates significance at the 1 percent level.
170
Table 17C
"Causality" Tests: GNP Deflator
GNP Deflator
L
G
CPR
0.34"
(0.10)
0.034
(0.11)
-0.0006
(0.001)
0.001*
(0.0004)
0.22*
(0.10)
0.095
(0.14)
-0.0002
(0.001)
-0.0003
(0.0004)
0.06
(0.10)
0.048
(0.13)
-0.0003
(0.001)
0.0003
(0.0004)
0.17
(0.10)
-0.03
(0.11)
0.0001
(0.0013)
0.0006
(0.0004)
F
24.92**
24.92*"
1.04
0.07
2.36
LR
80.79"
4.79
0.32
10.55*
Table 18A
"Causality" Tests: GNP Deflator
Lags
1
GNP Deflator
0.42"
M2
0.02
(0.10)
(0.07)
2
0.22*
(0.10)
0.03
(0.08)
3
0.14
(0.10)
0.009
(0.08)
4
0.11
-0.02
(0.10)
(0.06)
46.88**
0.21
117.45**
0.90
F
LR
Notes: Sample period: 1959-1 to 1989-1.
Variables other than the commercial paper rate are measured as quarterly rates of change.
The figures in brackets are standard errors.
F is the F-statistic and LR the likelihood ratio statistic for the hypothesis that four lags of a variable are
insignificant in explaining changes of the GNP deflator.
* indicates significance at the 5 percent level.
** indicates significance at the 1 percent level.
171
Table 18B
"Causality" Tests: GNP Deflator
Lags
F
LR
GNP Deflator
M2
0.35"
(0.10)
0.09
(0.08)
0.002"
(0.0004)
0.23*
(0.10)
0.03
(0.08)
0.0002
(0.0005)
0.07
(0.10)
0.09
(0.08)
0.0008
(0.0005)
0.17
(0.09)
-0.10
(0.07)
0.0009
(0.0005)
37.08"
1.33
4.03"
103.48"
5.86
16.87"
CPR
Table 18C
"Causality" Tests: GNP Deflator
Lags
GNP Deflator
CPR
M2
0.35"
(0.10)
0.09
(0.08)
-0.0004
(-.0013)
0.0013
(0.0004)
0.23*
(0.10)
0.02
(0.09)
0.0001
(0.0013)
0.0002
(0.0005)
0.07
(0.10)
0.09
(0.09)
-0.0002
(0.0013)
0.0008
(0.0005)
0.17
(0.01)
-0.10
(0.07)
0.0004
(0.0013)
0.0009
(0.0005)
F
33.82"
1.23
0.06
3.64*
LR
99.92"
5.63
0.28
15.93"
Notes: Sample period: 1959-1 to 1989-1.
Variables other than the commercial paper rate are measured as quarterly rates of change.
The figures in brackets are standard errors.
F is the F-statistic and LR the likelihood ratio statistic for the hypothesis that four lags of a variable are
insignificant in explaining the rate of change of the GNP deflator.
* indicates significance at the 5 percent level.
** indicates significance at the 1 percent level.
172
inflation. These results document a significant deterioration in the predictive power of the monetary
aggregates that occurred in the early eighties, at the same time that money demand equations were
breaking down. Before that time, Ml and L, and to a lesser degree M2, did convey significant
information about future movements in nominal GNP, while the commercial paper rate did not.82
V. Liquid Asset Measures as Constraints
A variable can play a role as a constraint in the policy process if it is possible to specify ex ante a
level, ratio, or rate of change for the variable that if breached in one direction or the other will signify
an egregious policy error. For example, consider ex post real short-term interest rates: negative real
rates are taken to signal unambiguously that policy is too expansionary.83 In general, "price
variables" show more promise than quantity variables in this role. A liquid asset measure may be
useful as a nominal anchor for monetary policy, but it is doubtful that one could find upper and lower
limits for its rate of growth that should never be violated.
VI. Conclusion
Three main rationales can be advanced to justify the use of liquid asset measures as indicators or
intermediate targets for monetary policy: First, it can be argued that the theoretical concept of money
has no well-defined empirical analogue. There exists a wide variety of financial assets that provide
monetary services, some of which are close substitutes for traditional transactions assets.
Consequently, policies aimed at controlling a narrow monetary aggregate may be ineffectual in
controlling final objectives, and policy makers may be misled in judging the stance of policy on the
basis of a narrow aggregate. Moreover, over time, financial innovations change the relative
attractiveness of different monetary assets and the extent to which particular assets provide monetary
services. Thus, since the set of assets that constitutes money is not clearly delineated and could
change rapidly over time, it would be prudent to use a broad monetary aggregate that internalizes
changes in the relative attractiveness and functions of different monetary assets. Second, capital
82
The F-statistic for the null hypothesis that four lags of (the rate of change of) a monetary
aggregate do not contribute significantly when added to a regression of four lagged changes in
nominal GNP itself and the six-month commercial paper rate is 4.54 for Ml, 4.38 for L, and 2.72 for
M2, compared with a critical value, at the 5 percent level, of less than 2.53.
83
It is not intuitively obvious, however, that a symmetrical upper bound for real rates can be
specified.
173
market imperfections imply that private agents' portfolio and spending decisions cannot be separated
and that households and nonfinancial business firms may be liquidity constrained. Thus, households'
and nonfinancial businesses' liquidity positions may provide information, which is not captured by
market interest rates and wealth positions, about their planned spending and their ability to finance
spending out of their own resources and through borrowing. A related justification for the use of
liquid asset measures may be that they may provide information about spending decisions, not because
liquid assets provide monetary services, but because the aggregates provide timely information about
lending to households and firms by financial institutions. Finally, some observers have argued that
there is no a priori reason to presume that a single monetary aggregate or asset price will capture all
the information pertinent to monetary policy decisions. Thus, policy makers can reduce the risk of
policy errors by using a number of policy indicators and intermediate targets, including a broad liquid
asset measure.
Recently a number of studies have provided empirical evidence for the existence of liquidity
constraints/credit rationing. Such constraints appear to play a significant role in explaining the
empirical failures of the permanent income hypothesis and the neoclassical theory of investment.
Moreover, for individual firms and households, credit rationing is closely correlated with liquidity
positions, while in the aggregate, the extent of credit rationing seems to be closely related to the stance
of monetary policy. The existence of capital market imperfections that give rise to credit rationing and
liquidity constraints could motivate the use of either liquid assets or credit aggregates or both in the
formulation of monetary policy. There is, nevertheless, no a priori reason to expect credit aggregates
and liquid asset measures to convey the same information; holdings of liquid assets may convey
information about spending decisions even when firms and households are not rationed. Whether a
credit aggregate or a liquid asset aggregate would be more useful is an empirical matter.
The Board's liquid asset measure, L, is currently defined to include the assets in M3 plus the
nonbank public's holdings of U.S. savings bonds, short-term Treasury securities, commercial paper,
and bankers' acceptances. There is considerable overlap between L and other high-level financial
aggregates, both quantitatively and in terms of the rationale for their use in the formulation of
monetary policy. L is only 20 percent larger than M3 and essentially represents the liabilities
associated with the short-term component Total Nonfinancial Sector Debt. Consequently, L is highly
correlated with these aggregates and may convey little information not provided by these aggregates.
Two measurement problems complicate the definition of a liquid asset aggregate corresponding to
the theoretical motivation for its use in formulating policy: the assets to be included in a liquid asset
174
measure and the aggregation principle to be used in compiling the aggregate. If a liquid asset
aggregate is to measure readily available spending power in the economy, a potentially serious
measurement error arises from the exclusion of prearranged lines of credit from L. Exploratory
calculations indicate that unused lines of credit currently amount to about 25 percent of L and that this
percentage is growing. However, we do not have sufficient data to assess whether unused lines of
credit have a stable relationship with L and with plausible determinants or whether their increased use
might limit the usefulness of L.
Currently, L is calculated as a simple sum aggregate; that is, the aggregate is compiled by adding
the dollar values of holdings of its component assets. Simple sum aggregation is, strictly speaking,
only a valid procedure if assets are perfect substitutes. If the assets included in an aggregate are
imperfect substitutes, both the demand for the aggregate and its relationship with economic activity
will not be independent of its composition--consequently, it may be of limited value to policy makers.
The Board's staff has experimented with monetary-services, or Divisia, indexes as alternatives to the
conventional simple sum aggregates. In a Divisia index each asset is weighted proportionately to the
value of the monetary services that it provides. The value of the monetary services equals the
opportunity cost of holding the asset rather than a higher yielding benchmark asset. Although
theoretically elegant, Divisia aggregation has foundered on practical problems associated with the
measurement of the (unobservable) ex ante return on the benchmark asset. The Divisia measure of L
performed well in comparison with various simple sum aggregates in empirical tests based on data
from the 1970s, but this promising performance was not maintained with post-1980 data. In fact, the
Divisia aggregates fared so poorly compared with simple sum aggregates that the Board has
discontinued their compilation.
A successful intermediate target is related in a stable and predictable manner to both policy
objectives and policy instruments. To assess whether a liquid asset measure would satisfy this
requirement, we conducted an empirical analysis using the Board's liquid asset measure L. We found
a stable long-run demand for real L as a function of real GNP and the Moody's AAA corporate bond
yield. The relationship that captures the dynamic adjustment of L to changes in its determinants, in
contrast, has shifted in 1980. The nature of this shift accords well with our priors about the potential
effects of the financial changes that occurred in the early eighties: the volume of L outstanding
adjusts less rapidly than it did earlier, especially to changes in long-term interest rates. Our
interpretation is that financial deregulation and innovations have freed own rates on liquid assets to
equate supply and demand, thus requiring less pronounced quantity adjustment. Since we cannot rule
175
out further financial innovation, we cannot rule out future shifts in this relationship. A disappointing
out-of-sample simulation performance indicates that the short-run dynamic relationship between real L
and its determinants is captured poorly. This poor performance, the likelihood of future shifts in the
dynamic relationship, and the one-month time lag in the availability of data on L are probably
sufficient to preclude its use as an intermediate target.
The L multiplier is less volatile and seems easier to forecast than, for example, the M2 multiplier.
It is not clear, however, that this advantage will be of practical consequence or can be exploited by a
policy regime in which the base is adjusted to hit an intermediate target for L.
A liquid asset measure can be used as a monetary policy indicator in two conceptually distinct
roles: (a) as an indicator of the state of the economy, that is, as a source of information on current and
potential future economic developments that may necessitate policy changes, and (b) as an indicator of
the stance of monetary policy. The potential usefulness of a measure such as L in the latter role
would depend on the specifics of the monetary policy-making setup in place, but evidence from the
period of monetary tightening in 1980 and 1981 suggests that L may at times provide misleading
signals about the stance of policy. L also does not seem to hold much promise as an information
variable: In vector autoregressions, four lagged values of L have only marginal predictive value for
nominal GNP once the commercial paper rate is included in the information set. L does not predict
real GNP or the general price level once lagged values of the dependent variable and of the
commercial paper rate are taken into account in making forecasts. L's poor performance in the VAR
tests, at least as far as nominal GNP and the price level are concerned, may be attributable to the fact
that the GNP deflator and L have different time series properties. Tests conducted in a cointegrationerror correctionframeworkdo indicate that L improves predictions of both real GNP and the inflation
rate, although the improvement in predictive ability is rather small.
Appendix: The Radcliffe View on Monetary Policy
The findings of the Radcliffe Committee are often invoked to justify the use of liquid asset
measures in designing monetary policy. On a closer reading, however, the "Radcliffe view" could as
easily be construed as a justification for using credit measures or interest rates. Ironically, the Radcliffe
Committee findings bearing most directly on liquid assets, those that relate to the substitutability of the
liabilities of bank and nonbank liabilities, have already been incorporated in U.S. monetary policy
176
through the redefinition of the monetary aggregates. In this appendix, we provide a summary of some
of the main conclusions of the Radcliffe Report to support our interpretation of it.
Five elements are central to the "Radcliffe Committee view" of the role of money and financial
institutions in a modern industrial economy:
(i)
Monetary policy is transmitted to demand through the availability of financing, not
through interest rates: "In theory, monetary action works upon total demand by
changing the interest incentive; we believe that only very limited reliance can be placed
on this" (paragraph 397); "The monetary authorities may bring to bear another influence
which can be altogether more peremptory. This is the availability of funds through
particular channels" (paragraph 387); "We may call the first effect the interest incentive
effect, and the second the general liquidity effect" (paragraph 385).
(ii)
The control of the (narrowly defined) money supply is of secondary importance because
private agents can easily substitute credit for money and because nonbank financial
institutions create credit on the basis of interest-bearing near-money deposits: "It is the
whole liquidity position that is relevant to spending decisions, and our interest in the
supply of money is due to its significance in the whole liquidity picture. A decision to
spend depends not simply on whether the would-be spender has cash or money in the
bank. . . . There is the alternative of raising funds either by selling an asset or by
borrowing. . . . The ease with which money can be raised depends on the one hand upon
the composition of the spender's assets and on his borrowing power and on the other
upon the methods, moods and resources of financial institutions and other firms which
are prepared . . . to finance other people's spending" (paragraph 389).
(iii)
Control over the (narrowly defined) quantity of money outstanding is ineffective,
because nonbank financial institutions create credit on the basis of interest-bearing nearmoney deposits: "We would nevertheless emphasize that the amount of money, in the
sense of the amount of notes and bank deposits, is of considerable significance. The
other classes of liquid assets . . . are inferior in convenience to the holders, and this
inferiority will have to be compensated by the payment of interest. If there is less
money to go round. . . , rates of interest will rise. But they will not have to rise by
much, because in a highly developed financial system . . . there are many highly liquid
assets which are close substitutes for money, as good to hold and only inferior when the
actual moment of payment arrives" (paragraph 392).
(iv)
Because the structure of interest rates has an important influence on the lending behavior
of financial institutions and hence on the institutions' overall liquidity position, it should
be the focus of monetary policy. "A rise in interest rates, quite apart from any direct
effect it may have on the demand for investment goods, may have appreciable effects on
the behavior of financial institutions. . . . A movement of interest rates implies
significant changes in the capital values of many assets held by financial institutions. A
rise in rates makes some less willing to lend because capital values have fallen, and
others because their own interest rates are sticky. A fall in rates, on the other hand,
strengthens balance sheets and encourages lenders to seek new business. This is an
important conclusion for policy, for it implies . . . that movements in interest rates have
an effect apart from any influence they have on the incentive to hold capital goods"
177
(paragraph 393). As this passage illustrates, the committee implicitly argues that credit
is not rationed by interest rates alone and that some borrowers may be constrained by
the willingness of financial institutions to lend. Indeed, the Committee goes on to
emphasize the importance of bank credit: "In the liquidity structure as we have described
it, the banks have a special position, in that they are, for most borrowers and for most
purposes, much the most convenient institutional source of funds and often the only
source" (paragraph 395).
(v)
Finally, the Committee advocates the use of selective credit controls-primarily on
banks, because of the perceived importance of the banks in providing credit that would
otherwise be unavailable—as a means of monetary control for emergency situations. "In
times of emergency they [the monetary authorities] may choose to exercise a direct
control over the lending operations of the banks. We emphasize that any such special
concern, and any such extreme measures, are to be aimed at the banks as key lenders in
the system, and not at the banks as creators of money. It is the level of bank advances
rather than the level of bank deposits that is the object of this special interest; the
behavior of bank deposits is of interest only because it has some bearing, along with
other influences, on the behavior of other lenders."
Bibliography
Auerbach, Alan J., and Kevin Hassett. "Corporate Savings and Shareholder Consumption."
University of Pennsylvania, February 1989, mimeo.
Anderson, Leonal C, and James L. Jordan. "Monetary and Fiscal Actions: A Test of Their
Relative Importance in Economic Stabilization." Federal Reserve Bank of St. Louis Review, vol. 50
(1968), pp. 1-24.
Avery, Robert B., Gregory E. Elliehausen, Arthur B. Kennickel, and Paul A. Spindt. "Changes
in the Use of Transactions Accounts and Cash from 1984 to 1986." Federal Reserve Bulletin. March
1987, pp. 179-96.
Barnett, William A. "Economic Monetary Aggregates: An Application of Aggregation and Index
Number Theory." Journal of Econometrics. 1980, pp. 11-48.
. "The Optimal Level of Monetary Aggregation." Journal of Money. Credit, and
Banking, vol. 14, no. 4 (November 1982), part 2, pp. 687-710.
Barnett, William A., Edward K. Offenbacher, and Paul A. Spindt. "The New Divisia Monetary
Aggregates." Journal of Political Economy, vol. 92, no. 6 (1984), pp. 1049-85.
. "Empirical Comparisons of Divisia and Simple Sum Monetary Aggregates." Board
of Governors of the Federal Reserve System, Special Studies Staff Paper no. 158, July 1981.
178
Belongia, Michael T., and James A. Chalfant. "The Changing Empirical Definition of Money:
Some Estimates from a Model of the Demand for Money Substitutes." Joumal of Political Economy,
vol. 97, no. 2 (1989), pp. 387-97.
Bennett, David C , Flint Brayton, Eileen Mauskopf, Edward K. Offenbacher, and Richard D.
Porter. "Econometric Properties of the Redefined Monetary Aggregates." Board of Govemors of the
Federal Reserve System memorandum, February 1980.
Berkman, Neil G. "Abandoning Monetary Aggregates." Controlling Monetary Aggregates in.
Federal Reserve Bank of Boston, Conference Series no. 23, 1980.
Campbell John Y., and N. Gregory Mankiw. "Consumption, Income and Interest Rates:
Reinterpreting the Time-Series Evidence." Princeton University, April 1989, mimeo.
Canner, George C , Charles A. Luckett, and Thomas A. Durkin. "Home Equity Lending,"
Federal Reserve Bulletin, May 1989, pp. 333-44.
Canner, George C , Thomas A. Durkin, and Charles A. Luckett. "Recent Developments in the
Home Equity Loan Market." Journal of Retail Banking. Summer 1989, pp. 35-47.
Cohen, D. "A Comparison of Fiscal Measures Using Reduced-Form Techniques." Board of
Governors of the Federal Reserve System, Division of Research and Statistics, 1989, mimeo.
Cooley, Thomas F., and Stephen Leroy. "Atheoretical Macroeconometrics: A Critique." Journal
of Monetary Economics, vol. 16, pp. 283-308.
Engle, Robert F., and Clive W. J. Granger. "Cointegration and Error Correction:
Representation, Estimation, and Testing." Econometrica, vol. 55 (1987), pp. 251-76.
Engle, Robert F., and Byung Sam Yoo. "Forecasting and Testing in Cointegrated Systems,"
Journal of Econometrics, vol. 35 (1987), pp. 143-59.
Fair, R.C. The Specification. Estimation, and Analysis of Macroeconometric Models. Cambridge:
Harvard University Press, 1984.
Fazzari, Steven M., R. Glen Hubbard, and Brace C. Petersen. "Financing Constraints and
Corporate Investment." Brookings Papers on Economic Activity. 1: 1988, pp. 141-206.
Flavin, Marjorie. "The Adjustment of Consumption to Changing Expectations About Future
Income." Journal of Political Economy, vol. 89 (1981), pp. 974-1009.
. "Excess Sensitivity of Consumption to Current Income: Liquidity Constraint or
Myopia." Canadian Journal of Economics. February 1985, pp. 117-36.
Friedman, Benjamin M. "Lessons on Monetary Policy from the Eighties." Journal of Economic
Perspectives. Summer 1988, pp. 51-72.
179
. "Monetary Policy Without Quantity Variables." American Economic Review. May
1988 (Papers and Proceedings of the 100th Annual Meeting of the American Economic Association,
December 1987), pp. 440-45.
. "Targets and Instruments of Monetary Policy." National Bureau of Economic
Research, Working Paper no. 2668. July 1988.
. "The Inefficiency of Short-run Monetary Targets for Monetary Policy." Brookings
Papers on Economic Activity, 2: 1977, pp. 293-335.
. "Targets, Instruments and Indicators of Monetary Policy." Journal of Monetary
Economics, October 1975, pp. 443-73.
Friedman, Milton, ed. Studies in the Quantity Theory of Money. Chicago: University of Chicago
Press, 1956.
Fuller, Wayne A., Introduction to Statistical Time Series. New York: Wiley, 1976.
Gertler, Mark L. "Financial Structure and Aggregate Economic Activity: An Overview." Journal of
Money, Credit, and Banking, vol. 20, no. 3 (August 1988), part 2, pp. 559-88.
Gertler, Mark L., and R. Glen Hubbard. "Financial Factors in Business Fluctuations." First
Boston Working Paper no. FB-88-37. September 1988.
Goldfeld, Stephen A. "Comment on Barnett." Journal of Money, Credit, and Banking.
November 1982, part 2, pp. 716-20.
Granger, Clive W. J., and Paul Newbold. "Spurious Regressions in Econometrics." Journal of
Econometrics, vol. 2 (1974), pp. 111-20.
Hall, Robert E., and Frederic S. Mishkin. "The Sensitivity of Consumption to Transitory
Income: Evidence from Panel Data." Econometrica, vol. 50 (1982), pp. 461-81.
Jackman, Richard, and John Sutton. "Imperfect Capital Markets and the Monetarist Black Box:
Liquidity Constraints, Inflation and the Asymmetric Effects of Interest-Rate Policy." The Economic
Journal, March 1982, pp. 108-28.
Judd, John P., Brian Motley, and Bharat Trehan. "The Demand for Money: Where Do We Stand?"
Federal Reserve Bank of San Francisco, Working Paper no. 88-02, July 1988.
Kaldor, Nicholas. The Scourge of Monetarism. New York: Oxford University Press, 1982.
Lindsey, David E., and Paul A. Spindt. "An Evaluation of Monetary Indexes." Board of
Governors of the Federal Reserve System, Special Studies Paper no. 195. March 1986.
Lucas, Robert E., Jr. "Econometric Policy Evaluation: A Critique." In Karl Brunner and Alan
Meltzer, eds., The Phillips Curve and Labor Markets. 1976.
180
McCallum, Bennett T. "Targets, Indicators, and Instruments of Monetary Policy." Monetary
Policy in an Era of Change. American Enterprise Institute, October 1988.
Modigliani, Franco, and Lucas Papademos. "The Structure of Financial Markets and the
Monetary Mechanism." Controlling Monetary Aggregates III. Federal Reserve Bank of Boston,
Conference Series no. 23, 1980.
Morris, Frank E. "Do the Monetary Aggregates Have a Future as Targets of Federal Reserve
Policy?" New England Economic Review, March-April 1982, pp. 5-14.
. "Monetarism without Money." New England Economic Review, March-April 1983,
pp. 5-9.
Offenbacher, Edward K., and Richard D. Porter. "Empirical Comparisons of Credit and
Monetary Aggregates using Vector Autoregression Methods." Board of Governors of the Federal
Reserve System, Division of Research and Statistics, Special Studies Paper no. 181, October 1983.
Patinkin, Donald. "Financial Intermediaries and the Logical Structure of Monetary Theory."
American Economic Review, March 1961, pp. 95-116.
Pissarides, Christopher A. "Liquidity Considerations in the Theory of Consumption." Quarterly
Journal of Economics, May 1978, pp. 279-96.
Poole, William A. "Monetary Policy Lessons of Recent Inflation and Disinflation." Journal of
Economic Perspectives, Summer 1988, pp. 73-100.
Said, Said E., and David A. Dickey. "Testing for Unit Roots in Autoregressive Moving Average
Models of Unknown Order." Biometrika, vol. 71 (1984), pp. 599-607.
Sayers, Richard S. "Monetary Thought and Monetary Policy in England." Economic Journal,
December 1960, pp. 710-24.
Scheinkman, Jose A., and Laurence Weiss. "Borrowing Constraints and Aggregate Economic
Activity." Econometrica, January 1986, pp. 23-45.
Simpson, Thomas D., and Richard D. Porter. "Some Issues Involving the Definition and
Interpretation of Monetary Aggregates." Controlling Monetary Aggregates HI. Federal Reserve Bank
of Boston, Conference Series no. 23, 1980.
Sims, Christopher A. "Money, Income and Causality." American Economic Review, vol. 62
(1972), pp. 540-52.
Spindt, Paul A. "Money is What Money Does: Monetary Aggregation and the Equation of
Exchange." Journal of Political Economy, vol. 93, no. 1 (February 1985), pp. 175-204.
Stiglitz, Joseph E. "Money, Credit, and Business Fluctuations." National Bureau of Economic
Research, Working Paper no. 2823, January 1989.
181
Stiglitz, Joseph E., and Andrew Weiss. "Credit Rationing in Markets with Imperfect
Information." American Economic Review, vol. 71, no. 3 (June 1981), pp. 393-410.
Tobin, James. "Money Capital and Other Stores of Value." American Economic Review, May
1961 (Papers and Proceedings of the 73rd Annual Meeting of the American Economic Association,
December 1960), pp. 26-37.
. "Commercial Banks as Creators of Money." In Dean Carson, ed., Banking and
Monetary Studies, pp. 408-19. Homewood, HI.: Richard D. Irwin, 1963.
. "A General Equilibrium Approach to Monetary Theory." Journal of Money, Credit,
and Banking, vol. 1, no. 1 (1970).
. "Money and Finance in the Macroeconomic Process." Journal of Money, Credit,
and Banking, vol. 14, no. 2 (May 1982), pp. 171-204.
Tobin, James, and William Brainard. "Financial Intermediaries and the Effectiveness of
Monetary Policy." American Economic Review, May 1963 (Papers and Proceedings of the 75th
Annual Meeting of the American Economic Association, December 1962), pp. 383-400.
Tobin, James, and Walter Dolde. "Wealth, Liquidity and Consumption." Consumer Spending
and Monetary Policy: the Linkages, Federal Reserve Bank of Boston, Conference Series no. 5, 1971.
Tobin, James, and Donald Hester. Financial Markets and Economic Activity. New York: Wiley,
1967.
White, Halbert. "A Heteroskedasticity-Consistent Covariance Matrix Estimator and a Direct
Test for Heteroskedasticity." Econometrica, vol. 48, pp. 817-38.
Wilcox, James A. "Consumption Constraints: The Real Effects of 'Real* Lending Policies."
Paper presented at the Federal Reserve System Committee Meeting on Business Conditions, Spring
1989.
Zeldes, Stephen P. "Consumption and Liquidity Constraints: An Empirical Investigation."
Journal of Political Economy, vol. 97, no. 2 (April 1989), pp. 305-46.
182
A REVIEW OF CREDIT MEASURES AS A POLICY VARIABLE
Lawrence J. Radecki
This paper will review the potential usefulness of measures of credit in formulating and
communicating Federal Reserve policy. To do so efficiently, the research on credit measures must be
divided at the outset into two groups. One group is focused on broad measures of credit; the analyses
and proposals of Henry Kaufman and Benjamin Friedman, among others, fall into this group. The
second is focused on bank credit, the work of Ben Bernanke, Mark Gertler, and Joseph Stiglitz is
representative of this second group.
The research on credit by both groups breaks from the tradition of focusing on money or
money and interest rates. The motivation for turning to credit measures is somewhat the same whether
a broad credit measure or a bank credit measure is the proposed alternative to the money supply.
Among the reasons given for moving to credit measures are: diversification of the financial quantities
on which to focus, instability in the money-income relationship, and a belief that financial markets
originate or amplify disturbances that subsequently affect the macroeconomy. Despite these common
reasons for studying credit measures, macroeconomic analysis utilizing broad measures of credit is
quite different in several regards from analysis utilizing bank credit. In the body of this paper these
areas of difference will be fully delineated. They include:
1.
The basic rationale for considering credit:
• Broad credit measures are an alternative or supplement to monetary aggregates, on which
inordinate emphasis has been placed in theoretical and applied economic research.
• Bank credit is "special" in the sense that it is pivotal in the transmission mechanism for
monetary policy and very useful in explaining macroeconomic fluctuations.
2.
The definition of the most relevant credit measure:
• It must be broad because credit is fungible, that I, one form easily substitutes for another.
• It must be narrow and focused on bank credit, the portion of total credit which is special
and vitally important. For some sectors of the economy, bank credit is the only type of
credit available to finance spending plans.
183
3.
The role of credit in the formulation of central bank policy:
• Total credit warrants equal weight with monetary aggregates in policy decisions, either as
an indicator for policymakers or as an intermediate target.
• Bank credit may be a good leading indicator of business cycle turning points, particularly
of downturns. It may be, however, too narrow to serve as an intermediate target. Besides,
the use of nonprice devices to allocate credit may be prevalent and thus may introduce
nonlinearities into the relationship between the volume of bank loans and economic
activity.
4.
Credit measures and the level of economic activity:
• Total debt has been shown to have had as tight a relationship with nominal GNP as money.
Moreover, the demand for money has been distorted by innovations and deregulation.
Thus, it may be a better financial quantity on which to base long-run policy decisions.
• The availability of bank credit may be a crucial factor in the amplification of disturbances
originating elsewhere in the economy, and crucial in transforming a mild recession into a
deep downturn. But it may be a poor variable to focus on for the purpose of long-term
control of inflation.
Not all economists who stress the role of the credit markets in the business cycle fit neatly into
either of the above groups. Some fit into neither, others straddle the two groups. Among these
somewhat harder-to-classify economists, Charles Kindleberger has analyzed the tendency of financial
markets to engage in speculation resulting in a subsequent crash.1 Hyman Minsky has emphasized the
fragility of the banking system and the pyramidal structure of debt.2 Albert Wojnilower has described
the role of credit "crunches" in the postwar economy.3 While these three economists have attached
great importance to the performance of the financial markets for the real sector of the economy, they
do not recommend reliance on credit quantities in formulating central bank policy.
It should be noted that the Federal Open Market Committee (FOMC) has at times during the past
thirty years used credit quantities in its policy making.4 At the strategy level of policy, the
Carles Kindleberger, Panics, Manias, and Crashes (New York: Basic Books, 1978).
2
Hyman P. Minsky, John Maynard Keynes (New York: Columbia University Press, 1975).
3
Albert M. Wojnilower, "The Central Role of Credit Crunches in Recent Financial History,"
Brookings Papers on Economic Activity, 1:1980, pp. 277-326.
4
Ann-Marie Meulendyke, "A Review of Federal Reserve Policy Targets and Operating Guides in
Recent Decades," in this volume.
184
Committee has been setting a monitoring range for a broad credit aggregate, Total Domestic
Nonfinancial Debt, since 1983; for a period in the 1970s it set a range for a proxy measure of bank
credit. And at the tactical level, during a period in the late 1960s the domestic policy directive to the
Federal Reserve Bank of New York contained a proviso clause stating that open market operations
should be conducted somewhat differently depending on the growth of bank credit. For example,
directives from late 1966 state:
To implement this policy,.. . System open market operations . . . shall be conducted
with a view toward maintaining firm but orderly conditions in the money market;
provided, however, that operations will be modified in the light of . . . any apparently
significant deviations of bank credit from current expectations.
To implement this policy, System open market operations . . . shall be conducted with a
view toward somewhat easier conditions in the money market, unless bank credit
appears to be resuming a rapid rate of expansion.5
I. Broad Measures of Credit
A. The Rationale for Their Use
There are several reasons why economists have argued for giving more consideration to broad
measures of credit in their applied research on macroeconomics. The reasons are interrelated and not
always easily distinguished from one another, and they represent a mixture of theoretical and empirical
justifications. Nevertheless, we will try to identify five separate reasons.
First, some economists argue that there is no indisputable justification for focusing on any
particular financial quantity to the exclusion of others in the modern economy (with the possible
exception of the monetary base). Financial instruments-including money—are privately issued and
their dollar volumes are determined jointly by the actions of the financial sector and the public in a
general equilibrium setting. They are all subject to the influence of the central bank, but none is under
its tight control. A particular monetary aggregate represents one specific collection of the numerous
financial instruments currently in use; a credit aggregate represents a different collection of
instruments. Neither is inherently superior regarding the information it provides concerning
5
Minutes of the Federal Open Market Committee, October 4, 1966, and November 12, 1966.
185
developments in the financial markets or about the direction of central bank policy.6 Thus, the
emphasis on money in economic analysis and policymaking seems unwarranted.
The notable exception to this general argument is the monetary base, or outside money. In
contrast to inside money or credit, this financial quantity is under the nearly complete control of the
central bank. Its potential usefulness for central bank policy is examined in one of the other papers
written as part of this study.7
Second, specific arguments have been made that credit quantities should be given equal
importance with monetary quantities in macroeconomic research. In the context of a short-run
macroeconomic model with a simple financial structure, "the determination and control of nominal
income can be achieved without exogenous control of the money supply and require only exogenous
control of some financial aggregate, or of some linear combination of such aggregates."8 In other
words, control by the central bank over any one of several financial aggregates would seem to be
sufficient to control nominal GNP--money is not crucial for this purpose. The variable the central
bank uses could be narrowly or broadly defined money, but it could also be the monetary base, total
debt, or bank credit. Which would be best among the financial aggregates for policy purposes would
depend on the exact structure of the financial system, key parameter values in the economy, and the
relative uncertainty among behavioral relationships.9
Third, there is considerable skepticism that a single quantity can sufficiently capture most of the
relevant information on developments in the financial markets, or that the dollar volume of either the
assets or liabilities of commercial banks is sufficiently comprehensive. This skepticism is based in
part on the obvious complexity of the domestic and international financial systems and their
6
Warren L. Smith, "A Neo-Keynesian View of Monetary Policy," Controlling the Monetary
Aggregates, Federal Reserve Bank of Boston, Conference Series no. 1, 1969, pp. 105-26; and
Benjamin M. Friedman, "The Role of Money and Credit in Macroeconomic Analysis," in James Tobin,
ed., Macroeconomics, Prices, and Quantities: Essays in Honor of Arthur M. Okun (Washington, D.C.:
Brookings Institution, 1983), pp. 161-99.
7
See Ann-Marie Meulendyke, "Possible Roles for the Monetary Base."
8
Franco Modigliani and Lucas Papademos, "The Structure of Financial Markets and the Monetary
Mechanism," Controlling Monetary Aggregates, Federal Reserve Bank of Boston, Conference Series
no. 23, 1980, p. 129.
9
See Ben S. Bernanke and Alan S. Blinder, "Credit, Money, and Aggregate Demand," American
Economic Review, May 1988, pp. 435-39; and Friedman, "The Role of Money and Credit in
Macroeconomic Analysis," p. 182.
186
connections with the real side of the economy. It is also based in part on experience and intuition
regarding the importance of financial developments for the performance of the real sector. For
example, Paul Volcker explicitly stated:
More broadly, I think the intellectual emphasis on monetary aggregates that
developed in the 1960's threatened to~and in some occasions did-go too far in
implying that credit markets—broadly defined~"don't count"—that they are never or
seldom a source of disturbance in the economy or a legitimate concern of policy.
Indeed, I suspect little attention directed toward serious and systematic analysis of the
role of the credit markets, toward the financial complexities of the economy generally,
and toward their disruptive potential is a common failing of most modern theorizing,
regardless of the intellectual starting point.
We have had many occasions in the 1970's to pay the closest possible attention to
particular financial problems and to the potential vulnerabilities of various credit
markets. I would remind you of the recurrent concerns about thrift institutions and the
mortgage market, Penn Central and commercial paper, Herstatt and the Euro-dollar
market, New York City and the municipal bond market, and the rising commercial bank
loan losses a year ago.10
The perceived importance of developments on the financial side of the economy is also
reflected in the construction of large-scale econometric models of the U.S. economy, in which several
financial variables are typically included. And the collapse of the banking system has long been
thought to have played a significant role in the Great Depression.
Fourth, the instability in the demand for various monetary aggregates during the past fifteen
years has reduced their value as an intermediate target for monetary policy and spurred interest in
exploring other financial quantity variables. At this time we cannot tell when or whether this
instability in money demand might end. The abrupt and sharp slowing of the growth of the money
supply (Ml and M2) during 1987 and the first half of 1988 was not soon thereafter followed by a
recession in 1988 or 1989.11
10
Remarks by Paul A. Volcker, "The Contributions and Limitations of Monetary Analysis," before
the Joint American Economic Association and the American Finance Association Luncheon,
September 16, 1976. Reprinted in the Federal Reserve Bank of New York Quarterly Review, 75th
Anniversary Issue, pp. 35-41.
11
Ml grew at a rate of 13.1 percent over the four quarters ending 1986-11 and at a rate of 14.3
percent over the four quarters ending 1987-11. Ml growth slowed markedly to 3.8 percent over the
next four quarters, a signal of an imminent recession. M2 showed a similar pattern. It grew 9.0
percent and 9.3 percent over the four quarters ending 1986-IV and 1987-IV, respectively. It then grew
only 4.2 percent in the following four quarters.
187
Fifth, for determining aggregate demand, some economists conclude that the availability of
credit is crucial, not the availability of money.12 That is, a constraint placed on the quantity of credit
would be more disruptive than a restriction on currency holdings or checking account deposits.
Aggregate demand was adversely affected when the ceilings imposed by Regulation Q on the interest
rates banks could pay on savings and time deposits became binding at various times in the 1960s and
1970s. The effect of deposit rate ceilings was substantively similar to a credit supply stoppage.13
In sum, there are several good reasons to pay closer attention to broad credit aggregates and
maintain a balance between money and credit in macroeconomic analysis. Why then has so much
attention been paid to money and relatively little to credit? There seem to be several factors:
• In widely used macroeconomic models, the market for money appears in the foreground
and the market for credit in the background. The credit market can be set aside by virtue
of the balance sheet constraint on total financial assets.
• The money supply and interest rates are more readily observable than credit quantities.
• There is less ambiguity as to what should be included in a narrow monetary aggregate,
defined in terms of the payments function of money.
• Money is thought to be exogenous to some degree and more readily under the control of
the central bank, whereas credit is unquestionably an endogenous variable.
B. Measurement Issues
Since credit is highly fungible, there is widespread agreement with the notion that a broad
measure of credit must be genuinely comprehensive to be useful in macroeconomic analysis. It should
definitely include all the short-term and long-term credit extended to domestic nonfinancial private
economic units (households and nonfinancial businesses) and to most if not all public economic units.
As such it would consist of the dollar volume of outstanding government securities, corporate bonds,
mortgages on residential, commercial, and farm properties, bank loans, consumer credit, and
commercial paper. Although the aim is to be comprehensive, certain exclusions are almost always still
12
Alan S. Blinder and Joseph E. Stiglitz, "Money, Credit Constraints, and Economic Activity,"
American Economic Review, May 1983, (Papers and Proceedings), pp. 297-302.
13
The relevance of the economy's behavior following the implementation of the Credit Control
Program of 1980 is questionable. The program did not directly impinge on consumer credit, but
consumption spending collapsed soon after its implementation as households seemed to cut back
voluntarily their demand for credit.
188
made. Total debt is netted for financial intermediation; that is, total debt excludes instruments issued
by banks, insurance companies, financial subsidiaries of nonfinancial corporations, and federally
sponsored credit agencies. Total debt is not usually netted, however, for the financial assets held by
individual households, businesses, and governmental units; therefore, gross liabilities are counted.
Total debt also usually excludes credit extended to foreigners since their borrowing is not directly
related to U.S. economic activity.
Applying these principles, but also making certain other judgments, the staff of the Board of
Governors developed the definition of Total Nonfinancial Sector Debt. As of the first quarter of 1989,
it was measured to be $9209.4 billion, of which 70 percent was the debt of the private sector and 30
percent was the debt of the public sector. As such, it was about six times as large as Ml, somewhat
more than double the size of M3, and almost exactly twice as large as L. If the debt of the federal
government is excluded, the resulting measure, nonfederal debt, is considerably smaller, standing at
$7053.7 billion (as of the first quarter 1989). Table 1 shows the composition of Total Nonfinancial
Sector Debt.
Two issues regarding the measurement of total debt are somewhat open. First, should the
aggregate include the debt issued by all levels of government (federal, state, and local), or should the
debt of the federal government be excluded? The exclusion of the federal portion of government debt
can be justified on the grounds that federal spending is unrelated to the cost of raising funds and that
the U.S. Treasury's ability to raise funds in the credit markets is virtually unbounded.14 Conversely,
it could be argued that the debt of state and local governments should be included since their spending
is constrained by both their ability to raise funds and the cost of borrowing. Despite the apparent
soundness of these a priori arguments, conclusions have generally been reached on empirical grounds.
Friedman pointed out that the credit-to-income ratio is much more stable when total debt is defined to
include federal as well as state and local government debt.15 It has since been customary to include
the debt of the federal government, as discussed further in the next section.
The second measurement issue is: Is a simple sum aggregate adequate for measuring total debt?
Total debt comprises an even more heterogeneous collection of assets than does M3. Economists have
doubted the usefulness of combining with equal weights the quantities of various bank deposits
14
Richard G. Davis, "Broad Credit Measures as Targets of Monetary Policy," Federal Reserve
Bank of New York Quarterly Review, Summer 1979, pp. 13-22.
15
Friedman, "The Role of Money and Credit in Macroeconomic Analysis."
189
Table 1
Credit Market Debt Outstanding
Levels at the End of
the First Quarter of 1989
(In Billions of Dollars)
Total Nonfinancial Sector Debt
(Total credit market debt
owed by domestic nonfinancial
sectors)
9,209.4
Alternative broad credit measures
Nonfederal Debt
(Total Nonfinancial Sector Debt
less the credit market liabilities
of the federal government)
7,053.7
Private debt
(Nonfederal debt less the credit
market liabilities of state and
local government units)
6,458.5
By sector
U.S. government
State and local
governments
2,155.7
595.2
Households
3,183.8
Nonfinancial businesses
3,274.6
Farm
Nonfarm noncorporate
Corporate
Source: Board of Governors of the Federal Reserve System.
190
140.1
1,213.6
1,920.9
(and their near substitutes) in a monetary aggregate and have tried instead more careful methods of
aggregation.16 These doubts concerning aggregation would have to be even stronger in the case of
broad debt measures, but there has been little discussion of the matter.17 The aggregation problem is
probably too complex to resolve; besides, the application of sounder aggregation methods has not
apparently led to greater usefulness of the monetary measures.18 Moreover, there are deeper
questions surrounding a broad credit aggregate, such as those raised by the new, complicated financial
contracts: Has credit in fact been extended?19 It is not always possible to tell in the case of certain
innovative interest rate swaps.
C. Empirical Research
The existence of an apparently stable relationship between a broad credit measure and nominal
GNP was first identified by Kaufman.20 He used a chart to illustrate a close relationship over the
period 1950-77 between nominal GNP and credit, measured by a proxy for total debt (as defined in
footnote 17). Soon thereafter, Richard Davis published a study applying statistical analysis to
investigate the relationship.21 (A brief summary of the specifications, econometric techniques, and
16
See William A. Barnett, "Economic Monetary Aggregation: An Application of Aggregation and
Index Number Theory," Journal of Econometrics, September 1980, pp. 11-48; William A. Barnett,
Edward Offenbacher, and Paul Spindt, "New Concepts of Aggregated Money," Journal of Finance,
1980, pp. 497-505; and William A. Barnett and Paul A. Spindt, "The Velocity Behavior and
Information Content of Divisia Monetary Aggregates," Economic Letters, 1979, pp. 51-57.
17
There is also the issue of the timeliness of credit data. Because they are collected from several
sources, credit data have a greater time lag than the monetary data. To shorten the time lag, Kaufman
suggested using a proxy, that is available earlier, namely the credit market instruments and deposits
held by the private domestic nonfinancial sector (households, businesses, and state and local
government). See the testimony by Henry Kaufman before the Committee of the Budget, U.S. House
of Representatives, February 6, 1978.
18
David E. Lindsey and Paul Spindt, "An Evaluation of Monetary Indices," Board of Governors of
the Federal Reserve System, Special Studies Paper no. 195, February 1986.
19
Shiller raised some additional questions regarding measurement, including whether market
values ought be used instead of par values, and whether preferred stock ought to be included. See the
comments of Robert J. Shiller, following Friedman, "The Role of Money and Credit in
Macroeconomic Analysis."
20
See the testimony of Kaufman before the House Committee on the Budget.
21
Davis, "Broad Credit Measures as Targets of Monetary Policy."
191
s
Table 2
Summary of Statistical Findings on Broad Credit Measures
Author
Definitions of
Credit Used
Statistical
Techniques
Sample
Period
Davis
(1979)
Total nonfederal
Debt proxy
Reduced-form
1961-77
Over the whole period, money and credit are equally correlated with
nominal GNP; over subperiods, correlations vary widely. In
regressions with both money and credit, credit is insignificant.
Friedman
(1983)
Total
Total nonfinancial
Total nonfederal
Private nonfinancial
Reduced-form,
St. Louis
approach,
VARs
1953-78
Total nonfinancial debt is as closely related as the monetary
aggregates to nominal GNP.
Brittian
(1981)
Bank deposits and
Reduced-form
credit market
instruments (United States,
Japan)
1951-75
1960-76
1964-76
1967-76
Main Results
(United States)
(Germany)
(Italy)
(Japan)
Money is the most suitable intermediate target in the United States
and Germany, credit is slightly more suitable in Italy, and
inconclusive results are found for Japan.
Total financial
instruments (Germany)
Total internal
credit (Italy)
Islam
(1982)
Total financial institutions Reduced-form
Total private
(Germany, Japan)
Total (United States)
1960-80 (United States)
1964-80 (Japan)
1969-80 (Germany)
Only moderate differences in fit and out-of-sample forecasts, but
total credit in the United States and total private credit in Japan
are among the most successful.
Table 2
Summary of Statistical Findings on Broad Credit Measures (continued)
Definitions of
Credit Used
Statistical
Techniques
Sample
Period
Total nonfinancial
Reduced-form
1959-80
1962-80
1962-80
1964-80
1963-79
Offenbacher- Total nonfinancial
Porter
(1983)
VAR
1954-82
The results favor monetary aggregates over credit aggregates but are
highly sensitive to slight modifications in specifications.
Hafer
(1984)
Total nonfinancial
St. Louis
approach
1960-81
Total debt makes no marginal contribution in an equation that
already includes Ml.
Fackler
(1985)
Total nonfinancial
VAR
1962-80
In a VAR system with money, credit, and interest rates, neither
money nor credit is significant; only interest rates are significant for
explaining nominal GNP.
Friedman
(1986)
Total nonfinancial
Bivariate
regressions
1891-82 (annual)
1952-82 (quarterly)
Overall, a lack of persistence over time in these simple relationships.
Author
Friedman
(1982)
NO
Main Results
Total nonfinancial debt has the most stable contemporaneous
(United States)
relationship with nominal income, while the narrow money supply
(Canada)
has the most stable dynamic relationship.
(Germany)
(Japan)
(United Kingdom)
findings of the several papers to be covered in this section is contained in Table 2.) Regressing
quarterly percentage changes in current dollar GNP from 1961 to 1977 on current and lagged
percentage changes in various monetary and credit aggregates, he found that the relationship between a
broad credit measure and GNP is roughly as close as the relationship between a monetary aggregate
and GNP.22 Davis* other results were less favorable to total debt. He found that in regressions
omitting the current value of the financial aggregate and using only lagged values, the explanatory
power of total debt falls to zero. And when he ran equations including both a monetary aggregate
variable and a credit aggregate variable, credit measures were not able to make a statistically
significant contribution to the explanatory power of the equation.23
A few years later Friedman published a more extensive study of broad credit aggregates.24 He
looked at several definitions: total nonfinancial debt, nonfederal debt, private debt, total debt (which
includes the credit market debt of U.S. financial intermediaries), and bank credit. Of these, total
nonfinancial debt was Friedman's preference. (This is the definition refined and adopted later by the
Board staff.) He produced a chart showing the ratio of total nonfinancial debt and its components to
nominal GNP over the 1946-80 period. The ratio of total nonfinancial debt to GNP, its income
velocity, shows surprising stability at a level just below 150 percent of GNP. The sector components
are, however, much less stable over the period as a whole. Most pronounced is the trend in federal
government debt, falling from about 100 percent of GNP in 1946 to around 25 percent in the first half
of the 1970s. The other components, in order to keep total debt near 150 percent of GNP, rose over
the same period from 50 percent of GNP to 125 percent. Friedman also noticed that bulges in federal
debt issuance during recessions have usually been accompanied by an abatement of private borrowing.
Friedman then investigated the relationship between the economy and total nonfinancial debt.
The statistical approaches used included estimating regressions similar to those run by Davis but using
22
He also uses bank credit, which will be discussed later.
23
Davis also ran the regressions over the subperiods 1961 to mid-1969 and mid-1969 to 1977 and
found for most of the monetary and credit aggregates that the closeness of their relationship to GNP
varies considerably between the two subperiods.
24
Friedman, "The Role of Money and Credit in Macroeconomic Analysis." Some of the
regressions were rerun over a later time period, see the results presented in "Monetary Policy with a
Credit Aggregate Target," Money, Monetary Policies and Financial Institutions, Carnegie-Rochester
Conference Series on Public Policy, Spring 1983, pp. 117-47.
194
data over a longer time period, regression equations based on the St. Louis reduced-form approach,
and vector autoregressions (VARs). Friedman summarized his findings in this way:
Results based on a variety of methodological approaches consistently indicate that the
aggregate outstanding indebtedness of all nonfinancial borrowers in the United States
bears as close and as stable a relationship to U.S. nonfinancial activity as do the more
familiar asset aggregates like the money stock (however defined) or the monetary
base.25
These results were not accepted uncritically. The most troublesome issue was whether the
observed constancy of the total debt-to-income ratio was a fluke, or whether it was the product of
economic forces and hence could be explained. Friedman offered three possible explanations for the
link between total debt and nominal income:
• An ultrarationality hypothesis: the public tries to maintain constant indebtedness relative to
its income, and in its calculations private debt substitutes perfectly for government debt.
• A capital leveraging hypothesis: because of constraints imposed by the credit markets, the
private sector can increase its total liabilities only to the extent that it has accumulated
tangible assets with which to back them.
• An asset demand hypothesis: the public tries to maintain a stable ratio of total financial
asset holdings to income, such that a relative decline in the supply of government debt is
offset by an increase in the demand for private liabilities.26
Robert Shiller was among those who were not convinced that the total debt-to-income ratio
would remain constant. He pointed out that the relative decline in government debt since World War
II was the consequence of a return to peacetime levels of defense expenditures.27 He did not see a
connection between lower defense spending and less federal government borrowing, on the one hand,
25
Friedman, "The Role of Money and Credit in Macroeconomic Analysis," p. 165. Using a
different econometric approach, Kopke showed that the velocity of total debt was more stable around
its trend over the period 1960 to 1982 than were the velocities of the monetary aggregates around their
trends. See Richard W. Kopke, "Must the Ideal Money Stock Be Controllable?" Federal Reserve Bank
of Boston New England Economic Review, March-April 1983, pp. 10-23.
26
Benjamin M. Friedman, "The Relative Stability of Money and Credit 'Velocities' in the United
States: Evidence and Some Speculations," National Bureau of Economic Research, Working Paper no.
645, 1981.
27
See the comments of Robert J. Shiller and Philip Cagan following Friedman, "The Role of
Money and Credit in Mafcrofcfconomic Analysis."
195
and higher private-sector spending and greater individual and corporate borrowing, on the other. If the
decline in federal government borrowing coincided, for example, with a curtailment of highway and
transportation expenditures by the public sector, offset by increases in private sector spending on
transportation facilities financed by borrowing, Friedman's argument would have been more
persuasive. Also skeptical was Philip Cagan, who thought that even if the capital stock-to-GNP ratio
were constant there was no reason to presume that the debt markets would provide a constant fraction
of the financing of capital expenditures. Thus the total debt-to-GNP ratio need not be constant.
Later studies of broad credit measures
Whatever the merits of these criticisms, interest in broad credit aggregates remained high.
Bruce Brittain and Shafiqul Islam extended the econometric work of Friedman on the U.S. economy
by comparing the money-to-income and credit-to-income relationships in other major economies.
Applying criteria of closeness-of-fit and exogeneity with respect to income, Brittain found mixed
results: in Germany and the United States broad money is the superior intermediate target, in Italy
total credit is marginally superior to broad money, and in Japan no firm conclusion can be drawn.28
In contrast, Islam's analysis of U.S., German, and Japanese data was more favorable to credit. He
found that credit aggregates in all three countries are as good as or slightly better than conventional
monetary aggregates in the stability and predictability of their relationships with nominal GNP.29
Taken together, the findings of Kaufman, Davis, Friedman, and Islam generally seemed to
support the view that broad credit aggregates should be given at least some weight in economic
research and central bank policy. And in 1983, amid deregulation of bank deposits and instability in
the demand for money, the Federal Open Market Committee (FOMC) began to set a monitoring range
for Total Domestic Nonfinancial Sector Debt. The Federal Reserve's decision to supplement the
monetary aggregates with a credit aggregate in its reports to Congress instigated further research on
28
Bruce Brittain, "Choosing between Money Targets and Targets for Credit," in Brian Griffiths
and Geoffrey E. Wood, eds., Monetary Targets (New York: St. Martin's Press, 1981), pp. 62-80.
29
Shafiqul Islam, "Monetary and Credit Aggregates and Economic Activity: Evidence from
Germany, Japan, and the United States," Federal Reserve Bank of New York, Research Paper no.
8203, January 1982. Friedman also analyzed data from foreign countries: Canada, Germany, Japan,
and the United Kingdom. Benjamin M. Friedman, "Money, Credit and Nonfinancial Economic
Activity: An Empirical Study in Five Countries," National Bureau of Economic Research, Working
Paper no. 1033, November 1982.
196
broad credit aggregates. None of the succeeding studies, however, proved to be particularly supportive
of total debt.
Edward Offenbacher and Richard Porter found that Friedman's empirical results are sensitive to
slight changes in data construction and the specifications of the VAR model.30 They concluded that
their estimates using the VAR methodology favor money aggregates over credit aggregates. Moreover,
they concluded that results from VARs cannot show the effects of the monetary authorities adopting a
specific policy for debt expansion.31
Friedman himself put his earlier work on credit in better perspective by extending the period of
his analysis back into the late ninteenth century.32 From this long-run study he confirmed that
economic fluctuations are reflected in the financial side of the economy. Financial quantities clearly
show procyclical behavior and have the tendency to lead turning points in the business cycle. But the
theme of his paper is that the quantitative relationships connecting monetary and financial variables to
the business cycle are not truly invariant. There appear to be strong and persistent regularities, but on
closer inspection the relationships turn out not to be so persistent or regular. The reasons he offered
were the evolution of monetary policy, changes in banking and securities regulations and in the tax
code, and financial innovations. Friedman felt that the implication for central bank policy was clear: it
is dangerous to base macroeconomic policy on the presumption that an observed ten- or twenty-year
regularity in the relationship between one or more financial variables and economic activity will
continue.
In 1985 it became clear that the previously observed relationship between credit aggregates and
30
Edward K. Offenbacher and Richard D. Porter, "Empirical Comparisons of Credit and Monetary
Aggregates Using Vector Autoregression Methods," Special Studies Paper no. 181, August 1983; and
Federal Reserve Bank of Richmond Economic Review, November-December 1983, pp. 16-29.
31
Others also showed the sensitivity of the results on broad credit aggregates: James S. Fackler,
"An Empirical Model of the Markets for Goods, Money, and Credit," Journal of Money, Credit, and
Banking, February 1985, pp. 28-42; and R.W. Hafer, "Money, Debt, and Economic Activity," Federal
Reserve Bank of St. Louis Review, June-July 1984, pp. 18-25, and "Choosing between Ml and Debt
as an Intermediate Target for Monetary Policy," Understanding Monetary Regimes, Carnegie-Rochester
Conference Series on Public Policy, Spring 1985, pp. 89-132.
32
Benjamin M. Friedman, "Money, Credit, and Interest Rates in the Business Cycle," in Robert
Gordon, ed., The American Business Cycle: Continuity and Change (Chicago: University of Chicago
Press, 1986), pp. 395-458.
197
Chart 1
Ratio of Total Nonfinancial Sector Debt to GNP
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
i
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988
Source: Federal Reserve System.
nominal GNP had deteriorated badly (Chart l). 33 Beginning in the second quarter of 1983, the ratio
of Total Nonfinancial Sector Debt to nominal GNP rose rapidly after exhibiting no trend previously.
The next year Kaufman listed the "close correlation between the growth of debt and the growth of
nominal GNP" as one of seven "fallen financial dogmas and beliefs."34 Friedman subsequently
remarked, after having seen the money-income and credit-income relationships deteriorate, that both
money and credit had become irrelevant for monetary policy.35
33
See John F. Wilson, Elizabeth M. Fogler, James L. Freund, and Guido E. van der Ven, "Major
Borrowing and Lending Trends in the U.S. Economy, 1981-85," Federal Reserve Bulletin. August
1986, pp. 511-24.
34
Henry Kaufman, Interest Rates, the Markets, and the New Financial World (New York: Times
Books, 1986), chap. 10.
35
Benjamin M. Friedman, "Monetary Policy Without Quantity Variables," American Economic
Review, May 1988, pp. 440^45.
198
The earlier doubts as to the constancy of the total debt-to-nominal GNP ratio appear to have
been correct. Changes in the banking system and the financial markets, related to both innovations
and deregulation, seem to have facilitated a financial deepening and widening-a rise in the amount of
marketable debt instruments plus credit created by financial intermediaries relative to income and
wealth in the economy. Hence the ratio has risen significantly since 1983. Instability in the demand
and supply of credit appears to have been as severe as instability in the demand and supply of money.
Updating the econometric results
To round out this review of broad credit aggregates, we repeat some of the earlier econometric
analysis, making use of additional years of data. Table 3 reports simple correlation coefficients
between the one-quarter growth rates of the monetary and financial aggregates over the period 1960 to
1989. The growth of Total Nonfinancial Sector Debt is less closely related to each of the monetary
aggregates during the second half of the period than in the first. Thus, the information on the
financial markets and the economy contained in total debt may have increased relative to the
information contained in the money supply. But the lower correlation could simply reflect instead a
looser relationship between total debt and the level of economic activity.
Table 4 reports the results from two sets of regressions. In the first set the growth of nominal
GNP is explained by current and lagged growth rates of a single financial aggregate and lagged values
of nominal GNP; in the second set only lagged growth rates of the financial variable are used. After
the regressions are run, a test of the significance of the coefficient estimates for the financial variable
is performed. Passing the test implies that the financial variable is able to make a statistically
significant contribution to the explanation of the movements in nominal GNP unaccounted for by past
movements in GNP itself.
Total debt does contribute to the explanation of nominal GNP over the entire period and over
both subperiods if its current value is included in the equation. However, if the current value of the
financial aggregate is dropped from the equation, total debt does rather poorly relative to the monetary
aggregates, a result similar to that found earlier by others. These findings are not particularly
favorable to total debt, but they are not favorable to the monetary aggregates either. The monetary
and credit aggregates are uniformly better at explaining nominal GNP over the 1960-74 period than
199
Table 3
Correlations among the Monetary and
Credit Aggregates
1960-n to
1989-1
Monetary
Base
Monetary base
Ml
M2
M3
L
Total Nonfinancial
Sector Debt
Tot Nonfin
Sec Debt
Ml
M2
M3
L
1.00
0.67
0.36
0.41
0.57
0.70
1.00
0.54
0.40
0.43
0.48
1.00
0.75
0.61
0.20
1.00
0.83
0.35
1.00
0.62
1.00
1.00
0.74
0.38
0.52
0.74
0.79
1.00
0.69
0.63
0.74
0.56
1.00
0.87
0.71
0.37
1.00
0.84
0.49
1.00
0.78
1.00
1.00
0.63
0.29
0.20
0.15
0.35
1.00
0.46
0.30
0.17
0.24
1.00
0.88
0.27
1.00
0.41
1.00
1960-n to
1974-m
Monetary base
Ml
M2
M3
L
Total Nonfinancial
Sector Debt
1974-IV to
1989-1
Monetary base
Ml
M2
M3
L
Total Nonfinancial
Sector Debt
200
1.00
0.68
0.49
-0.06
Table 4
Regressions of Nominal GNP on
Various Money and Credit Aggregates
Nominal GNP Explained by Current and Four Lagged Values of
the Financial Aggregate.
Variable
R-Bar Squared
(1960-11 to 1989-1)
R-Bar Squared
(1960-H to 1974-ffl)
R-Bar Squared
(1974-IV to 1989-1)
0.11***
0.08**
0.16***
0.14***
0.20***
019***
0.24***
0.30***
0.21***
0.22***
0.29***
0.33***
0.03
-0.01
0.03
0.17**
0.17**
0.12**
Monetary base
Ml
M2
M3
L
Total nonfinancial
sector debt
Nominal GNP Explained by Four Lagged Values of
the Financial Aggregate Only.
Variable
R-Bar Squared
(1960-n to 1989-1)
R-Bar Squared
(1960-H to 1974-HI)
R-Bar Squared
(1974-IV to 1989-1)
0.11***
0.08**
0.16***
0#i4***
0.12***
0.05***
017***
0.25***
0.22***
0.23***
0.30***
0.19***
0.05
0.01
0.05
0.10**
-0.02
-0.04
Monetary base
Ml
M2
M3
L
Total nonfinancial
sector debt
* Significant at the 10 percent level.
k
**
* Significant at the 5 percent level.
Significant at the 1 percent level.
201
over the 1975-89 period. Innovation and deregulation have apparently had roughly parallel effects on
all the financial aggregates.
D. The Potential Usefulness of Broad Credit Measures
Few economists who investigated broad credit aggregates argued that they should simply
replace the money supply in the Federal Reserve policy process. Instead, they suggested two ways in
which a broad measure of credit could be incorporated in Federal Reserve policy-setting procedures.
Total credit could be used as an intermediate target by the FOMC along with one or more measures of
the money supply.36 Or, if the FOMC did not adopt intermediate targeting, total debt could be one
of the key financial quantities, along with the money supply, on which the FOMC bases its decisions—
that is, it could be treated as an important indicator, either a "monitoring" or "information" variable.
Both Kaufman and Friedman at one time argued that their findings supported the use of a broad credit
measure in either of these policy frameworks, although they were skeptical of a mechanical approach
to intermediate targeting. Either way, the use of total debt was consistent with the position that central
bankers should lengthen the list of financial variables they take into consideration to include more than
the monetary aggregates.
To evaluate the potential usefulness of a broad credit aggregate for Federal Reserve policy,
Fackler and Silver estimated the marginal contribution that total nonfinancial debt would have made to
Federal Reserve policy during the period 1970 to 1982 if the FOMC had given equal weight to total
debt and the money supply.37
They attempted to identify those occasions when total debt signaled, rightly or wrongly, the
need for a policy different from that indicated by the money supply. They compared the signals given
by the growth of total debt with those given by the growth of the Ml money supply, noting when the
two signals differed markedly. There were three such instances: late 1973, mid-1980, and late 1980early 1981. Fackler and Silver concluded that the signal from total debt was not unambiguously more
accurate than that given by money in these instances.
36
Benjamin M. Friedman, "Time to Rethink the Monetary Targets Framework," New England
Economic Review, March-April 1982, pp. 15-23; and Henry Kaufman's discussion in Controlling
Monetary Aggregates III, Federal Bank of Boston, October 1980, pp. 66-68.
37
James Fackler and Andrew Silver, "Credit Aggregates as Policy Targets," Federal Reserve Bank
of New York Quarterly Review, Winter 1982-83, pp. 2-9.
202
The issue of controllability
An evaluation of a broad credit aggregate as an intermediate target for policy must also take
into consideration the degree of control the central bank can exert over credit aggregates. One could
argue that the Federal Reserve can maintain no more or less control over total debt than it can over
almost any other financial aggregate (with the exception of the monetary base). Financial quantities
are endogenously determined, with Federal Reserve actions only one of many forces working on the
total dollar volume of Ml, bank credit, or total debt. Thus, the FOMC has limited ability to control
the growth of financial aggregates. In this context, total debt has nothing in principle to recommend
for or against it relative to other financial quantities.
One could also argue, though, that Ml, and possibly M2, is under greater control than total
debt. Reserve requirements are imposed on demand deposits and other checkable deposits, thus
potentially allowing tighter control over Ml.38 Moreover, demand-for-money and supply-of-money
functions have been specified and estimated to some degree of satisfaction. The implication is that
economists have a better understanding of the factors determining the growth of the monetary
aggregates and that the Federal Reserve can use this knowledge to gain greater control over the growth
of the money supply. Similar equations have not been developed for total debt, one reason being the
difficulty of identifying the principal factors determining its growth.
Friedman tried to estimate the degree of control the Federal Reserve is likely to have in practice
over money and total debt. He related changes in these financial quantities to changes in two Federal
Reserve instruments, the federal funds rate and bank reserves.39 Total debt is significantly related to
both; in fact, it is more closely related to the federal funds rate than money is. The real question,
however, is whether the Federal Reserve has sufficient control over monetary and credit aggregates to
use them as an intermediate target, especially when slippage in the relationship between the
intermediate target and economic activity is taken into account.
38
But by calculating "shadow" reserve requirements, the tactics of open market operations can also
accommodate the use of credit aggregates at the strategy level of policy. See Marcelle Arak, "Control
of a Credit Aggregate," Federal Reserve Bank of New York Quarterly Review, Winter 1982-83, pp.
10-15.
39
Friedman, "Monefafy Policy with a Credit Aggregate Target."
203
Summary
Beginning in the latter half of the 1970s, economists took a greater interest in using broad
measures of credit in macroeconomic analysis. Theoretical arguments for these measures had
considerable appeal. Econometric studies, particularly those by Friedman, generally showed the
relationship between total nonfinancial sector debt and economic activity to be comparable, in terms of
stability and predictability, to the relationship between the money supply and economic activity.
These studies were used to support the position that a broad credit aggregate could be used jointly
with a monetary aggregate in setting central bank policy.
The theoretical arguments and empirical findings on total debt were well received by many
economists and central bankers. They had been facing instability in the demand for money since the
mid-1970s, if not earlier; and besides, some were skeptical of basing Federal Reserve policy too
heavily on a single financial variable. Nevertheless, enthusiasm for using broad measures of credit as
a policy variable was tempered by the lack of a convincing explanation of the observed stability of the
total debt-to-income ratio.
Since 1983, the potential value seen in using a broad measure of credit in formulating policy
has diminished markedly. Around that time, the previously observed relationship between a broad
credit measure and nominal GNP, which was stable over much of the postwar period, began to
deteriorate badly. The ratio of Total Nonfinancial Sector Debt to nominal GNP has risen rapidly since
then. Financial innovations and deregulation seem to have facilitated a rise in the amount of
marketable debt instruments and credit created by financial intermediaries relative to income and
wealth in the economy. Additional research on broad credit aggregates has also subsequently shown
that the relationship between the aggregates and economic activity is not as tight or as stable as the
earlier studies had led people to believe; the econometric results are sensitive to changes in the time
period studied and to seemingly minor changes in the specification of the statistical tests. Nor can it
be firmly established that total credit can make a significant marginal contribution to the explanation
of variation in GNP left unaccounted for by the growth of the money supply. Total debt thus does not
seem to be a suitable intermediate target for policy.40 But even if the relationship between total debt
40
In speaking of total debt as an intermediate target, we mean that the authorities would aim at
this target instead of the goal variables of Federal Reserve policy, real growth and price stability. To
use total debt as an intermediate target, the FOMC would establish a growth rate for total debt that
promises to lead to desirable outcomes for the goal variables. Then the Committee's decisions
concerning the discount rate and the open market directive would be designed to achieve the
established growth rate for total debt.
204
and output were tighter, the degree of control over total debt might well be inadequate for it to serve
as an intermediate target. While the money supply is thought to be under some degree of control by
the central bank, no convincing argument has been made that total debt could be placed under its
direct influence. This would appear to be true whether the tactics of policy are characterized by either
automatic or discretionary adjustments to instrument settings in response to growth of total debt that is
deemed too fast or too slow.41
The odd behavior of total debt during the past few years does not automatically rule out using
its growth, along with other financial variables, as an indicator of the need for a change in policy.42
While its overall relationship with nominal GNP is neither stable nor close, regression analysis does
indicate that a relationship exists, and it may well have a degree of reliability comparable to the
relationship between Ml or M2 and nominal GNP. Furthermore, it is still uncertain whether the
period of instability in the demand for money is over.
As for measuring the thrust of Federal Reserve policy, growth of total debt is flawed in the
same way that nominal interest rates and other financial variables are flawed. Total debt is an
endogenous variable and hence is procyclical; it cannot serve the same role that, for example, the fullemployment surplus can in analyzing fiscal policy. In any case, if there were a way to make a
measure of total debt exogenous, shifts in the demand and supply of credit seen in recent years might
defeat its usefulness.
All things considered, rapid acceleration or deceleration in the growth of total debt would still
seem to signify the need for a policy change in the context of confirming signals from other financial
market indicators. There is information to be gained by monitoring the growth of total debt as well as
41
Nor does there seem to be a way to devise a "fail-safe mechanism" for central bank policy based
on the growth of total debt. What is meant by a "fail-safe mechanism" is some self-imposed
constraint on the policy makers' decision set intended to prevent them from ever making a serious
error in setting policy. For instance, the imposition of a floor on the real federal funds rate at 2
percent (but in effect only during periods of economic expansion) would be a fail-safe mechanism.
42
One specific definition of "indicator" variable states that Federal Reserve policy must be
adjusted when designated indicators are unanimously signaling that the economy is either much too
strong or much too weak, or when price pressures are building. The indicator variables would be
chosen so that together they reliably convey the full range of information coming out of the financial
markets. The indicators are not chosen, however, on the basis of their incremental predictive abilities
and then used to help forecast the economy. They are of little value in forecasting when the usual
information set is available. See Bennett T. McCallum, "Targets, Indicators, and Instruments of
Monetary Policy" (Paper presented at the American Enterprise Institute Conference "Monetary Policy
in an Era of Change," November 16-17, 1988).
205
the money supply. Moreover, the theoretical arguments in favor of total debt continue to have appeal
and remain relevant. In all, there still seems to be a good case for maintaining a balance between the
use of money and credit measures in making policy.
II. Bank Credit
A. The Rationale for Its Use
A number of economists have maintained over the years that bank credit is "special" and more
important than money for understanding macroeconomic fluctuations. They argue, moreover, that
bank credit, rather than the money supply, is pivotal in the transmission mechanism of central bank
policy. The research of Ben Bernanke, Mark Gertler, and Joseph Stiglitz, whose approach to
macroeconomics is sometimes referred to as the New Keynesian Economics, has advanced such ideas
during the past ten years.43 But Robert Roosa stated the position almost forty years ago:
"It is principally through effects upon the position and decisions of lenders, and only
secondarily through effects upon the decisions of borrowers and savers, that central bank
action affecting interest rates achieves its significance."44
The recent interest in bank credit has been motivated by some of the same factors that have
spurred the interest in broad credit measures, such as the need for alternatives or supplements to the
poorly performing monetary aggregates. But this research also represents in some respects a revival of
an earlier interest in bank credit. In studies of the relationship between the money supply and nominal
GNP done almost twenty years ago, measures of bank credit were tried as an alternative to the
conventionally defined monetary aggregates. Bank credit was used because it was thought that the
items appearing on the asset side of the balance sheet could be measured more reliably than the
monetary assets appearing on the liability side. As such, bank credit was defined to be the sum of
43
Other economists working outside theframeworkof New Keynesian macroeconomics have also
proposed that bank lending is special and merits careful consideration. Fama concluded that since the
"tax" imposed by reserve requirements is paid by borrowers, meaning that they pay a higher interest
rate on a bank loan than they would on a loan from some individual or institution that did not pay
reserve requirements, there must be something unique about bank loans. Eugene F. Fama, "What's
Different about Banks?" Journal of Monetary Economics. 1985, pp. 29-39.
44
Robert V. Rosa, "Interest Rates and the Central Bank," Money. Trade, and Economic Growth
(New York: Macmillan, 1951), p. 293.
206
loans and investments. The current rationale for focusing on bank credit, however, is somewhat
different. This new rationale will be outlined in the rest of this section, followed by a section on the
measurement of bank credit.
The case that bank credit is "special"
Underlying the renewed emphasis on bank credit are two observations. First, as a practical
matter, households and most small and medium-size business borrowers have no access to the open
market for credit. For such borrowers, banks are likely to be by far the most important source of
credit, though other sources such as finance companies and, for small firms, trade credit are likely to
exist as well.45 If banks are unwilling to finance these borrowers' plans for purchasing inventory,
equipment, housing, or automobiles, aggregate spending is likely to be curtailed.46
The second observation is that bank credit is not allocated to potential borrowers simply
according to who is willing to pay the highest interest rate. Banks do not attempt to maximize their
return by lending to whoever walks into the lobby, applies for a loan, and promises to pay the highest
rate of interest. Instead, banks follow a strategy of charging a lower interest rate and using several
tactics each of which increases the likelihood of repayment.47 They screen loan applicants, monitor
borrowers, require collateral, give preference to long-term customers of the bank, make mostly short-
45
Bank lending can also be viewed as "special" because it serves as the backup source of liquidity
for both financial and nonfinancial businesses in the event of disruptions in the payments system, the
clearing of securities, or the financial markets. E. Gerald Corrigan, "Are Banks Special?" Federal
Reserve Bank of Minneapolis Annual Report, 1982.
46
Exactly the same argument has been made in a different context. Analyzing the effects of both
direct controls on bank credit and indirect control stemming from Regulation Q ceilings, Davis
contends that.. ."some small business and consumer borrowers are unlikely to be able to substitute
fully nonbank credit or other financial sources for the now unavailable bank credit... .Therefore, there
will be at least some cuts in total credit flows and in spending on output by businesses and consumers
whose access to bank borrowing has been reduced." Richard G. Davis, "An Analysis of Quantitative
Credit Controls and Related Devices," Brooking Papers on Economic Activity, 1:1971, pp. 75-76.
47
For related reasons, there are very few long-term bonds issued with a resettable interest rate.
They are unpopular because a rise in the interest rate reduces the firm's ability to pay and puts a
bond's principal at greater risk. In fact, although the bond's interest rate is supposed to be reset
annually so that the bond trades at par, there may be no interest rate that will get the bond's price to
par. See The Wall Street Jotimal May 2, 1989, p. C2.
207
term loans, and so forth.
In sum, banks allocate credit or restrict its availability through various
means, creating an excess demand for credit at the prevailing interest rate.49
The fundamental reasons giving rise to these banking practices will not be reviewed here in
detail. They have been analyzed by economists interested in explaining the origins of various credit
instruments, the widespread use of particular practices in finance, and the existence of financial
intermediaries.50 We will simply mention that imperfect information about the probability of a
borrower defaulting is a major factor underlying the existence of banks. These institutions specialize
in lending to small and medium-size firms and households, two groups whose creditworthiness is
difficult to assess.51 Problems of adverse selection and incentives are fundamental to the adoption of
many of the bank lending practices mentioned.
An implication of widespread use of the practice of setting more restrictive loan terms rather
than raising the interest rate to allocate credit is that short-term bank lending rates may be a very
unreliable guide to policy. The noninterest rate terms of credit (collateral, maximum time length of a
loan, conditions for rolling over the loan, acceptable uses of the loan) would seem to be as important
in their ultimate effect on aggregate demand as the level of the interest rate. These terms either raise
the effective cost of the loan well above the interest charges or cause the small business firm to be
48
Several studies have attempted to show that banks actively use nonprice devices to allocate
credit. See Stephen R. King, "Monetary Transmission: Through Bank Loans or Bank Liabilities?"
Journal of Money, Credit, and Banking, August 1986, pp. 290-303; Dwight M. Jaffee and Franco
Modigliani, "A Theory and Test of Credit Rationing," American Economic Review, December 1969,
pp. 850-72; and William R. Keeton, "Deposit Deregulation, Credit Availability, and Monetary Policy,"
Federal Reserve Bank of Kansas City Economic Review, June 1986, pp. 26-42.
49
We are deliberately avoiding the use of the word "ration," which many associate with
government-imposed and administered ceilings on interest rates.
50
For an analysis of the role of banks, see Joseph E. Stiglitz and Andrew Weiss, "Banks as Social
Accountants and Screening Devices for the Allocation of Credit," National Bureau of Economic
Research, Working Paper no. 2710, prepared for A. Courakis and C. Goodhart, eds., The Monetary
Economics of John Hicks (London: Macmillan, 1988). For a partial survey of the work on credit
rationing, Stiglitz and Weiss cite: D.M. Jaffee and J.E. Stiglitz, "Credit Rationing," in F.H. Hahn and
B. Friedman, eds., Handbook of Monetary Economics (New York: North-Holland, forthcoming).
51
In the case of households, another way of stating this is that the bank cannot estimate a family's
permanent income, it can only verify its current income. Hence, it imposes a ceiling on the ratio of
loan repayment to income. This argument is made by James A. Wilcox, "Consumption Constraints:
The Real Effects of 'Real1 Lending Policies," Federal Reserve Bank of San Francisco Review, Autumn
1989, pp. 39-52.
208
disqualified. For those households that, like small firms, are heavily dependent on banks for credit,
the availability and all-in cost (interest rate charges plus costs imposed by the other terms of the loan)
are again relevant, perhaps for home mortgages in particular. In contrast, only for large established
firms that have ready access to both bank and nonbank sources of credit at all times, even during
periods of tight money and great economic uncertainty, is the interest rate sufficient for assessing
credit conditions. Interest rates by themselves may thus be an acceptable indicator of the effect of
central bank policy on the corporate business sector, but a poor indicator outside that sector.
Several issues discussed in this section and in the next were raised nearly twenty years ago in
connection with the Federal Reserve's Regulation Q. At that time, the substitutability of nonbank
sources of credit for bank sources and the usefulness of the money supply, credit quantities, and
interest rates in policy making were also analyzed against a background of fluctuations in credit
availability. It was pointed out then that a serious weakness in any argument that bank credit is
"special" is the fact that banks are not the sole source of credit available to households and small
businesses. Significant nonbank alternatives exist for households seeking a home mortgage or an
automobile loan; and finance company loans and trade credit from suppliers provide inventory
financing for small firms, as discussed further in a later section.
Research supporting the view that bank credit is "special"
Before examining the link between bank credit and the macroeconomy, it may be useful to
review some evidence that bank lending is a determinant of spending in the economy. Several studies
show that the nonprice terms of home mortgage loans have had a very significant effect on the
housing market, although innovations in housing finance may have made those findings obsolete.52
For consumption spending, Duca produced evidence that he found to be a clear indication that the
availability of bank credit has been influential.53 He showed that an index of banks' willingness to
lend to households, derived from the Senior Bank Officer Lending Survey, is a statistically significant
factor determining durable goods demand. The results indicated to him that in the past (the study
covered the period 1966-86) households could not readily switch to alternative lenders when
commercial banks were unwilling to lend to them.
See the references in Keeton, "Deposit Deregulation, Credit Availability, and Monetary Policy."
53
John Duca, "The Effects of Credit Availability on Consumer Durable Expenditures" (Paper
presented at a meeting of Federal Reserve System economists, October 1987).
209
In the business sector, Fazzari, Hubbard, and Petersen maintain that investment spending by
small firms is generally more sensitive to internal financial positions than investment spending by
large firms in the same industry.54 The explanation they give is that smaller firms face more
financial constraints than do larger firms. Consistent with this point, Gertler and Hubbard show by
using cross-sectional time-series data that cash flow is a factor determining investment spending by
individual manufacturing firms.55
Notwithstanding these findings, it has yet to be established that loan availability is important
for the working of the macroeconomy, or even that smaller businesses as a group can have a
significant effect on aggregate demand.56 How large is the small business sector and does it produce
a significant portion of GNP? Is the dollar volume of investment spending conducted by small-tomedium size firms a major portion of total business fixed investment?
We can strengthen the case that a significant volume of bank credit is still special by drawing
on Commerce Department data collected from manufacturing firms on their short-term borrowings,
those with original maturities of one year or less. As of the third quarter of 1988, these data show that
manufacturing corporations with assets of less than $250 million are highly dependent on bank loans;
they obtain an insignificant amount of funds from the commercial paper market (Table S.) Firms
holding assets of more than $1 billion have good access to the commercial paper market, using the
commercial paper market much more than their banks. Firms holding between $250 million and $1
billion of assets have limited access to the commercial paper market.
On the basis of these data we would estimate that between $26 billion and $32 billion of bank
loans, out of $46 billion of loans outstanding to manufacturing firms, is "special" in the sense that the
54
Steven M. Fazzari, R. Glenn Hubbard, and Bruce C. Petersen, "Financing Constraints and
Corporate Investment," Brookings Papers on Economic Activity, 1:1988, pp. 141-206.
55
Mark L. Gertler and R. Glenn Hubbard, "Financial Factors in Business Fluctuations," Financial
Market Volatility, a symposium sponsored by the Federal Reserve Bank of Kansas City, August 1988;
also, First Boston Working Paper Series FB-88-37, September 1988.
56
At the macro level, the business fixed investment equation developed by Eckstein and Siani
gives an important role to the financial health of the business sector. This econometric equation is
consistent with the idea that bank credit is special, although it does not directly test the hypothesis that
firms face financial constraints. See Otto Eckstein and Allen Sinai, "The Mechanisms of die Business
Cycle in the Postwar Era," in Robert Gordon, ed., The American Business Cycle: Continuity and
Change" (Chicago: University of Chicago Press, 1986).
210
Table 5
Distribution of Short-Term Borrowing by
Manufacturing Corporations according to Firm Size
Firm Size,
as Measured
by the Total
Assets of the
Corporation
Borrowings as of 1988-Third Quarter
(In Billions of Dollars)
Loans from
Commercial Banks
Commercial
Paper
Other
Borrowings
$250 thousand
to $25 million
14.9
2.r
$25 million to
$50 million
3.8
0.3
$50 million to
$100 million
3.8
0.7
$100 million to
$250 million
3.4
0.2
0.8
$250 million to
$1 billion
6.4
1.4
1.2
13.8
25.9
13.9
More than
$1 billion
Source: U.S. Department of Commerce, Bureau of the Census, Quarterly Financial Report for Manufacturing,
Mining, and Trade Corporations, third quarter, 1988.
"Other borrowings" include commercial paper for the category $250 thousand to $25 million.
#
Less than $50 million.
211
borrower may not have a good substitute available. By dollar volume, between 56 percent and 70
percent of bank loans to manufacturing firms is "special." We also see that the net sales revenue of
those firms apparently dependent on bank lending is between 32 percent and 44 percent of the sales
revenue of all manufacturing firms. Thus the portion of the business sector relying on banks for credit
may be sufficiently large to matter for macroeconomic analysis.57
The link between bank credit and the macroeconomy
Bank credit is supposed to play a prominent role in the economy because the cost and
availability of bank credit can directly and substantially affect aggregate demand and the level of real
economic activity. A shift in the supply of bank reserves is usually thought of as the principal
underlying cause of exogenous changes in the price and availability of bank credit, but there are two
other ways such changes can come about. First, a softening of the real economy, or the expectation of
a softening, not only lowers the value of collateral that a business firm can offer, but also raises the
probability that a firm will be unable to repay. In response, banks may reduce their exposure by
shrinking the credit lines of their business borrowers. Then the volume of bank loans would contract
or its growth would slow, causing capital spending and inventory investment to slacken and possibly
to decline.58 Movements in interest rates and the money supply may not give signs of the developing
weakness in aggregate demand.
Second, a financial disruption, such as a bank failure or disintermediation caused by deposit
rate ceilings, reduces the flow of bank credit from the affected banks to their customers, causing their
investment spending to decline. Again, interest rates and the money supply may not warn of the
effects on aggregate demand. This is precisely the finding of Bernanke: the collapse of the banking
system disrupted the flow of credit to a large segment of the economy and contributed to the decline
57
Keeton pointed out, however, that the nonprice terms of a loan seem to be used to a lesser
degree by banks for the purpose of allocating credit during the past several years, as might be
expected with less regulation of the banking sector. From the Loan Officer Survey he found that since
1978 there has been relatively more variation in bank lending rates and less variation in the credit
standards for new customers. This may indicate that banks are relying more on the loan rate to
allocate credit—and by doing so bank credit would no longer be so special. See Keeton, "Deposit
Deregulation, Credit Availability, and Monetary Policy."
58
The farm and energy sectors are specific examples from the 1980s of small business borrowers
facing reduced credit availability. The fall in the prices of farm land and drilling equipment reduced
the value of the collateral that farmers and oil companies could offer creditors. The problems of banks
and other lenders to these sectors compounded the reduced credit availability.
212
in aggregate demand during the Great Depression, and interest rates and the money supply failed to
indicate the extent to which this was occurring.59
Thus the volume of bank loans may provide additional information concerning the impact of
the financial sector on the real economy. Interest rates are important; as borrowing costs rise,
consumption spending is deferred and investment projects are postponed. Moreover, a wealth effect
from the stock market may be felt. But the impact on aggregate demand of policy moves by the
central bank may be better gauged by watching bank loans as well as the money supply since it may
matter how banks create additional deposits and thus additional money. The growth of bank loans
could indicate that the effect of policy on the economy was greater or less than indicated by the
money supply.
If loan growth is stronger than money growth, banks have probably responded to a gain of
reserves from open market operations by making credit more available. Aggregate demand would then
be bolstered by more than the money supply or interest rates indicated because bank credit would
become available to finance spending that otherwise would not occur. In contrast, if loan growth is
weaker than money growth, banks have probably responded to the increase in bank reserves by adding
securities to the asset side of their balance sheets and have not made credit more available to small
business customers. Thus the impact on domestic aggregate demand would be weaker.
To demonstrate these points, Bernanke and Blinder built a small macroeconomic model in
which money and bank loans have separate roles.60 Their model, a variant of the textbook IS-LM
model, is specified to include three financial instruments: money, bonds, and bank loans. There are
two interest rates, one on bonds and another on loans. Shocks to aggregate demand affect the quantity
of loans and the quantity of money in the same direction. An autonomous fall in aggregate demand
would cause both quantities to shrink, and the signals from the financial aggregates to policy makers
would be the same. Suppose instead that a more expansionary central bank policy is implemented,
shifting the LM curve to the right. This will have the usual effects on money, the interest rate, and
output; and because the volume of credit demanded is endogenous, the volume of bank loans will
expand accordingly. Here too the signals from the financial aggregates would be the same.
59
Ben S. Bernanke, "Nonmonetary Effects of the Financial Crisis in the Propagation of the Great
Depression," American Economic Review, June 1983, pp. 257-76.
60
Ben S. Bernanke and Alan S. Blinder, "Credit, Money, and Aggregate Demand," American
Economic Review, May 1988, pp. 435-39.
213
More revealing than these cases is that of a shift in the loan-demand or the loan-supply
functions. Suppose that, coincident with looser policy, loan supply shifts because of an anticipated
increase in the profitability and hence creditworthiness of the small business sector. Because of
greater availability of credit to bank borrowers and greater ability to finance inventories and capital
spending, the IS curve also shifts to the right.61 This would cause the level of output to rise further
than indicated by the shift of the LM curve alone. In this case, the impact of policy, not in direction
but in magnitude, depends on a reinforcing shift in loan supply. The impact will be stronger than
indicated by the money supply if bank loans become available such that more firms can now borrow
and proceed with their planned spending.
The Bernanke-Blinder model also implies that incorrect policy moves may be avoided by taking
into consideration the growth of bank loans. For example, suppose that there is a shift in money
demand. As a result, the money stock, endogenously determined and not under the complete control
of the central bank, would increase. If the policy makers' goal were to hit a money supply target,
policy would have to be tightened to prevent an undesired expansion in the money stock. If policy
makers were also watching the volume of bank loans, however, they would see it contract or its
growth slow as the money stock expands. If appreciable weight were given to the growth of bank
loans as well as money, the policy makers' response to the increase in the money stock would be
tempered.62
B. Measurement Issues
Measures of bank credit are readily available; the Federal Reserve System collects extensive
data on credit extended by commercial banks. The data are similar in coverage and timeliness to the
monetary data, and they are collected by category: real estate, commercial and industrial, securities,
61
Stiglitz and Weiss make this point in "Banks as Social Accountants and Screening Devices for
the Allocation of Credit," p. 27.
62
Bernanke and Blinder show how the optimal combination interest rate-money supply policy is a
function of the elasticities of credit demand and supply and the variance of their disturbance terms.
Duca and VanHoose also use this model to show how the optimal policy is affected by the prevalence
of floating rate loan commitments. See John V. Duca and David D. VanHoose, "Loan Commitments
and Optimal Monetary Policy," Board of Governors of the Federal Reserve System, Finance and
Economics Discussion Paper no. 44, August 1988.
214
individual, agricultural, and so on.63 A problem occurs, however, in obtaining data on that portion of
bank credit that is "special"—those bank loans made to households and businesses that either have no
alternative source of financing or have only alternatives more costly than bank loans. As mentioned
earlier, it appears that the size of a business firm for which bank credit is special is one whose assets
are less than $250 million. The Federal Reserve does not request that banks report loan data
according to the size of the borrower and, of course, certainly not by the degree of access the bank
perceives that the borrower has to the credit markets. Hence, no currently available measure of bank
credit directly corresponds to the concept of "special" bank credit.
Another problem in measuring the dollar volume of credit that is "special" arises from the use
of lenders other than the commercial banks. Small businesses also borrow from nonbank depository
institutions, finance companies, and other nonfinancial firms. Table 6 shows that finance companies
make a significant amount of credit available to small firms.64 Some finance companies concentrate
on lending to firms considered to be too small and uncreditworthy for commercial banks; others
concentrate on factoring, a line of business many banks consider too risky to enter. Accordingly, the
short-term business credit made available by finance companies may be just as relevant as bank credit
in macroeconomic analysis. Table 6 also shows the net amount of trade credit outstanding. Trade
credit, often extended by large firms supplying parts or materials to smaller firms, substitutes for shortterm bank borrowing and could be a nontrivial fraction of the credit extended to small firms. But
because trade credit is direct lending from one party to another and the terms are different from a bank
loan, it is unclear whether trade credit should be included in a measure of bank credit.
There are at least three ways to proceed in obtaining data in order to conduct econometric
research on bank credit. We could use total bank loans or just its commercial and industrial (C&I)
component, keeping in mind that these measures include loans that are not special. Alternatively, we
could adjust the bank loan figures by adding or subtracting items that should or should not be
included. For example, "bank credit" as defined by Bernanke and Blinder is the sum of short- and
63
If economic research can show the importance of bank loan availability for the macroeconomy,
greater effort can be devoted to collecting the bank loan data.
^Some of their lending may be to the parent company. For instance, General Electric Capital
Corporation may at times float commercial paper and then lend the proceeds to its parent. General
Electric Capital could also lend to some large firms. These loans, too, would not represent special
credit.
215
Table 6
Short-Term Liabilities of Nonfinancial Corporations
Amounts Outstanding at Year-End 1983
(In Billions of Dollars)
Type
Amount
Fraction of Total
Bank loans
Commercial paper
Bankers' acceptances
Finance company loans
Net trade credit
Profit tax payable
401.9
37.7
9.4
100.5
106.1
8.2
61
6
1
15
16
1
Total
663.8
100
Source: Federal Reserve Board, Flow of Funds Outstanding, September 1984.
medium-term credit extended to households and nonfinancial businesses by financial intermediaries, as
derived from the flow-of-funds data. Apparently, this would be the sum of home mortgages, consumer
installment credit, bank loans (not elsewhere classified), and other loans, mortgages, and installment
credit extended to the household sector; and mortgages, bank loans (not elsewhere classified), bankers'
acceptances, and nonbank finance loans made to nonfinancial businesses. As a second alternative, we
could try a proxy for special bank credit-loans extended by smaller banks (possibly with the addition
of finance company lending). The reasoning would be that large banks lend primarily to large firms
and small banks lend almost exclusively to small firms. Much of the credit extended by large banks is
not special and it may be better to exclude their loans; nearly all the credit extended by smaller banks
(and finance companies) qualifies as special. Fortunately, the Federal Reserve does report separately
loan data from the largest banks (the weekly reporting banks) and the U.S. branches and agencies of
foreign banks.65 The residual, total bank loans less the loans of the weekly reporters and the loans of
U.S. branches and agencies of foreign banks, provides an estimate of special bank credit. Of course,
65
Weekly reporting banks are basically the 169 largest banks in the country, mostly those holding
greater than $2 billion of assets as of January 1, 1988.
216
Chart 2
Bank Loans and Total Nonfinancial Sector Debt
Change from Four Quarters Earlier
Percent
. 5
- J 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 I 1 1 1 1 1 1 1 I 1 1 1 I 1 1 1 I 1 1 1 I 1 1 1 I 1 1 1 I 1 1 1 I 1 1 1 1 1 1 1 I 1 1 1 I
1974
1975 1976
1977 1978
1979 1980 1981
1982 1983 1984
1985
1986 1987
1988
Source: Federal Reserve System.
this measure will be distorted if large banks manage to gain a larger share of the lending to mediumsize businesses, a change which may well be occurring, or if small banks take more participations in
loans to large borrowers.66
In Chart 2, the four-quarter growth rates of these measures of "special" credit are plotted: total
loans at all banks, total loans at small banks, and intermediated short- and medium-term credit to
households and nonfinancial businesses. On the chart, Total Nonfinancial Sector Debt is also drawn
for comparison. For each measure of "special" loans, the movements in the four-quarter growth rate
appear correlated with the movements in the growth of total credit, but the amplitude of the observed
cycles in each of these measures is much greater. It is not clear whether any of these bank loan series
66
For a report on large banks' attempts to gain small-business customers, see The Wall Street
Journal, December 13, 1988, p. B2.
217
would be useful, but the growth rate of each does show considerably more variation than total debt in
the period 1980-84, when the economy went through two sharp recessions and the growth of total debt
was relatively smooth.
In Table 7 we report the simple correlation coefficients between the one-quarter growth rates of
various monetary, broad credit, and bank loan measures over the period 1973-89. Some conelations are
surprisingly weak, especially between measures of bank loans and the narrower monetary aggregates.
These weak correlations imply often divergent growth rates and raise the possibility that useful
information that is not already contained in the monetary data can be extracted from bank loan data.
Prearranged lines of credit
Another question surrounding measurement of bank credit is whether it is more relevant to
measure the actual amount of credit extended by banks or the amount of "potential" credit, that is, the
sum of credit extended and the unused portions of prearranged credit lines. Normally, the amount of
credit outstanding is thought to be more relevant for the measurement of any credit aggregate.
Extending credit creates purchasing power, whereas granting a line of credit does not.
What the establishment of a line of credit clearly does is enhance liquidity. And so,
prearranged credit lines may create a greater problem for the measurement of a broad monetary
aggregate than for the measurement of a credit aggregate.67 Only if granting a credit line affects the
spending of a firm or household in a manner similar to the extension of a loan should unused credit
lines be included in a bank credit aggregate.
Prearranged credit lines, however, may affect the timing in the relationship between the growth
of a bank credit aggregate and changes in central bank instruments. The aggregate amount of unused
credit lines could be a key determinant of the length of the lag between changes in policy instruments
and the effect on the volume of bank credit outstanding. A tightening of Federal Reserve policy may
initially induce firms and households to make use of their lines of credit as sales revenue or personal
income falters, and only sometime later would the amount of credit demanded grow more slowly or
contract. Hence, prearranged credit lines may be a determinant of the lag in policy effects on
economic activity and of the degree of short-term control that the System has over the quantity of
bank credit.
67
A monetary aggregate of liquid assets is the topic of Lawrence Radecki and Gabriel de Kock,
"Liquid Asset Measures as Intermediate Targets and Indicators for Monetary Policy," in this volume.
218
Table 7
Correlations among the Monetary and
Credit Aggregates
1960-IIto
1989-1
Ml
M2
M3
L
Total Nonfinancial
Sector Debt
Intermediated credit
1974-IVto
1989-1
Total
Nonfinancial
Sector Debt
Ml
M2
M3
L
1.00
0.54
0.40
0.43
0.48
1.00
0.75
0.61
0.20
1.00
0.83
0.35
1.00
0.62
1.00
0.31
0.04
0.25
0.43
0.61
Total
Nonfinancial
Sector Debt
Intermediated
Credit
1.00
C&I
Loans
C&I at Small
Loans Banks
Total
Loans
Total
Loans
at Small Intermediated
Banks
Credit
Ml
M2
M3
Ml
M2
M3
L
Total Nonfinancial
Sector Debt
1.00
0.46
0.30
0.17
0.24
1.00
0.68
0.49
-0.06
1.00
0.88
0.27
1.00
0.41
1.00
C&I loans
C&I loans at
small banks
Total loans
Total loans at
small banks
0.08
-0.27
0.26
0.39
0.34
1.00
0.06
0.15
0.16
-0.25
0.56
0.30
0.66
0.44
0.34
0.68
0.51
0.83
1.00
0.51
1.00
0.15
0.07
0.37
0.49
0.61
0.33
0.77
0.61
1.00
Intermediated Credit
0.32
-0.04
0.32
0.37
0.70
0.39
0.30
0.68
0.51
to
1.00
C. Empirical Research on Bank Credit
There have been several econometric studies of the relationship between bank credit and the
level of economic activity. (Brief summaries of the papers cited in this section are presented in Table
8.) The earlier studies, conducted between 1970 and 1975, defined bank credit as total loans and
investments at commercial banks,68 not as total loans or as bank loans to small-to-medium size
business firms. Thus, the measure was broader than that recommended by later analysts of the
importance of bank credit in the macroeconomy. Nonetheless, most of these earlier studies by
Hamburger, Andersen, Hunt, Shadrack, and Levin showed that current and lagged changes in bank
credit were capable of explaining as much or more of the changes in nominal GNP than the money
supply.69 Studies conducted a little later, however, were less favorable toward bank credit. For
instance, Davis found that bank credit did not do as well as either broad credit measures or money
measures in explaining movements in GNP over the period 1961-77.70 Similarly, Friedman found
that bank credit had lower explanatory power than either money or total credit.71
While the importance of bank credit measured as the sum of all loans and investments was
placed in doubt by these studies, the importance of the bank loan portion of bank credit was stressed
by Bemanke's paper examining the nonmonetary effects of the banking and financial crisis during the
Great Depression.72 Bernanke estimated regressions relating the rate of growth of industrial
68
Adjusted to exclude domestic interbank loans and to include net sales of loans to commercial
bank affiliates.
69
Michael J. Hamburger, "Indicators of Monetary Policy: The Arguments and the Evidence,"
American Economic Review, May 1970, pp. 32-39; L.G. Andersen, "Selection of a Monetary
Aggregate for Use in the FOMC Directive," Open Market Policies and Operating Procedures, Board of
Governors of the Federal Reserve System, Staff Studies, July 1971; L.H. Hunt, "Bank Credit and the
Money Stock: Their Roles in the Determination of Income in the Post Accord Period," Journal of
Finance, December 1973, pp. 941-54; Frederick C. Shadrack, "An Empirical Approach to the
Definition of Money, Monetary Aggregates and Monetary Policy. Federal Reserve Bank of New York,
1974, pp. 28-34; and Fred J. Levin, "The Selection of a Monetary Indicator: Some Further Empirical
Evidence," Monetary Aggregates and Monetary Policy. Federal Reserve Bank of New York, 1974,
pp. 35-39.
70
Davis, "Broad Credit Measures as Targets for Monetary Policy."
71
Friedman, "The Role of Money and Credit in Macroeconomic Analysis" and "Monetary Policy
with a Credit Aggregate Target."
72
Bernanke, "Nonmonetary Effects of the Financial Crisis in the Propagation of the Great
Depression."
220
Table 8
Summary of Statistical Findings on Bank Credit Measures
Definitions of
Bank Credit Used
Statistical
Techniques
Sample
Period
Hamburger
(1970)
Bank credit
Reduced-form
1953-68
Bank credit and total reserves are more closely related to
movements of GNP than are the monetary aggregates.
Andersen
(1971)
Bank credit
St. Louis
approach
1953-69
Bank credit produces a slightly better fit than the other aggregates.
Hunt
(1973)
Bank credit
Reduced-form
1953-71
The correlation between bank credit and GNP is generally higher than
that between money and GNP.
Shadrack
(1974)
Bank credit
Reduced-form
1953-71
The results favor M2 over bank credit.
Levin
(1974)
Bank credit
Reduced-form
1960-72
The results favor bank credit slightly over monetary aggregates.
Davis
(1979)
Bank credit
Reduced-form
1961-77
Bank credit has less explanatory power than either money or total credit.
Friedman
(1983)
Bank credit
Reduced-form,
St. Louis
approach,
VAR
1953-78
Same as Davis.
Bernanke
(1983)
Proxy for the
banking crisis
Regression
1921-41
The collapse of the financial system had an independent effect on aggregate
demand.
King
(1984)
Bank loans
VAR
1950-79
The relationship between GNP and demand deposits is stronger than that
between GNP and bank loans.
Bernanke
(1986)
Bank loans
VAR
1954-84
Money and bank loans have about equal impact on GNP.
Author
K
Main Results
to
Summary of Statistical Findings on Bank Credit Measures (continued)
Definitions of
Bank Credit Used
Statistical
Techniques
Sample
Period
Main Results
Lown
(1988)
Bank loans
Reduced-form
1959-87
The results reverse King's findings.
BlinderBemanke
(1988)
Bank loans
Regression
1974-85
The estimated demand-for-bank-loans equation has coefficients as sensible and
significant as those of an estimated demand-for-money equation,
and the residuals are smaller in the credit equation.
Author
production to "unanticipated" changes in money and prices in the current quarter and in previous
quarters, the rate of growth of industrial production in previous quarters, and proxies for the financial
crisis (the deposits of failing banks and the liabilities of failing businesses). The proxies for the
banking crisis taken jointly have a high level of statistical significance in the regressions. Bernanke
argued that this was confirmation of what was generally believed to be true but until then had not been
shown: the collapse of the financial system and the reduced availability of credit had an independent
effect on aggregate demand and the economy that was not reflected in conventional monetary data.
Nor was the effect seen in the general level of interest rates, although the unusually large spread
between the rates on Treasury bonds and low-grade corporate bonds was suggestive.
The findings from postwar data, however, are somewhat ambiguous regarding the importance of
bank loans. King used VAR techniques to show that the relationship between demand deposits and
future output is far stronger than that between output and C&I loans.73 But Bernanke subsequently
produced a paper that identified a role for bank loans in the postwar economy comparable to what he
found during the Great Depression.74 Looking at the impulse-response functions derived from VARs,
Bernanke found that credit shocks have a strong immediate effect on output lasting a year or more,
while money shocks have little contemporaneous effect on output (although they do have a lagged
effect). Bernanke concluded that"... money and credit are parallel forces of approximately equal
importance." Cara Lown updated some of the King and Bernanke work.75 Over the 1959-87 time
period, she found that both total bank loans and demand deposits Granger-cause nominal GNP, but
total bank loans have a slightly higher level of significance. Moreover, when the interest rate is paired
in a regression equation with bank loans or demand deposits, bank loans are still highly significant in
predicting nominal GNP, but not demand deposits.
Recently, Blinder and Bernanke estimated the demand for money and the demand for bank
loans using comparable specifications for the regression equations and the same sample period.76
73
King, "Monetary Transmission: Through Bank Loans or Bank Liabilities?"
74
Ben S. Bernanke, "Alternative Explanations of the Money-Income Correlation," Real Business
Cycles, Real Exchange Rates, and Actual Policies, Carnegie-Rochester Conference Series on Public
Policy no. 25, 1986, pp. 49-100.
75
Cara S. Lown, "The Credit-Output Link vs. the Money-Output Link: New Evidence," Federal
Reserve Bank of Dallas Economic Review, November 1988, pp. 1-10.
76
Bernanke and Blinder, "Credit, Money, and Aggregate Demand."
223
They found that the coefficients are equally sensible and significant in the credit equation as in the
money equation; and the residuals are smaller in the credit equation than in the money equation.77
Extending the econometric research on bank credit
In sum, the econometric research on various measures of bank credit has produced mixed
results regarding its importance for the behavior of the U.S. economy. The role of the bank loan
component is clearest in the case of the Great Depression. The econometric evidence from postwar
data, rather limited when compared with the volume of research on broad measures of credit, provides
weak confirmation of the usefulness of bank credit measures for policy making in normal times. To
supplement what has been done previously, we conduct some analysis of our own.
Regression equations, specified in the same way as those used earlier in the paper to study total
credit, were estimated using several measures of bank loans. Table 9 reports the estimates of the
regression equations in which the growth of nominal GNP is explained by current and lagged values
of various monetary, broad credit, and bank credit aggregates. In an equation which includes both the
current and lagged values of the financial aggregate (shown in the upper half of the table),
intermediated credit (as defined above) has the most explanatory power over the period 1974-89,
followed by M3, L, and total bank loans. (The sample period begins in 1973, the starting date for the
loan data.) In equations from which the current value of the financial measure is dropped (lower half
of the table), most financial aggregates are no longer statistically significant. Only M3 and
intermediated credit show statistically significant explanatory power over the 1974-89 period.
Another set of regressions was estimated, this time using two financial variables
simultaneously: intermediated credit and either a monetary aggregate or Total Nonfinancial Sector
Debt. The purpose was to see whether a significant contribution to the explanation of GNP
movements could be made by intermediated credit, the strongest performer among the measures of
bank credit, in a regression equation that already included a conventional aggregate. The results in
Table 10 show that intermediated credit is as likely to be statistically significant as the monetary or
broad credit aggregate; in a few of the equations, both financial aggregates are significant.
77
The indicator approach to the analysis of macroeconomic fluctuations has for a long time used
bank credit measures extensively. The Bureau of Economic Analysis uses nine credit measures among
its 112 principal cyclical indicators. The Credit Flows series are all classified as leading indicators of
peaks and troughs in the business cycle. Business Conditions Digest. U.S. Department of Commerce,
Bureau of Economic Analysis.
224
Overall, bank credit measures do as well as the monetary aggregates in these regressions, if not
a little better. But because the bank credit data appear to be rather "noisy" on a quarter-to-quarter
basis, possibly obscuring the relationship between economic activity and bank credit, an additional set
of regressions was run in which the four-quarter growth rates of nominal GNP were explained by fourquarter growth rates of the aggregates. The results are shown in Table 11. In these regressions, Ml
and M2 do very poorly; in fact, the correlation between Ml and nominal GNP is negative. M3, L,
and Total Debt, however, do make statistically significant contributions to the explanation of nominal
GNP at the 1 percent level (the time period is 1974-89).
Three measures of special bank credit-total loans at small banks, commercial and industrial
loans at small banks, and intermediated credit-have considerably more explanatory power than the
narrow scope of these financial aggregates would have suggested. The growth rates of total loans,
total loans at small banks, and intermediated credit are plotted in Chart 3 along with nominal GNP,
and the positive relationship between the bank loan measures and GNP can be seen. These regression
results would seem to indicate that there may be some validity to the argument that bank credit is
special and that bank credit measures could be useful to central bankers in formulating policy.
A study of the relationship in the Canadian economy between monetary and credit aggregates
and gross domestic product has recently been completed by an economist at the Bank of Canada.78
This study, too, found that measures of household credit and business credit contribute to the
explanation of movements in the level of economic activity. It was concluded that, despite
innovations in the financial markets, monetary and credit aggregates have a place in the formulation of
central bank policy.
Perhaps it should be kept in mind that the continued availability of bank credit provides an
interesting explanation for the economy's resilience following the 1987 stock market crash. In 1987IV and 1988-1, when bank credit remained readily available according to the Senior Loan Officer
survey, consumption and business fixed-investment spending continued to be strong, and inventories
were not liquidated. Apparently, durable goods purchases, business expansion, and inventories could
all be readily financed.
78
Patrice Muller, "Might Financial Aggregates Still Have a Role in Policy Formulation?" (Paper
presented at the Annual Meeting of the Canadian Economics Association, June 1989).
225
Table 9
Regressions of Nominal GNP on
Various Money and Credit Aggregates
Nominal GNP Explained by Current and Four Lagged Values of
the Financial Aggregate
Variable
Ml
M2
M3
L
Total Nonfinancial
Sector Debt
C&I loans
C&I loans at
small banks
Total loans
Total loans at
small banks
Intermediated credit
R2
(1974-IV to 1989-1)
R2
(1960-n to 1989-1)
R2
(1960-H to 1974-ffl)
0.08**
0.16***
0.14***
0.20***
0.30***
0.21***
0.22***
0.29***
-0.01
0.03
0.17**
0.17**
0.19***
0.33***
0.12**
....
....
0.04
—
—
—
—
0.06
0.15**
—
—
0.06
0.17***
0.13**
0.21***
Nominal GNP Explained by Four Lagged Values of
the Financial Aggregate Only
Variable
Ml
M2
M3
L
Total Nonfinancial
Sector Debt
R2
(1960-H to 1989-1)
R2
(1960-H to 1974-m)
R2
(1974-IV to 1989-1)
0.08**
0.16***
0.14***
0.12***
0.25***
0.22***
0.23***
0.30***
0.01
0.05
0.10**
-0.02
0.05**
0.19***
-0.04
C&I loans
C&I loans at
small banks
Total loans
Total loans at
small banks
....
....
0.02
—
—
—
—
0.00
0.02
—
—
-0.04
Intermediated credit
0.06**
0.10**
Significant at the 10 percent level.
** Significant at the 5 percent level.
*** Significant at the 1 percent level.
226
0.07*
Table 10
Regressions of Nominal GNP on Intermediated Credit
and Another Money or Credit Aggregate
Nominal GNP Explained by Current and Four Lagged Values of
Intermediated Credit (IC) and Another Financial Aggregate
Money and Credit Variables
Found to Have Statistical Significance
Financial Variable
Included Along with IC
Sample Period
1960-n to 1989-1 1960-n to 1974-ffl
Ml
IC***
Ml***
M2
1974-IV to 1989-1
IC***
IC*
M3
M3**
M3*
IC*
L***
IC**
L***
.....
TNFSD**
IC*
TNFSD***
TNFSD*
IC***
ic***
Total Nonfinancial
Sector Debt
(TNFSD)
Nominal GNP Explained by Four Lagged Values of
the Financial Aggregate Only
Money and Credit Variables
Found to Have Statistical Significance
Financial Variable
Included Along with IC
Sample Period
1960-n to 1989-1 1960-n to 1974-ffl
Ml
Ml*
Ml***
M2
M2***
M2**
M3
M3***
M3**
L
L***
IC*
L***
TNFSD*
TNFSD**
Total Nonfinancial
Sector Debt
(TNFSD)
1974-IV to 1989-1
IC*
* Significant at the 10 percent level.
** Significant at the 5 percent level.
*** Significant at the 1 percent level.
227
Table 11
Explanatory Power of Various Financial Quantities
When Growth Is Measured over Four-Quarter Intervals
(Sample Period: 1974-1989)
R2 of the Regression
Independent Variable
Current Value
Only
Current and One
Lagged Value
Ml
0.02i;
0.02i;
M2
0.04*
0.11**
M3
0.24***
0.30***
L
0.25***
0.39***
Total Nonfinancial
Sector Debt
0.05*
0.22***
Commercial and industrial
loans
-0.02
0.09**
Commercial and industrial
loans of small banks
0.14***
0.21***
Total bank loans
0.03*
0.19***
Total loans of small
commercial banks
0.19***
0.22***
Intermediated credit to the
household and nonfinancial
business sectors
0.15***
0.27***
Notes: In the first column, the regression equation is Y = a + bX + e, where Y is the percentage change of
nominal GNP measured from four quarters earlier, and X is the percentage change of the monetary or credit
quantity measured from four quarters earlier. In the second column, the regression equation is: Y = a + bX +
cX(-l) + e.
- Negative correlation
* Significant at the 10 percent level.
** Significant at the 5 percent level.
*** Significant at the 1 percent level.
228
Chart 3
Credit Measures and Economic Activity
Change from Four Quarters Earlier
Percent
20
i
i
.5
I
I
i
1 1 1 I
;
•
1 i
I
i
• •
1 I
i
1 » •
I
i
•
•
•
I
i
1 •
•
I
i
1 •
1 1 • •
i
•
I
1 1 •
i
i
I
1 1 1 I
1 1 t
i
i
I
1 1 1 I
i
i
i
1 1 I
1 1 1 I
i
• •
1 I
1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988
Source: Federal Reserve System.
D. The Potential Usefulness of Bank Credit Measures
Research in macroeconomics has usually focused on the monetary aggregates and, to some
extent, on broad credit aggregates. Less consideration has been given to bank credit. During the past
ten years, however, the argument that bank credit is "special" in the transmission of central bank
policy has been revived and recast in terms of new theories in banking and finance, and some
econometric work supports this view. Bernanke's work on the effect of the banking crisis during the
Great Depression is thought to show that, at least in that historical episode, the volume of bank
lending has an independent effect on the economy, one not accounted for by movements in the money
supply and interest rates.
It is maintained that banks are, practically speaking, the main source of credit for many
households and for small to medium-size business firms. Cross-sectional data from the manufacturing
sector suggest that currently about three-fifths of bank loans outstanding to manufacturing firms are
229
"special" in the sense that the borrower obtains credit mostly from banks rather than the financial
markets. These same firms generate one-third or more of the total sales revenue of manufacturing
firms. The figures suggest that changes in the availability of bank credit could potentially affect the
level of macroeconomic activity, especially if a similar fraction of firms in other sectors also relies on
banks for credit. Recent innovations in banking practices, however, may be diluting or reducing the
special nature of bank credit.
Two problems arise in obtaining comprehensive time series data on the portion of total credit,
provided by banks or other financial intermediaries that is "special." First, data are not collected on
the volume of bank loans made to households and business firms that have no good alternative source
of financing to bank loans. Second, borrowers use lenders other than commercial banks; firms also
borrow from nonbank depository institutions, finance companies, and other nonfinancial firms.
Nonetheless, researchers have obtained econometric results from postwar data on the U.S. and
Canadian economies that are thought to indicate that bank credit is at least somewhat useful for
explaining changes in economic activity and thus may convey information to central bank policy
makers.
In this paper, the ability of some measures of bank credit to explain movements in nominal
GNP was tested. These measures—including total loans at small banks, C&I loans at small banks, and
total credit obtained from financial intermediaries-had some success, showing basically the same
power to explain movements in GNP that the money supply or total debt had in the 1974-89 period.
The explanatory power demonstrated is rather low, however, in absolute terms. At best about onequarter of the quarterly movements in nominal GNP can be explained by the various measures of bank
credit.
Bank credit thus does not seem to be a suitable intermediate target—its relationship to economic
activity is simply too weak. Moreover, the degree of short-term control over bank credit exercised by
the central bank may be inadequate. Little research has been undertaken on the issues surrounding the
central bank's ability to control the volume of bank credit. One suspects that tight control is not
achievable, in part because prearranged lines of credit can be drawn upon when policy is tightened.
Despite the weakness of the overall relationship between bank credit measures and economic
activity, bank credit may still play a key role in the transmission mechanism of monetary policy and in
the amplification of shocks originating elsewhere in the economy. Hence there may well be some
place for a measure of bank credit in the formulation of Federal Reserve policy.
230
Perhaps the best way to make use of bank credit is to have it supplement the information
provided by other important financial market variables. Interest rates appear to be sufficient only for
assessing the credit conditions facing large established firms that have ready access to bank and
nonbank sources of credit, even during periods of tight money and great economic uncertainty. More
indicative of financial conditions experienced outside the corporate sector may be the growth of bank
credit, properly measured. Furthermore, data show that the correlation between quarterly growth rates
of the money supply and bank credit measures is rather weak, in some cases negative. Information
may be contained in the bank credit data that is absent from money supply data. Bank credit may
prove to be an unreliable indicator, however, if firms borrow more heavily from banks when their
sales begin to lag and their accumulating inventories have to be financed. At the onset of a recession,
a bulge in bank credit would give a very misleading signal to policy makers.
231
TARGETING NOMINAL GNP
Spence Hilton and Vivek Moorthy
By definition, nominal GNP is the product of two important variables of ultimate concern to
monetary policymakers: the aggregate price level and total real output.1 Thus, nominal income
cannot be considered a "determinant" of the final policy objectives in the sense that other intermediate
target candidates are.2 Also, because nominal GNP is observed simultaneously with these final
objectives, it is not considered a "leading indicator" of the ultimate goals.3 As the following
discussion illustrates, many of the arguments for and against using nominal income as an intermediate
target for monetary policy can be tied to this unique relation between nominal GNP and the final
policy objectives.
Nominal GNP already receives considerable attention in setting policy. Monetary authorities
routinely update forecasts of nominal income and its major components for the forthcoming one to two
years, and a "central tendency" of policy makers' expectations is published twice a year. Moreover,
when target paths for money aggregates are established, an accompanying trajectory for nominal
income is at least implicitly set. While important, these practices do not constitute a formal targeting
of nominal GNP. This study only examines regimes where nominal GNP is the principal, if not sole,
intermediate target variable (although its use in conjunction with operating targets is considered) and
where target levels or growth ranges for nominal income are announced publicly.
The discussion is organized into three main sections. First, the major reasons advanced for
targeting nominal GNP are presented. Some of these are theoretical, while others are more practical in
nature. As will become apparent, not all advocates of nominal GNP targeting subscribe to the same
underlying logic for choosing to target this measure. Second, possible strategies for targeting nominal
*In this paper, the expressions "nominal GNP," "nominal income," and "aggregate income" are used
interchangeably.
2
For example, interest rates, another potential intermediate target, are believed to be an important direct
determinant of the level of economic activity.
3
For instance, some argue that commodity prices are good predictors of inflation.
232
GNP are considered. For the most part the discussion focuses on existing proposals for nominal
income targeting, although other targeting strategies certainly could be devised. Some important
unresolved issues raised by these proposals are also addressed, including their potential value for
stabilizing economic activity. Third, the difficulties with controlling nominal GNP are examined,
along with some proposed operating procedures. A summary and some conclusions are then
presented.
I. Rationale for Targeting Nominal GNP
A number of both economic and political reasons have been advanced for using nominal GNP
as an intermediate target. On purely economic grounds, nominal income targets are believed by some
to offer an effective way to absorb economic shocks or control inflation over time. As a public policy
tool, nominal GNP is sometimes seen as a way to limit the policy options available to the monetary
authorities and to communicate effectively the final goals of monetary policy. The major arguments
provided by advocates for targeting nominal GNP are presented in this section.
A. Absorbing the Impact of Economic Disturbances
Many analysts believe that under a nominal GNP target the impact of a wide variety of
economic shocks on prices and output could be effectively absorbed or deflected. The basis for this
view is fully developed in Appendix 1, where a textbook IS-LM model and a standard aggregate
supply and demand framework are used to examine the effects of economic shocks on price and real
output under nominal GNP targeting. For comparison, the impacts of shocks under money stock
targeting are also derived. The results of this static analysis are summarized in this section.4
When nominal GNP is targeted, demand shocks are neutralized fairly quickly and have no
lasting impact on prices and real output. The simple explanation for this result is that monetary policy
operates by moving aggregate demand (via changes in the underlying money stock); consequently, any
unexpected disturbance to aggregate demand is immediately countered by a policy-induced change in
the money stock that restores demand to its initial level. Thus, the price level and real output are
unaffected, and nominal GNP is maintained at its original (target) level. Of course, other important
economic variables, such as interest rates or the various components of GNP, might be altered in the
4
In order to examine the implications of nominal GNP targeting, Bradley and Jansen (1989) adopt an
approach similar to that used in Appendix 1 and summarized here.
233
process.
Unlike demand shocks, supply shocks have a lasting impact on prices and real output, at least
in the short to intermediate run when real output might deviate from its full-employment level. A
negative supply shock places upward pressure on prices and reduces real output. Under a nominal
GNP target, monetary policy is aimed at ensuring that these price and output movements are exactly
offsetting (in percentage terms), leaving nominal GNP unchanged. This is accomplished by using
monetary policy to move aggregate demand to where the original level of nominal GNP is restored.
With the new aggregate supply schedule, each level of real output is now associated with a higher
price level; therefore, at the new equilibrium point where nominal income is unchanged, prices will be
higher and output lower than they were originally. Advocates of nominal income targeting typically
maintain that splitting the impact of supply shocks on prices and output represents a good compromise
between the ultimate objectives of price stability and adequate real output. In contrast, attempts to
maintain the initial level of output (prices) will lead to excessive swings in prices (output). It can be
argued, however, that other combinations of price and output, ones that do not keep nominal GNP
fixed, might be preferable. For example, the comparative static results in Appendix 1 show that
targeting the money stock could leave output higher, but prices higher as well, than targeting nominal
income would, in response to supply shocks.
The results just summarized are derived using a static theoretical framework within which the
monetary authorities operate with complete certainty. Policy makers possess enough knowledge about
the underlying structure of the economy to be able to control aggregate demand with sufficient
precision to keep nominal income on target. In fact, the information required to target nominal GNP
successfully undoubtedly exceeds what is needed under other targeting regimes (see Section III).
There also are no lags or price dynamics in the simple static analysis. Policy makers can pursue a
chosen target without concern about its impact on inflationary dynamics or the formation of inflation
expectations. But even some advocates argue that strictly targeting nominal income can lead to
considerable instability in levels of real output if there is significant inertia in inflation. In such a
case, the monetary authorities might have to develop more flexible strategies for targeting nominal
GNP, a subject explored more fully in Section i n and in Appendix 2. Nonetheless, this simple
analytical framework illustrates the general theoretical arguments for targeting nominal GNP.
B. Control of Inflation
Targeting nominal GNP is often advanced as a way to control if not eliminate inflation over
234
time. This point is most often made by referring to the simple identity relating nominal output to
price and quantities:
(1)
nominal GNP = P*Q,
where P is the general price level and Q measures real output. The neutrality of monetary policy
actions on real output in the long run is assumed. Demographic and technological factors, which are
outside the control of monetary authorities, determine the "natural" or "potential" rate of real income
growth. Potential growth, it is argued, can be reliably estimated from long-run average rates of real
growth, and this rate is believed to be reasonably stable because its determinants are slow to change.
With potential output largely predictable and predetermined, from a policy perspective, the output
variable in the above equation, Q, is seen as fixed. Thus, targeting nominal GNP is equivalent to
controlling the general price level, and for this reason it is advanced as an effective strategy for
placing a lid on inflation over time.
The time horizon to which this argument for targeting nominal GNP applies is much longer
than the horizon implicit in the framework outlined in the preceding section.5 This difference in time
dimension often reflects different perspectives among the advocates of nominal income targeting on
the proper conduct of monetary policy. Generally speaking, advocates of nominal GNP targeting who
focus on the effectiveness of such a regime in responding to economic disturbances also tend to hold
the view that it takes a long time before monetary policy ceases to have an effect on real variables.
Within this time frame, policy makers are seen as having the ability to steer the economy to some
degree. Thus, a more "activist" approach in general should be taken in conducting monetary policy,
but a nominal GNP target would be helpful in guiding policy during these periods.
According to many others, however, the determinants of the price-output split of nominal GNP
are poorly understood, unpredictable, and largely outside the control of policy makers. While
monetary policy can have an impact on the interaction between nominal and real variables, the effects
of monetary policy on real variables are not long-lasting. Moreover, monetary authorities lack the
knowledge to exploit any systematic relation between nominal and real variables. Consequently,
policy should be aimed solely at longer term objectives like controlling price inflation. Seen from this
perspective, the analysis in Section IA and Appendix 1 overstates the ability of policy makers to
manage the economy in the short run. Nonetheless, nominal GNP targeting can still be useful as a
5
Kahn (1988) analyzes targeting nominal GNP in the "short run" using a model similar to the one
developed in Appendix 1 and summarized in the previous section. His description of targeting nominal
GNP in the "long run" follows the argument outlined in this section.
235
means for controlling inflation over longer time horizons for the reason outlined in this section.
The different arguments for targeting nominal GNP mirror a longstanding debate about the
proper conduct of monetary policy. As discussed in Section n, the strategies and operating procedures
advocated for implementing a nominal GNP target reflect these differing viewpoints.
C. Changes in Velocity
Interest in targeting nominal GNP arose during the early 1980s after unpredictable changes in
velocity complicated money stock targeting procedures. In fact, much of the analysis of targeting
nominal GNP is made in an explicitly comparative context to highlight its superiority over setting
targets for monetary aggregates when velocity is unstable. For advocates of this view, nominal GNP
is a substitute for a money aggregate target and shares many of the benefits, but not the drawbacks, of
a money target. This point is most often made using the simple money-income identity:
(2)
nominal GNP = M*V,
where M is a monetary aggregate and V is its corresponding income-velocity measure.
The "proof of the superiority of targeting nominal GNP is trivial in this framework. On the
assumption that nominal GNP is closely related to the final policy goals, it is better to target nominal
GNP directly than to target a monetary aggregate because unexpected velocity shocks can result in an
undesirable outcome for nominal GNP, with potentially adverse consequences for real output and
prices. That is, by targeting nominal GNP, the impact of changes in the income velocity of money on
the economy is, by definition, cushioned. While this argument makes no distinction between the
various possible sources of a change in velocity, during the past several years many analysts have
claimed that financial innovation and deregulation have led to deviations in the behavior of velocity
from longstanding trends. On this basis, they have argued that monetary authorities should target
nominal GNP directly rather than a money aggregate.6
6
Velocity movements induced by an evolving financial structure can be considered a special type of
demand shock, one that—in the framework presented in Appendix 1— shifts the LM curve. The appendix
shows that nominal GNP targeting is superior to money stock targeting for all types of demand
disturbances. Thus, arguments for targeting nominal GNP that focus on the behavior of velocity may be
considered derivatives of the arguments outlined in that appendix. However, focusing directly on the
money-income identity and velocity presupposes no knowledge about the split between prices and
quantities. Moreover, targeting the money supply might be preferred to targeting nominal income if
velocity shifts are caused by supply shocks and if there are certain real output objectives. As shown in
Appendix 1, targeting the money stock may be more consistent with policy objectives for real output than
(continued...)
236
The preceding demonstration of the relative merits of targeting nominal GNP generally
assumes that nominal GNP and a monetary aggregate can be controlled equally well. Alternatively,
the variability in velocity may be taken as evidence of the relative uncontrollability of income. Seen
from this perspective, the variability in velocity, instead of being an argument for targeting nominal
GNP directly, merely underscores the practical difficulties of implementing such a targeting regime, at
least through use of a monetary aggregate.
D. Constraints on Policy Objectives
Some advocates of nominal GNP targeting maintain that a nominal income objective would
impose a needed constraint on the monetary authorities. According to these analysts, policy makers
place undue importance on maintaining a certain level of employment and usually attempt to deflect
the effects of negative disturbances away from real output at the cost of accepting price increases.
This, it is argued, imparts an inflationary bias to the economy. Targeting nominal GNP, as Kahn
(1988) writes, "makes it impossible for policy makers to engineer a short-run increase in real output by
allowing inflation to rise." Instead, policy would be forced to maintain a given level of nominal GNP
without reference to the behavior of its price and real output components. Thus, the virtue of nominal
GNP targeting is that it prevents the kind of myopia sometimes associated with direct targeting of real
output.7
This argument for targeting nominal income is similar in spirit, but differs in detail, from the
argument made using a static model in Section LA when there are price shocks. In the static
framework, splitting the impact of disturbances evenly between prices and real output is seen as a
satisfactory outcome for these final goal variables. In a dynamic setting, an equal trade-off between
inflation and employment may not be a satisfactory outcome at all times, but over time, a nominal
income target will constrain policy so as to prevent an eventual or unintended sacrifice of price
stability.
6
(...continued)
targeting nominal income is, in response to supply shocks, depending on the slope of the aggregate
demand curve.
7
Of course, this assumes that policy makers do not adjust their targeted level of nominal income to
reflect changing objectives for prices and real output.
237
E. Communication of the Final Monetary Policy Goals
Most proposals call for initially specifying a target path for nominal GNP that lasts indefinitely,
based on a consensus long-run inflation and estimates of potential real growth in the economy (see
Section HA). Such a long-run nominal GNP target requires setting an explicit final objective for price
stability. However, the principal concern of many analysts is how (or whether) to set interim targets
when economic disturbances move nominal GNP away from its long-run targeted path. This section
considers the different views on the likely impact that announced nominal income targets drafted under
these circumstances could have on the policy debate.
Analysts disagree about the political consequences of using an intermediate target as closely
linked as nominal GNP to the final price and real output objectives of policy. Some believe that the
identity between nominal GNP and these final goals would enhance the communication of the
objectives of policy. Others are concerned that this same identity raises the likelihood that a nominal
GNP target will lead to public pressure to specify targets for prices and output—a development, it is
feared, that would ultimately undermine discipline in establishing realistic or appropriate objectives for
monetary policy.
One purpose of an announced intermediate target is to convey information about the goals and
purposes of monetary policy. Some argue that a nominal GNP intermediate target would be easy to
understand and would better communicate the purposes of monetary policy because the variable is so
clearly associated with the final policy objectives. Brittan (1983) believes that a target for nominal
GNP would be "much easier for people to understand, once they got used to it, than the bewildering
variety of money and credit measures" sometimes used to describe policy objectives. And Taylor
(1985) writes that "the aims of policy makers would be much easier to interpret if their goals for
nominal GNP were clearly stated." Moreover, the definitional relationship between nominal GNP and
prices and output could affect the nature of the policy debate surrounding the setting of intermediate
target values in a positive way. McNees (1987) argues:
A nominal GNP objective, reflecting both desired real growth and inflation, focuses attention
clearly on the basic policy choice: an "easier" policy will encourage faster nominal GNP
growth, which will promote real economic growth, but only at the risk of higher inflation; a
"tighter" policy will slow nominal GNP growth as is needed to reduce inflation, but only at the
risk of insufficient real growth.. . . Under nominal GNP targeting, the difficult social choice
between conflicting objectives receives primary attention.
Not all analysts agree that policy debates would be elevated in this way by using a nominal
GNP target. The close association between nominal income and the final objectives of policy might
238
suggest that by targeting income the monetary authorities possess an unrealistic degree of direct
control over prices and real output, contributing to equally unrealistic expectations for policy.
Furthermore, a number of policy makers have written that as a practical matter it would be difficult to
avoid specifying objectives for prices and output when setting a nominal GNP target. Axilrod (1985)
argues that "setting a nominal GNP target is unlikely in practice to spare the central bank from taking
a view on the mix between real activity and prices," and he expresses concern that strong pressures
"would work toward an upward bias in an announced real GNP objective." In a similar vein, Solomon
(1984) believes that under a policy of targeting nominal income it would be difficult to resist an
evolution "toward setting separate objectives for the price and real output components of GNP," and
"the tendency to set GNP goals chronically too high would be very strong." Volcker (1983) further
maintains that "the independent status of the Federal Reserve that makes a longer term view possible
might well be compromised with GNP targeting, since the Federal Reserve could be under great
pressure to conform its target to some immediately attractive number."
F. International Coordination
Nominal income targets figure prominently in several recent proposals for international
coordination of economic policy. Frankel (1989) and Williamson and Miller (1987) advocate
cooperatively setting nominal GNP targets. These proposals also often call for international
cooperation in setting fiscal policy and for establishing targets for external variables.8 One issue
raised by these studies concerns the choice of target variable. Frankel, like Williamson and Miller,
argues that it would be inappropriate to pursue an income objective at the expense of trading partners,
say, through protectionism. Furthermore, Brittan (1983) writes that the spillover of nominal demand
into imports is largely outside the control of monetary authorities. For these reasons, in an
international environment, a measure such as domestic demand (GNP less net exports) may be better
suited than nominal income as an intermediate target. However, these proposals are not examined
further in this study because the basic rationale for using a nominal income target is the same as in a
closed economy context, and because most of these proposals also require setting explicit objectives
for fiscal policy.
8
Williamson and Miller assign to fiscal policy the task of hitting the nominal income target, while
monetary policy is used to achieve the target set for either the exchange rate or the external balance.
239
II. Strategies for Targeting Nominal GNP
Proponents of nominal GNP targeting, like advocates for most other target variables, typically
write that the "details" for implementing their proposals remain to be worked out. However, even
among the advocates of such a policy regime, there is no general consensus about the guiding strategic
principles to be applied when targeting nominal income. Perhaps the most important difference
centers on how (or whether) target values for nominal GNP should be reset in response to economic
disturbances—particularly supply shocks—that move the economy away from its previous target path.
This section begins with an overview of some income targeting proposals. Next, the dynamic
properties of some of these proposals are evaluated using a simple economic model with unexpected
supply disturbances. Finally, some practical and conceptual issues raised by these proposals are
discussed.
A. Overview of Proposals
There is little disagreement about the general nature of an intermediate target for nominal GNP.
Among virtually all proponents, nominal income is seen as a long-term intermediate target.
Recommended target periods usually fall between one and two years and are sometimes much longer.
Because of long lags, the potential value of nominal GNP as a leading indicator or an information
variable is limited, although some advocates believe that nominal income could play a feedback role,
triggering changes in settings for operating instruments.
Upon adoption of nominal GNP as an intermediate target, most proposals call for setting a
long-run path for nominal income that is consistent with ultimate inflation goals and with expectations
about potential real GNP growth. Hall (1983) recommends that a path be set "once and for all," as
does McCallum (1987). Gordon (1985) suggests setting a desired path for nominal GNP ten years
ahead. This target path is designed to serve over an extended period, perhaps indefinitely, or at least
until new estimates of potential growth or changing inflation objectives require its modification. The
significance of this initial step is that it binds the monetary authorities to a permanent objective for
nominal GNP, one that is intended to prevent them from arbitrarily casting any immediate price and
real output objectives in terms of nominal income and labeling the practice "targeting nominal GNP."
In the subsequent discussion, this path is referred to as the "initial" or "long-run" nominal GNP target.
There is some difference of opinion about whether the long-run path should be constructed in
terms of levels or growth rates of nominal GNP. A target path specified in level terms has the
advantage of ensuring over time the target price level (assuming the path for potential output is
240
unchanged) and, therefore, the value of economic contracts established years earlier. Thus, much of
the uncertainty under which agents operate would be eliminated. Others, however, advocate pursuing
a long-run growth rate for nominal GNP. The economic costs of returning to a predetermined level of
nominal income might be unacceptably high following inflationary shocks. Moreover, following
deflationary shocks, the monetary authorities might see no purpose in pursuing temporarily high
inflation rates just to return nominal income to its earlier level path. The question whether a long-run
path for nominal GNP should be set in terms of levels or growth rates to some degree mirrors the
debate over rebasing under money targeting. The preferences of some advocates of nominal GNP
targeting are spelled out in the following overview; however, the distinction between setting long-run
targets for levels or growth rates does not affect the basic conclusions reached in the later analysis of
the dynamic properties of income targeting.
Most proponents of income targeting recognize that economic disturbances or policy mistakes
can take nominal GNP far away from its long-run target path. Axilrod (1985) writes that "the focus of
a GNP target will inevitably shift away from the intermediate or longer run" to stress the short run,
and advocates have proposed various strategies for determining interim targets or objectives for
nominal GNP. The importance of this issue is that a rapid return of nominal GNP to any long-run
path could induce unacceptable instability in real output-whether this path is defined in terms of levels
or growth rates. This overview now turns to the specifics of some of these proposals and their
differences.
In perhaps the simplest statement of how a nominal income target might operate, Hall (1983)
proposes that
once and for all, Congress would adopt a target path for nominal GNP. In the future, if
nominal GNP were above the path, monetary policy would be judged excessively expansionary
and would be required to contract as necessary to bring nominal GNP back to the path. If the
economy slipped below the path, monetary expansion would be called for.
Under Hall's proposed target path, which is specified in levels, when nominal income falls below its
long-run path, higher growth targets of nominal GNP would be set temporarily. In terms of policy
strategy, this leaves unanswered how quickly to return to the long-run target path when off course.
McCallum (1987,1988a,1988b) advocates targeting nominal GNP levels in a series of papers.
Unlike most other analysts, McCallum is concerned more with the practical problems of achieving
such targets, which he argues are best solved by using the monetary base as an operating instrument.
His operational rule is detailed fully in Section IIIB, but its implications for setting near-term nominal
GNP targets are summarized here. Although McCallum does not actually propose to announce
241
publicly nominal GNP targets on an interim basis when income deviates from its long-run path, a
strategy for returning nominal income to its desired long-run level can be inferred from his operating
rule.
McCallum sets growth in the operating instrument (the monetary base) each period equal to the
desired long-run growth rate of nominal GNP-the estimated potential rate of growth of real output
plus desired long-run inflation—with an adjustment for past deviations of nominal GNP from its target
level. This partial adjustment mechanism will eventually return nominal income to its long-run target
path, but at a slow enough pace so as to avoid the kind of dynamic instability that can arise when
feedback rules are too strong. McCallum recommends that the most recently observed gap between
the level of nominal GNP and its long-run target level be used to adjust growth in the operating
instrument, thereby raising base growth whenever nominal income is below its initial path. Because of
lags in the adjustment of nominal GNP to changes in the operating instrument, it may take some time
before nominal GNP returns to its long-run target level.
McCallum also proposes that monetary base growth be adjusted in response to past changes in
the income velocity of the base. In this way his operating rule is akin to regimes that target monetary
aggregates but periodically revise the money objectives in response to past or expected changes in
velocity. An important theme underlying McCallum's proposal is that policy makers cannot know,
much less determine, how changes in nominal income will be divided between inflation and output
growth in the short run. Furthermore, he maintains that smoothing movements in nominal income by
only slowly adjusting changes in the operating instruments will reduce swings in real output around its
potential level.9
McCallum's temporary adjustments to the target path for his operating instrument are based
solely on past deviations of nominal GNP from its long-run target level. In contrast, some other
advocates of income targeting suggest that explicit interim targets for nominal GNP should be
constructed to take into account available information on current real economic conditions and
knowledge about the trade-off between inflation and real output. For instance, Tobin
(1983a, 1983b, 1985) calls for periodic revisions in targeted income levels to lessen the impact of some
disturbances on real output. And in two articles, one by Gordon (1985) and another by Taylor (1985),
more detailed strategies of nominal GNP targeting are proposed. An overview of these targeting
9
In a subsequent paper, McCallum uses simulations from different models of the relationship between
nominal and real GNP to support his contention that his rule would help stabilize real GNP as well (see
National Bureau of Economic Research, Working Paper no. 3047).
242
strategies follows, and the dynamic implications of some of the strategies for real output behavior are
examined in the next section using a simple dynamic model.
Gordon focuses on the need to modify nominal income targets when faced with economic
disturbances.10 An initial long-run desired path for the level of nominal GNP should be set on the
basis of potential growth and desired inflation, and, in general, monetary expansion is called for
whenever income is below this path. However, Gordon mostly examines how monetary policy should
be conducted when unemployment is away from its natural rate, although much of his discussion is
conducted in general terms. According to Gordon, because of aggregate supply constraints, a rapid
return to the targeted path of nominal income or strict adherence to a policy of constant nominal GNP
growth under these circumstances could induce cycles in real GNP growth. To avoid this kind of
oscillating behavior when unemployment is away from its natural rate, monetary authorities should
decelerate the growth rate of nominal GNP as real output nears its potential level. Gordon does not
specify a procedure for setting interim nominal GNP targets, and it is not clear from his statements
how the eventual return of nominal GNP to its long-run target level would be guaranteed.
Taylor also focuses on the implications of targeting nominal GNP for business cycles, and he
too finds that after supply shocks, some conditions targeting nominal income can lead to instability in
real output because of inflation inertia.11 To reduce oscillations, Taylor proposes a rule for setting
nominal GNP targets:12
(3)
qt - q*t = Dp* - Dpp
where q is the log level of real output, q* is potential output, and p is the log level of prices. D is the
change operator, and Dp* represents the long-run desired rate of inflation. After some algebraic
manipulation, this expression may be rewritten as:
(4)
Dxt = Dx* - ( qM - q*M ).
This states that the current period's target rate of growth in nominal GNP (Dx,) is equal to the
10
Gordon advocates the use of a nominal final sales target (nominal income less changes in
inventories) in place of an income target. The series for final sales is less erratic than income because it
is not subject to short-term inventory cycles, and for that reason it is perhaps a better measure of
underlying demand.
n
The "elastic price standard" proposed by Hall (1984) is similar to Taylor's rule, except that Hall uses
levels of nominal income rather than rates of growth.
12
If desired inflation is set to zero, then this expression reduces to (qt - q*t) = -Dpt, the formulation
used by Taylor.
243
long-run desired rate of growth (Dx*) minus the percentage deviation of real output from its potential
level in the preceding period. Thus, nominal income growth is raised (lowered) whenever real output
is below (above) potential. Unlike other proposals, this rule includes no mechanism that returns
nominal income to a predetermined level but only one that returns it to a long-run desired rate of
growth.13
B. Dynamic Properties of Targeting Strategies
As noted in the preceding section, Gordon and Taylor examine the potential dynamic
implications of strictly targeting either the level or growth rate of nominal GNP, and they find that
some income targeting policies can induce destabilizing oscillations in real output following supply or
price disturbances.14 They argue that interim nominal GNP targets should be designed to avoid or
dampen these cycles. A summary of their analyses is presented in this section. A more detailed
presentation of the model and simulation results is in Appendix 2.
A simple aggregate supply relationship or price adjustment mechanism states that current
inflation (Dpt) depends on its past values and the ratio of real output to potential output:
(5)
Dpt = DpM + a*( q, - q*t) + ut, a > 0,
where p and q are defined as above, and ut represents the impact effects of supply shocks on
inflation.15 Inflation tends to rise or fall according to whether output is above or below potential, and
the equation incorporates the "natural rate" hypothesis that output above potential cannot be sustained
without accelerating inflation. In the simulations reported in Appendix 2, a distributed lag on past
inflation replaces Dpt_i. This relation assumes considerable inflation inertia, and therefore it is
probably more useful for analyzing the impact of shocks over a somewhat limited time horizon.
13
Taylor presents a more general version of equation 3 to allow for a different desired trade-off
between inflation and stability in real output:
qt - q*t = R*( Dp* - Dpt), R>0.
With smaller values of R, there is less tolerance for fluctuations in real output, but there is greater
tolerance of inflation. This equation may be rearranged:
Dx, = Dx* - ( q^ - qV, ) + <R-1)*< Dp* - Dp t ).
If, for example, R is below 1 and inflation is above its desired level, then the targeted growth of the
nominal GNP target for the period will be above its long-run desired rate of Dx*.
14
The effect of demand shocks are also evaluated by Taylor, but these have a much smaller impact
on business cycles, and they are not considered in this review.
15
This model is just a dynamic version of the static aggregate supply relation used in Appendix 1.
244
To facilitate the presentation of the dynamic implications of the above supply constraint, the
definition of aggregate demand is broadened to include the monetary policy reaction function that
keeps nominal income on its targeted path.16 For example, if the level of nominal income is targeted
each period to remain on its long-run path, then aggregate demand becomes:
(6)
p, + qt = x*t.
Equations 5 and 6 are solved to determine the level of prices and output resulting from this targeting
strategy on the assumption that target values of nominal GNP are hit each period except when there
are unexpected shocks. Equations representing aggregate demand under other targeting strategies are
presented in Appendix 2.
The results reported in the graphs in Appendix 2 show that if the targeted level or growth rate
of nominal GNP is not adjusted during transition periods, then an inflation shock will cause
considerable variability in real output around its potential level. Intuitively, remaining on the target
path after prices jump initially requires a large reduction in real output because of inflation inertia.
Output must then remain below potential while inflation slowly decelerates. Gradually, the pace of
deceleration in inflation quickens and output growth rises, but then the inertia in the process forces
output above potential while inflation continues to drop. A period of slowly dampening cycles ensues.
This cyclical behavior remains even when McCallum's rule is used in these simulations.17 Adjusting
nominal income targets on the basis of past deviations of nominal income from its long-run target path
does not guarantee that instability in real output can be avoided.
Gordon writes that avoiding this type of instability requires a gradual deceleration in the
growth rate of nominal GNP as real output approaches its potential level. Taylor's proposal provides
one way of doing this: following an inflationary price shock, targeted nominal GNP growth is raised
and only gradually lowered as the economy nears full output.18 Oscillations in real output are
dampened considerably when this rule is used.
Although these results suggest that nominal income targets should be adjusted to reflect current
16
fti Appendix 1, aggregate demand is derived by collapsing the IS and LM curves, and the target rule
guiding monetary policy is a separate function. Here, the definition of aggregate demand incorporates the
monetary policy reaction function.
17
For this exercise it is assumed that the income velocity of money is fixed. The results are not
affected by expanding the model to account for velocity changes.
18
Taylor's rule is not entirely consistent with Gordon's proposals. As the charts in Appendix 2 show,
under this rule nominal GNP will not be returned to its initial target level as is advocated by Gordon.
245
conditions of unemployment or real output, a number of criticisms are often leveled against this kind
of analysis. Successfully targeting income each period requires first calculating the impact of
movements in real output on inflation and then controlling real output so as to hit that period's
nominal GNP target. This, critics maintain, assumes an unrealistic degree of knowledge about the
structure of the economy and of control over real output. Moreover, doubts are raised about the
stability of the price adjustment mechanism and the expectations formation process. A proven longterm commitment to a predetermined target might better ensure that inflation expectations remain
consistent with policy goals and are not upset by temporary disturbances. This, it is argued, is better
achieved by adhering as closely as possible to a predetermined path of nominal income rather than
periodically revising policy objectives in light of current real economic conditions.
C. Strategic Issues in Nominal Income Targeting
The overview of proposals to target nominal GNP demonstrates that the structure of such a
regime could take many forms. Some questions and issues raised by these proposals are discussed in
this section.
Establishing a long-run path for nominal GNP
Adherence to a permanent or semipermanent path of nominal GNP, whether stated in levels or
growth rates, is an important feature of most proposals. Establishing this long-run path requires
estimation of the potential rate of real growth and agreement upon a long-run target for prices or
inflation.
As for prices, in most academic studies the desired or targeted long-run rate of inflation is
assumed to be zero. However, there is a commonly held perception that as the rate of inflation falls,
the benefits of lowering inflation further begin to be offset by the economic sacrifices that might be
needed to achieve further reductions. Thus, a zero rate of inflation may be resisted by some as an
impractical goal. Instead, an "acceptable" rate of inflation may become the standard, but for this it
might be difficult to reach a consensus. While this problem always exists in setting policy, the
difficulty it presents is enhanced under most proposals for nominal GNP targeting because of the
identity linking nominal GNP to the price level and the "once and for all" nature of the long-run target
path.
For the most part, these proposals assume that reliable estimates of potential real output growth
are available, and that nominal GNP targets will be reset when changes in the natural rate of growth
246
are detected. But there exist different views about potential output growth and the natural
unemployment rate, and considerable controversy would undoubtedly accompany any revisions to
nominal GNP targets based on perceived changes in this variable. Current estimates of potential
growth generally range from less than 2 1/2 percent to a little above 3 percent per annum. When
potential growth is overestimated, successively raising a nominal income target in an attempt to reach
full employment could lead to accelerating inflation.
Authority for setting targets
Many advocates of nominal income targeting argue that authority for setting targets should not
rest just with the Federal Reserve System. Hall, who is quoted above, believes that Congress should
have sole responsibility for setting these targets, while Taylor writes that "the Fed in conjunction with
the Congress and the administration should state realistic forecasts of nominal GNP growth conditional
on their intended plan for monetary and fiscal policy."
Several reasons are given why the Federal Reserve should not be granted sole authority for
setting targets under nominal GNP targeting. Because of the close relation between nominal GNP and
the final objectives for price and real output, all government branches that help set economic policy
have a natural interest in establishing such an important benchmark standard. This would be
particularly true if, as some proposals suggest, a target value for nominal GNP were to be set over
very long time horizons. As it is, the executive and legislative branches currently present forecasts for
nominal GNP growth in shaping fiscal policy for the coming years, along with expected inflation and
real growth. It may be inappropriate for the Federal Reserve to have a different nominal income
objective.
Ideally, a consensus between the Federal Reserve and the federal government on a nominal
income objective would imply improved coordination between monetary and fiscal policy. In practice,
difficulties could easily arise in jointly setting nominal GNP targets. At the outset, persistent
differences in estimates of potential real output growth between various agencies charged with setting
policy might emerge. This, combined with any difficulties in reaching agreement on an acceptable
long-run goal for inflation, would complicate setting initial targets for nominal GNP. Furthermore,
many proposals for nominal GNP targeting address the need to revise target objectives in response to
economic disturbances that move the economy away from its long-run target path. The difficulties in
achieving a consensus between the Federal Reserve and the government on the objectives of policy
would undoubtedly increase with the frequency of these revisions. Moreover, several policy makers
247
have voiced concern that as a practical matter it would be unrealistic to assume that the Federal
Reserve could maintain its independence if it were required to set explicitly objectives for broad
economic measures like nominal GNP that are of such direct interest to government officials.
Time horizon for nominal GNP targets
Most proposals call for establishing a long-term target path of nominal GNP, based on
estimates of potential real GNP growth and a consensus inflation goal, with modifications made only
to reflect changes in potential growth and perhaps slowly evolving long-run objectives for inflation.
While many of these proposals also focus on appropriate ways to return nominal GNP gradually to its
long-run target after disturbances have moved it away from this path, few advocates are specific about
the time frame over which these targets would be set. Most proposals call for short-run targets that
begin anywhere from one to four quarters in the future and remain in effect for a period of one to two
years. This is done out of recognition of the long lags commonly believed to exist between the
operating instruments and nominal GNP. Still, some critics question whether it would be realistic to
set nominal income targets for even a year or two ahead because of the uncontrollability of nominal
income, an issue examined in further detail in Section in.
Short-term considerations in conducting policy
Advocates of nominal GNP targeting are not in agreement about the use of rules versus
discretion in implementing policy or in setting near-term income targets. In particular, they differ on
the extent to which short-term considerations should influence the setting of interim nominal GNP
targets. Nominal income targeting in some form is compatible with very different views on the proper
conduct of monetary policy.
McCallum writes that "a rule exists when the policymaker chooses not to attempt optimizing
choices on a period-by-period basis, but chooses rather to implement in each period a formula for
setting his instrument that has been designed to apply to periods in general, not just the one currently
at hand."
McCallum's rule for setting his operating instrument is in keeping with this view. While some
consideration is given to potentially undesirable output dynamics, for the most part the rule is
motivated by the view that a trade-off between prices and real output cannot be exploited by policy
makers, and that in the long run the credibility of the monetary authorities would be undermined by
248
attempts to do so.19 Short-term nominal income objectives cannot be constructed under this rule to
take advantage of a perceived trade-off.
Many other proposals are premised on the view that there is sufficient knowledge about the
structure of the economy to justify setting shorter term nominal GNP targets with explicit underlying
objectives for real growth and inflation. Short-run considerations do matter in setting monetary policy,
and nominal income targets should be periodically modified to reflect these considerations. From this
perspective, an operational rule like McCallum's is not seen as providing sufficient flexibility for
achieving near-term objectives.
Advocates such as Gordon and Tobin argue for "discretion" in setting interim nominal income
targets. In contrast, Taylor proposes an explicit rule for setting income targets that depends at any
time upon the values of real variables. However, it is shown in Appendix 2 that in its modified form,
Taylor's rule tends to function as a period-by-period optimizing solution to a standard loss function
instead of optimizing over a longer horizon. Thus, Taylor's proposed rule is not a "rule" in the sense
used by McCallum, and it may not differ significantly from the kind of short-term optimizing
strategies often decried by some analysts.20
In sum, intermediate targets are often proposed as a way to prevent authorities from focusing
narrowly on near-term objectives, such as a real output or an unemployment goal, to the potential
detriment of longer term objectives such as price stability. However, any combination of objectives
for inflation and real growth can be framed in terms of a nominal GNP target. Proposals that call for
frequent modifications to income targets in order to, say, dampen fluctuations in real output may not
prevent longer run goals such as price stability from being ultimately sacrificed in the process. On the
other hand, rigid operating rules may preclude the flexibility in setting intermediate targets needed to
moderate cyclical movements.
19
The paper by A. Steven Englander in this volume deals generally with the time inconsistency
problem in setting monetary policy.
Baylor's formula does not conform to McCallum's idea of a "rule" for several other reasons as well.
McCallum (1987) writes that a rule should apply to a variable under the direct control of the monetary
authorities. Moreover, it should be designed in recognition of the limits of macroeconomic knowledge
and not be based on any particular model of the interaction between real and nominal variables in the
economy.
249
HI. Controlling Nominal GNP
Proponents often argue that the monetary authorities ultimately can control nominal GNP
because it is a nominal magnitude, but many critics question whether this control can be exerted over
time horizons relevant to policy makers. This criticism pertains especially to strategies to set income
targets covering periods other than just the very long run. To compensate for any lack of control,
some suggest using target ranges or working with moving averages of nominal GNP. Even in these
circumstances, it is questionable whether nominal income could be controlled accurately enough to
serve as an intermediate target. This section explains why controlling nominal GNP is difficult,
considers possible operating procedures for targeting nominal income, examines the possible role of
forecasts in controlling income, looks at some empirical evidence of controllability, and discusses the
possible consequences of targeting a variable that cannot be adequately controlled.
A. Sources of Uncontrollability
Controlling nominal GNP is difficult because the lags between the policy instruments and
income are long, imperfectly known, and probably variable. Also, many factors other than monetary
policy determine nominal income.
Impact lags measure the length of time that elapses before changes in policy have their effect
on targeted variables. By most estimates, between one and two years must elapse before changes in
policy instruments have their median impact on nominal GNP. Typically, it is believed that between
six and nine months will pass before changes in monetary policy have a substantial effect on real
output, and up to another year will pass before inflation is noticeably affected. In addition to being
long, these lags are also of uncertain duration. This makes it difficult to measure the impact of past
policy changes on the target variable and reduces the value of the targeted variable as a feedback
source of information on policy actions. Some analysts maintain, however, that by understanding the
ultimate implications of a policy action, market participants can accelerate its impact on the economy
and significantly reduce the lags in the policy process, as long as policy actions are seen as "credible."
A proven adherence by monetary officials to a nominal GNP target could effectively shorten this
impact lag.
Setting target objectives and operating procedures for nominal GNP is complicated by
uncertainty about the current level of the target variable. Initial quarterly estimates of nominal GNP,
along with a breakdown into real output and price measures, are available about a month following the
end of the quarter. Sometimes substantial revisions to this data appear later. Lags in the availability
250
of official GNP statistics contribute to the practical difficulties in targeting nominal GNP, although
alternative information sources may help reduce this lag. Much of the underlying data used to
construct quarterly GNP statistics is available on a more timely basis, including payroll employment
statistics, consumer and producer price data, and measures of industrial output and shipments.
Nominal income is a broad measure of economic activity, and for that reason may be affected
by a variety of factors in addition to monetary policy. A host of economic disturbances as well as
sustained shifts in the spending behavior of major economic agents such as consumers, businesses, or
the government can greatly alter the level of nominal GNP, and the lags in recognizing these changes
may be long. For all these reasons, nominal GNP may remain largely outside the control of the
monetary authorities for many quarters.
B. Operating Procedures for Targeting Nominal GNP
Proponents of nominal GNP targeting often do not discuss in detail the operational procedures
that would be appropriate in this regime. Yet because of the delays and uncertainties involved in
measuring the impact of policy changes on nominal GNP, the target variable itself will provide little
guidance for policy operations.
Tobin (1985) advocates establishing a hierarchy of targets, each consistent with the other,
starting with settings of nominal income for periods of one year or more and extending back to the
operating targets of monetary control, such as bank reserves or short-term interest rates. But his
proposals do not address which specific lever to use to implement policy.
McCallum offers a specific operating rule for implementing a strategy of nominal GNP
targeting. His rule for the quarterly setting of his operating instrument, the monetary base, is:
(7)
Dbt = Dx* - (1/16)( VM - VM7 ) + B*( x*M - x,.,), 0<B<1,
where b is the log of the monetary base, V is the log of the velocity of the base, x is the log of
nominal GNP, x* is its long-run target level, and D is the change operator. The growth rate in the
monetary base each quarter equals the long-run desired rate of growth of nominal income, adjusted for
past changes in the income velocity of the base and for past deviations in income from its long-run
target path. The partial adjustment term, B, is chosen so as to ensure that nominal income will
eventually return to its long-run targeted path, but without the dynamic instability that abrupt changes
in the operating instrument could induce.
In many ways, McCallum's rule is similar to procedures that modify growth targets of
monetary aggregates to offset movements in income velocity, but it differs in its use of the monetary
251
base as the controlling instrument. According to McCallum, the strength of the proposal rests upon
the fact that the base is under the direct control of the monetary authorities.21 Moreover, McCallum
cites a close historical correlation between the monetary base and nominal income, although the exact
linkages between these variables are not spelled out. Critics argue that the monetary base is a flawed
operating tool of policy because its large currency component is subject to erratic changes in demand.
B. Friedman (1988) writes that changes in income have been important in driving movements in
currency, not vice versa, and he finds that excluding the currency component eliminates any statistical
relation between nominal GNP and the monetary base.
C. Use of Forecasts in Targeting Nominal GNP
Intermediate targets are intended in part to alert policy makers to needed changes in their
operations. Nominal GNP may not be well suited for this purpose because of the long delays in data
availability. To compensate for the lags in data availability, some advocates of nominal GNP targeting
propose to use income forecasts in their operations. A path for instrument settings, such as interest
rates, would be set on the basis of periodic GNP forecasts made by the Federal Reserve and possibly,
as some advocate, using forecasts made by others. Instead of reacting to past (observed) income
movements, policy makers would respond to expected movements in future nominal GNP based on
forecasted values. Gordon (1985) writes that "there would be nothing novel in the Federal Reserve's
using its own forecasts, since it does this already." However, their use in this suggested manner would
represent a significant increase in their policy role.
Opinions vary about the reliability of forecasts of nominal GNP. Volcker (1983) comments
that "the uncertainty and unreliability of economic forecasts have been amply demonstrated over recent
years" and expresses reluctance to grant forecasts any formal role in the execution and formulation of
monetary policy. Forecasts might be used not only to provide point estimates of future income, but
also to flag turning points in the economy. In summarizing two studies on the forecastability of
nominal GNP, Kahn (1988) writes that "forecasts of nominal GNP have been good enough to predict
several important cyclical turning points in the economy." However, this conclusion is based on an
analysis of composite indexes of forecasts. Any single forecasting model might not consistently
anticipate turning points.
21
The assumption that the total base, as opposed to the nonborrowed reserves portion of the base, is
strictly controllable over the relevant period is open to debate. The paper by Meulendyke in this volume
examines the money base as an intermediate target.
252
Many who advocate using forecasts are unclear whether separate forecasts are needed for prices
and real output or whether an outlook for nominal income alone is required. Some proposed targeting
strategies make use of expectations for real output and inflation. Separate forecasts for these variables
might be needed, but these are usually less reliable than forecasts for nominal income alone. In an
examination of a few large econometric models, McNees (1985) finds that errors in underpredicting
inflation tend to offset errors in overpredicting real growth.
Not all proponents of nominal GNP targeting advocate the use of forecasts. In keeping with a
view that the links between policy variables and output are poorly understood, especially as they
pertain to the price adjustment mechanism, McCallum's operational rule calls for modifying control
instruments solely on the basis of incoming data for nominal GNP. This, he argues, is better than
basing policy on fallible forecasts. At the same time, such a "backward-looking" policy, that is, one
that responds just to current conditions and past errors, is frequently criticized because it risks
accentuating business cycles by, say, stimulating activity only after output is already beginning to
expand.
D. Empirical Evidence on Controllability
Empirical evidence provides a mixed picture of how well nominal GNP can be controlled with
the instruments available to the monetary authorities. Large macroeconomic models are often used to
study the effects of monetary policy on the economy, and these usually indicate that nominal GNP is
difficult to control because the links between the instruments of monetary policy and nominal income
are indirect and complex, and because nominal GNP is affected by many factors that cannot be
predicted or controlled. However, some critics maintain that the imposed structure of these models is
often arbitrary and so should not be used to make policy inferences.
Evidence of a statistically significant relation between nominal income and control instruments
is often found in smaller scale models where direct links are estimated. However, usually such
equations explain just a small fraction of the total variation in nominal GNP. This is true, for
example, of the model of GNP reported by McCallum (1987). More formal tests of "causality" also
confirm that monetary policy actions are correlated with nominal GNP. Kahn (1988) performs
bivariate Granger causality tests between changes in nominal GNP and a control instrument, both
lagged four quarters. He finds that the hypothesis that the federal funds rate does not cause nominal
GNP can be rejected. The hypothesis that the monetary base does not cause nominal income is also
rejected by the same tests. However, Kahn questions whether the relation, while statistically
253
significant, is "economically strong." That is, his results indicate that large movements in policy
instruments might be required to make relatively small adjustments in nominal income. Overall, this
evidence lends support to the conclusion that nominal GNP may be affected by monetary policy in a
somewhat predictable way in the long run, but the degree of control over shorter time horizons is
limited.
E. Consequences of Uncontrollability
The available evidence on the controllability of nominal GNP is not conclusive, but even
among its advocates the view holds that nominal income is less subject to the control of the monetary
authorities than other potential intermediate targets. Whether this undermines the usefulness of
nominal GNP as an intermediate target may depend upon the nature or source of this uncontrollability.
Nominal GNP targets might still be useful if the "noise" preventing the monetary authorities
from hitting those targets with any precision is caused mostly by random and unexpected changes to
the many variables affecting nominal income. This would be like adding an error term to the
aggregate supply or demand relationships in the simple analyticalframeworkpresented in the
appendixes. As long as the impact of monetary policy actions on the economy remains known with
sufficient certainty, then the basic conclusions from these sections would carry through in terms of
expectations. Targeting nominal GNP may still be justified on the grounds of the expected outcome of
this strategy.
Many policy makers are concerned that even in this environment the value of a nominal GNP
target as a standard of accountability is undermined because market participants would not know
whether to interpret a "miss" as signaling a shift in policy or as reflecting the effect of an
unanticipated random shock. A persistent inability to deliver on a nominal GNP target is seen as
destroying the usefulness of the target as a tool for evaluating Fed performance. Axilrod (1985) writes
that "because it is not attainable with a reasonable degree of certainty, a nominal GNP target would
almost inevitably call the central bank's credibility into question." But proponents see in this
argument merely an attempt to avoid accountability. Gordon (1985) observes that adopting a relatively
uncontrollable nominal income target would not add a new dimension to the public aspects of setting
policy because the Fed "has already become expert at giving excuses" for misses in its targets.
Ultimately, they maintain, the performance of the economy is the only measure of success of monetary
policy that matters, and it remains the only guarantee of credibility for monetary authorities.
Uncertainty about the structure of the economy, including the timing and ultimate impact of
254
policy changes on nominal GNP, also may undermine the value of nominal GNP targets.
Inappropriate or "noisy" instrument or target settings may be made when the basic structure of the
economy is unknown or misspecified. Under nominal GNP targeting, these errors may be identified
only after a substantial period of time has elapsed because long lags prevent a timely monitoring of the
impact of policy on income. Improperly set instruments or targets represent a type of
"uncontrollability" that injects a persistent bias into monetary policy.
IV. Summary and Conclusions
Several reasons are typically advanced for using nominal GNP as an intermediate target in
monetary policy. Advocates frequently maintain that under a nominal income target, economic
disturbances affecting aggregate demand will be entirely offset, while the impact of aggregate supply
or price shocks will be spread evenly between prices and real output. This view rests on a textbook
macroeconomic analysis in which the monetary authorities act to keep nominal income at its targeted
level. Some proponents of income targeting claim that had nominal GNP been targeted directly, the
complications that unexpected velocity movements caused under money stock targeting could have
been avoided. This conclusion follows from the identity that links money and velocity to income.
Taking a somewhat longer perspective, some argue that setting targets for nominal GNP would better
enable the monetary authorities to control inflation over time. Simply put, because the level of real
output is outside the control of the monetary authorities over the long run, the identity between
nominal income and the product of prices and real output means that setting target values for nominal
income translates into setting targets for prices or inflation. This argument for targeting nominal GNP
is more compatible with a view that policy makers should focus solely on the long run and should not
attempt to set near-term real output objectives, however acute the apparent need.
Reservations have been expressed about using nominal GNP as an intermediate target because
of its relative uncontrollability. Long lags in data availability and policy effects, the sensitivity of
nominal income to many economic variables, and the general lack of knowledge about the structure of
the macroeconomy combine to place nominal GNP largely outside the direct control of the monetary
authorities. Yet in the simple models typically used to demonstrate the value of targeting nominal
income, almost complete control is assumed. Even in the most basic models it is clear that the
information that would be needed to target nominal income successfully far exceeds what would be
required to target, say, the money stock. The volatility displayed in recent years by the income
velocity of money, a development often cited as a reason for targeting income directly, can be
255
regarded as evidence that nominal GNP is difficult to control, at least through a monetary aggregate.
The issue of controllability raises a general point about nominal GNP as an intermediate target.
Intermediate targets are largely valued because they stand somewhere between the controlling
instruments and final objectives of policy and thereby provide needed information and guidance to
policy makers. Nominal GNP, by virtue of the identity that links it to prices and real output, is
skewed towards the ultimate policy objectives, a condition that leads to offsetting considerations. On
the one hand, it may be possible to frame a nominal income target that in theory could be used to
achieve desirable outcomes for these final objectives; but on the other hand, the ability to control the
intermediate target may be relatively low in the case of nominal GNP because its links to the operating
instruments are remote.
Some arguments for targeting nominal income rest more on its symbolic and disciplinary value.
Policy makers are frequently criticized for instilling an inflationary bias in their actions by focusing on
the short-run real output gains of an expansionary policy at the cost of adding to inflationary pressures.
A nominal income target, it is argued, would effectively constrain the monetary authorities from
pursuing such policies by forcing them to maintain a given level of nominal income without regard to
the split between its price and real output components. Furthermore, nominal income is seen as a
relatively comprehensible measure, one that could more readily express the objectives of monetary
policy.
At the same time there is some concern that because of the close identification of nominal
GNP with prices and real output, a highly visible nominal income target could place counterproductive
pressures on policy makers. Several policy makers have expressed the view that, as a practical matter,
it would be difficult to avoid setting separate objectives for real output and inflation under nominal
GNP targeting. Moreover, the economic benefits that in principle might arise from targeting nominal
GNP may be outweighed by the policy makers' loss of credibility in targeting a variable that often
remains largely outside their control. These arguments underscore the public policy dimension of
targeting nominal income that arises from the close association between the measure and the final
policy goals. This dimension needs to be considered when assessing the merits or drawbacks of
adopting nominal GNP as an intermediate target.
A major part of this review addresses how, in practice, nominal GNP targets would be set.
Most advocates maintain that it would be useful at the outset to establish a long-run target path for
nominal income, either in levels or growth rates, based on estimates of potential output growth and
long-run inflation goals. One purpose of such a long-run path would be to prevent the authorities
256
from arbitrarily constructing a target value based solely on short-run objectives. There is some
disagreement, however, over whether such a path would be set by the monetary authorities alone, or in
conjunction with Congress or the Administration.
The monetary authorities would be expected to adhere strictly to the long-run target over time,
although analysts disagree about the frequency or manner in which interim targets would be set when
nominal GNP has been forced off its long-run target path by economic disturbances. As some simple
dynamic models demonstrate, a rapid return to the long-run target path of nominal income, especially
following supply shocks, could induce excessive fluctuations in real output. To lessen this kind of
instability, some advocates of nominal GNP targeting have designed strategies for setting interim
income targets that specifically consider short-run real output or employment objectives. For instance,
following an inflationary shock, a more accommodative monetary policy might be tolerated in order to
prevent a significant downturn in real output. And if real output is already well below its full
employment level, policy makers might aim for faster nominal GNP growth until output returns near
its potential level. This kind of strategy is in keeping with the view that the monetary authorities
should take a more active role in conducting policy by making use of their knowledge about the
structure of the economy in setting and achieving their nominal GNP targets.
An alternative view holds that a closer adherence to a predetermined long-run path of nominal
income should be maintained because policy makers' knowledge about the trade-off between prices
and output in the short run is limited, and any trade-off may not be invariant over time. Strategies for
targeting nominal GNP that use short-term targets to meet real output objectives lose the benefits of a
regime based on a precommitment to a rule because they inevitably rely heavily on policy makers'
discretion.
Appendix 1: Targeting Nominal GNP and the Money Stock in a Static Theoretical Framework
The impact of economic shocks under a nominal GNP target is developed in this appendix
using a textbook IS-LM model and a standard aggregate supply and demand framework. The
implications of money stock and interest rate targeting are also assessed.
257
The Model
The basic model is presented graphically in Figures la and lb. In the charts, the logarithmic
values of all variables are plotted so that the slopes of the curves may be interpreted as elasticities. In
Figure la, the IS and LM curves represent equilibrium combinations of real output and interest rates in
the goods and money markets, respectively. In keeping with standard practice, the IS curve slopes
down while the LM curve has a positive slope. The monetary aggregate underlying the LM curve is
the real supply of money: nominal money deflated by the price level. Thus, changes in either the
nominal money supply or in prices will shift the LM curve. The IS and LM curves are assumed to
take the following simple functional forms:
(1)
IS: log(r) = Io - I/logCQ)
(2)
LM: log(M/P) = Lo - L,*log(r) + L2*log(Q),
where Q measures the level of real output, r is the interest rate, M is the size of the nominal money
stock, and P is the price level. The equations are written so that all variable coefficients are greater
than zero.
Aggregate supply and demand relations are represented by the curves AS and AD, respectively,
in Figure lb. Aggregate demand is derived by measuring the effect that changes in the price level
have on real output through their impact on the real money supply, holding the nominal money supply
constant.22 The AD schedule will have a negative slope because higher prices reduce the real money
supply, causing the LM curve to shift to the left and lowering real output. The exact slope of the
aggregate demand curve can be derived algebraically by solving the above equations for prices as a
function of real output. The resulting expression is:
(3)
slope AD = -(Ii^+Lj),
which, while always negative, may be above or below negative one.
The AS curve describes the response of output to changes in prices.23 The curve will have a
positive slope if contracts or other practices fix costs of supply factors—such as labor-in nominal
22
The AD curve is derived by tracing the impact that changes in the price level have on the LM curve
and thereby on real output in the IS-LMfiramewoiicin Figure la. Shifts in the LM curve brought on by
changes in the nominal money supply and shifts in the IS curve lead to shifts in the AD curve.
Movements in the LM schedule that are induced by changes in the price level correspond to movements
along the AD curve.
23
If the initial equilibrium point corresponds to potential output, then the AS curve can be interpreted
as describing the price changes associated with deviations of real output from its potential level.
258
Targeting Nominal GNP - Demand Shock
Figure 1a
Interest rate
Real output
Figure 1b
Price
Q.Q.
Real output
terms for extended periods. In this case, higher prices will lower real factor costs and more of the
productive inputs will be employed, raising real output. It is generally accepted that in the long run
the aggregate supply function will be vertical so long as economic agents do not suffer from any form
of money illusion. This means that in the long run, actions of the monetary authorities cannot have
any impact on the level of real output. However, in the short to intermediate run, monetary policy
decisions can either exacerbate or moderate swings in real output caused by economic disturbances.
The transition from the short run to the long run is not spelled out in this analysis.
The model is completely deterministic. The monetary authorities fully control their operating
instrument, the money supply, and all structural parameters are known with certainty. Thus, nominal
GNP, or any single target variable, can always be hit except immediately following unexpected
disturbances. The economic relationships in this model are developed largely in real terms, and all
prices are specified in levels. Inflation expectations are not incorporated in thisframework;thus,
among other things, there is no distinction between real and nominal interest rates. The general
absence of inflationary dynamics, coupled with the upward sloping supply curve, makes this model
more suited for analyzing targeting in the short run. Nonetheless, this static framework highlights
some of the principal features that some associate with nominal GNP targeting, and it is a standard
tool, at least implicitly, of many discussants.24
Targeting Nominal GNP
When nominal GNP is the intermediate target, the monetary authorities change the money
stock following economic disturbances (that is, they shift the LM curve) so as to keep the economy at
a point where prices and real output yield the target value of nominal income. Let the line labeled
NGNP* in Figure lb represent all combinations of P and Q that yield this target level of nominal
income. By definition this curve has a slope (elasticity) of negative one. In the following examples,
the economy is assumed to begin at a point of equilibrium, and the targeted level of nominal GNP
remains unchanged following economic disturbances.
Under nominal GNP targeting, demand shocks corresponding to movements in the IS and LM
curves have no impact on prices and real output since these are offset by money supply changes. For
Baylor (1985) uses this approach to examine what he labels the "impact" effects of nominal GNP
targeting, and Kahn (1988) adopts a similarframeworkin his analysis of the "short-run" consequences of
nominal GNP targeting. Bean (1983) analyzes nominal GNP targeting using a more complex theoretical
model.
260
instance, a decline in business investment shifts the IS curve in Figure la from IS0 to IS! and the
aggregate demand schedule in Figure lb from AD0 to AD,. The resulting change in prices from P0 to
Px will lead to a shift in the LM curve to LM, through its impact on real money balances. At P, and
Qlf nominal GNP is below the target level. Thus, the money supply expands, simultaneously shifting
the LM curve to the right and returning AD to its starting position. In final equilibrium, output and
prices are returned to Q0 and P0, respectively.25 Similarly, an increased desire by consumers to hold
money balances shifts both the LM and AD curves to the left in Figure 1. This is countered by an
increase in the money supply that moves these schedules back to their initial positions, restoring prices
and output to their original levels. Under a nominal GNP target, all demand disturbances are
completely offset, leaving aggregate prices and real output unchanged.
Unlike demand shocks, supply or price shocks—such as increases in oil prices or declines in
labor productivity-have a lasting impact under nominal GNP targeting, but monetary policy distributes
their impact evenly between prices and real output.26 In Figure 2b, a negative price shock shifts the
aggregate supply schedule leftward, from AS0 to AS,. The higher prices (P,) caused by this shift
reduce real money balances, and the LM curve moves to LM,. At Px and Qlf nominal GNP is above
target. Authorities respond by lowering the nominal money supply, a change which shifts the
aggregate demand curve to AD,, where it intersects aggregate supply at a point along the NGNP*
curve. At P2 and Q2, nominal GNP is again on target, and the impact of the negative supply
disturbance is split evenly between higher prices and lower real output.27
Even in this simpleframework,the exact policy response needed to keep the economy on the
line NGNP* following a price shock may be unclear. Inflationary supply shocks will always ultimately
lead to a decline in the real money supply, but the direction of change in the nominal money supply
required to hit the target level of nominal income depends on the slope of the AD curve. The
graphical treatment in Figures 2a and 2b presents the situation when the AD curve is steeper than the
25
The rise in the nominal money supply needed to move aggregate demand back to AD0 initially
moves the LM curve to the right of LMj, but the rise in prices from Px to P0 shifts the LM curve back to
LM2 through the real balance effect.
26
That is, the percentage rise (fall) in prices will exactly offset the percentage fall (rise) in real output,
on the assumption that nominal GNP is restored to its initial level.
27
The shift in the LM curve from LMt to LM2 in Figure 2a reflects both the drop in the nominal
money supply needed to shift the AD curve to its new position and the impact of permanently higher
prices on the real money supply corresponding to a movement along the aggregate demand curve.
261
Targeting Nominal GNP - Price Shock
Figure 2a
Interest rate
Real output
Figure 2b
Price
AS!
^ ^ ^ ^ ^
\
^
262
\
^ ^
ASo
NGNP*
ADi ADo
^
Q2Q,
Q0
Real output
unit-elastic target line for nominal GNP. The alternative case, where the NGNP* line is steeper than
the AD schedule, is presented in Figures 3a and 3b. Here the initial disturbance to aggregate supply
shifts that schedule from AS0 to ASj. With output at Q! and prices at Plf nominal income is below
target. (The price rise moves the LM curve to LMj.) In this example, returning nominal income to its
targeted level requires an increase in the nominal money supply that shifts aggregate demand to the
right, to ADX, from its initial position and moves the LM curve to LM2. At P2 and Q2, nominal GNP
is again on target. Despite the rise in the nominal money supply, real money balances are lower in
final equilibrium.28
Nominal GNP versus Money Stock Targeting
The impact of supply and demand shocks under money stock targeting is presented in order to
provide a basis for comparing these regimes with a policy of targeting nominal GNP. Figures 1
through 3 can be used to illustrate the impact of fixing the money stock in nominal terms. Only the
final step in the previous examples of targeting nominal GNP--where the nominal money stock is
changed to return to the original level of nominal income-is deleted in the analysis when the money
supply is fixed.
Negative demand shocks lower real output and prices when the money supply is targeted. For
example, using Figures la and lb, a decline in real investment spending shifts the IS curve to ISX and
aggregate demand to ADj. The decline in prices from P0 to P! raises real money balances and the LM
curve shifts to LMj. Unlike targeting nominal GNP, however, money targeting produces no policy
reaction and final output remains below its initial equilibrium level at Qv A similar result holds in the
case of a shock such as an increased desire to hold money balances, which shifts the LM schedule to
28
This result can be derived by combining the total derivatives of equations 1 and 2 and substituting
expressions of (negative) changes in prices for changes in real output, because under nominal income
targeting the percent changes in real output and prices are of equal size but opposite sign. In percentage
terms, the expression for the change in the nominal money supply required to hit the nominal GNP target
reduces to:
Percent change in money = (Percent change in price)*(l-I1*L1-L2).
The final expression in parentheses must be under one because all coefficients are positive. Thus, the
nominal money stock rises less than prices do following a supply disturbance that raises prices, so the real
money supply must fall. However, whether the nominal money supply rises or falls depends on the sign
of the final expression in parentheses. This expression is simply one plus the slope of the AD curve.
Thus, the nominal money supply must fall (rise) following a supply shock that raises prices if the AD
curve has a slope greater (lesser) than negative one in absolute value.
263
Targeting Nominal GNP - Price Shock (alternative case)
Figure 3a
Interest rate
Q
1Q2
Q
Real output
0
Figure 3b
Price
. ASi
^ / ^
AS0
R
r
2
H
H>
^ w
AD0
NGNP
264
Q,Q2
Q0
Real output
the left. With money targeting, prices and real output are lower following a negative demand shock
than they are under nominal GNP targeting.
The steepness of the AD schedule will determine whether, as compared with nominal income
targeting, targeting the money supply will have a greater or lesser impact on real output or prices
when there are supply disturbances. In Figure 2b, where the AD curve is steeper than the NGNP* line,
a supply shock that shifts the aggregate supply curve from AS0 to AS! leaves output and prices at Qx
and Pj, respectively, when the nominal money supply is held constant. Compared with final
equilibrium under nominal GNP targeting, where policy shifts the AD schedule back to where it
intersects aggregate supply on the NGNP* line, real output and prices are higher. The opposite result
holds when the aggregate demand schedule has a slope below one in absolute value, as seen in Figures
3a and 3b. In this case, under nominal money targeting, a disturbance that shifts aggregate supply to
ASj will leave output and prices at Q! and Plf respectively-below the values both variables have at
their final equilibrium point under nominal GNP targeting. Taylor (1985) argues that in the short run
the responsiveness of real output to interest rate changes is likely to be small; that is, the value of the
coefficient Ij is large. This corresponds to a relatively steep IS curve and, by equation 3, a steep AD
curve. In this case, money stock targeting leaves prices and output at higher levels than nominal GNP
targeting when there are negative supply disturbances, and so targeting the money supply is described
as more "accommodative" than nominal GNP targeting when there are price shocks.
Conclusions of the Static Analysis
The preceding analysis is intended to highlight some of the comparative benefits often
associated with targeting nominal GNP. With nominal income targeting, aggregate demand
disturbances are entirely offset by policy moves that keep nominal GNP on target while the impact of
supply shocks is distributed evenly between offsetting movements in prices and real output. By
comparison, with money stock targeting, negative demand shocks reduce both prices and the level of
real output. However, the net effect of supply shocks is unclear, and money targeting might be either
more or less accommodative to real output than nominal GNP targeting, depending on the underlying
parameters of aggregate demand. To proponents of nominal income targeting, these results
demonstrate that targeting nominal income not only provides the monetary authorities with enough
flexibility to respond to all demand disturbances so as to offset their impact on prices and real output,
but also promises to split the impact of supply shocks evenly between prices and output-an outcome
believed to represent a satisfactory compromise between the final objectives of sustaining adequate real
265
growth and maintaining price stability.
Appendix 2: Targeting Nominal GNP in a Dynamic Framework
In this Appendix, a quarterly price adjustment model is used to evaluate the dynamic properties
of various nominal GNP targeting strategies. Some of these results are summarized in Section IIB.
The model is essentially a dynamic version of the aggregate supply framework laid out in Appendix 1,
and it is similar to models used by Gordon and Taylor in their analyses of the effects of targeting
nominal income on business cycles.
Price adjustment, or aggregate supply, is represented by:
(1)
Dpt = DpM + a*( q, - q*t) + ut, a > 0,
where p is the log level of prices, and q and q* are the log levels of real output and potential output,
respectively. D is the change operator. The variable DpM is intended to represent the impact of all
past inflation rates on the current inflation rate. With the coefficient on this term set to unity,
equation 1 incorporates the "natural rate" hypothesis that output above potential cannot be sustained
without accelerating inflation. The variable ut captures the effect of unexpected disturbances on price
inflation. In the subsequent exercises, the coefficient of quarterly adjustment of inflation to output, a,
is set to .06, which is the adjustment rate estimated by Taylor. Also, the following distributed lag of
past inflation rates is substituted for the term Dp^ in equation 1:
(2)
.4*DpM + .3*Dpt_2 + .2*Dpt_3 + .l*DpM.
As a general rule, the amplitude of the oscillations to real output increase as adjustment lag to past
inflation is shortened.
For these exercises, the definition of aggregate demand is broadened to include the monetary
policy reaction function that keeps nominal income on its targeted path.29 Thus, aggregate demand is
represented by the rule that is used to set the nominal GNP target each period, and several possibilities
are considered here.30 Most rules make reference to a long-run targeted level or growth rate of
nominal GNP, based on estimated potential real growth and ultimate inflation goals, but in most cases
29
In Appendix 1, aggregate demand was derived by collapsing the IS and LM curves, and the target
rule guiding monetary policy was a separate function. Here, the definition of aggregate demand
incorporates the monetary policy reaction function.
3(V
Thus, in this framework, as in Taylor's analysis, aggregate demand corresponds to the line labeled
NGNP* in the static framework.
266
near-term nominal GNP targets may differ from this long-run objective. The log level of the long-run
path is labeled x*t, and the corresponding growth rate is the constant Dx*.
If in each period the target level of nominal GNP is set to the long-run targeted level, then
aggregate demand is:
(3)
pt + 4 = x*t.
If a constant growth rate for nominal GNP is targeted each period, then aggregate demand is:
(4)
Dpt + Dqt = Dx*.
The approximation to McCallum's rule used in this analysis ignores changes in velocity (see
Section HA). Thus, targeted changes in the operating instrument equal targeted changes in nominal
GNP, and his rule collapses to:31
(5)
Dxt = Dx* + .25*( x*M - x ^ ) ,
which states that percentage changes in nominal income are set equal to the desired long-run rate of
change plus an adjustment term for lagged deviations in the (log) level of nominal GNP from its longrun target path.
Under Taylor's proposed formula, aggregate demand is:
(6)
Dxt = Dx* - ( qM - q*M ).
Finally, these rules are compared with a policy that minimizes on a period-by-period basis a
loss function with the following formulation:
Loss = (q, - q*t)2 + B*(Dpt - Dp*)2.
(7)
The parameter B gives the relative weights attached to deviations of inflation and real output from
their long-run targeted levels or rates of change. The strategy that minimizes this function each period
is:
(8)
q t -q* t = B*a*(Dp*-Dp t ).
This is functionally equivalent to Taylor's general formula for setting nominal GNP targets, cited in
footnote 13 of Section HA. For this exercise, B is set to one.
The economy begins in initial equilibrium, with real output growing at an assumed 2 1/2
percent potential rate of growth and inflation equal to an assumed long-run desired pace of 3 percent.
The effects of a temporary 1 percent (4 percent annualized) shock to inflation are evaluated on the
assumption that nominal GNP can be controlled in each subsequent period. The impact under each of
31
McCallum's rule only states how the operating instrument should be reset each period, and he does
not address whether or how formal short-run nominal GNP targets should be set.
267
the alternative nominal GNP targeting strategies is graphically illustrated for five variables: the level
of nominal GNP (Chart 1), the growth rate for nominal income (Chart 2), growth in real output
(Chart 3), deviations of the level of real output from potential output (Chart 4), and the inflation rate
(Chart 5).
Nominal GNP remains close to its initial target path when either its level or the growth rate is
directly targeted (Chart 1). A feedback rule such as McCallum's returns nominal income to its target
path after a short while. But nominal GNP steadily moves away from its initial targeted level when
Taylor's rule or a period-by-period optimizing strategy is used. In terms of growth rates, McCallum's
feedback mechanism quickly returns nominal income growth to its long-run targeted rate (Chart 2).
Taylor's rule only gradually restores nominal GNP growth to this rate, and the short-run optimizing
strategy puts nominal income growth on an almost permanently higher path.
Both Taylor's rule and the period-by-period optimizing strategy keep real output growing near
its potential rate of 2 1/2 percent in all but the very short run (Chart 3). In contrast, strategies that
strictly target either the level or the growth rate of nominal income cause considerable instability in
Chartl
Log Level of Nominal GNP
Under Alternative Targeting Procedures With a Price Shock
Log (bil. $)
E> •
D
8.6
8.5
84 h
0--T*"
D 1 *
g 2
S
D
D
A
9
9
A
A
A
A
Q
i^-*^
oA- --
S**-
°
°
A
8
-i-ji^^
8.2
a 5J*-*
8.1
"%****
JU'
8
:£_]
i
L
2
4
6
8
10
12
— Long-run target path
Nominal GNP level targeted
o Nominal GNP growth rate targeted
268
14
16
18
20
22
24
26
* McCallum's rule
* Taylor's rule
• Loss function minimized
28
30
32
Period
Chart 2
Growth Rate of Nominal GNP
Under Alternative Targeting Procedures With a Price Shock
Percent
10 h
8 \-
i \
: :
/ : a
D
2
A
n
A
4
9
R
o
D
*
6
D a D D D n n D D n D
* * * * A A A A A A
A
8
10
12
14
16
Long-run target path
Nominal GNP level targeted
o Nominal GNP growth rate targeted
18
n
A
n
n
n
n
A A A A
20
22
n
D
a
D
D
D
D
D
D
* ^ A A A A A A A
24
26
28
30
32
Period
30
32
Period
* McCallum's rule
A Taylor's rule
o Loss function minimized
Chart 3
Growth Rate of Real Output
Under Alternative Targeting Procedures With a Price Shock
Percent
6
4
h
2
h
-2
0
2
4
6
8
10
J
I
I
I
I
12
14
16
18
20
Long-run target path
Nominal GNP level targeted
o Nominal GNP growth rate targeted
I
22
'
24
* McCallum's rule
A Taylor's rule
a Loss function minimized
l
26
l
28
269
Chart 4
Percent Deviation of Real Output from Potential
Under Alternative Targeting Procedures With a Price Shock
Percent
*
* *
"6"'0'"Q.. $
.*•'
1h
..-*•
o
.*•'
o
O
-•—*
n u u—u u u u—•—n—o—n—u u u—u u 7 T
\
A
:
A
A
A
A
£
A
A
A
A
A
A
/'* °
O
*
-n—a—n—n—n—n—u u—n—n—n—cr
A A A A A A A A A A
A A A A G A A A
A
.-•' *
O
*
o
\ *
o
o
.*•::....$..
"••.'*''
•o•••..o*' °*
* *
-3
8
10
12
14
16
18
20
22
24
26
28
30
32
Period
* McCallum's rule
A Taylor's rule
n Loss function minimized
- Long-run target path
— Nominal GNP level targeted
o Nominal GNP growth rate targeted
Chart 5
Inflation Rate
Under Alternative Targeting Procedures With a Price Shock
Percent
o
*
7
6
5
_...:....
D
"A" • a. . D . . . C O . . . D - D .
'»
O
A
A
A
A
A
.a.
A
. . Q . . Q . •DA
A
.*
4
3
A
O
*\*
__.*
A
A
A
A
A
A
O
- a- • • a• • D • - ••• • • • • • Q • • ••• • • • • • n • • cr • • a* • •
A . A . . . A . . . A .. A . .. A . .. A. .. .A
A
A
^
A
..* °
2
*
8
L
JL
6
270
I
8
A
A
o
_^-* .^..°
...••*
..-• * o
O
_L
A
..-*•
o
^ -J
„
A-
I
10
JL
12
- Long-run target path
••• Nominal GNP level targeted
o Nominal GNP growth rate targeted
I
14
o
. .O-.-O" "9" '9
I
16
* *
J
JL
18
20
I
22
24
* McCallum's rule
A Taylor's rule
n
Loss function minimized
26
28
30
32
Period
real growth. Real output growth also fluctuates significantly when using McCallum's rule, which has
a feedback mechanism from past misses in nominal GNP. A similar set of results holds when
examining percentage deviations in the level of real output from its potential level (Chart 4). Under
Taylor's rule the level of real output gradually returns to its potential level, and with an optimizing
strategy real output never moves far away from potential. But under McCallum's rule, and when
levels or growth rates of nominal income are targeted inflexibly, real output oscillates around its
potential level.
Finally, the inflation rate moves around its long-run targeted value of 3 percent when the level
or growth rate of nominal GNP is targeted and when McCallum's rule is used (Chart 5). Inflation
only gradually falls back towards the 3 percent level with Taylor's rule, and under a period-by-period
optimizing policy the inflation rate jumps to a permanently higher level.
Bibliography
Axilrod, Stephen H. 1985. "The Conduct of Domestic Monetary Policy: Comments." In Albert Ando
et al., eds., Monetary Policy in Our Times. Cambridge: MIT Press.
Bean, Charles. 1983. "Targeting Nominal Income: An Appraisal." Economic Journal, December.
Bradley, Michael, and Dennis W. Jansen. 1989. "Understanding Nominal GNP Targeting." Federal
Reserve Bank of St. Louis Economic Review, November-December.
Brittan, Samuel. 1981. How to End the "Monetarist" Controversy. Hobart Paper no. 90, Institute of
Economic Affairs.
. 1983. The Role and Limits of Government: Essays in Political Economy. Maurice
Temple Smith, Ltd.
Brunner, Karl. 1983. "The Politics of Myopia and Its Ideology." Shadow Open Market Committee,
Policy Statement and Position Papers, September 18-19.
Dudler, Herman. 1985. "The Conduct of Domestic Monetary Policy: Comments." In Albert Ando
et al., eds. Monetary Policy in Our Times. Cambridge: MIT Press.
Enzler, Jared, and Lewis Johnson. 1981. "Cycles Resulting from Money Stock Targeting." New
Monetary Control Procedures, vol. 1, Board of Governors of the Federal Reserve System Staff
Study.
271
Frankel, Jeffrey. 1989. "A Modest Proposal for International Nominal Targeting." Harvard University,
Kennedy School of Government, Discussion Paper no. 176D.
Friedman, Benjamin M. 1988. "Conducting Monetary Policy by Controlling Currency Plus Noise:
A Comment." Carnegie-Rochester Conference Series on Public Policy, no 29. Amsterdam: NorthHolland.
Gordon, Robert J. 1985. "The Conduct of Domestic Monetary Policy." In Albert Ando et al., eds.,
Monetary Policy in Our Times. Cambridge: MTT Press.
Hall, Robert E. 1983. "Macroeconomic Policy under Structural Change." Industrial Change and Public
Policy, Federal Reserve Bank of Kansas City.
. 1984. "Monetary Strategy with an Elastic Price Standard." Price Stability and Public
Policy. Federal Reserve Bank of Kansas City.
Hall, Thomas E. 1988. "McCallum's Base Growth Rule: Results for the United States, West
Germany, Japan, and Canada." U.S. Department of State, PAS Working Paper no. 11.
Kahn, George. 1988. "Nominal GNP: An Anchor for Monetary Policy?" Federal Reserve Bank of
Kansas City Economic Review, November.
Lombra, Raymond E. 1984. "Monetary Strategy with an Elastic Price Standard: Commentary." Price
Stability and Public Policy, Federal Reserve Bank of Kansas City.
McCallum, Bennett T. 1987. "The Case for Rules in the Conduct of Monetary Policy: A Concrete
Example." Federal Reserve Bank of Richmond Economic Review, September-October.
. 1988a. "The Role of Demand Management in the Maintenance of Full
Employment." National Bureau of Economic Research, Working Paper no. 2520.
. 1988b. "Robustness Properties of a Rule for Monetary Policy" and "Reply to
Comments by Benjamin Friedman." Carnegie-Rochester Conference Series on Public Policy, no. 29,
Amsterdam: North-Holland.
McNees, Stephen K. 1985. "Which Forecast Should You Use?" Federal Reserve Bank of Boston
New England Economic Review. July-August.
. 1987. "Prospective Nominal GNP Targeting: An Alternative Framework for
Monetary Policy." Federal Reserve Bank of Boston New England Economic Review. SeptemberOctober.
Poole, William. 1980. "Stabilization Policy After Ten Years: Comments." Brookings Papers on
Economic Activity. 1.
Rasche, Robert H. 1985. "What Would Nominal GNP Targeting Do to the Business Cycle? A
Comment." In Karl Brunner and Allan H. Meltzer, eds., Understanding Monetary Regimes.
Amsterdam: North-Holland.
272
Solomon, Anthony M. 1984. "Unresolved Issues in Monetary Policy." Federal Reserve Bank of New
York Quarterly Review. Spring,
Taylor, John B. 1985. "What Would Nominal GNP Targeting Do to the Business Cycle?" In Karl
Brunner and Allan H. Meltzer, eds., Understanding Monetary Regimes. Amsterdam: North-Holland.
Tobin, James. 1980. "Stabilization Policy After Ten Years." Brookings Papers on Economic Activity,
1.
. 1983a. "Macroeconomic Policy Under Structural Change: Commentary." Industrial
Change and Public Policy, Federal Reserve Bank of Kansas City.
. 1983b. "Monetary Policy: Rules, Targets, and Shocks." Journal of Money, Credit, and
Banking," vol. 15, no. 4 (November).
. 1985. "Monetary Policy in an Uncertain World." In Albert Ando et al., eds, Monetary
Policy in Our Times. Cambridge: MIT Press.
Volcker, Paul A. 1983. "Statement before the House Committee on Banking, Finance and Urban
Affairs." Reprinted in Federal Reserve Bulletin, August.
Williamson, John, and Michael Miller. 1987. Targets and Indicators: A Blueprint for the International
Coordination of Economic Policy, Institute for International Economics.
273
INTEREST RATES AS TARGETS AND INDICATORS FOR MONETARY POLICY
Charles Steindel
There has been renewed interest in the use of market indicators in the formulation of monetary
policy. Other papers in this study evaluate the pros and cons of using commodity prices, exchange
rates, and the yield curve. This paper discusses what may be the most obvious set of price indicators
for monetary policy-nominal and real interest rates.
The paper first focuses on the use of interest rates in intermediate targeting. There is almost
universal agreement that nominal and real interest rates have serious problems as long-term
intermediate targets. Rates can have very weak and unstable, or even perverse, relationships to final
goals-especially inflation. Furthermore, the Open Market Desk's ability to control the types of
interest rates most relevant to inflation and real growth is limited. For these reasons, rates are not
proposed as the sole targets of monetary policy in the same sense as monetary aggregates, with the
aim of policy to be the maintenance of nominal rates at certain levels or in certain ranges. However,
given the empirical evidence that real interest rates have a significant effect on economic activity, there
is some justification for bringing real rates into the intermediate targeting process—at least as
information variables helping to evaluate the signals about final goals provided by monetary
aggregates. This conclusion is reinforced by recent theoretical arguments that support a role for
nominal or real rates in intermediate targeting systems.
Interest rates may also be used as operating instruments to control other intermediate targets,
such as monetary aggregates. Short-term nominal interest rates, such as the federal funds rate, have
been considered—and used—as operating targets. The literature suggests that given a monetary
aggregate as an intermediate target, it will rarely if ever be optimal to ignore money market rates in
the formulation of open market operations.
The final section of this paper traces some of the historic uses of interest rates in Federal
Reserve policy making. The various shifts in the use of rates in the policy process help to shed light
on the issues discussed in other sections.
Before we proceed to the main body of the paper, one further point should be mentioned. At
times, interest rate stability has been viewed as either a final goal or an intermediate target of policy.
274
The rationale for smoothing interest rates is that excessive volatility may hamper the efficient
operations of the banking and financial system, and an inefficient financial system may hamper the
workings of the overall economy. Despite these concerns, the precise relationship between interest
rate volatility and the smooth workings of thefinancialsystem is by no means obvious. To be sure,
large daily swings in securities prices may strain payments and settlements systems and make it
difficult for savers and borrowers to evaluate investment opportunities. However, enforced interest
rate stability in the face of shocks to the economy may well exacerbate macroeconomic instability,
which will in turn create problems for financial markets. The existence of such a trade-off between
limiting highfrequencyinterest rate volatility and increasing macroeconomic stability would weaken
the case for interest rate stability as a target or final goal.
I. Interest Rates as Intermediate Targets
A. General Considerations
Interest rates have not been proposed as intermediate targets in the same sense as money and
credit aggregates. There are no serious proposals that monetary policy be geared toward keeping a
single interest rate (nominal or real), or an average of interest rates, at some optimal level or in some
predetermined range. The evidence produced in this section implies that there are good reasons for
this lack of interest; interest rates (nominal or real) have ambiguous relationships to final goals, such
as real growth and inflation, and the ability of the monetary authorities to control the interest rates
relevant to final goals through the usual operating instruments is limited.
The analysis of interest rates as intermediate targets is complicated by the potential need to
analyze four different types of rates: nominal short-term rates, real short-term rates (that is, the
difference between nominal rates and expected inflation), nominal long-term rates, and real long-term
rates. In principle, eight sets of linkages may be analyzed-the linkage of Desk operations to each rate
type and the linkage of each type to final goals.
The next subsection will discuss the control issue, followed by subsections on the conceptual
and empirical linkages of rates to final goals, with a special focus on the incremental information
value of rate moves when changes in monetary aggregates are taken into account. A final subsection
discusses recent technical literature bearing on the issue of the optimal role of interest rates in an
aggregate-based intermediate targeting strategy.
275
B. Interest Rate Control
Federal Reserve open market operations consist essentially of the purchase and sale (temporary
and permanent) of Treasury securities. Purchases involve the creation of bank reserves, while sales
remove reserves from the banking system. The large scale of Desk operations relative to the reserves
market means that the Desk exerts a great deal of influence on the federal funds rate, the overnight
interest rate for reserves. There is little doubt that the Desk can achieve nearly complete control over
the funds rate. (For a discussion of control issues in the current financial market environment, see
Radecki and Reinhart 1988.) Arbitrage between the funds market and the markets for other short-term
securities, such as Treasury bills and bank CDs, implies that the Desk can greatly influence the general
level of short-term nominal interest rates. Although there may not be perfect control of an average of
short-term rates, the Federal Reserve's ability to control nominal short-term interest rates can be
viewed as at least as great as its ability to control any other proposed intermediate target.
By contrast, the control of real short-term rates and long-term rates, nominal and real, raises
some significant issues. The problems generally center on in the relationship of inflation expectations
and short-term interest rate expectations to monetary operations.
The problems are clearest in the control of nominal and real long-term rates. It is reasonable
to assume that nominal long-term interest rates are affected by expectations of the future course of
short-term interest rates (the expectations hypothesis of the term structure asserts that such expectations
can fully explain long-term rates). A restrictive monetary policy, which has the immediate impact of
raising nominal short-term rates, may create expectations that higher short-term rates will persist and
thereby increase nominal long-term rates.
Alternatively, however, a restrictive policy may generate expectations that future levels of
inflation will be lower, consequently putting downward pressure on future short-term interest rates.
Furthermore, a currently restrictive policy may encourage expectations of a future policy change,
which could also eventually put downward pressure on short-term rates. Thus, the unpredictability of
the relationship between current policy actions and expectations of future interest rates may limit the
Desk's ability to control nominal long-term interest rates.1
*In a recent study, Timothy Cook and Thomas Hahn (1989) found that changes in the federal
funds operating target in the 1970s had a progressively smaller impact on Treasury yields as the time
to maturity advanced—smaller in both an economic sense (a given change in the funds rate target had a
smaller estimated impact on the longer lived maturities) and in a statistical sense (less of the variation
in yields on longer lived instruments was explained by changes in the targeted funds rate).
276
Real long-term interest rates are defined as nominal long-term rates less expected long-term
inflation. One fundamental problem involved with utilizing real long-term interest rates in policy is
the lack of any agreed-upon way to measure long-term inflation expectations and thus to observe real
long-term interest rates. Going beyond this problem is the difficulty of controlling real long-term rates
through the usual instruments of monetary policy. Conceivably, as suggested above, a restrictive
policy may generate expectations of both lower nominal short-term rates and lower inflation in the
future. The effect of the policy on real long-term rates can not be predicted without considerable
knowledge of its effect on expectations of future nominal short-term rates and inflation.
One can well doubt that monetary policy can play a predominant role in the movement of real
long-term interest rates, even if the effect of policy on expectations at any time were perfectly
understood. In the long run, real interest rates will move in tandem with the productivity of the
capital stock, which is determined by factors such as the capital-labor ratio and the development of
technology. Changes in expectations about these real factors may have more to do with movements in
real long-term interest rates than do changes in monetary policy.
Inflation expectations are also not directly observed even over short horizons, meaning that
real short-term rates are not observed. In principle, the same interaction between monetary policy
moves and inflation expectations that affects long-term interest rate control may be operative at short
horizons, complicating real short-term interest rate control. However, there may plausibly be little
direct linkage between Desk actions and near-term inflation expectations (say, less than six months),
implying that there could be a reasonably stable linkage between Desk actions and real short-term
interest rate movements.2
Overall, it appears that the monetary authorities can achieve a high degree of control over
short-term nominal interest rates and may exert some systematic influence over short-term real interest
rates. The effect of monetary policy on long-term interest rates, nominal and real, is problematic.
C. Interest Rates and the Economy
The standard simple IS/LM model of the economy implies a straightforward relationship
between interest rates and real output. In the standard models, the aggregate demand for goods and
services is negatively related to the one interest rate considered (that is, the IS curve traces the
2
Eugene Fama (1975, 1977) argued that the real ninety-day Treasury bill rate is constant. His
conclusion was widely contested (Carlson 1977; Joines 1977; and Nelson and Schwert 1977). Robert
Shiller (1981) found that the Desk can, indeed, exert some control over real short-term rates.
277
negative relationship between real demand and the interest rate). Under these assumptions, in the
absence of uncertainty, the monetary authorities can achieve a real output goal by targeting the interest
rate, increasing the rate target to reduce output and reducing it to increase output.
The simple IS/LM model has fixed prices. It is not directly applicable for analysis of the
implications of nominal interest rate moves for real output, or the implications of real or nominal
interest rate movements for inflation and nominal output. Furthermore, the model has only one
interest rate. It is generally believed that the most interest-sensitive components of demand consist of
spending on long-lived products such as capital goods, consumer durables, and structures, so the
simple IS/LM model appears most relevant for analyzing the influence of real long-term rates on real
output.3
The IS/LM model must be extended to analyze the relationship of interest rates to inflation and
the influence of real short-term rates and nominal rates (short- and long-term) on real output and
inflation. This is often done by appending to the basic model a Phillips Curve relationship between
the level of real output (relative to the full employment level) and inflation, fixed term-structure
equations relating nominal long-term interest rates to realized short-term interest rates, and an
assumption that inflation expectations are related to actual current and past inflation. With these
additions, common to the standard large econometric models, there will be at any time a predetermined
relationship among all interest rates, real and nominal, short- and long-term, and among movements in
rates, real output, and inflation. Any interest rate will in principle have at any time a predetermined
relationship to inflation and real output; a given level of the rate will be associated with given levels
of output and inflation. The relationship between any rate and final goals will still be subject to
statistical uncertainty and some rates will be more closely associated statistically with changes in the
real economy and inflation than others.
The structure of many large-scale models implies that it would be possible to set short-term
nominal interest rate intermediate targets for monetary policy. The target could be connected to final
objectives subject only to the stochastic errors present in the model. In practice, however, such an
approach would appear to be seriously flawed. For one thing, the term structure equations used in
these models tend to have problems tracking the actual behavior of long-term interest rates (Shiller,
Campbell, and Shoenholtz 1983; Blanchard 1984; and Steindel and Palash 1986). This suggests that
3
This conclusion is not self-evident. If there were perfect secondary markets for capital goods, the
profitability of physical investment would be judged by short-run holding period returns, and thus
short-term interest rates. For a discussion of this issue, see Robert Hall (1977).
278
even given predictable, stable relationships between nominal long-term rates and real long-term rates,
and between real long-term rates and real output and inflation, the relationship of nominal short-term
interest rates to real output and inflation may well be unstable (although under these circumstances
nominal long-term rates might have a stable relationship to final goals).
Even in the absence of uncertainty, a policy of nominal interest rate targeting would need to
be more sophisticated in practice than simply attempting to peg a short- or long-term rate at a
predetermined level and maintaining the target over a period of time. In particular, if inflation, and
inflation expectations, happen to be increasing, the maintenance of a fixed nominal rate target may
lead to a further acceleration of inflation. If the nominal interest rate target does not change, the
upward pressure on the targeted rate stemming from the increase in inflation expectations will be
offset by a more expansionary monetary policy, resulting in declines in real interest rates and
ultimately upward pressure on real output and inflation. Thus, even in the most favorable case,
nominal interest rate targets would have to be continually updated in the light of incoming information
on the economy and inflation; a strategy of simply pegging a nominal rate could put the economy on a
continuing inflationary (deflationary) spiral if the authorities set the rate too low (high). This problem
is sometimes referred to as the lack of a "nominal anchor" for nominal rate targets.
Instability in the relationship among interest rates and between interest rates and the economy
causes further problems for a nominal rate target. For example, monetary policy may inappropriately
react to random fluctuations in pressure on a targeted rate stemming from shifts in yield curve
relationships or changes in risk premiums. Random shifts in real aggregate demand will also influence
nominal (and real) interest rates, with an increase in demand putting upward pressure on rates. A
policy of literally pegging a nominal (or real) interest rate will add to output instability if such shifts
are common (this is, as Poole 1970 showed, analogous to the problems money demand shifts add to
targeting monetary aggregates).
Furthermore, a revelation that the central bank is focusing attention on one rate (or a narrow
set of rates) could well lead to a concentration of speculative market attention on the rate, leading to
the degradation of its relationship to other rates and to final goals. A policy maker would need to
have considerable detailed knowledge about markets and the economy to use nominal interest rates
(short- or long-term) as intermediate targets.4
The general problem of possibly unstable relationships among interest rates and between
4
Milton Friedman (1968) well describes the hazards of nominal rate targeting.
279
interest rates and final goals is as true for real as for nominal interest rate targets. The linkage from
real rates to inflation is quite indirect and possibly unstable, relying on the links from real rates to real
output and from real output to inflation. It is conceivable that a targeted real rate may come under
downward pressure as inflation increases, leading to an inappropriate policy response.
There would be fundamental problems with real rate targeting in the areas of both observation
and control (particularly severe for real long-term rates), as well as in the possibility of instability in
the links among real rates and in the links between real rates and output and inflation. In practice,
policy makers (1) may not observe changes in real rates, (2) may not be able to control satisfactorily
those real rates most closely linked to final goals, (3) may react to idiosyncratic shifts in the targeted
rate, (4) may accommodate rather than offset shifts in aggregate demand, and (5) may accommodate
rather than offset inflation.
Nominal interest rates are observable, but nominal rate targeting shares all
of the other four problems-especially the last.5
D. Empirical Evidence on Interest Rates and the Economy
Clearly, intermediate targeting regimes directed solely at controlling nominal or real nominal
interest rates are rather unattractive. A secondary, but still important issue, is the empirical importance
of interest rate movements: Do movements in nominal and real interest rates systematically lead
movements in economic activity? If rate moves lead changes in the economy, then rates are worth
monitoring, even if they are not used as intermediate targets.
This issue can be addressed by examining the empirical evidence on the relationship between
interest rates and the economy. The first body of evidence comes from the standard large econometric
models. Real interest rate effects are pervasive in these models, which are constructed along the lines
of the extended IS/LM framework; changes in real long-term interest rates will follow changes in
5
In contrast, some views of the economy suggest that nominal rate targeting could be used to help
control inflation. This approach centers on the "Fisher equation" defining the nominal interest rate as
the sum of the real rate and the expected inflation rate. If the real rate is constant, then all fluctuations
in nominal rates are due to changes in inflation expectations. Taking the argument to extremes, the
monetary authorities might use nominal rate targets to control inflation expectations directly (and
ultimately inflation) or, more modestly, use nominal interest rate movements as uniquely qualified
information variables about future inflation.
In fact, much research on the issue has failed to detect evidence that nominal rate fluctuations
can be used to forecast inflation with any accuracy for periods longer than six months or a year, even
at these intervals, there are probably superior forecasting techniques. See Lawrence Summers (1983)
and Frederic Mishkin (1990).
280
operating instruments and precede changes in goal variables, such as real GNP and inflation. In the
MIT-Penn-SSRC (MPS) model, increases in long-term real interest rates raise the "cost of capital" and
thus reduce investment in consumer durables, housing, nonresidential fixed capital, and inventories.6
In this model, increases in long-term real interest rates also reduce the value of the stock market,
depressing consumer spending on nondurable goods and services. While a "reduced-form" relationship
relating monetary aggregates and economic activity may be constructed from the equations of this
model, the actual structure of the model is that monetary policy operates by first changing nominal
short-term interest rates, then nominal and real long-term rates through fixed term-structure equations
(and backward-looking inflation expectations). These changes in real long-term rates ultimately
influence economic activity, with changes in the aggregates a byproduct of the general change in
interest rates (Modigliani 1971).
The MPS model is largely based on research originally done in the 1960s. More recent work
continues to find a leading relationship between real interest rates and economic activity, along much
the same lines as outlined in the model (Akhtar and Harris 1987, Boswoith 1989, and Friedman 1989).
However, the specific sectoral relationships are unstable statistically and have shifted over time. The
looseness of these relationships suggests that it may be inadvisable to give extensive weight to moves
in real long-term rates as predictors of future output and inflation.
A second strand of the empirical literature uses vector autoregressions (VARs) to make explicit
measurements of the timing relationships between interest rates, monetary aggregates, and economic
activity. In 1980, Christopher Sims produced evidence that current and lagged innovations in nominal
Treasury bill rates were negatively related to and explained a large fraction of the variation in
industrial production during the postwar era, dominating the influence of innovations in Ml. Robert
Litterman and Laurence Weiss (1984) and James Stock and Mark Watson (1989) have basically
reaffirmed Sims' result.
An important impression one draws from the vector autoregression literature is that it is
difficult, if not impossible, to accept the hypothesis that the total effect of interest rates on economic
activity is captured by movements in money and credit aggregates. It is somewhat surprising, though,
that these interest rate effects seem to be captured in nominal, rather than real, interest rate
6
Flint Brayton and Eileen Mauskopf 1985.
281
movements.7
A final strand in the empirical literature arises from direct estimates of "reduced forms"
relating interest rates to final goals. This approach is analogous to that used in investigations of
money and credit aggregate targeting. John Paulus and Steven Roach (1984) calculated that during the
1952-82 period a sustained 1 percentage point increase in short-term real interest rates (measured ex
post as the commercial paper rate less the growth in the GNP implicit price deflator) led to roughly a
1 percentage point decline in real GNP after three years.
Raymond Lombra (1984) carried out some further explorations of the Paulus-Roach model.
He found that the basic specification was disappointing; the overall fit of the equation was quite low,
and the coefficients on the interest rate terms showed some instability.
On the whole, the empirical evidence suggests that movements in interest rates do lead
changes in real economic activity (and by inference, inflation, since inflation changes tend to lag
changes in real activity). As the earlier discussion of the general issue of interest rate targeting
suggested, there is no striking evidence of a strong, stable relationship between any particular rate and
the economy.
E. Interest Rates as Information Variables
The direct relationship between interest rates and the economy appears to be unstable. For
example, the weakness of Lombra's results suggest that there is little value in simply reestimating the
relationship between interest rates and economic activity.8 However, one may still be interested in a
direct, simple test of the hypothesis that interest rates provide information about the economy over and
above that provided by a monetary aggregate.9 If interest rates have some independent explanatory
power over and above an aggregate, they might be used to monitor the signals about future activity
given by changes in an aggregate target.
7
George Kahn (1989) has produced evidence from VAR models that the aggregate effect of shortterm nominal interest rates sharply diminished in the 1980s.
8
More recently, a spate of studies have appeared examining the relationship between the yield
curve and the economy. See Robert Laurent (1988, 1989) and Arturo Estrella and Gikas Hardouvelis
(1989, 1990).
*VARs can be used to provide a formal, rigorous test of this hypothesis. However, the results of
VARs are difficult to place in a standard targeting framework-for example, it would be very hard to
use the results of a VAR to determine the near-term effect of an interest rate increase on velocity.
282
The procedure used in the exercise was straightforward. Simple relationships were estimated
between nominal and real economic growth and nominal and real M2 growth. Added to these
relationships were various combinations of nominal and ex post real federal funds rates. On a formal
level, if the rate terms add any information about economic activity, their addition will increase the
overall explanatory power of the regressions. More informally, their addition may improve the fit of
the regressions at certain critical turning points.
Table 1 presents the results of the regressions. The first regression relates nominal GNP
growth to nominal M2 growth. The long-run stability of M2 velocity is upheld~the constant term is
effectively zero, and in the long run a sustained 1 percentage point increase in M2 growth will result
in a 1 percentage point increase in nominal GNP growth.
The second regression adds a distributed lag on the real federal funds rate to the first equation.
The real funds rate is measured on an ex post basis as the nominal rate less the contemporaneous rate
of increase in the GNP implicit price deflator. The sum of the coefficients on the real funds rate is
less than zero, which suggests that the level of the real rate has a long-run restraining influence on the
growth rate of current-dollar spending.10
The third regression adopts the idea that fluctuations in the real federal funds rate within
"normal" bounds give no particular signal for the course of the economy. In the specification adopted,
levels of the real funds rate between 0 percent and 6 percent are constrained to have no effect—the
distributed lag is solely on the outlying values of the real funds rate (an alternative range of 0 percent
to 4 percent had an inferior fit).
This specification accords with some views of the role of interest rates in intermediate
targeting that were advanced in the early 1980s, both within and without the Federal Reserve system
(Hester 1982 and Walsh 1982). Two episodes may have sparked this interest. In the late 1970s, real
ex post short-term interest rates were generally negative. At the same time, the growth of the
monetary aggregates was generally above target, and inflation accelerated. During much of the 198182 recession, despite strong growth of the monetary aggregates, real ex post short-term interest rates
were extremely high and the economy contracted sharply. The differing response of the economy to
rapid money growth in the two episodes could have been the result of many factors, but the very
different real interest levels suggest something straightforward for the Federal Open Market
10
The regression was also estimated using the level of the nominal funds rate and the changes in
both the nominal and real funds rates. The specification using the level of the real rate had the lowest
standard error of estimate.
283
Table 1
GNP Growth Regression
(Estimation Period 1963-1 to 1986-IV; Standard Errors in Parentheses)
Nominal
GNP Growth
(1)
Nominal M2 growth
.985
(.0472)
Real
GNP Growth
(2)
(3)
1.105
(.0556)
1.017
(.0449)
Real M2 growth
(4)
.895
(.0920)
(6)
1.009
(.117)
.918
(.0911)
-.276
(.170)
-.566
(.169)
Real federal funds rate
(5)
-.912
(.274)
Extreme real funds rate
-.548
(.267)
Standard error
4.178
3.945
3.909
3.809
3.794
3.708
Durbin-Watson
1.779
1.984
2.070
1.907
1.973
2.070
Notes: M2 coefficients are sums of twelve-quarter second-degree polynomials with a far period constraint.
Federal funds coefficients are sums of twelve-quarter second-degree polynomials.
Committee (FOMC) to observe. Thus, these authors suggested that the monitoring of real short-term
interest rates might be used as a sort of "firewall" to prevent an egregious outcome: when real shortterm rates breach the preset bounds the FOMC should override any constraints placed by aggregate
targets and instruct the Desk to bring the real funds rate back into the normal range.
The specification that includes only the extreme real funds rates has the smallest estimated
standard error of the three-in other words, all the explanatory power that the real funds rate adds to
M2 growth comes from the outliers. Equations 4, 5, and 6 repeat the first three specifications but
substitute real GNP and M2 growth for the current-dollar growth rates. The results for the real growth
equations are qualitatively similar to those from the current-dollar equations: the real funds rate adds
explanatory power to real GNP growth, over and above that given by real M2 growth, with the
extreme real funds rate observations providing the explanatory power.
284
The extra explanatory power given by the real funds rate seems small—a matter of a few tenths
of 1 percent on the standard errors of the regression. However, especially in the specifications
including only the extreme real funds values, the additional fit makes a major difference in tracking
economic growth at certain times. Charts 1 to 4 show the actual and predicted values for real and
nominal GNP growth for two important periods: 1977-79 and 1982. They show that there was a
tendency for M2 growth to underestimate real and nominal growth in the late 1970s11 and to
overestimate real and nominal growth in 1982. The equations with the real rate added, particularly
those with only the extreme values admitted to the specification, do a better job of tracking the
economy in these periods.12
The upshot of this empirical exercise is that there may well be value in monitoring real interest
rates within an aggregate intermediate targeting framework. Periods in which real short-term rates go
beyond "reasonable" bounds do appear to be those in which the M2-GNP relationship weakens.13
Thus, it may be worthwhile to structure the nominal funds rate proviso in the open market directive to
the Desk to keep the monitoring range for the nominal funds rate at least within the extreme ranges for
the real funds rate. For instance, if the inflation rate can be reasonably forecast at around 5 percent
for an intermeeting period, the monitoring range for the funds rate could be set within 5 percent to 11
percent, subject to incoming information on inflation. This strategy should keep the real funds rate
within the normal bounds of 0 percent to 6 percent.
H
The substantial underestimate of growth in 1978-11 reflects abnormal growth in that quarter as the
economy sprang back from an unusually severe winter.
12
Chow tests were performed to test whether any of these regressions show a break in specification
between the third and fourth quarters of 1979. In all cases the hypothesis of changed coefficients
could be rejected at a 5 percent level. Before the 1980s, the only cases of "extreme" real federal funds
rates were those with negative rates. The hypothesis that real or nominal growth reacts differently to
negative and positive extremes in the real funds rate could also be rejected at a 5 percent level for the
whole sample period.
13
These results appear roughly consistent with what might be predicted by money demand
considerations. In 1977-79, nominal interest rates were rising, a development which would normally
suggest an increase in M2's velocity, so it is not a surprise to find that M2 growth, unadjusted for rate
effects, underpredicts nominal GNP growth. In 1982 nominal rates fell, suggesting a decline in
velocity and an overprediction of nominal GNP growth for an equation relying solely on M2 growth.
However, the regressions estimated here are intended to make the point that extreme real rate levels
can provide a clear signal to modify growth targets; they are not intended to provide the detailed
explanation of velocity that can come from a full-blown money demand function.
285
Chart 1
Nominal GNP Growth
Percent
25
20 4
15 4
1CM
-J
I
1977
I
L
1978
1979
Chart 2
Real GNP Growth
286
1977
1978
1979
Chart 3
Nominal GNP Growth
Percent
10
Predicted by equation 1
Predicted by equation 2
1982
Chart 4
Real GNP Growth
Percent
4
Predicted by equation 4
Predicted by equation 6
287
-8
1982
IV
F. Summary of Findings on the Intermediate Targeting of Interest Rates
There is a tension between the ability of the monetary authorities to influence interest rates and
the influence of interest rates on economic activity and inflation. Open market operations can be
expected to have a well-defined effect on short-term nominal interest rates and probably to have some
fairly stable relationship to real short-term interest rates, but they bear a very problematic relationship
to nominal and real long-term interest rates. The major channels through which interest rates influence
real output appear to run through changes in long-term real rates. Moreover, the relationship of interest
rates-real or nominal, short- or long-term-to inflation and nominal output is at best complex and
convoluted. The inability to control the rates most critical for real output, the instability of the link
between interest rates and output, and the weakness of the link between interest rates and inflation
imply that interest rates should not be the primary intermediate targets of monetary policy.
However, empirical evidence suggests that real interest rates (and perhaps nominal rates as
well), even at the short end of the maturity spectrum, contain information about the economy that is
not captured in the movement of monetary aggregates. This finding suggests that there is value in
monitoring interest rates as information variables in a monetary targeting framework. Short-term real
interest rates would appear to fit well into such a setup.
Such a dual targeting procedure accords well with some of the recent theoretical literature on
the relative merits of aggregate and rate targets. The next section of the paper reviews some of this
literature.
G. Modern Theoretical Literature
The modern theoretical literature on the choice between aggregate and rate targets stems, to a
large extent, from William Poole's seminal 1970 article.14 As mentioned above, Poole worked with a
standard fixed-price IS and LM model with only one interest rate and one monetary aggregate,
variables which the monetary authorities can control perfectly. The structure of the IS curve and the
LM curve are known, except for random shocks of zero means and known variances.
In Poole's model the monetary authority's goal is to minimize the variance of real GNP
around a predetermined optimal level. The authority can, at any point in time, control either the
money supply or the interest rate but can not simultaneously control both. Poole showed that the
14
Robert Holbrook and Harold Shapiro (1970) and John Karekan (1970) made many of the same
points as Poole.
288
optimal procedure generally involves systematically varying the money supply in response to
observations on the interest rate. If the LM curve is stable, the optimal schedule reduces to pegging
the money supply at the level that will produce the interest rate expected to yield the GNP objective,
given the structure of the IS curve.15 If the IS curve is stable, the optimal policy is to peg the
interest rate at the level consistent with the GNP objective.
The extensive work of James Tobin and his colleagues on the money transmission mechanism
and the structure of financial markets (Tobin and Brainard 1963; Brainard 1964; Tobin 1969, 1978)
does not directly deal with the issue of intermediate targeting but illuminates some of the issues
involved in choosing between aggregates and rates as targets. One thrust of Tobin's work is that
given a complex financial structure, where financial instruments (including ownership claims for
physical capital) can substitute for one another, it is exceedingly difficult to single out any particular
financial aggregate or interest rate as having a well-specified relationship to real economic activity.16
Although Tobin does not focus on the issue, this result suggests the possibility that a targeting setup
combining aggregates and rates may be superior to targeting either alone.
Franco Modigliani and Lucas Papademos (1980) combined the insights of Poole's and Tobin's
approaches. In their model, a financial system consisting of a multiplicity of financial assets,
somewhat like Tobin's, interacts with a structure of real demands for goods and services (also see
Tobin 1982). As in Poole's model, the equations of the Modigliani-Papademos model are subject to
stochastic disturbances, and the authors consider the issue of choosing the optimal instrument to target.
The Modigliani-Papademos paper reaches no firm conclusion; the criteria for choosing
intermediate targets depend upon a maze of structural parameters and variances about which little can
be known or reasonably speculated. Thus they conclude, "The choice among the alternative
intermediate targets becomes a substantive problem in the presence of uncertainty" (p. 152).
All of the above models ignore the generation of inflation and inflation expectations. They
implicitly assume that the Desk is able to monitor real interest rates and is able to measure the impact
of proposed actions on inflation expectations. In contrast, the literature on rational expectations
15
Because of the stochastic nature of the IS curve, the actual interest rate realized will differ from
the expected interest rate.
16
"In this complex situation, it is not to be expected that the essential impact of monetary policies
and other financial events will be easy to measure.. . .There is no reason to think that the impact will
be captured in any single exogenous or intermediate variable, whether it is a monetary stock or market
interest rate" (Tobin 1969).
289
explicitly allows inflation expectations to reflect anticipations of monetary policy.
The classic paper in this tradition was written by Thomas Sargent and Neil Wallace (1975).
Sargent and Wallace argued that a policy of nominal interest rate targeting, ignoring monetary
aggregates, will result in an "undetermined" price level.
Sargent and Wallace's nominal interest rate targeting policy consists of a feedback mechanism
relating the interest rate objective to observations on final goals. Monetary policy in this world
consists of varying the monetary base to keep the interest rate at its targeted level. The assumption of
rational expectations, however, means that the private sector knows the monetary policy rule and that
the difference between actual and expected prices would be due only to stochastic errors. Sargent and
Wallace also assume that the real interest rate is given, up to some stochastic error.
These assumptions lead to the conclusion that the economic system can operate at any price
level. The combination of knowledge of the central bank's interest rate rule and a fixed real interest
rate allows the private sector to make a rational estimate of the expected rate of inflation. Any price
level today can be consistent with that expected rate of inflation. The Sargent-Wallace model will pin
down a real money supply (the ratio of the nominal money supply to the price level), but this does not
pin down the price level, because given the interest rate rule, the nominal money supply can assume
any value.
Recent work, however, has shown that a policy of combining nominal interest rate and
monetary targeting will not lead to the troubling Sargent-Wallace result, even while maintaining the
hypothesis of rational expectations. Several authors have demonstrated that a policy of targeting the
nominal rate in the short run combined with a policy of targeting the money supply in the long run
can be made consistent with an arbitrary degree of short-term smoothness in interest rates while
maintaining a deterministic price level (Parkin 1978; Canzoneri, Henderson, and Rogoff 1983;
McCallum 1981 and 1986; Dotsey and King 1986; Goodfriend 1987; VanHoose 1989 and Barro
1989).17
The key assumption of these models is that future levels of the money supply are on a
predictable path. If the private sector can forecast the future level of the money supply, it can forecast
17
James Hoehn (1987) claims that these sorts of policies are simply special cases of money stock
targeting. Strictly speaking, he is correct, but this seems to be a rather fine distinction; a money stock
rule related to the nominal rate is a rather different animal than one related solely to final goals. The
papers in Porter (1990) demonstrate how, theoretically, interest rate targets may be constructed under a
variety of assumptions and goals.
290
the future price level, and if it can forecast the future price level, it can pin down today's price level
(from the expected inflation path).
In most of these models, the desirability of smoothing nominal interest rates is assumed,
along with inflation control, to be one of the final goals of monetary policy. In the paper by
Canzoneri, Henderson, and Rogoff (which may be viewed as an extension of Poole's work to a
flexible-price, rational expectations model), the monetary authorities are attempting to control real
output and inflation; they show that under a number of alternative assumptions about price-setting
behavior by private agents it would be optimal for the monetary authority to use nominal interest rates
as information variables to change aggregate targets.
All these rational expectations models are highly stylized, but they show that nominal rates
may be successfully combined with aggregates in an intermediate targeting system as long as long-run
control of the aggregates is maintained to provide the system with a "nominal anchor."
In conclusion, the theoretical literature surveyed here tends to be skeptical of the notion that
targeting monetary aggregates alone, and ignoring interest rate fluctuations, will be an optimal
policy.18 This literature does not yield much in the way of practical suggestions for the
implementation of policy. The Poole model would suggest combining aggregate targets with real
interest rates, while the rational expectations literature suggests combining aggregates with nominal
rates, and neither group pays much attention to the yield curve. Nonetheless, this body of work
indicates that it is sensible to pay attention to interest rates in evaluating the movement of aggregates.
II. Interest Rates as Operating Targets
The preceding portion of the paper dealt with the issue of intermediate targeting, in part in
terms of rates versus aggregates. A separate, if somewhat similar, set of issues exists in the operating
target area: Should the Federal Reserve use interest rates or the elements of the monetary base to
18
There are some unconventional arguments that bear on the issue of nominal rate targeting.
Public choice considerations suggest that nominal rate targeting is inadvisable, simply because nominal
rates are so visible (Kane 1980, 1983). If the authorities acknowledged using nominal rates in
intermediate targeting—whether as the sole targets or blended with others—sharp public pressure might
be focused on every movement on rates, and pressure would be exerted to keep rates low.
In contrast, the time-consistency literature seems to suggest that nominal rate targets have
some desirable properties. As discussed in Steven Englander's paper in this volume, "Optimal
Monetary Policy Design," policy objectives stated in terms of easily grasped and monitored targets are
desirable in their own right. The credibility policy makers gain by adhering to a fixed nominal (or ex
post real) interest rate target may overcome other problems with such a policy.
291
control an aggregate intermediate target? In this discussion, the rates in question have almost always
been short-term nominal money market rates, most notably the federal funds rate.
A simple example where rates are good operating targets is a world where the intermediate
target is a monetary aggregate, the demand for which is completely explained by the federal funds rate
and lagged nominal GNP. The targeted level of the aggregate for a quarter, in combination with the
value of nominal GNP in the previous quarter, yields a level of the funds rate that may be used as an
operating target. In this world, Desk operations would consist of adding or subtracting reserves or
adjusting the discount rate to peg the funds rate at the optimal level for the quarter.
In reality, of course, the operational use of interest rates would never be so simple. Some
problems stem from uncertainties in money demand—uncertainties arising from questions of
specification (such as which interest rates belong in a money demand function), stochastic shocks to
demand, shifts in the structure of money demand, and imperfect observations of factors affecting
money demand. These uncertainties mean that interest rate control alone will not ordinarily be
sufficient to achieve satisfactory control of the targeted aggregate. As a general rule, the optimal
control procedure would consist of the Desk varying its provision of reserves in response to interest
rate signals (Pierce and Thomson 1972; LeRoy 1975, 1979; LeRoy and Waud 1977; Bryant 1983;
Axilrod and Lindsey 1981; and Dotsey 1989).19
This technical point may be misunderstood. Defending the usefulness of partial or total
control of some money market rate in helping to control a monetary aggregate is not equivalent to
defending the intermediate targeting of interest rates. In the money control problem, interest rate
movements can be used to provide information about the near-term growth of the aggregates rather
than the economy.20
19
The optimal schedule depends in part on the relative variances of the public's demand for money
and the banking system's demand for reserves. If reserve demand is highly uncertain, then the optimal
policy is to peg the interest rate affecting money demand at the level consistent with the optimal
money stock, given the structure of the money demand equation-the quantity of reserves would
increase with interest rates. If money demand is highly uncertain, the optimal policy is to vary reserve
supplies inversely with interest rates, in order to maintain the combination of reserves and interest
rates suggested by the structure of reserve demand that will keep the money supply at its optimal
level.
^Obviously, anything which provides information about the intermediate target also provides some
indirect information about final goals. However, in a formal intermediate targeting framework this
information might be disregarded. The inefficiencies of ignoring additional information about final
(continued...)
292
Even if the questions about the structures of money and reserve demands are answered,
problems arise in the selection and use of an operational rate instrument. The ideal rate to target is
one that is closely associated both with the demand for the aggregate and with reserve demand and
that is also subject to Desk control. The federal funds rate is not perfect for this use, despite the
Desk's ability to exert great control over it, because the association between the funds rate and the
demand for any of the usual aggregates is rather indirect. The theoretically correct rate relevant to
money demand measures the opportunity cost to households and businesses of holding money, as
opposed to other assets. A plausible proxy for this ideal rate might be the spread between the threemonth Treasury bill rate and deposit rates (when the money aggregates in question are Ml, M2, or
M3). Open market operations—which are usually carried out in the bill market, either for outright
purchases or sales or in the form of repurchase agreements (RPs)-do affect the bill rate, but the
relationship between the funds rate and the bill rate can vary unpredictably.21 Moreover, in today's
world of deregulated interest rates, deposit interest rates may move in hard-to-predict ways with the
funds rate. Thus, the funds rate may be a poor control variable for the monetary aggregates, since its
relationship to the opportunity cost of holding the aggregates could be unpredictable.
Even if interest rates are not used as formal instruments to control intermediate targets, they
can play a role in the operation of monetary policy. Unlike data on financial stocks and flows, interest
rate data are available continuously and can be used to gather information about operational targets
less frequently observed. The federal funds rate, for instance, has been and can be used to get a sense
of the strength of the demand for reserves during a maintenance period.22
In general, interest rates may have a useful role in the control of intermediate aggregate
targets. Uncertainties in the demand for aggregates and the inability of the Desk to tightly control the
opportunity cost of holding aggregates make it inadvisable to use rates as the only operating
instruments. At the same time, however, these uncertainties make it inadvisable to ignore the
^(...continued)
goals in a rigid intermediate targeting system have been described by John Karekan, Thomas Muench,
and Neil Wallace (1970) and Benjamin Friedman (1975, 1977).
21
An example of such a break in the funds-bill relationship was the exceptionally low rates on bills
on a number of occasions in the summer of 1987, as recurrent federal debt ceiling crises resulted in
cancellations of a number of weekly bill auctions and technical shortages of bills in the marketplace.
See Federal Reserve Bank of New York, Open Market Analysis Division (1988).
22
This procedure will be discussed in more detail below.
293
information contained in rates. Interest rates also have a useful supplemental role in providing very
high frequency information about other operating instruments.
in. Past Experiences with Rate Targeting23
The Federal Reserve System has often made explicit use of interest rates in the policy process.
The System's most explicit and prolonged experience with rate "targeting" was, of course, during the
rate pegging period of the 1940s. The policy was simplicity itself: the Desk stood ready to buy
government securities to prevent nominal rates breaching ceilings. The secondary market yield on
three-month bills was maintained at 0.38 percent for the entire period from the spring of 1942 through
the middle of 1947, then increased in steps to about 1 3/8 percent by the start of 1951. The yield on
U.S. long-term bonds never rose above 2.49 percent during the entire 1942-50 period and dropped as
low as 2.08 percent in early 1946.
The initial aim of the pegging policy was to aid in thefinancingof World War II: since the
peg put a floor on government securities prices, it enabled the Treasury to calculate accurately the
minimum nominal proceeds and maximum interest expense of debt issues.24 It is hard to view the
peg as either an operating or an intermediate target, because there was no explicit linkage of the peg to
the behavior of any financial or economic variable. Nevertheless, the gradual increase in the short-run
peg in the postwar period did reflect some concern about inflation, and therefore the peg may best fit
into the intermediate target framework.
Perhaps the most interesting developments of the pegging era were the Desk's ability to
stabilize both short- and long-term rates and the maintenance of the pegs through the early postwar
period. Because inflation rates were high in the late 1940s after price controls ended, ex post real
interest rates were negative. During this period, postwar federal surpluses reduced the supply of
Treasury issues, and long-run inflation expectations were probably held down by a number of factors,
such as fears of a postwar depression, the long-term U.S. history of price-level stability, and the
23
For some general overviews of this subject, see Milton Friedman and Anna Schwartz (1963),
Ann-Marie Meulendyke (1988, 1990), and Larry Mote (1988).
24
The modern perspective would argue that the Treasury should have been concerned with the real
proceeds and real interest expense of the issues. A nominal peg would not necessarily reduce this
uncertainty, since the inflation rate would continue to be uncertain. However, the interaction of the
nominal rate peg with effective price controls during World War II meant that the Treasury could
implicitly calculate the real proceeds and real cost of debt issues, at least for the short run.
294
continued connection of the dollar to gold.
The pegged structure of rates came under increasing pressure after the start of the Korean War,
as government debt expanded again, inflationary pressures increased, and depression fears faded. The
pegging policy was abandoned with the 1951 Federal Reserve-Treasury Accord.
The pegging experience would seem to have few lessons for today—for instance, fears of a
severe depression are not likely to emerge as major elements in the financial markets. Nonetheless,
the ability of the System to cap rates (both short- and long-term) for nearly a full decade, with the
economy showing little sign of exceptional instability, provides some challenge to the more dogmatic
rational expectations view discussed in the section on recent theoretical targeting models.
In the generation following the accord, the System tended to avoid policy adjustments in the
periods surrounding Treasury refundings~a policy known as "even keeling" (see Axilrod 1971;
Kaufman 1973; and Struble and Axilrod 1973). For practical purposes, this meant that the Desk tried
to prevent policy from influencing interest rates from the time the refundings were announced until the
issues were distributed (interest rates could and did fluctuate for other reasons during these periods).
During the period of even keeling, the Treasury sold coupon issues through subscriptions. A
coupon and minimum price were preannounced, and the issues were allocated to the highest bidders.
If market interest rates rose following the issue announcement, there was the chance that the issue
could fail (the Treasury would not be able to sell the full amount of the issue at the announced
coupon), while if rates rose between the time of the auction and the distribution, successful bidders
would suffer capital losses. The even keel helped to prevent both of these events; in effect the Federal
Reserve was acting as the Treasury's underwriter.
The even keeling policy ended in the early 1970s. The Treasury switched to the current
coupon auction procedure, which effectively eliminates the possibility of a failed auction. The
increased frequency of coupon issues as the government debt rose may have also increased the
likelihood that keeping an even keel would hinder the conduct of basic monetary policy. Furthermore,
the increasing liquidity of the government securities market enabled purchasers to sell off exposure in
the secondary and derivative markets between the time of auction and distribution.
The even keel approach treated rate stability as a target, with the successful sale of the
Treasury issue as the goal. There was some criticism that the procedure may have hindered the Desk's
5
Karl Brunner and Allan Meltzer (1966, 1989) discuss the pegging experience and its relationship
to debt management and monetary policy. Barry Eichengreen and Peter Garber (1990) put the episode
in a modern theoretical framework.
295
ability to control the aggregates, but hard evidence that this was the case is scant (Struble and Axilrod
1973). Given today's very different Treasury auction procedures, this experience offers few direct
lessons, but it does illustrate effectively the importance the System has at times placed on limiting
interest rate volatility.
In the early 1960s many believed that a flatter yield curve would be desirable. Lower longterm interest rates would supposedly help to increase U.S. capital spending, while higher short-term
rates would supposedly attract foreign investment in dollar assets and ease balance of payments
problems. To help flatten the yield curve, the System made some efforts to lengthen the maturity of
its portfolio ("operation twist").
In practice, the "twist" policy was not strongly pursued. Yield curve considerations were by
no means the only factors then governing Desk actions, and analysis suggests that the policy had little
effect on the yield curve (Modigliani and Sutch 1966). Conceptually, though, operation twist may be
viewed as intermediate targeting. A target (a reduced spread between short- and long-term rates) was
chosen as a stand-in for ultimate goals (faster capital formation, increased foreign willingness to hold
dollars), and policy was at least partly directed at the target.
The most sustained and systematic use the System has made of interest rates in the last
generation has been the role of the federal funds rate as either an outright operating instrument or a
guide to Desk operations. During the 1970s, following a period when operations were directed toward
affecting credit market conditions (which were in turn believed to be heavily influenced by short-term
nominal rates), the Desk tended to peg the funds rate in a narrow range during intermeeting periods.
The desired funds range was an operating target supposed to be consistent with the desired growth
rates of the monetary aggregates, which were the primary intermediate targets (Axilrod and Lindsey
1981). The usefulness of this type of procedure was discussed above in the section on operating
instruments.
This operating policy changed in October 1979. Nonborrowed reserves replaced the funds rate
as the primary operating instrument. (The FOMC continued to post ranges for the funds rate in its
directive, but these were many times wider than the earlier ranges, applied only to weekly averages,
and viewed as signals for consultation, rather than as binding constraints on the Desk.) The view at
the time was that high and variable rates of inflation had downgraded the value of interest rates as
control variables for the aggregates: the optimal level of the funds rate for monetary control had
become excessively volatile and hard to compute.
In 1982, a deep recession, coupled with problems in various parts of the financial system,
296
persuaded the FOMC that the observed strong growth of the monetary aggregates, particularly Ml,
was not providing a correct signal about the future course of economic activity. The ongoing
introduction of variable interest rate accounts had led to large, unpredictable swings of funds into and
out of the various aggregates.26 In these circumstances, the FOMC downgraded the aggregates
(especially Ml) as intermediate targets, in favor of placing more emphasis on direct observation of
economic indicators.
Although interest rates have gained no further role as formal intermediate targets, they have
become more important on an operating level (Wallich 1984). With less emphasis placed on control
of the aggregates, the Desk has tended to allow short-term shocks to reserve demand to be absorbed
by fluctuations in the volume of nonborrowed reserves—the level of borrowings has been viewed as a
key operating target. The borrowings target, coupled with the discount rate, suggests a likely range for
the federal funds rate during a reserve maintenance period. (The FOMC directive includes a wider
monitoring range for the entire intermeeting period.) A tendency for the federal funds rate to breach
the top end of such a range, along with high levels of borrowing, implies that reserve demand may be
fundamentally stronger than anticipated. If this strength is coupled with unusually strong economic
data or rapid growth of the broader aggregates, the FOMC may decide to increase the degree of
pressure on reserve markets by raising the borrowings objective or the discount rate.
Thus, under procedures adopted in 1983, the federal funds rate has been used to provide
information for setting operating targets. It has not been used as the primary operating instrument, as
it had been before 1979, but it has been viewed as a useful indicator of the demand for reserves. In
recent years, particularly since the 1987 stock market crash, the relationship between borrowings and
the funds rate appears to have shifted or to have become more uncertain than it was previously. In
these circumstances, the Desk has often given increased weight to the funds rate in operations and
lesser weight to the borrowings target, particularly at times when the markets were sensitive to
potential policy shifts.
In sum, the history of interest rates in the formal monetary policy process has been tangled.
There has been a gradual evolution away from treating rate levels as either ultimate goals of policy or
intermediate targets for final goals. However, the System continues to put weight on interest rates as
indicators of reserve market pressure.
26
These large shifts of funds ended after the introductory period for the new accounts. However,
the changes in the financial system meant that the demand for the aggregates became more sensitive to
various interest rate spreads (John Wenninger 1986).
297
IV. Conclusions
There is little or no reason to believe that using interest rates as the only intermediate targets
of monetary policy is desirable. While the Desk can exert great control over nominal short-term
interest rates and can probably exert strong influence over short-term real interest rates, its ability to
influence nominal and real long-term interest rates in a predictable way is problematic (even assuming
that real long-term rates can be observed on an ex ante basis). Since it is generally believed that
fluctuations in real long-term interest rates exert more influence on real economic activity than
movements in alternative types of rates, interest rate targeting cannot reliably achieve a real output
goal. The likely instability in the relationship between interest rates and real output further weakens
the case for interest rate targeting.
The relationship between interest rates and inflation is particularly uncertain. In principle, the
implications of rate moves for inflation may be derived from their implications for real growth, but
making such inferences adds another potentially weak link to an already weak chain.
The empirical literature concludes that real long-term interest rates (and perhaps nominal shortterm rates as well) do lead real output movements and ultimately inflation, but the relationship is
unstable. It may be possible, however, to use interest rate movements as supplemental information
variables in an aggregate targeting system. Some limited empirical evidence suggests that ex post real
short-term interest rates would be useful in this context. Such a targeting system seems to be in rough
accord with some of the recent theoretical discussion.
The role of interest rates as operating targets is more clearcut than their role as intermediate
targets. Under the current operating procedure, the federal funds rate provides information about
conditions in the reserve markets.
The System has at times brought interest rates into the forefront of policy making, most
notably in the rate pegging period during and after the Second World War, and to a much more
limited extent during "operation twist" and in the 1970s, when the federal funds rate acted as the main
operational tool. Inflationary pressures led to the end of pegged rates and contributed to the
downgrading of the funds rate in 1979 (the funds rate has been more important operationally since
1983). It is likely that unless rate targets are highly flexible, any explicit rate targeting system, on an
operational or intermediate level, that ignores the information provided by reserve, money, and credit
aggregates would also fall victim to inflation.
298
References
Akhtar, M. A., and Ethan S. Harris. 1986-87. "Monetary Policy Influence on the Economy-An
Empirical Analysis." Federal Reserve Bank of New York Quarterly Review, vol. 11 (Winter),
pp. 19-34.
Axilrod, Stephen H. 1971. "The FOMC Directive as Structured in the Late 1960's: Theory and
Appraisal." In Open Market Policies and Operating Procedures-Staff Studies. Board of Governors
of the Federal Reserve System, pp. 1-36.
Axilrod, Stephen H., and David E. Lindsey. 1981. "Federal Reserve System Implementation of
Monetary Policy: Analytical Foundations of the New Approach." American Economic Review, vol.
71 (May), pp. 246-52.
Barro, Robert J. 1989. "Interest-Rate Targeting." Journal of Monetary Economics, vol. 23 (January),
pp. 3-30.
Blanchard, Olivier J. 1984. "The Lucas Critique and the Volcker Deflation." American Economic
Review, vol. 74 (May), pp. 211-15.
Bosworth, Barry. 1988. "Institutional Change and the Efficacy of Monetary Policy." Brookings
Papers on Economic Activity, 1, pp. 77-124.
Brainard, William C. 1964. "Financial Intermediaries and a Theory of Monetary Control." Yale
Economic Essays, vol. 4 (Fall), pp. 431-82.
Brayton, Flint, and Eileen Mauskopf. 1985. "The Federal Reserve Board MPS Quarterly
Econometric Model of the U.S. Economy." Economic Modelling, vol. 2 (July), pp. 170-292.
Brunner, Karl, and Allan H. Meltzer. 1966. "A Credit Market Theory of the Money Supply and an
Explanation of Two Puzzles in U.S. Monetary Policy." In T. Bagiotti, ed., Investigations in
Economic Theory and Methodology, Essays in Honor of Marco Fanno, vol. 2. Padua, Italy: Cedam.
Reprinted in Brunner and Meltzer, eds., Monetary Economics. Oxford and London: Basil Blackwell,
1989.
Bryant, Ralph C. 1983. Controlling Money: The Federal Reserve and its Critics. Washington, D.C:
Brookings Institution.
Canzoneri, Matthew B., Dale W. Henderson, and Kenneth S. Rogoff. 1983. "The Information
Content of the Interest Rate and Optimal Monetary Policy." Quarterly Journal of Economics,
vol. 98 (November), pp. 545-66.
Carlson, John A. 1977. "Short-Term Interest Rates as Predictors of Inflation: Comment." American
Economic Review, vol. 67 (June), pp. 469-75.
299
Cook, Timothy, and Thomas Hahn. 1989. "The Effect of Changes in the Federal Funds Rate Target
on Market Interest Rates in the 1970s." Journal of Monetary Economics, vol. 24 (November),
pp. 331-51.
Dotsey, Michael. 1989. "Monetary Control under Alternative Operating Procedures." Journal of
Money, Credit, and Banking, vol. 21 (August), pp. 273-90.
Dotsey, Michael, and Robert G. King. 1986. "Informational Implications of Interest Rate Rules."
American Economic Review, vol. 76 (March), pp. 33-42.
Eichengreen, Barry, and Peter Garber. 1990. "Before the Accord: U.S. Monetary-Financial Policy
1945-51." National Bureau of Economic Research, Working Paper no. 3380. June.
Englander, A. Steven. 1990. "Optimal Monetary Policy Design: Rules vs. Discretion Yet Again."
Federal Reserve Bank of New York.
Estrella, Arturo, and Gikas A. Hardouvelis. 1989. "The Term Structure as a Predictor of Real
Economic Activity." Federal Reserve Bank of New York, Research Paper no. 8907. May.
. 1990. "Monetary Policy and the Yield Curve." Federal Reserve Bank of New York.
Fama, Eugene F. 1975. "Short-Term Interest Rates as Predictors of Inflation." American Economic
Review, vol. 65 (June), pp. 269-82.
. 1977. "Interest Rates and Inflation: The Message in the Entrails." American Economic
Review, vol. 67 (June), pp. 487-96.
Federal Reserve Bank of New York, Open Market Analysis Division. 1988. "Monetary Policy
and Open Market Operations during 1987." Federal Reserve Bank of New York Quarterly Review,
vol. 13 (Spring), pp. 41-58.
Friedman, Benjamin M. 1975. "Targets, Instruments, and Indicators of Monetary Policy." Journal of
Monetary Economics, vol. 1 (October), pp. 443-73.
. 1977. "The Inefficiency of Short-Run Monetary Targets for Monetary Policy." Brookings
Papers on Economic Activity, 2, pp. 293-335.
. 1989. "Changing Effects of Monetary Policy on Real Economic Activity." Monetary
Policy Issues in the 1990s, pp. 55-112. Federal Reserve Bank of Kansas City, Jackson Hole
Conference.
Friedman, Milton. 1968. "The Role of Monetary Policy." American Economic Review, vol. 58
(March), pp. 1-17.
Friedman, Milton, and Anna Jacobson Schwartz. 1983. A Monetary History of the United States,
1867-1960. Princeton: Princeton University Press and National Bureau of Economic Research.
300
Goodfriend, Marvin. 1987. "Interest Rate Smoothing and Price Level Trend-Stationarity." Journal of
Monetary Economics, vol. 19 (May), pp. 335-48.
Hall, Robert E. 1977. "Investment, Interest Rates, and the Effects of Stabilization Policies."
Brookings Papers on Economic Activity, 1, pp. 61-103.
Hester, Donald D. 1982. "Panel Discussion." Interest Rate Deregulation and Monetary Policy,
pp. 231-39. Federal Reserve Bank of San Francisco, Asilomar Conference.
Hoehn, James G. 1987. "Interest Rate Rules are Infeasible and Fail to Complete Macroeconomic
Models." Federal Reserve Bank of Cleveland, Working Paper no. 8719, December.
Holbrook, Robert, and Harold Shapiro. 1970. "The Choice of Optimal Intermediate Economic
Targets." American Economic Review, vol. 61 (May), pp. 40-46.
Joines, Douglas. 1977. "Short-Term Interest Rates as Predictors of Inflation: Comment." American
Economic Review, vol. 67 (June), pp. 476-77.
Kane, Edward J. 1980. "Politics and Fed Policymaking: The More Things Change, the More They
Remain the Same." Journal of Monetary Economics, vol. 6 (April), pp. 199-211.
. 1982. "Selecting Monetary Targets in a Changing Financial Environment." Monetary
Policies in the 1980s, pp. 181-206. Federal Reserve Bank of Kansas City, Jackson Hole Conference.
Karekan, John H. 1970. "The Optimum Monetary Instrument Variable." Journal of Money, Credit,
and Banking, vol. 2 (August), pp. 385-90.
Karekan, John H., Thomas Muench, and Neil Wallace. 1972. "Optimal Open Market Strategy: The
Use of Information Variables." American Economic Review, vol. 63 (March), pp. 156-72.
Kaufman, Henry. 1973. "Federal Debt Management: An Economist's View from the Marketplace."
Issues in Federal Debt Management, pp. 155-74. Federal Reserve Bank of Boston, Conference
Series no. 10, June.
Laurent, Robert D. 1988. "An Interest Rate-Based Indicator of Monetary Policy." Federal Reserve
Bank of Chicago Economic Perspectives, vol. 12 (January-February), pp. 3-14.
. 1989. "Testing the 'Spread.1" Federal Reserve Bank of Chicago Economic Perspectives,
vol. 13 (July-August), pp. 22-34.
LeRoy, Stephen F. 1975. "Efficient Use of Current Information in Short-Run Monetary Control."
Board of Governors of the Federal Reserve System, Division of Research and Statistics, Special
Studies Section, Paper no. 66.
. 1979. "Monetary Control Under Lagged Reserve Accounting." Southern Economic Journal,
vol. 46 (October), pp. 469-70.
301
LeRoy, Stephen F., and Roger N. Waud. 1977. "Applications of the Kalman Filter in Short-Run
Monetary Control." International Economic Review, vol. 18 (February), pp. 195-207.
Lombra, Raymond E. 1984. "The Changing Role of Real and Nominal Interest Rates." Federal
Reserve Bank of Kansas City Economic Review (February), pp. 12-25.
McCallum, Bennett T. 1981. "Price Level Determinacy with an Interest Rate Policy Rule and
Rational Expectations." Journal of Monetary Economics, vol. 8 (November), pp. 319-29.
. 1986. "Some Issues Concerning Interest Rate Pegging, Price Level Determinacy, and the
Real Bills Doctrine." Journal of Monetary Economics, vol. 17 (January), pp. 135-60.
Meulendyke, Ann-Marie. 1988. "A Review of Federal Reserve Policy Targets and Operating Guides
in Recent Decades." Federal Reserve Bank of New York Quarterly Review, vol. 13 (Autumn),
pp. 6-17.
. 1990. U.S. Monetary Policy and Financial Markets. Federal Reserve Bank of New York,
1990.
Mishkin, Frederick S. 1990. "What Does the Term Structure Tell Us About Future Inflation?"
Journal of Monetary Economics, vol. 25 (January), pp. 77-95.
Modigliani, Franco. 1971. "Monetary Policy and Consumption." Consumer Spending and Monetary
Policy: The Linkages, pp. 9-85. Federal Reserve Bank of Boston, Conference Series no. 5, June.
Modigliani, Franco, and Richard Sutch. 1966. "Innovations in Interest Rate Policy." American
Economic Review, vol. 56 (May), pp. 178-97.
Modigliani, Franco, and Lucas Papademos. 1980. "The Structure of Financial Markets and the
Monetary Mechanism." Controlling Monetary Aggregates HI. pp. 111-55. Federal Reserve Bank of
Boston, Conference Series no. 23, October.
Mote, Larry R. 1988. "Looking Back: The Use of Interest Rates in Monetary Policy." Federal
Reserve Bank of Chicago Economic Perspectives, vol. 12 (January-February), pp. 15-28.
Nelson, Charles R., and G. William Schwert. 1977. "On Testing the Hypothesis that the Real Rate of
Interest is Constant." American Economic Review, vol. 67 (June), pp. 478-86.
Paulus, John D., and Stephen S. Roach. 1984. "Real Interest Rates and the Economic Recovery."
Economic Perspectives, Morgan Stanley Economics Department, New York, May 11.
Parkin, Michael. 1978. "A Comparison of Alternative Techniques of Monetary Control Under
Rational Expectations." Manchester School of Economics and Social Studies, September,
pp. 252-87.
Pierce, James L., and Thomas D. Thomson. 1972. "Some Issues in Controlling the Stock of Money."
Controlling Monetary Aggregates II: The Implementation, pp. 115-36. Federal Reserve Bank of
Boston, Conference Series no. 9, September.
302
Poole, William. 1970. "Optimal Choice of Monetary Policy Instruments in a Simple Stochastic
Macro Model." Quarterly Journal of Economics, vol. 84 (May), pp. 197-216.
Porter, Richard D., ed. 1990. Proceedings of the Monetary Affairs Workshop on Asset Prices and the
Conduct of Monetary Policy. Board of Governors of the Federal Reserve System, Finance and
Economics Discussion Series, no. 125.
Radecki, Lawrence J., and Vincent Reinhart. 1988. "The Globalization of Financial Markets and the
Effectiveness of Monetary Policy Instruments." Federal Reserve Bank of New York Quarterly
Review, vol. 13 (Autumn), pp. 18-27.
Sargent, Thomas J., and Neil Wallace. 1975. "'Rational' Expectations, the Optimal Monetary
Instrument, and the Optimal Monetary Supply Rule." Journal of Political Economy, vol. 83 (April),
pp. 241-54.
Shiller, Robert J., John Y. Cambell, and Kermit L. Shoenholtz. 1983. "Forward Rates and Future
Policy: Interpreting the Term Structure of Interest Rates." Brookings Papers on Economic Activity,
1, pp. 173-225.
Sims, Christopher A. 1980. "Comparison of Interwar and Postwar Business Cycles: Monetarism
Reconsidered." American Economic Review, vol. 70 (May), pp. 250-57.
Steindel, Charles, and Carl Palash. 1986. "The Shifting Term Structure of Interest Rates." Federal
Reserve Bank of New York, Research Paper no. 8613. December.
Stock, James H., and Mark W. Watson. 1989. "Interpreting the Evidence on Money-Income
Causality." Journal of Econometrics, vol. 40 (January), pp. 161-81.
Struble, Frederick M., and Stephen H. Axilrod. 1973. "Even Keel Revisited." Issues in Federal Debt
Management, pp. 235-53. Federal Reserve Bank of Boston, Conference Series no. 10, June.
Summers, Lawrence H. 1983. "The Nonadjustment of Nominal Interest Rates: A Study of the Fisher
Effect." In J. Tobin, ed., Macroeconomics, Prices and Quantities. Washington, D.C.: Brookings
Institution.
Tobin, James. 1969. "A General Equilibrium Approach to Monetary Theory." Journal of Money,
Credit, and Banking, vol. 1 (February), pp. 15-29.
. 1978. "Monetary Policies and the Economy: The Transmission Mechanism." Southern
Economic Journal, vol. 44 (January), pp. 421-31.
. 1982. "Money and Finance in the Macroeconomic Process." Journal of Money, Credit, and
Banking, vol. 14 (May), pp. 171-204.
Tobin, James, and William C. Brainard. 1963. "Financial Intermediaries and the Effectiveness of
Monetary Controls." American Economic Review, vol. 53 (May), pp. 383-400.
303
VanHoose, David. 1989. "Monetary Policy and Price Level Non-Trend-Stationarity." Journal of
Money, Credit, and Banking, vol. 21 (May), pp. 232-39.
Wallich, Henry C. 1984. "Recent Techniques of Monetary Policy." Federal Reserve Bank of Kansas
City Economic Review. May, pp. 21-30.
Walsh, Carl E. 1983. "Should the Federal Reserve Establish a Real Interest Rate Target?" Federal
Reserve Bank of Kansas City Economic Review, June, pp. 22-33.
Wenninger, John. 1986. "Responsiveness of Interest Rate Spreads and Deposit Flows to Changes in
Market Rates." Federal Reserve Bank of New York Quarterly Review, vol. 11 (Autumn), pp. 1-10.
304
COMMODITY PRICES AS INTERMEDIATE TARGETS
AND INFORMATION VARIABLES
Spence Hilton
Commodities have played a central role in many monetary systems through history. Perhaps
the most notable commodity-based monetary regime was the international gold standard upon which
many countries based their economies until early in the twentieth century. Since that time,
commodities have played a greatly reduced role in monetary policy. Recently, however, there has
been a revived interest in using commodity prices to help guide monetary policy, but within the
general framework of the current monetary structure. This study reviews current arguments and
evidence on the use of commodity prices in monetary policy. The principal focus is on the
information content, or leading indicator properties, of commodity prices, although some consideration
is also given to the usefulness of commodity prices as an intermediate target variable. The discussion
is organized into four major parts. First, the various possible uses of commodity prices in monetary
policy are examined, beginning with a brief comparison of commodity standards, commodity price
targets, and commodity prices as information variables. A simple theoretical examination of the
implications of adhering to a price target is included. Second, the behavior of commodity prices is
reviewed. The underlying determinants and several key behavioral characteristics of commodity prices
are examined, with particular attention given to the reasons that commodity prices might be a leading
indicator of economic developments. Empirical evidence on the leading indicator properties of
commodity prices, much of which is of recent vintage, is critically reviewed in a third section. A final
section discusses the ability of the monetary authorities to control commodity prices.
I. The Role of Commodities in Monetary Policy
This study considers only proposals to utilize commodity prices in the context of the existing
financial framework-that is, where monetary policy is implemented using conventional instruments of
control, such as open market operations in government securities, within the current regulatory and
structural environment. Excluded are proposals that call for a return to a commodity standard.
However, a commodity standard is the historical forerunner of proposals to use commodity prices as a
305
target or leading indicator, and despite many institutional differences, there are underlying similarities
in the way these different regimes transmit economic disturbances. The essential ingredients of a
commodity standard are briefly reviewed, and then recent proposals to use an index of commodity
prices as an intermediate target or as an information variable are described. Some implications of
using price targets in monetary policy are also examined.
A. Commodity Standards
Under a commodity standard, the currency unit (dollars) is defined as a specific quantity of a
commodity or basket of commodities.1 In principle, any commodity or group of commodities may
serve as the underlying resource unit, although historically gold has most often been the basis of a
commodity standard.2 The commodity forming the standard typically has legal tender status, but
otherwise any number of institutional arrangements may apply. A practice commonly associated
with—but not truly essential to~a commodity standard is for the central bank to intervene directly in
commodity markets and to hold commodity stockpiles. In fact, some argue that such intervention can
distort important price relations critical to the smooth functioning of a commodity standard,
particularly when there are economic disturbances unique to that commodity, by artificially affecting
the supply or demand of the commodity.
To its advocates, a commodity standard offers stability in prices and, thereby, stability in real
economic activity as well. Aggregate price changes under a commodity standard would be largely the
result of shifts in supplies or demand for the commodity relative to total supply and demand. If, say,
the commodity constituting the standard becomes more plentiful, then prices for most goods and
services will rise in order to maintain the legally defined value of the currency unit. Similarly, if the
value of the commodity rises, perhaps because of a supply shortfall, then the general price level will
decline.3
Proponents of a commodity standard usually argue that the likelihood of price changes arising
^ u c h of this summary of the workings of a commodity standard is adapted from Hall (1984).
See the U.S. Gold Commission (1982) and Bordo and Schwartz (1984) for a retrospective on the gold
standard and an evaluation of the possible uses of gold in monetary policy today.
2
For many years the U.S. dollar was defined as being worth .04838 ounces of gold. Hall (1984)
describes how a multicommodity standard might be devised.
3
Under an international commodity standard, the commodity flows between nations arising from
international payments imbalances lead to price level adjustments between countries.
306
from disturbances affecting relative supplies and demand for the commodity is small when compared
with the risk that monetary authorities will pursue inflationary policies under a fiat money regime.
Critics respond that the historical record of aggregate price stability under commodity standards is not
as good as its advocates suggest. Moreover, as with all monetary regimes, the adjustment of the
general price level to changes in the relative value of the commodity can lead to large real economic
costs when there exist structural rigidities, such as labor contracts, that fix some nominal prices.4
Because general price movements under a commodity standard arise largely from the changing relative
value of the commodity, a demonstrated stability in this relative price is perhaps the most important
characteristic of any commodity chosen to make up the standard.
B. Current Proposals to Use Commodity Prices
Currently, most advocates of an expanded role for commodities in monetary policy suggest
using their prices either as an intermediate target or much more commonly as a key information
variable, as a way of significantly enhancing the stability of aggregate prices. This section summarizes
the main elements of recent proposals to use commodity prices in monetary policy.
Commodity prices as an intermediate target
Under a commodity price target, the central bank attempts to maintain the value of some
commodity or basket of commodities within a predetermined range. Unlike a commodity standard, the
commodity itself is not legal tender, and the monetary authorities attempt to hit their target by
adjusting reserves and interest rates.5 Most proposals call for targeting an index of "sensitive"
commodity prices rather than a single price in order to avoid the complications that can arise when
individual commodity markets are disrupted.6 As an exception, the price of gold is sometimes singled
out as a possible target variable. In a variation on direct targeting of a commodity price index,
Genetski (1982) proposes to automatically recalibrate the targeted growth range of a money
aggregate-the money base—each month, according to a rule based on the rate of growth of an index of
4
Hall (1984) argues that when the relative value of a commodity changes, leading to incipient
movements in the general price level, the central bank could head off any inflation by adjusting the
legal definition of the quantity of the commodity in the currency unit.
5
Directly intervening in commodity markets is also occasionally proposed as a way to target
commodity prices.
6
Relative price movements and aggregate commodity price indexes are discussed in Section EC.
307
sensitive commodity prices.
If the monetary authorities could strictly control the targeted commodity price, then economic
disturbances would affect aggregate prices much as they would under a commodity standard. For
example, suppose that the central bank hits its targeted level for commodity prices with reasonable
accuracy at all times. In this case, general price movements would arise only when the relative value
of the targeted commodity changed, as they would under a commodity standard. In the absence of
shocks that cause these kinds of relative price changes, a realized commodity price target could
contribute to overall price stability.
Most proposals for targeting commodity prices do not detail the time horizon over which these
targets should be defined or the circumstances under which they should be changed. Some see
commodity price targets as an alternative to intermediate monetary aggregates, suggesting that target
ranges covering periods of up to a year would be appropriate. Others emphasize the advantages of
targeting a variable whose value is reported daily, indicating that commodity prices might be well
suited for a shorter term operational role. This latter possibility, however, seems limited by the
volatility of commodity prices even over short intervals (see Section ED). Rarely do these proposals
address how target ranges should be revised following relative price movements or changes in real
economic conditions during business cycles. Implicit in most proposals is the view that targets should
not be modified in order to meet short-term objectives for real variables.
Commodity prices as an information variable
Recognizing some of the practical difficulties in setting formal price targets, many analysts
propose to use commodity prices, including gold prices, as an information variable or leading indicator
to help guide monetary policy.7 In this capacity, some index of commodity prices would be carefully
monitored because of its historical correlation with aggregate inflation or, sometimes, real output. By
anticipating future price movements, commodity prices can provide policy makers the opportunity to
combat inflationary pressures at a relatively early stage. Proponents maintain that such a system,
properly applied, would ensure overall price stability without being excessively rigid. The current
tools of monetary control would be used to implement policy, but there would be no inflexible rule or
target value for setting commodity prices or the operating instruments. Instead, the monetary
7
In this study, the phrase "leading indicator" is used interchangeably with "information variable" to
describe a variable that possesses some predictive power for other variables.
308
authorities would be given considerable leeway in responding to movements in commodity prices,
allowing them to react to economic disturbances according to the particular circumstances.
Some proposals would combine the adoption of commodity prices as an information variable
with other institutional changes. Angell (1987) proposes a "policy adjustment matrix" for modifying
target ranges of a money aggregate on the basis of commodity price behavior but eschews specific
feedback rules. Johnson (1988) advocates monitoring commodity prices in conjunction with other
sensitive indicators of economic conditions, such as exchange rates and interest rates. Following
several sensitive economic measures simultaneously, it is argued, might better enable policy makers to
identify the nature of underlying economic disturbances.
Evaluating proposals that recommend monitoring certain variables but do not call for specific
policy rules is difficult because it is typically assumed that the judgments needed to correctly apply the
proposal will be made. Moreover, to some degree the cental bank's current policy procedure, often
characterized as one that "looks at everything," already conforms to these kinds of proposals. In
testimony, Chairman Greenspan (1987) indicated that the Federal Reserve already uses information
contained in commodity prices in conducting policy, but he stopped short of advocating a more formal
role for commodity prices. This review only evaluates proposals that substantially enhance the role of
commodity prices in monetary policy.
C. Targeting Aggregate Prices
The maintenance of general price stability is the overriding goal of most proposals to expand
the use of commodity prices in monetary policy. Moreover, in most of the proposals it is argued that
long-term price goals should not be sacrificed to pursue short-term output or employment objectives.
Even the proposed use of commodity prices as an information variable is intended to constrain
aggregate price movements. In the extreme, expanding the role of commodity prices in monetary
policy may be likened to targeting aggregate prices. This section outlines the implications for real
output of economic disturbances when commodity prices are targeted on the assumption that this
translates into successfully pegging aggregate prices. The analysis shows that under some
circumstances a policy that targets prices could aggravate the impact of economic shocks on real
output.
A textbook model of aggregate supply and demand is used in this discussion. The basic model
is presented graphically in Figure 1. An upward sloping aggregate supply (AS) curve is used to
describe supply relations in the short to intermediate run. A supply function might have a positive
309
Price Targets with a Demand Shock
Figure 1a
Price
Real output
Figure 1b
Interest rate
310
Real output
slope if, say, nominal wages respond with a lag to disturbances that affect the price level. In the long
run, the supply curve may be vertical. The downward sloping aggregate demand (AD) schedule is
derived from the LM and IS schedules presented in Figure lb, which describe equilibrium in the goods
and the money markets, respectively.8 In keeping with standard practice, the IS curve slopes
downwards while the LM curve has a positive slope. The money stock, which can be used to change
the LM curve and thus aggregate demand, is the only instrument of control available to the monetary
authorities. When the price level (P) is targeted, the monetary authorities change the money supply
following economic disturbances so as to keep prices fixed.
The simplicity of the model obviously limits this analysis. The economic relations are entirely
static, and there are no price expectations in the model. In fact, prices themselves play only a very
limited role in the underlying behavioral relations represented in the IS-LMframeworkand in
aggregate supply. The mechanism connecting commodity prices to aggregate prices is assumed to be
known and stable, allowing the monetary authorities to maintain complete control over commodity
price movements. Despite these shortcomings, theframeworkhighlights some of the principal
theoretical motivations for targeting commodity prices while also presenting some potential drawbacks.
Under a price rule, aggregate demand shocks corresponding to unexpected movements in the IS
and LM curves have no impact on prices and real output. For example, a decline in investment
spending will shift the IS curve in Figure lb from IS0 to IS,, lowering real output and interest rates
and causing aggregate demand to shift to AD,. (The resulting change in prices from P0 to P, will lead
to a shift in the LM curve to LM, through its impact on real money balances.) Under the price rule,
the nominal money supply is expanded, shifting the LM curve to the right and returning aggregate
demand to AD0, where prices and real output are restored to their initial levels.9 Disturbances to the
^ e AD curve is derived by tracing the impact that changes in the price level have on the LM
curve by changing real money balances and thereby on real output in the IS-LMframeworkin Figure
lb. Shifts in the LM curve brought on by changes in the nominal money supply and movements in
the IS curve lead to shifts in the AD curve. Movements in the LM schedule induced by changes in
the price level correspond to movements along the AD curve.
^ e rise in the nominal money supply needed to move aggregate demand back to AD0 will
initially move the LM curve to the right of LM2, but the rise in prices from P, to P0 will shift the LM
curve back to LMLj through the real balance effect.
It is assumed that economic disturbances shift the schedules but do not affect their underlying
slope parameters. If the slopes of the aggregate demand function were to rotate, maintaining aggregate
prices at the level P0 might cause the level of real output to change.
311
LM curve are countered in a similar manner. Under a price rule, demand disturbances are completely
offset, leaving real output as well as aggregate prices unchanged.
With supply or price shocks, the monetary authorities magnify the total impact of these
disturbances on real output by adhering to a price target, at least in the short to intermediate run when
aggregate supply is not vertical. Consider a negative supply disturbance, such as an adverse change in
factor productivity, that shifts the aggregate supply schedule in Figure 2a leftward, from AS0 to ASj.
With the nominal money supply initially fixed, the price rise induced by this shift causes the LM
curve to move to LMt as real money balances decline. To combat higher prices, policy makers reduce
the nominal money supply, shifting the LM curve to LM2 and aggregate demand to AD2, where prices
are restored to their initial level. But this places additional downward pressure on real output when
aggregate supply is upward sloping.10 The cost of returning the price level to P0 is a further
reduction in real output to Q2.
This simpleframeworksuggests that targeting commodity prices could effectively counter
demand disturbances, but unless sufficient flexibility is built into its structure, a commodity price
target could aggravate the effects of aggregate supply or price shocks on real output. The analysis
assumes that lags in the recognition of disturbances and implementation of policy do not interfere with
the policy makers' ability to control aggregate prices. If these lags are significant or if there is some
inertia in price movements following an initial disturbance, then attempts to adhere strictly to an
aggregate price target could induce some instability in real output as the monetary authorities belatedly
attempt to maintain their targeted price level. An important benefit to monitoring or targeting
commodity prices, according to its proponents, is that these prices respond relatively quickly to
economic disturbances. Consequently, policy makers have sufficient opportunity to respond before
these shocks become embedded in the general price structure.
II. Commodity Price Behavior
This section discusses some of the key behavioral characteristics of commodity prices,
beginning with an overview of their proximate determinants, and explains why commodity prices, may
act as a leading indicator of important economic trends. Also examined are the volatility of
commodity prices, which can make it difficult to distinguish between temporary and permanent
10
In the long run, with a vertical AS curve, restoring prices to their original level induces no
further decline in output beyond that caused by the initial supply shock.
312
Price Targets with a Supply Shock
Figure 2a
Price
/
, AS
A D
Q
o
Q
Q
,
2
o
1
1
1
Real output
0
Figure 2b
Interest rate
Q
Q
2
Q
1
0
Real output
313
movements, and the importance of shifts in the equilibrium relation between commodity prices and the
aggregate price level.
A. Proximate Determinants of Commodity Prices
The immediate causes of movements in the overall level of commodity prices presented in this
section are identified from empirical attempts to model commodity prices. These "proximate"
determinants of commodity prices may differ from the "ultimate" determinants, which include
monetary and fiscal policies. This overview looks only at determinants of aggregate commodity price
measures and does not consider the many factors that can cause movements in individual commodity
prices.
Empirical research on the behavior of the general level of commodity prices focuses mostly on
the determinants of demand for commodities. In empirical work, real output, interest rates, exchange
rates, and aggregate prices are typically found to be the most significant determinants of broad-based
commodity price movements.11 As a measure of real economic activity, industrial production is
considered a good proxy for demand arising from the direct production requirements of manufacturing
industries, the principal users of primary commodities. Interest rates are seen as capturing both the
financing costs to manufacturers and the opportunity costs to speculators of holding commodities.
Interest rates may also affect commodity prices indirectly through their impact on the aggregate level
of economic activity. Commodity prices often respond to exchange rate movements because of the
important role commodities play in international trade. As a general rule, the sensitivity of commodity
prices denominated in any one country's currency to exchange rate changes is inversely related to that
country's importance (measured in market share) in world supply and demand for the commodity. For
example, exchange rate changes should have little impact on the domestic currency price of those
commodities that are primarily produced and consumed domestically. Finally, aggregate prices are
empirically useful for explaining commodity prices as a proxy for production costs.
Prices of individual commodities, of course, may be greatly affected by the many decisions that
affect the exploration, extraction, or cultivation of particular commodities. However, few attempts
have been made to capture supply-side factors in estimation because of the difficulties in identifying
satisfactory proxy variables.12 As discussed in Section ED, it is usually assumed that aggregate
"Examples of this research include Chu and Morrison (1984) and Englander (1985).
12
Chu and Morrison (1986) attempt to estimate supply-price dynamics.
314
measures of commodity prices are largely unaffected by supply shocks to individual commodities,
either because these shocks are offsetting or because their impact falls on just a small number of
commodities.
B. Leading Indicator Properties of Commodity Prices
Proposals to monitor commodity prices are based largely on the view that movements of these
prices precede broader economic developments, especially general inflation trends. This section
outlines the reasons that commodity prices are believed to be a good leading indicator. A survey of
empirical evidence on the information content of commodity prices appears in Section HI.
Commodity prices are flexible and thus respond quickly to underlying shifts in aggregate
demand or supply, as well as to all new information distilled in the marketplace. This flexibility may
reflect the competitive market structure in which many commodity prices are set, the relative
homogeneity of commodities, and their storability. By contrast, many other important price or cost
measures, such as wage rates, are fixed for long periods, perhaps because they are set by long-term
contracts.
The importance of commodities as an input in production is sometimes used to explain why
they can be used to predict changes in real output and aggregate prices. Primary commodities account
for about 10 percent of the value of total U.S. output, and their consumption occurs at a relatively
early stage of production. Economic disturbances affecting production translate quickly into changes
in demand for primary inputs, and the ability to hold commodities in inventory magnifies this
response. Given their flexibility, commodity prices can respond rapidly to these demand pressures. In
this way, commodity prices might anticipate future changes in final output. Commodity prices are
included in the Commerce Department's index of leading economic indicators for this reason. In
addition, commodity prices will signal subsequent changes in prices of final goods induced by
sustained movements in real output. It can be argued that aggregate demand pressures may show up
first at the retail level and only eventually work their way back towards commodity prices in the early
production stages. Even so, if prices for final goods and services are sticky, commodity prices may
still respond relatively quickly to these aggregate pressures.
In explaining the predictive power of commodity prices, some analysts focus on the "asset"
nature of the claims represented by ownership of commodity stocks and conclude that, like prices of
315
other assets, commodity prices incorporate market expectations about future price developments.13
Most commodities are nonperishable and can be stored reasonably cheaply, and so their possession
represents a kind of real claim on future output and provides a hedge against future inflation.
Moreover, because commodity prices are flexible, they can quickly reflect the expected impact of
economic developments. Together, these features suggest that commodity prices may respond rapidly
to inflationary shocks and be reliable indicators of future general price movements.
The availability of commodity prices on a timely basis is also seen as enhancing their value as
a leading indicator. Prices for many commodities are available daily and sometimes are quoted
virtually continuously. In addition, several commodity price indexes are compiled regularly and are
available at least monthly if not more often. The early availability of commodity price information is
seen as enabling policy makers to recognize and respond more quickly to evolving aggregate price
pressures.
C. Relative Price Changes and Commodity Price Indexes
The effectiveness of commodity prices as a target or information variable may be undermined
by secular changes in the relative price between the target commodity price and aggregate prices. In
many ways, the kinds of complications that relative price changes create are analogous to those
presented by unexpected income velocity shifts when a money aggregate is targeted. In the case of
commodity prices, when the relation between the leading indicator variable and the final policy
objectives changes, the ability to infer future price performance from commodity price changes is
impaired. The use of broad commodity price indexes to circumvent this problem and the historical
importance of relative price shifts are reviewed in this section.
Prices for many individual commodities or commodity groups have been greatly affected by
market-specific economic shocks. As a result, basing monetary policy on a narrow set of commodity
prices would entail substantial risks that their relative prices could change unexpectedly. To avoid this
potential problem, most proposals call for targeting or monitoring the price index of a broad basket of
commodities on the assumption that such a price index is unlikely to be unduly disturbed by shocks
specific to individual commodity markets.
In choosing a commodity price index to target or monitor, consideration must be given to the
commodities included in the index and to the weights used to aggregate the commodities. To reduce
13
See Section IID and the work of Frenkel (1986) and Boughton and Branson (1988) for examples.
316
the likelihood that shocks to a particular commodity category will dominate the entire index, the
commodity basket should include goods from most of the major commodity groups, such as
agricultural materials, metals, and chemicals. Some prefer to exclude petroleum products from any
index, despite its importance in world production, because of the tremendous oil price movements
caused by the actions of the suppliers' cartel. Agricultural commodities are also sometimes excluded
in proposals because of their sensitivity to weather patterns and their limited use in manufacturing
production. Consequently, much of the research on the predictive properties of commodity prices has
used price indexes of non-oil raw industrial commodities only.
Most existing commodity price indexes are weighted according to the relative importance of
each commodity in world or domestic output, or by its relative importance in trade. Others are simply
unweighted averages of the commodities in the index. Ideally, for purposes of forecasting general
price changes, commodities are best weighted by their information content about future inflation,
which may not be related to their relative importance in output or trade. However, few attempts have
been made to estimate optimum weights, and most analysts find it more convenient to rely on existing
commodity price indexes.14
A number of well known commodity price indexes are plotted in Chart 1. These differ by the
commodities included and in the weights used to compile each index.15 As Chart 1 shows, most
commodity price indexes tend to move together over long periods of time, but over shorter periods
they can diverge significantly. Currently, no one measure of commodity prices has been identified as
being clearly superior for use as a target or leading indicator variable.
There is strong empirical evidence that even broad-based commodity price indexes do not move
closely with the general price level for extended periods of time. The series plotted in Chart 2 show
that there have been sustained swings in the relative price of commodities and final outputs over the
postwar period. At the same time, there has been a secular downward trend in the relative value of
commodities. Additional evidence on the historical importance of relative price shifts comes from a
series of statistical tests of cointegration between commodity prices and the general price level. Two
variables are said to be cointegrated if there exists an underlying equilibrium relationship that prevents
14
Boughton and Branson (1988) report limited success in their attempts to estimate optimum
weights in constructing a commodity price index. The Journal of Commerce publishes an index that
reportedly is weighted by each commodity's value in forecasting inflation.
15
Rosine (1987) summarizes the purposes and construction techniques behind many aggregate measures
of commodity prices and output.
317
Chart 1
Indexes of Commodity Prices
Index, 1970=100
350 i
50 U
I
I
1969
I
I
1971
I
I
1973
I
I
1975
I
I
1977
I
I
1979
I
I
1981
i
i
1983
i
i
1985
i
1987
Producer price index of crude materials, includes oil
Journal of Commerce index of non-oil commodities, inflation forecasting weights
Commodity Research Bureau spot price index of non-oil raw industrials, production weights
IMF index of non-oil commodities, world export weights
Chart 2
Ratio of Commodity Prices to Consumer Prices
Index, 1970=1.00
1.6 i
g l i
i
i
i
i
i
i
i
i
i
i
i
i
i
i
i
i
i
i
i
i
i
i
i
i
i
i
i
i
i
i
i
i
i
i
i
1953 1955 1957 1959 1961 1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987
Ratio of P P I of crude materials to C P I
Ratio of Commodity R e s e a r c h Bureau to C P I
them from drifting apart indefinitely. Using these tests, a number of analysts report that they cannot
empirically reject the hypothesis that the levels of commodity prices and aggregate prices are not
cointegrated.16
Even if no long-run equilibrium relation exists between the levels of commodity prices and
aggregate prices, commodity prices could still play a useful role in monetary policy. For example,
tests for cointegration indicate that changes in commodity prices still may be related to general
inflation rates, that is, to changes in the general price level. Moreover, as long as the change is
gradual or predictable, a secular trend in relative prices need not complicate policy procedures any
more than, say, smooth velocity shifts complicate targeting a money aggregate. However, Chart 2
suggests that historically these relative price shifts may not always be smooth. On balance, these
results raise important questions about the nature of the relation between commodity prices and
aggregate prices, and they indicate that inflexible commodity price policy rules may be
counterproductive at times.
D. Volatility of Commodity Prices
Compared with the economic variables to be predicted, commodity prices are very volatile.
Furlong (1989) calculates that during the 1980s changes in many monthly aggregate commodity price
indexes were anywhere from three to twenty times more volatile than inflation in consumer prices,
depending on the commodity price index used. Chart 3, which plots monthly percent changes of a
representative commodity price index and consumer prices, illustrates this point. Changes in general
inflation expectations, seasonal factors, or supply shocks cannot account for much of this volatility.
This volatility reduces the ability to discriminate between temporary and permanent price movements
and makes daily or weekly commodity price data of little value for policy purposes.
Economists have advanced two theories to explain why a price index for even a broad basket of
commodities may vary far more than aggregate prices. The first focuses on the inelastic nature of
supply and demand for many commodities, at least over short periods of time. Aggregate disturbances
that, say, shift demand schedules for all products alike will often initially have a disproportionate
impact on commodity prices because their supply elasticities are so low. Over time, as supplies
become more plentiful, price pressures ease. Compared with aggregate prices, commodity prices
16
See the work by Boughton and Branson (1988), Garner (1989), Durand and Blondal (1988), and
Cody and Mills (1988) reported in Section IIIA.
319
Chart 3
Commodity Price Volatility
Monthly Percent Change
Percent
20
CRB commodity index
15
:
/
•
10 h
•
• » •
• i
• i
• i
i,
i,
i,
..»
••
'
L
-10 h
-15
1971
1973
1975
1977
1979
1981
1983
1985
1987
undergo much wider swings following economic disturbances. However, this reasoning does little to
explain large weekly or even daily commodity price swings.
A model of commodity price behavior that treats commodities as akin to financial assets better
explains the sometimes erratic nature of commodity price movements over even short time periods.
The basic model is a derivative of the popular exchange rate overshooting models, and examples are
found in Frenkel (1986) and Boughton and Branson (1988). Thisframeworkis notable for treating
commodity markets as though they were like any other asset market. In such asset markets, the
following arbitrage condition always holds:
(1)
i = dPc + b,
where i is a nominal interest rate, dPc is the expected rate of change of commodity prices, and b
represents the intrinsic real rate of return to holding commodities net of any storage costs, which is
320
typically assumed constant. In long-run equilibrium, dPc is zero, and i and b are equal to one
another.17
In the long run, changes in the money stock are neutral. An increase in the money stock, for
instance, will eventually cause commodity prices and all other prices to rise proportionately. However,
with final goods prices relatively inflexible in the short run, a rise in the money stock leads to a
temporary increase in the real money stock and places downward pressure on interest rates. By the
arbitrage condition (1), lower interest rates are consistent with an expected decline in commodity
prices, which can take place only if commodity prices first jump to a level above their new long-run
equilibrium rate. Thus, commodity prices initially overshoot their new long-run equilibrium level.
In sum, commodity prices respond quickly to economic shocks, and this information is readily
available. Consequently, commodity price data may be helpful to policy makers in predicting future
inflation, so long as these disturbances do not alter the underlying relation between commodity prices
and aggregate prices. However, the volatility of commodity prices, which can cause delays in
recognizing and reacting to underlying inflation pressures, and the possibility of significant relative
price changes militate against an important policy role for commodity prices.
HI. Leading Indicator Properties of Commodity Prices: Empirical Evidence
Most empirical work that examines the relation between commodity prices and macroeconomic
variables, especially aggregate prices, falls into one of two general categories: econometric evidence
using more or less conventional estimation techniques and qualitative work comparing major turning
points in the behavior of commodity prices and other economic variables of interest. This section
surveys a number of these studies, most of which focus on the usefulness of commodity prices for
predicting inflation.
A. Econometric Evidence18
Heightened interest in the possibility of targeting or monitoring commodity prices has prompted
a number of econometric investigations into the predictive content of commodity prices in the past few
17
A theory for a positive return to holding commodity stocks, or inverse carrying charge, is
presented in Working (1977). A positive real return to holding commodity stocks may arise from the
convenience and cost savings to producers of having commodity inputs on hand. Other possible
explanations are found in Frenkel and Hardouvelis (1985).
18
See the specific authors cited for more detailed presentations of the methods used and results
obtained in estimation.
~01
years. From the studies surveyed in this review, five general conclusions about the relation between
commodity prices and inflation emerge: (1) there is no apparent long-run equilibrium relation between
commodity and aggregate price levels; (2) commodity price movements often precede inflation
changes; (3) commodity prices apparently are not important direct determinants of aggregate prices; (4)
the marginal contribution of commodity prices to inflation forecasts appears small; and (S) the relation
between commodity price movements and inflation is not stable. The basis for each of these
conclusions is discussed below, following a brief examination of the methodological features common
to many of the studies reviewed.
Much of the recent econometric research has been conducted along similar methodological
lines. The sample period is usually restricted to the postwar era, and the data used are either monthly
or quarterly time series. Most of the studies are restricted to the U.S. case, although some look more
generally at countries within the Organization for Economic Cooperation and Development (OECD).
A consumer price index (CPI) is the most commonly used measure of aggregate prices, but sometimes
a producer price index (PPI) or a general price deflator is substituted.19 Estimated results do not
appear to be particularly dependent on the choice of aggregate price index.
Commodity prices are most often measured using one of the readily available broad-based
indexes. These include the Journal of Commerce measure of industrial commodity spot prices (JOC),
the producer price index of crude materials (PPICM), the Commodity Research Bureau index of
industrial materials spot prices (CRB), and an International Monetary Fund index of commodity prices
(IMF). These indexes are slanted toward coverage of raw materials used in manufacturing production,
and only the PPICM includes petroleum prices. As an exception, gold prices are sometimes included
separately in estimation. Percentage changes of these indexes and broad inflation measures are
generally used in estimation, and to reduce seasonality, twelve-month movements are calculated.
Moving averages of inflation rates are sometimes used to further reduce the volatility of commodity
prices. For the most part, the empirical results reported do not appear to be particularly sensitive to
the composition of the commodity price index.
In many of these studies, preliminary statistical work designed to ensure the proper
specification of estimated equations reveals that there is no long-run equilibrium relation between the
levels of commodity prices and aggregate prices. To guard against the kind of spurious correlation
19
A CPI is perhaps a better indicator of overall price trends for final output, but a more direct link
may exist between commodity prices and a PPI, because the PPI excludes services, which require few
commodities for their production.
322
that can arise when nonstationary variables are included in estimation, logged first differences are
typically taken of all price variables used in estimation. This approximates percentage changes, and
for most variables this transformation ensures that the series has a stable mean and variance over the
sample period, that is, the series is stationary.
To be sure, two nonstationary variables may still be used in estimation if they are cointegrated,
that is, if they tend to move together over time. However, tests performed by Boughton and Branson
(1988), Garner (1989), Durand and Blondal (1988), and Cody and Mills (1989) on levels of
commodity and aggregate prices do not support the hypothesis that these variables are cointegrated.20
This result suggests that there is no stable equilibrium between the levels of these variables, a finding
consistent with the observation that there are often permanent relative price shifts. Most researchers
conclude that it is appropriate to use rates of change for the commodity and aggregate price series in
regression analysis. One practical implication of this finding is that rules or inferences about
aggregate prices based on commodity prices may only be properly framed in terms of inflation rates,
not levels. Further clouding this picture, tests performed by Boughton and Branson and by Cody and
Mills indicate that an equilibrium relationship may only exist between changes in commodity price
levels and changes in aggregate inflation rates.21
According to standard statistical tests, many estimated relations show that commodity price
movements historically have preceded changes in overall inflation. This finding generally supports
claims that monitoring commodity price movements may be of some help to policy makers in
forecasting aggregate price trends. Again, the estimated equations point to a systematic relation
between commodity price inflation and subsequent total inflation. Much of the statistical support for
this conclusion is based on a series of Granger causality tests performed by Horrigan (1986), Webb
(1988), Garner (1989), Durand and Blondal (1988), and Cody and Mills (1989).22 In these tests,
inflation is specified as a function of its own past values and of lagged values of commodity price
inflation. The longest lag on the commodity price variables is typically restricted to no more than one
20
Many of the empirical studies cited in the bibliography contain a brief description of tests for
cointegration and their interpretation. See, for example, Durand and Blondal (1988) and Garner
(1989).
21
These researchers find that commodity prices are integrated of order 1 while consumer prices are
integrated of order 2.
22
Additional econometric support comes from estimates between inflation and commodity price
changes reported by Whitt (1988), Boughton and Branson (1988), and Klein and Horvath (1988).
323
year. Granger causality is determined by measuring the statistical significance of the estimated
parameters on the commodity price variable, which in these studies is usually found to support the
hypothesis that commodity price changes can be used to foretell changes in overall inflation.23
The tests described above actually only reveal whether commodity price changes generally
precede movements in inflation. The underlying structural relation between commodity and aggregate
prices may merely be one where both series respond independently to the same economic forces, but
where commodity prices respond more quickly. If the relation between commodity prices and total
prices arises from this kind of mutual correlation with other economic variables, then commodity
prices may provide little additional information if those other variables can be monitored directly. To
test for this possibility, Horrigan and Garner extend the standard bivariate Grangerframeworkto
include measures of real output, interest rates, money aggregates, and exchange rates. Commodity
prices remain statistically significant in these estimates, indicating that they retain some explanatory
power for overall inflation even after accounting for the information contained in these other variables.
Furlong (1989), Horrigan (1986), and Garner (1989) perform "variance decomposition" tests to
examine whether commodity prices are economically important, in addition to being statistically
significant, in explaining future inflation. Their results are somewhat mixed but suggest on balance
that commodity prices may play a relatively small role in actually causing subsequent inflation. Their
methodology attributes forecast errors (innovations) in a vector autoregression model (VAR) to various
explanatory variables, including commodity prices, aggregate prices, a money aggregate, and real
output.24 The results are then used to make inferences about the economic relevance or causality of
each explanatory variable. In this case, the proportion of the inflation forecast errors generated by the
VAR that can be attributed to "shocks" in each variable is believed to correspond to the relative
importance of that variable in determining inflation.25 Horrigan finds that innovations in commodity
prices account for only a small portion of the prediction errors of inflation. Furlong finds that the
23
Strictly speaking, the hypothesis that commodity price changes do not "Granger cause" aggregate
inflation can be reliably rejected. Durand and Blondal also find evidence of Granger causality running
from total price inflation to commodity price inflation.
^Furlong's measure of real output is the deviation of the unemployment rate from its full
employment level, while Horrigan and Garner use industrial output. Also, Furlong and Garner
transform price variables into percent changes while Horrigan uses their logged levels in estimation.
25
Researchers disagree about the conclusions or inferences one may properly draw from these tests.
In addition, the results of variance decomposition may be sensitive to the causal ordering that
researchers must impose on the variables.
324
importance of commodity prices in explaining inflation is sensitive to the choice of commodity price
index used in estimation, but in any event shocks to the money aggregate always account for a much
larger portion of inflation innovations. In contrast, Garner finds that innovations in commodity prices
may explain as much as 25 percent of the prediction error variance of inflation.
The interpretation that these findings point to a small economic role for commodity prices in
explaining aggregate inflation, if correct, need not undermine the predictive usefulness of commodity
prices. Instead, when combined with the findings from Granger causality tests, the results of variance
decomposition may shed some light on the channels by which commodity prices signal future
inflation. In general, these findings are consistent with the view that prices of commodities have only
a small direct impact on prices of final goods through the production process; nonetheless, commodity
prices are a leading indicator of inflation, perhaps because the long-run implications of underlying
economic disturbances are quickly distilled in commodity markets.
Several of the researchers cited above find that, despite its statistical significance in estimation,
the marginal contribution of commodity price information to inflation forecasts is modest. Furlong
(1989) and Webb (1988) extend their analyses to measure the impact of commodity prices on the
mean squared error of out-of-sample inflation predictions extending out from one to two years. After
they have accounted for the information contained in past inflation movements, money aggregates, real
economic activity, and various financial measures, the accuracy of their inflation forecasts is little
improved by the addition of commodity price data. A similar result is reported by Boughton and
Branson (1988) for inflation forecasts generated solely by past inflation movements. To be sure,
commodity price data may be available sooner than some other data used in these studies, but this is
not true for the monetary and financial indicators. Thus, the incremental improvement to inflation
forecasts gained by monitoring commodity price movements may be slight.
The usefulness of commodity prices for predicting aggregate inflation, as distinct from their
statistical significance, may also be examined by direct reference to a conventional inflation
forecasting equation. In recent work performed at this Bank, a standard expectations augmented
Phillips curve, designed to forecast CPI inflation, was estimated using quarterly data over a sample
period beginning in 1960 both with and without a variable measuring changes in commodity prices
(lagged one year).26 Year-ahead forecast rates of CPI inflation were compared to actual inflation for
^Quarterly CPI changes were specified as a function of the level of the unemployment rate and
recent changes in the unemployment rate-to capture aggregate demand pressures-along with past CPI
inflation as a proxy for inflation expectations. Quarterly commodity price changes were measured
325
the 1980s, and the results indicated that including commodity prices in the forecast equation, although
statistically significant, made only a small improvement in the forecasts. This same finding held when
the energy component of the CPI was excluded from the estimated equations.
Several researchers test for stability in their estimated relations between commodity price
changes and general inflation, and their results raise doubts about the reliability of using these
estimated relations to forecast inflation.27 Whitt (1988) estimates inflation as a function of past
growth in a monetary aggregate and of commodity price changes. When the sample period is
restricted to the 1960s and early 1970s, the monetary aggregate "performs" much better than
commodity prices, but in the subsequent period ending in 1987 commodity prices are a much superior
precursor of general inflation. This finding, of course, reflects at least in part the deteriorating
performance of the monetary aggregates during the later period. Furlong (1989), using his estimated
VAR framework over two subperiods, measures the response of inflation to specified commodity price
shocks. He finds that the response of inflation to innovations in growth rates of commodity prices is
much more pronounced when the model is estimated over a period from the mid-1970s to late 1980s
than when the model is estimated over an earlier period. These reported results possibly suggest that
the performance of commodity prices as a leading indicator of inflation may have improved over time.
Alternatively, the discrepancy in results over the two time periods may have arisen because policy
makers in the earlier period responded more effectively to economic disturbances affecting commodity
prices before these disturbances became embedded in the general price structure.
Finally, most recent econometric research focuses on the relation between commodity prices
and general inflation, but Furlong (1989) and Cody and Mills (1989) use a VARframeworkto
examine whether commodity prices are a leading indicator of real economic activity.28 Although the
body of evidence is much smaller, statistical results are generally the same as those for commodity
prices and inflation. In both studies, commodity price inflation is found to be a statistically significant
leading indicator of movements in economic activity. Furthermore, using variance decomposition
using the average of nine non-oil commodity price indexes.
27
In addition to the two studies cited, Durand and Blondal (1988) test for stability in their
estimated equations over a sample period split in 1973. They also find that these relationships are
generally unstable.
28
The difference between the actual unemployment rate and the "full employment" unemployment
rate is Furlong's measure of economic activity, while Cody and Mills use growth in industrial
production.
326
tests, Furlong finds that his commodity price index "explains" real economic activity better than
money aggregates do; however, past values of real activity outperform commodity prices.
B. Commodity Prices and Inflation Turning Points
The econometric techniques used in the studies cited above are designed to determine whether a
stable quantitative relation exists between the regression variables. A number of analysts argue that
even in the absence of such a relation, major turning points in commodity price inflation signal
important changes in underlying inflation trends. Empirical evidence reviewed in this section
demonstrates that sustained shifts in general inflation usually have been preceded by a significant
change in commodity price inflation. However, false signals of impending inflation changes do
periodically arise, the lead time of commodity price movements varies, and there is no consistency
between the size of these commodity price movements and subsequent changes in inflation. Thus, at
best a loose qualitative relation seems to exist between these variables.
Turning points in commodity and aggregate price inflation are correlated simply by visually
inspecting the data. Correlating rates of change of commodity and aggregate prices, rather than their
levels, seems appropriate in light of the results of the cointegration tests reported in the previous
section. Because of their volatility, commodity price movements must be heavily smoothed to aid in
the identification of peaks and troughs. The necessity of such procedures underscores that, in practice,
turning points in commodity price inflation might be recognized only with considerable hindsight. To
preserve the timing between movements in all price series, the measure of aggregate price inflation is
smoothed in an identical manner. The specific criteria used to smooth prices and to define turning
points are to some degree arbitrary. A major drawback to this kind of analysis is that it does not
allow for the impact of other variables on inflation.
Major changes in commodity price inflation have consistently preceded important turning points
in aggregate inflation over the past few decades. Based on work performed at the Federal Reserve
Board of Governors, Angell (1987) reports that from the late 1960s through 1987 virtually all of eight
turning points in CPI inflation were preceded by significant changes in commodity price behavior.29
29
Inflation for all price series is measured as the three-month average of twelve-month growth rates.
327
Percent
Chart 4
CPI Inflation and Identification of Peaks and Troughs
1953 1955 1957 1959 1961 1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987
Three-month average of twelve-month percent changes
His measure of commodity prices is the average of nine separate indexes, but the result is reportedly
insensitive to the choice of index. The basic findings reported by Angell are upheld when the period
is extended back to the 1950s, although the relationship, already loose, becomes less precise.
Consumer price inflation and its major peaks and troughs are plotted in Chart 4. These turning points
are superimposed over alternative measures of commodity price inflation in Charts 5 through 8. In
most cases, turning points in commodity price inflation precede changes in consumer price inflation.
In related work, Boughton and Branson (1988) report that since 1970, commodity price turning points
led six of nine turning points in an aggregated measure of OECD inflation.30
30
The authors used several measures of commodity prices; this is their best reported result.
328
Chart 5
Commodity Prices and Consumer Price Inflation Turning Points
Percent
Journal of Commerce Index of Non-oil Commodities
1953 1955 1957 1959 1961 1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987
Three-month average of twelve-month changes
Chart 6
Commodity Prices and Consumer Price Inflation Turning Points
Percent Commodity Research Bureau Spot Price Index of Non-oil Raw Industrials
1953 1955 1957 1959 1961 1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987
Three-month average of twelve-month changes
329
Chart 7
Commodity Prices and Consumer Price Inflation Turning Points
Percent International Monetary Fund Index of Non-oil Commodities
1953 1955 1957 1959 1961 1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987
Three-month average of twelve-month changes
Chart 8
Commodity Prices and Consumer Price Inflation Turning Points
Percent Producer Price Index of Crude Materials
50
40
30
20
10
-10
330
-20
1953 1955 1957 1959 1961 1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987
Three-month average of twelve-month changes
Commodity prices appear more likely to provide false signals of inflation turning points than to
miss actual turning points. For example, most of the commodity price indexes presented in the charts
incorrectly signaled first an inflation peak and then a trough during 1977, and during the early 1960s
several turning points in commodity price inflation were not followed by a change in overall inflation
trends. In their study, Boughton and Branson reported two such false signals. In part, this result may
arise from the volatility of commodity prices, which remains even after the series is smoothed. Thus,
despite a track record of correctly anticipating most major inflation turning points, movements in
commodity price inflation would still have to be interpreted cautiously, a condition which dilutes some
of their value as a leading indicator.
The lead time between commodity price and CPI inflation turning points varies considerably,
but on average appears sufficiently long to be of value to policy makers. The average lead time of the
commodity price index in the work reported by Angell is about seven months, much less than his
reported average estimated lag between changes in the monetary aggregates and changes in inflation.
The lead times of the series plotted in Charts 5 through 8 are on balance somewhat longer than those
reported by Angell, averaging about ten months, and they vary considerably from episode to episode
as well as across commodity price measures.31 On balance, these results suggest that commodity
prices could in principle provide an opportunity for policy makers to offset inflationary shocks;
however, the instability in these lead times, as well as the propensity to send false signals, indicates
that forecasts of major inflation turning points must remain highly uncertain.
Finally, there is no clear association between the size of movements in commodity prices and
the size of changes in total CPI inflation. In the work reported by Angell, commodity price inflation
changes are on average four times as large as CPI inflation changes, but the variation in this ratio is
large. This same finding holds for the commodity price indexes presented in the charts. The
instability of the ratio between changes in commodity prices and CPI inflation is of more consequence
than its absolute magnitude. Thus, an acceleration in commodity price inflation of a given magnitude
may provide little guidance about the degree of monetary tightening that is appropriate.
All the empirical work surveyed in this section and in the preceding section may be subject to
the "Lucas critique," which holds that the observed relation between, say, commodity prices and
inflation may be sensitive to, among other things, the monetary policy procedures followed during the
31
Lead times found by Branson and Boughton tend to be significantly shorter than those reported by
Angell or presented here.
331
sample period. Consider a simple example: as market participants came to recognize that the
monetary authorities were adjusting policy in order to maintain commodity prices within certain
bounds, they might themselves bid commodity prices to within the perceived policy range ahead of
any actual policy adjustments. Taken to an extreme, this kind of market reaction could alter the
observed relation between commodity prices and general inflation. Furlong (1989) speculates that, to
some degree, the differences in his estimates between commodity prices and inflation across different
sample periods may reflect this kind of phenomenon.32
IV. Controlling Commodity Prices
To successfully target commodity prices, policy makers must be able to keep these prices
within certain specified bounds. If policy makers were instead to monitor commodity prices solely for
their information content, they might still wish to prevent commodity prices from moving too far
outside some broad ranges. This section describes the links between the operating instruments of the
monetary authorities and commodity prices and examines the empirical evidence on the controllability
of commodity prices.
A. Channels of Control
Interest rates represent the principal channel through which monetary policy can exert some
control over commodity prices. Industrial production, a direct determinant of the demand for
commodities as an input in manufacturing, is sensitive to interest rate swings. As noted above, interest
rate changes may also affect commodity prices by altering the cost of holding commodity inventories.
Exchange rates, which affect prices of many commodities traded internationally, are also sensitive to
interest rate changes.
Direct intervention in commodity markets is sometimes recommended as a way to control
commodity prices.33 This would represent a significant departure from the current means of
monetary control, and a thorough consideration of this possibility is not undertaken here. An
important drawback to such proposals is that intervention affects commodity prices by changing the
32
This issue is also discussed in Richard Davis' introduction to this volume.
33
Miles (1984) calls for intervention in commodity futures markets to control their price. It can be
argued that restricting intervention to futures instead of spot markets makes it unnecessary to maintain
commodity stockpiles.
332
available supply, and thus can distort the relative prices of commodities.
While the means for controlling commodity prices appear to be available, there is considerable
skepticism about the precision with which policy makers could control these prices. Long lags exist
between changes in monetary policy and changes in manufacturing activity, and the degree of control
policy makers have over industrial production or exchange rates—important channels through which
policy affects commodity prices-is uncertain.34 The problems of measuring the impact of monetary
policy actions on commodity prices are compounded because economic activity, interest rates, and
other direct determinants of commodity prices are influenced by a number of factors outside the
control of the monetary authorities. For example, exchange rates are influenced by the stance of
monetary policy abroad as well as by domestic actions. The next section surveys some of the
available evidence on the ability of the monetary authorities to control commodity prices.
B. Empirical Evidence of Controllability
A few researchers have attempted to examine the ability of policy makers to control commodity
prices through conventional means. Overall, their findings point to a weak link of questionable
stability between the instruments of monetary policy and commodity prices.
Hafer (1983) finds that the simple correlation between past (lagged one quarter) Ml growth and
an index of raw materials commodity price inflation is statistically significant over the period 1960-82.
However, the correlation coefficient is not large in absolute magnitude and varies considerably when
calculated over different subperiods. A similarly weak result holds when a moving average of Ml
growth over the preceding three years is correlated with commodity price inflation. DeFina (1988)
extends Hafer's analysis to consider instruments under the more direct control of monetary authorities:
the federal funds rate and growth in nonborrowed reserves.35 Correlations between these variables
and commodity price inflation are not statistically significant, even when longer lag lengths on the
monetary variables are introduced. These results are supported by simple regression analysis
performed on commodity price changes and the monetary variables.
Garner (1989) presents the results of a number of Granger causality tests linking commodity
34
To the extent that interest rates affect future expected industrial production, they could have a
more immediate impact on current commodity prices.
35
In addition, both DeFina and Garner (1989) examine the relation between monetary variables and
gold prices and find little evidence of a direct link.
333
price changes to movements in a variety of monetary instruments and other variables. He uses four
alternative measures of commodity price inflation, and his control instruments include Ml, the
monetary base, nonborrowed reserves, and the federal funds rate. Overall, little evidence is found of a
statistically significant relation between these variables. Finally, Garner reports the results of some
variance decompositions of prediction errors of commodity price inflation derived from his estimated
VAR framework (outlined in Section IDA). These show that the innovations in the monetary variables
never explain a large percentage of the prediction error variance of commodity price inflation.
V. Conclusions
Renewed interest has been expressed in the role that commodity prices could play in setting
and executing monetary policy. Some analysts have recommended using commodity prices as an
intermediate target, but most advocates of an enhanced role for commodity prices envision using them
as a key leading indicator or information variable, especially of inflation trends. This study has
examined the major behavioral features of commodity prices and surveyed recent empirical
examinations of their predictive content. Several important points emerge from the body of evidence
reviewed.
Setting formal intermediate targets for commodity prices seems unrealistic for several reasons.
Significant shifts have occurred in the relative price of commodities to final goods, commodity prices
are highly volatile, and policy makers have only a limited ability to control commodity prices.
These drawbacks would pose much less of a problem to a monetary policy regime that uses
commodity prices as an information variable. However, the contribution that commodity prices can
make to policy in this role hinges on how valuable and reliable the information contained in
commodity prices is and how efficiently this information can be extracted from actual commodity
price movements. Only qualified conclusions can be reached on the basis of the evidence surveyed,
but skepticism about a dominant indicator role for commodity prices in monetary policy is warranted.
A considerable body of evidence points to some real predictive content in commodity prices.
Standard empirical tests typically show that commodity price indexes have a statistically significant
property for explaining overall inflation. This result holds even when the information content of, say,
exchange rates and financial variables is accounted for, suggesting that commodity prices might be
worth monitoring even in conjunction with other economic variables. Impressionistic comparisons of
inflation turning points seem to confirm that commodity prices have some ability to anticipate overall
price developments, although false signals have occurred on occasion.
334
Much of the same evidence, however, suggests that the relation between commodity prices and
the general price structure is too loose to justify using commodity prices as a primary basis for making
policy decisions. Cointegration tests, tests for stability in estimated relations, and the visual
examination of relative price movements all suggest that the link between commodity prices and
aggregate prices may not be predictable enough to be exploited by policy makers. At most, only rates
of change in commodity prices and in broader inflation measures may be correlated through time. As
the simple turning point analysis demonstrates, this uncertainty extends to both the length of the lag
and the quantitative correspondence between commodity price movements and subsequent inflation
changes. In addition, some researchers find that the marginal contribution of commodity prices to
inflation forecasts, although statistically significant, is small. Finally, while most empirical research
assumes that important changes in commodity prices are quickly recognized for what they are, policy
makers may only become sure about the duration of significant commodity price shifts with a
considerable lag because of their volatility. For these reasons, developing reliable operating
procedures based largely on the behavior of commodity prices could be very difficult.
References
Angell, Wayne D. 1987. "A Commodity Price Guide to Monetary Aggregate Targeting." Speech
before the Lehrman Institute.
. 1989. "Commodity Prices and Monetary Policy: Empirical and Theoretical
Considerations." Speech before the Sixteenth Annual Meeting of the Virginia Association of
Economists.
"Baker Suggests Role for Gold in Setting World Economic Policy." 1987. Wall Street Journal
October 1.
Bordo, Michael D., and Anna J. Schwartz. 1984. A Retrospective on the Classical Gold Standard,
1821-1831. Chicago: The University of Chicago Press.
Bosworth, Barry P., and Robert Z. Lawrence. 1982 Commodity Prices and the New Inflation.
Brookings Institution.
Boughton, James M., and William H. Branson. 1988. "Commodity Prices as a Leading Indicator of
Inflation." National Bureau of Economic Research, Working Paper no. 2750.
335
Boughton, James M., William H. Branson, and Alphecca Muttardy. 1989. "Commodity Prices and
Inflation: Evidence from Seven Large Industrial Countries." National Bureau of Economic Research,
Working Paper no. 3158.
Brunner, Karl. 1984. "From the 'Upper Tail Theory of Inflation1 to the 'Lower Tail Theory of
Deflation/" Shadow Open Market Committee, Policy Statement and Position Papers,
September 30-October 1.
Chu, Ke-Young, and Thomas K. Morrison. 1984. "The 1981-82 Recession and Non-Oil Primary
Commodity Prices." IMF Staff Papers. March.
. 1986. "World Non-Oil Primary Commodity Markets." IMF Staff Papers. March.
Cody, Brian J., and Leonard O. Mills. 1989. "Evaluating Commodity Prices as a Gauge for Monetary
Policy." Federal Reserve Bank of Philadelphia, Working Paper no. 89-5.
DeFina, Robert H. 1988. "Commodity Prices: Useful Intermediate Targets for Monetary Policy?"
Federal Reserve Bank of Philadelphia Business Review, May-June.
Durand, Martine, and Sveinbjorn Blondal. 1988. "Are Commodity Prices Leading Indicators of OECD
Prices?" Organization for Economic Cooporation and Development, Department of Economics and
Statistics, Working Paper no. 49.
Englander, A. Steven. 1985. "Commodity Prices in the Current Recovery." Federal Reserve Bank of
New York Quarterly Review, vol. 10, no. 1 (Spring).
Frankel, Jeffrey A., and Gikas A. Hardouvelis. 1985. "Commodity Prices, Surprises and Fed
Credibility." Journal of Money, Credit, and Banking, vol. 17, no. 4.
Frankel, Jeffrey A. 1986. "Expectations and Commodity Price Dynamics: The Overshooting Model."
American Journal of Agricultural Economics, May.
FuUerton, Thomas M., Jr., Richard A. Hirth, and Mark B. Smith. 1989. "Inflationary Dynamics and the
Angell-Johnson Proposals." University of Pennsylvania Working Paper.
Furlong, Frederick T. 1989. "Commodity Prices as a Guide for Monetary Policy." Federal Reserve
Bank of San Francisco Economic Review, Winter.
. 1989. "Commodity Prices and Inflation." Federal Reserve Bank of San Francisco
Weekly Letter, June 16.
Garner, C. Alan. 1985. "Commodity Prices and Monetary Policy Reform." Federal Reserve Bank of
Kansas City Economic Review, February.
. 1989. "Commodity Prices: Policy Target or Information Variable?" Journal of Money,
Credit, and Banking, November.
Genetski, Robert J. 1982. "The Benefits of a Price Rule." Wall Street Journal, December 10.
336
Greenspan, Alan. 1987. Statement before the House Committee on Banking, Finance, and Urban
Affairs, December 18.
Hafer, R. W. 1983. "Monetary Policy and the Price Rule: The Newest Odd Couple." Federal Reserve
Bank of St. Louis Economic Review.
Hall, Robert E. 1982. "Explorations in the Gold Standard and Related Policies for Stabilizing the
Dollar." In R. Hall, ed., Inflation: Causes and Effects. Chicago: University of Chicago.
Hardouvelis, Gikas A., and Scott W. Barnhart. 1988. "The Evolution of Federal Reserve Credibility:
1978-1984." Review of Economics and Statistics, forthcoming.
Horrigan, Brian R. 1986. "Monetary Indicators, Commodity Prices, and Inflation." Federal Reserve
Bank of Philadelphia, Working Paper no. 86-7.
Johnson, Manuel H. 1988. "Current Perspectives on Monetary Policy." Speech before Cato Institute.
Klein, L.R., and Matild Horvath. 1988. "Early Warning Signals for Inflation and Exchange Rate
Fluctuations." University of Pennsylvania Working Paper.
Klein, Philip A. 1986. "Leading Indicators of Inflation in Market Economies." International Journal of
Forecasting, vol 2.
Lown, Cara S. 1989. "Interest Rate Spreads, Commodity Prices, and the Dollar: A New Strategy for
Monetary Policy?" Federal Reserve Bank of Dallas Economic Review, July.
McCallum, Bennett T. 1989. "Targets, Indicators, and Instruments of Monetary Policy." National
Bureau of Economic Research, Working Paper no. 3047.
Miles, Marc A. 1984. Beyond Monetarism, Finding the Road to Stable Money. Basic Books.
Reynolds, Alan. 1982. "The Trouble with Monetarism." Policy Affairs, vol. 21 (Summer).
Rosine, John. 1987. "Aggregative Measures of Price and Quantity Change in Commodity
Markets." Board of Governors of the Federal Reserve System, Working Paper no. 81.
"Storm Warning on Inflation's Barometer?" 1988. Economist, April 2.
United States Gold Commission. 1982. Report to the Congress of the Commission on the Role of Gold
in the Domestic and International Monetary System. Washington, D.C.: Government Printing Office.
von zur Muehlen, Peter. 1990. "Predicting Inflation with Commodity Prices." Board of Governors of
the Federal Reserve System, Finance and Economics Discussion Series, no. 118.
Webb, Roy H. 1988. "Commodity Prices as Predictors of Aggregate Price Change." Federal Reserve
Bank of Richmond Economic Review, vol. 74, no. 6.
337
Whitt, Joseph A., Jr. 1988. "Commodity Prices and Monetary Policy." Federal Reserve Bank of
Atlanta, Working Paper no. 88-8.
Working, Holbrook. 1977. Selected Writings of Holbrook Working. Chicago Board of Trade.
338
POSSIBLE ROLES OF THE YIELD CURVE IN MONETARY POLICY
Arturo Estrella and Gikas Hardouvelis
In many macroeconomic models, and sometimes in discussions of monetary policy, economists
refer to the interest rate, as if there were only one fixed income asset in the economy. This
simplification is often adequate. However, when the intention is to investigate in detail the
mechanisms through which monetary policy operates, it is necessary to examine a wider array of fixed
income securities with various terms to maturity.
In the economic literature, most studies of the term structure of interest rates are designed to
test whether the "expectations hypothesis" holds. That is, they ask whether long-term rates are
weighted averages of expected short-term rates, possibly with an adjustment for various types of risks.
By contrast, how monetary policy affects or might make use of the term structure is a question that is
rarely addressed directly.
The yield curve has long been known to have important connections with the monetary policy
process, although explicit formulations of these connections, either in a policy or a research context,
are extremely rare. Recently, however, a member of the Federal Reserve Board of Governors affirmed
that the yield curve is one of three major indicators that may be useful in formulating monetary
policy.1
The purpose of this paper is to examine how and where the yield curve fits in the framework
of monetary policy. We ask: Can the yield curve serve as a policy target or indicator? Can the yield
curve be used in a routine or even mechanical way in the conduct of policy?
The yield curve could conceivably play various roles—as a final target, an intermediate target,
or an economic indicator. A final target of monetary policy, the ultimate object of influence for the
monetary authority, is a variable that describes aggregate economic performance. Natural candidates
are real GNP, inflation, or perhaps a combination of the two. The slope of the yield curve cannot be a
final target, however, because it has no direct link to aggregate economic welfare. Nevertheless, the
!
Governor Manuel Johnson (1988) has stated that the Federal Reserve Board looks at the yield curve,
as well as commodity prices and foreign exchange rates, in setting monetary policy.
339
yield curve could potentially serve as an intermediate target. An intermediate target is a variable that
is related to the final target in a stable, predictable, and causal way, and that can be readily controlled
by policy makers.
For the yield curve to serve as a macroeconomic indicator, and hence to provide monetary
authorities with information that will guide their decisions, the requirements are less stringent. It
suffices that the yield curve can predict future macroeconomic developments without necessarily
causing them.2
This paper is organized into two major sections. Section I examines the relationship between a
direct instrument of monetary policy, the federal funds rate, and the slope of the yield curve. The
empirical evidence suggests that the federal funds rate does have a partially predictable negative effect
on the slope of the yield curve. However, the effect varies over time, suggesting that the Federal
Reserve can influence-but not control-the slope of the yield curve.
Section II focuses on the relationship between the slope of the yield curve and two important
final policy targets: real GNP growth and inflation. The evidence shows that the slope of the yield
curve is positively associated with future changes in output and inflation. Furthermore, this positive
association is not attributable to the negative effect that restrictive monetary policy can have on both
the contemporaneous slope of the yield curve and on future changes in real output and inflation. A
safe interpretation of the evidence is that the yield curve is a useful addition to the tool kit of policy
makers and others as they try to assess future exogenous macroeconomic developments.
I. The Effect of Monetary Policy on the Yield Curve
Economic theory does not provide a precise and unambiguous link between monetary policy
and the slope of the yield curve. The initial effect of a monetary tightening is to raise real short-term
interest rates. The effect on long-term rates, however, is more complex, since these rates depend more
heavily on expectations about future real output and inflation. A tighter policy implies a period of
contraction in real output growth, so that the original increase in short-term rates would be followed
by a decrease in future real rates. A tighter policy also implies that the inflation premium component
^ e term "indicator" has often been used in the monetary policy literature to refer to variables
that provide information on the stance of policy. The analysis of Section I, which examines the
relationship between the funds rate and the yield curve slope, is a test of the indicator properties of the
slope in this sense. A "macroeconomic indicator," on the other hand, provides information on the state
of the macroeconomy, a state which is determined partly, but not exclusively, by monetary policy.
340
of long-term rates would decline. The final effect on long-term rates depends also on whether the
monetary tightening is perceived to be permanent or transitory and on whether it comes as a surprise
to market participants. Overall, theory suggests that the effect on long-term rates and hence on the
slope of the yield curve is ambiguous. How monetary policy relates to the slope of the yield curve
thus remains an empirical issue.
This section presents an empirical investigation of the effects of monetary policy on the slope
of the yield curve. It examines the relationship between unanticipated changes in the federal funds
rate and unanticipated changes in the slope of the Treasury yield curve. Although the funds rate is not
a perfect index of monetary policy, unanticipated changes in policy are reflected clearly and promptly
in the funds rate.
A. Econometric Framework
We measure the slope of the yield curve as the difference between the ten-year Treasury bond
rate and the three-month Treasury bill rate. These rates, as well as the funds rate, are monthly
averages of daily rates from January 1955 to May 1989.3
In order to construct the unanticipated component of each interest rate, we estimate a vector
autoregressive (VAR) model. The model is a system of three ordinary least squares regressions with
the funds, bill, and bond rates, respectively, as dependent variables, and the first six lags of each of
these three variables as regressors (with a constant term). We model the behavior of the bond and bill
rates separately because a system of three equations extracts more precise information from the data
than a system of two equations with only the funds rate and the slope. Moreover, it is straightforward
to calculate the results for the spread once we know the results for the bond rate and bill rate. The
basic vector autoregressive system is of the form:
FUNDSt = oc10 + Z6i ocH FUNDS' + 1\ pH BILL^ + Z\ yu BONDM + elt
BDLLt = 0C20 + I 6 ! 0^ FUNDS^ + I 6 , P2i BILL^ + I 6 , y2i BONDM + e*
BONDt = (Xjo + I 6 ! 0C3J FUNDSM + X6, (33i BILL^ + I 6 , y3i BONDM + e3t,
3
The difference between the ten-year and three-month Treasury rates is proportional to the
difference between the nine-year nine-month forward rate and the three-month rate. See Shiller,
Campbell, and Schoenholtz (1983) or Estrella and Hardouvelis [1989] 1991).
341
Table 1
Funds, Bill, and Bond Rates:
Significance Level of Granger Causality Tests*
1955 to 1989
Dependent
Variable
Funds
Progressors
Bill
Funds
.000
.000
.011
Bill
.000
.000
.000
Bond
.014
.009
.000
Bond
* Significance level of F test of the hypothesis that all six lags of a given regressor have zero coefficients.
where FUNDS is the federal funds rate, BILL is the three-month Treasury bill rate, and BOND is the
ten-year Treasury bond rate.
Table 1 presents exclusion tests for all lags of each of the explanatory variables in the above
regressions, known as Granger causality tests.4 Specifically, it shows the probability that a given
variable has no explanatory power for a particular dependent variable. As indicated by the low
probabilities in the table, there is evidence of causality flowing in every direction in the model. Every
test statistic is highly significant, and all but two are significant at the 1 percent level. These results
confirm the importance of allowing the three interest rate variables to enter independently.
Our focus in the remainder of this section is on the contemporaneous relations among the
unexpected components e^ £2, and e3 of the previous model. Econometric methods cannot determine
the contemporaneous cause and effect relationships among these variables. It seems natural to
suppose, however, that contemporaneous causation runs from the funds rate to the bill rate to the bond
rate. Under this assumption, the previous system of equations can be rewritten as follows:
4
For a discussion of Granger causality, see Granger and Newbold (1977) or Sims (1972).
342
(1)
BILLt = 81FUNDSt + a \ 0 + XjOt^FUNDS^ + Xip'^BILL^ + I ^ O N D ^ + %
(2) BONDt = 82FUNDSt + a'^ + XiCC^FUNDS^ + Xip^BELL^ + Xtf^BOND^ + r\2t
(3) SPREADt = 83FUNDSt + a'30 + Iia,3iFUNDSt.i + Zfi'JBILL^ + XiY^BOND^ + i\3v
In other words, in this new set of equations, current values of the bill rate, bond rate, and spread,
respectively, are made functions of that part of their value that is anticipated on the basis of lagged
movements in the three interest rates (that is, the content of the first set of equations) and, in addition,
the influence of the current movement in the funds rate. The new formulation highlights the
contemporaneous effect of the funds rate on the bill and bond rates and on the spread (given by the
coefficients 8t). The estimated coefficient of the funds rate in these equations reflects the effect of an
unanticipated change in the funds rate on the unanticipated change in the bill rate, the bond rate, and
the spread. The reason is that the remaining regressors control for anticipated movements in these
rates.5
We begin the empirical analysis by running simple constant-coefficient regressions.
Subsequently, we allow the regression coefficients to vary in progressively more complex ways.
These procedures are consistent with the presence of specific or unspecified factors (reflecting, for
example, the state of the economy, policy credibility, and operating procedures) that affect the
sensitivity of the spread to changes in the funds rate.
B. The Basic Evidence
Table 2 examines the sensitivity of the bill and bond rates and of the spread to changes in the
federal funds rate under the assumption that the relationship is stable and constant over the period
from 1955 to 1989. The table reports the principal results of estimating the reduced form equations
(l)-(3). Observe that both Treasury rates move in the same direction as the funds rate. Columns 1
and 2 show that the bill rate is much more responsive than the bond rate to movements in the funds
rate. This is evidenced both by the larger regression coefficient and by the higher marginal R2 of the
bill rate. The higher sensitivity of the bill rate implies a flattening of the yield curve in response to a
rise in the funds rate.
The third column of Table 2 shows that during the sample period from 1955 to 1989, an
5
If the coefficients a, p, and y are assumed to be constant, the 8 coefficients in the reduced form
equations can also be obtained from regressions involving only the residuals e from the basic vector
autoregression. See Granger and Newbold (1977, Section 7.3).
343
Tabic 2
Bill Rates, Bond Rates, and Spread:
Contemporaneous Effects of Funds Rate
from Vector Autoregression
1955 to 1989
Bill
Contemporaneous
funds rate
Marginal R2
SEE
Dependent Variable
Bond
Spread
.619
(21.17)
.189
(8.24)
-.431
(-17.03)
.537
.149
.428
.989
.993
.948
.319
.250
.276
Notes: t-statistics in parentheses. Marginal R2 is the proportion of the variance of the dependent variable that is
not explained by lags of the funds, bill, and bond rates but is explained by the contemporaneous funds rate. It is
the R2 analogue of the t-statistic for the funds rate.
unanticipated rise of 1 percentage point in the funds rate caused an average drop in the spread of about
43 basis points.6 The usefulness of this average sensitivity is limited, however, if the sensitivity does
not remain constant over time. We now proceed to relax the assumption of constancy in various ways.
C. Tests of Parameter Stability
To test the hypothesis of parameter stability, we partition the sample into two periods with a
break point in October 1979. This is a natural break point because the October 1979 change in
6
As we noted earlier, theory provides little guidance on the magnitude of these effects. In a
recent paper, Turnovsky (1989) presents a simple standard macroeconomic model to which he adds a
"term structure" equation relating short- and long-term interest rates. He then investigates the effects
of monetary and fiscal policy on the variables of the model. The Turnovsky model does not
correspond exactly to the empirical model in the text, but like the latter it implies that the long rate
reacts less to monetary policy than does the short rate. The ratio of the effect on the long rate to that
on the short rate, however, seems to be lower in the Turnovsky model than in the empirical estimates
reported in Table 2.
344
Table 3
Effect of Funds Rate on
Bill and Bond Rates and Spread
1955 to 1989
Sample Split: October 1979
Bill
Dependent Variable
Bond
Spread
.619
(21.17)
.189
(8.24)
-.431
(-17.03)
Pre-October 1979
.409
(8.91)
.149
(5.47)
-.260
(-6.51)
Post-October 1979
.761
(14.03)
.242
(4.81)
-.519
(-11.19)
Full Period
Significance level
of F-statistic
for change
.0001
.00004
.0001
Note: t-statistics in parentheses.
operating procedure for monetary policy produced significant changes in many empirical relationships
in the securities markets. Table 3 shows that the response of Treasury rates to changes in the federal
funds rate is no exception. After October 1979, the bill and bond rates, as well as the spread, became
more sensitive to unanticipated movements in the funds rate. Although all three coefficients were
significant in the earlier period, their absolute values showed a statistically significant increase after
October 1979. Observe that an unanticipated 1 percentage point rise in the funds rate causes a decline
in the spread of 26 basis points before October 1979 and 52 basis points after October 1979. The
instability of the response suggests that both explanatory and predictive accuracy may be improved by
allowing the regression coefficients to change over time.
The selection of October 1979 as a break point is, to a certain extent, arbitrary. Although it is
an important date for monetary policy, it is by no means the only significant event in recent years.
For example, after the summer of 1982, monetary policy operating procedures were modified again.
345
Developments of this sort suggest that, in addition to October 1979, there may be other important
breaks in the sample.7 To uncover those possible breaks, we continue the empirical analysis by
allowing the regression coefficients to vary smoothly over time.
D. Time-varying Regression Coefficients
A relatively simple way of allowing for time variation in the coefficients of a regression is to
reestimate the relationship periodically. In this section we consider the results of estimating equations
(l)-(3) using the technique of rolling regressions. Specifically, we estimate each equation over fiveyear intervals, advancing the estimation period one month at a time. Thus, a measure of the sensitivity
of the Treasury rates to the funds rate is obtained for each five-year estimation period ending between
July 1960 and May 1989. An important feature of this method is its "limited memory." Events that
occurred before the beginning of a given five-year period have no statistical influence on the estimated
coefficients for that period.8
Chart 1 plots the estimated time-varying sensitivity of the slope of the yield curve to
unanticipated changes in the funds rate, together with a time-varying 95 percent confidence band for
each point estimate. Observe that an unanticipated increase in the funds rate tends to lower the slope
of the yield curve, but on certain occasions, notably in 1970, the relationship is reversed. However,
none of the sporadic positive estimates is statistically significant, whereas many of the negative
estimates are highly significant. The simple average of the coefficient estimates obtained from the
rolling regressions is -25 basis points, compared with the estimate of -43 basis points when the
coefficient is assumed to be constant.
Chart 2 plots the actual and predicted values of the spread. Clearly, the method of rolling
regressions produces reasonably good results in predicting the slope of the yield curve. During the
sample period from July 1960 to May 1989, the prediction root mean squared error is only 47 basis
7
Another possible break point is 1966. Until 1965, the federal funds rate did not exceed the discount rate
and was relatively unresponsive to money market conditions. The Federal Reserve watched the repurchase
agreement rate, the three-month Treasury bill rate, and dealer lending rates as gauges of reserve availability
and money market conditions. See Meulendyke (1990).
8
We also experimented with a model in which the sensitivity of the slope with respect to the funds rate is
assumed to follow a random walk of the form:
8t = 8^ + vt,
where 8t is the sensitivity in month t and vt is white noise. The results were fairly similar to those obtained
from the rolling regressions.
346
Chart 1
Effect of Funds Rate on Yield Curve Slope
Percentage points
0.4
*S \\\^
-0.2
-0.4 4
'%
«
-0.8
^
Rolling
regressions
I
I
U
I
I
I
l_
% significance level
VVM
1960 1962 1964 1966 1968 1970
I
kXXfc
I
1972
I
1974
1976
I
L
1978 1980 1982 1984 1986 1988
Chart 2
Predicted Yield Curve Slope Using Rolling Regressions
Percentage points
347
1960 1962 1964 1966 1968 1970
1972
1974
1976
1978 1980 1982 1984 1986 1988
Chart 3
Predicted Yield Curve Slope Using Cumulative Regressions
Percentage points
1960 1962 1964 1966 1968 1970
1972
1974
1976
1978 1980 1982 1984 1986 1988
points, whereas the standard deviation of the spread is 126 basis points.9
Although the rolling regressions method produces good forecasting results, it discards the data
from earlier time periods. If there is persistence in the structure of the model, this method wastes
potentially valuable information. In such a case, a superior method is a cumulative regression, which
adds one observation at a time but does not dispose of older observations.
Chart 3 presents the predictions of the cumulative regressions. These predictions are
qualitatively similar to the predictions of the rolling regressions. However, the prediction root mean
squared error is now only 34 basis points, compared with 47 basis points when the method of rolling
regressions is used. The slightly superior performance of cumulative regressions suggests that the
^ e scale of Figures 2 and 3 may make the estimates appear more accurate than the root mean squared
error would indicate. Note, however, that there are some predictive difficulties at turning points.
348
importance of structural shifts in the VAR model, while statistically significant, should not be
exaggerated.
E. Coefficients as Functions of the Macroeconomy
We now explore the possibility that the time-varying sensitivity of the yield curve to monetary
policy may be systematically related to the state of the macroeconomy. Many variables influence the
effect of monetary policy on interest rates, but of these variables, inflation is the most important.
When the monetary authority assumes a more restrictive policy stance, it is frequently out of
concern that the economy is overheating and that inflation rates may exceed acceptable levels. In
these circumstances, the policy action would have a noticeable effect on future inflation and hence on
inflation expectations. Thus, when an unanticipated increase in the funds rate occurs during a period
of relatively high inflation, the tightening would probably be perceived as intentional and sustained
and would therefore have a stronger negative effect on longer term inflationary expectations. Hence,
the negative reaction of the yield curve to unanticipated increases in the funds rate would probably be
more pronounced at times when inflation is high.
To test the importance of the level of inflation in the relationship between the federal funds
rate and the slope of the yield curve, we may pose the following question: Is the effect of an
unanticipated rise in the funds rate on the unanticipated change in the yield curve slope greater in
magnitude when inflation is high? To measure the unanticipated component of the interest rates, we
estimate an enhanced VAR of the sort described earlier using the funds, bill, and bond rates and the
rate of inflation. Six lags of each variable are included as regressors. The difference between this
VAR and the one reported earlier is the introduction of the inflation rate. Inflation is included mainly
for consistency, since the contemporaneous rate will be introduced in a second stage of the estimation,
but this addition does not affect the results in any appreciable way.
The main regression equation then has the following form:
SPREADut = p0 + (p, + p2INFLATIONt) * FUNDSut,
where the superscript u identifies the unanticipated component of the variable, which is the residual
from the appropriate VAR equation. Observe that the coefficient of the unanticipated component of
the funds rate is a function of the inflation rate. The parameter estimates for the sample period from
August 1955 to May 1989 are as follows:
p! = -.303 (-6.17)
P2 = -.181 (-2.81),
349
where t-statistics appear in parentheses. The negative and significant p2 implies that the inflation rate
has an important effect on the sensitivity of the slope to the funds rate. For every additional
percentage point of inflation, the decline of the spread in response to a percentage point rise in the
funds rate increases by 18 basis points. Thus, the evidence is consistent with the hypothesis that
higher inflation makes contractionary policy more "credible."
Furthermore, there is evidence that the inflation rate explains a large portion of the overall
variability in the sensitivity of the spread to the fiinds rate. When the coefficient of the fiinds rate is
allowed to vary with inflation, there are no further strong signs of instability in the model. For
example, a structural break in October 1979 is rejected at the 5 percent level of significance.
Chart 4 plots the predicted sensitivity of the slope of the yield curve from the previous
regression with the time-varying sensitivity from the earlier rolling regressions. It is clear from the
chart that the contemporaneous effect of inflation identified in this section does not fully explain the
statistical variability of the sensitivity of the yield curve slope to the fiinds rate. Nevertheless, the
significance of the (32 coefficient shows that inflation does explain a large portion of that variability.
Chart 4
Effect of Funds Rate on Yield Curve Slope
Percentage points
I960
350
1962 1964
1966 1968 1970
1972
1974
1976
1978 1980 1982 1984 1986 1988
F. Summary
The results of this part of the paper suggest that monetary policy has significant and
predictable effects on the slope of the yield curve. The magnitude of these effects has varied over
time, however, and has been highest in periods of high inflation. Over the entire sample period from
19SS to 1989, an unanticipated increase in the federal funds rate of 1 percentage point decreases the
spread between the ten-year Treasury bond rate and the three-month Treasury bill rate by about 40
basis points. Currently, the effect is weaker, however, and is in the neighborhood of 15 to 20 basis
points.
II. The Yield Curve, Future Output, and Future Inflation
The yield curve could be an intermediate target of monetary policy if it were to exhibit a stable
and causal relationship with the ultimate targets of policy, namely, real output and inflation. The more
causal and direct the relationship, the better the variable as an intermediate target. In the case of the
yield curve, it is difficult to conceive of a strong causal relationship with the final targets, but there is
evidence of a predictive relationship between these variables.
Estrella and Hardouvelis ([1989], 1991) and Laurent (1988, 1989) present evidence that the
yield curve predicts future real output. Mishkin (1987) and Fama and Bliss (1988) show that the yield
curve predicts future inflation. Overall, the yield curve is a predictor of real activity in the short run
(within two years) and of inflation in the long run (beyond two years). The remainder of this section
summarizes and evaluates the evidence on the predictive power of the slope of the yield curve.
A. The Yield Curve as a Predictor of Real Activity
Economic theory—the expectations hypothesis—provides a clear connection between present
interest rates of different maturities and future expected spot rates. It does not, however, establish a
precise connection between current spot interest rates and the future level of real economic activity.
Theoretically, for example, an inverted yield curve may predict either a decrease or an increase
in real GNP. The spot short-term rate may be expected to fall in the future because of a decline in
real economic activity (an inward shift of the textbook IS curve). Alternatively, the interest rate may
be expected to fall because an expansionary monetary policy is anticipated (an expected future outward
shift in the textbook LM curve, so long as the policy is not perceived to be inflationary). In the latter
case, the implicit prediction in the term structure is that output would expand. Another possibility is
that an inverted yield curve may reflect an expected decrease in the rate of inflation attributable to an
351
expected expansion in aggregate supply. Here too an increase in output is anticipated.
Because theory offers no clear resolution of this issue, the contention that the slope of the yield
curve predicts future changes in real economic activity in a specific direction can only be judged on
empirical grounds. We find that the empirical relationship between the slope of the yield curve and
future GNP growth is positive. Chart 5 plots the growth rate of GNP from four quarters earlier and
the spread between the ten-year Treasury bond and the three-month Treasury bill four quarters earlier
(that is, at the beginning of the period over which growth is measured). The slope of the yield curve
tracks the future realization of output growth impressively well, especially in the 1970s and 1980s.
To present more precise empirical evidence, we run regressions of the form:
Y
w+k = a + p SPREAD, + e,,
where y, = real GNP in quarter t,
Yw+k = (400/k)[log(yt+k/yt)] = cumulative real GNP growth over the next k quarters at an annual
rate,
SPREAD, = ten-year Treasury bond rate minus three-month Treasury bill rate on a quarterly
average basis in quarter t.
Chart 5
Growth of Real GNP and the Slope of the Yield Curve
1955 1957 1959 1961 1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989
The value of k, the prediction horizon, is allowed to run from one quarter to five years. We estimate
the model using quarterly data from 1955 to 1988 and using overlapping observations for k-quarter
GNP growth when k is 2 or more.
For forecast horizons beyond one quarter, the overlapping horizons produce econometric
problems that have been discussed by Hansen and Hodrick (1980). The data overlap creates a moving
average error term of order k-1, where k is the forecast horizon. This does not affect the consistency
of the ordinary least square (OLS) coefficient estimates but does affect the consistency of the OLS
standard errors. To perform hypothesis tests, we adjust the standard errors using the Newey-West
(1987) method.
Table 4 presents the basic regression results. Empirically, the yield curve slope and future real
growth are positively correlated. A steeper yield curve slope implies faster future growth in real
output. The R2 provides a measure of in-sample forecasting accuracy, while the statistical significance
of the SPREAD coefficient provides information on the reliability of the equation in predicting the
direction of a future change in output. Observe that accuracy in predicting cumulative changes is
highest five to seven quarters ahead. For these horizons, the spread explains more than one-third of
the variation in future output changes. This result is especially impressive because, as shown below,
the lagged value of real GNP growth has very little predictive power.
The yield curve seems to make particularly good predictions when large changes in output take
place. To illustrate this, we use the SPREAD variable, as defined above, to predict whether or not the
economy will be in a recession four quarters ahead. (We employ the National Bureau of Economic
Research dating of recessions.) Chart 6 shows the periods when a recession occurred, together with
the probability of a recession based on a model that uses the spread of four quarters earlier as the only
explanatory variable. The dependent variable in the equation assumes a value of one in quarters
within a recession and zero otherwise. The resulting equation, estimated using quarterly data from
1956 to 1988, is:
Probability(recession) = N[-.56 - .78 SPREADtJ,
(.16) (.16)
where N denotes the standard normal cumulative distribution function and standard errors appear in
parentheses.10 As Chart 6 shows, the only time the yield curve gave a wrong signal was in 1966-67,
10
The model was estimated using the probit method, a maximum likelihood technique that assumes
that the errors in the equation are normally distributed.
353
Table 4
Predicting Future Change in Real Output Using the Slope
of the Yield Curve
Sample: Quarterly, 1955 to 1988
(400/k)(log yt+k - log yt) = oto + a, SPREADt + e,
k (Quarters
Ahead)
Nobs
Oo
SEE
a.
135
1.74*
(0.64)
1.23*
(.29)
.13
3.75
134
1.63*
(0.56)
1.35*
(.28)
.24
2.82
133
1.64*
(0.54)
1.35*
(.28)
.31
2.39
132
1.70*
(0.52)
1.30*
(.27)
.35
2.C
131
1.79*
(0.50)
1.24*
(.24)
.38
1.86
130
1.89*
(0.48)
1.15*
(.22)
.38
1.70
129
1.99*
(0.46)
1.05*
(.19)
.37
1.59
128
2.11*
(0.44)
0.93*
(.16)
.33
1.54
12
124
2.50*
(0.37)
0.53*
(.14)
.18
1.34
16
120
2.75*
(0.30)
0.33*
(.11)
.09
1.16
20
116
2.86*
(0.24)
0.23
(.14)
.05
1.00
1
Notes: yt is the level of real GNP of quarter t. SPREADt is the difference between the ten-year Treasury bond
and three-month Treasury bill rates. The interest rates are annualized quarterly-average bond-equivalent yields.
In parentheses are Newey-West (1987) corrected standard errors that take into account the moving average
created by the overlapping of forecasting horizons as well as conditional heteroskedasticity. Nobs denotes the
number of quarterly observations, R2 the coefficient of determination adjusted for degrees of freedom, and SEE
the regression standard error.
•Significant at 5 percent level.
354
Chart 6
Probability of Recession (within Sample)
Based on Yield Curve Slope for Quarter 1-4
Probability
1
0.8 4
0.6 4
0.4 4
0.24
1955 1957 1959 1961 1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989
when a slowdown occurred instead of a recession.
B. The Yield Curve, Other Information Variables, and Real Activity
In this section, we add to the basic regression equation a number of information variables that
are widely thought to predict future real economic activity. Our aim is to determine whether the slope
of the yield curve continues to have predictive power in the presence of these other variables. We
also examine how the yield curve slope performs relative to survey forecasts of real GNP growth.
The information variables we choose are: the real federal funds rate, the recent growth in the
Commerce Department's index of leading indicators, lagged growth in real output, and the lagged rate
of inflation. The real federal funds rate is included as a proxy for the tightness of monetary policy.
The nominal funds rate is adjusted for inflation expectations obtained from a twelfth-order
autoregressive model for the inflation rate. Lagged inflation and lagged output growth are included in
the regression primarily because they describe the state of the economy.
355
Our reasons for including the index of leading indicators, which consists of eleven
macroeconomic variables, are obvious. The variables, both real and financial, are termed leading
indicators precisely because they have been found to have predictive power. The index provides a
convenient way of summarizing their aggregate information without entering each variable separately
in the regression equation. Some of the components do not become known until a month or more
after the statement month. Since we want to use only regressors that are known to market participants
at the time the slope of the yield curve is determined, we define the rate of growth in the index from
the first month of the previous quarter to the first month of the current quarter.
The regression results, presented in Table 5, point to several interesting conclusions. First, the
spread continues to have explanatory power over the entire range of forecasting horizons. Its
regression coefficients are statistically significant up to three years into the future. Second, an increase
in the real federal funds rate predicts a drop in real GNP, ceteris paribus, for about six quarters into
the future. Third, an increase in the index of leading indicators predicts a future increase in real GNP;
however, the predictive power lasts for only three quarters. This performance is very weak when
compared to that of the yield curve. Fourth, the lagged growth in output has a negative coefficient,
showing a slight degree of reversion towards the mean. Fifth, the lagged rate of inflation also shows a
negative coefficient; this coefficient is statistically significant at horizons beyond two quarters.
C. Comparison with Survey Forecasts
Another way to test the quality of the information in the yield curve slope is to compare its
forecasting performance with that of survey forecasts. We use data from a midquarter survey of real
GNP forecasts conducted by the American Statistical Association and the National Bureau of
Economic Research since the beginning of 1970. The data are the survey medians and refer to the
current and the following two quarters. Starting with 1981, we also have forecasts for three quarters
ahead.
Table 6 presents the results of regressing the realized percentage change in GNP on the spread
and on the survey prediction. It is clear that the spread is a better predictor than the survey median.
The survey has predictive power for only one and two quarters ahead, but not for three quarters ahead.
This is evidenced by the level of the R2s and by the significance of the coefficient (Jlf which
corresponds to the survey forecast. Observe that the R2s for the yield curve alone are consistently
higher than those for the survey alone. In addition, the coefficient of the survey forecast is not
significant beyond one quarter if the spread is present in the regression.
356
^
Table 5
Predicting Future Change in Real Output Using the Slope
of the Yield Curve and Other Information
Sample: 1955 to 1988
(400/k)(log yt4lt-log yt) = cto + a, SPREAD, + otj RFFt +
+ (X3 GLI, + a 4 LDEP t . k + 0I5 ntJkJL+ e.
k (Quarters
Ahead)
a.
q>
<*»
<*.
ots_
R2
SEE
1
.55*
(.26)
-.24*
(.11)
.16*
(.04)
-.03
(.10)
-.14
(.11)
.28
3.40
2
.73*
(.24)
-.31*
(.07)
.10*
(.03)
-.04
(.07)
-.17
(.12)
.40
2.51
3
.78*
(.26)
-.27*
(.07)
.07*
(.03)
-.15*
(.07)
-.26*
(.12)
.46
2.12
4
.81*
(.26)
-.24*
(.07)
.04
(.024)
-.18*
(.05)
-.29*
(.12)
.49
1.84
5
.82*
(.24)
-.20*
(.06)
.02
(.02)
-.21*
(.08)
-.31*
(.12)
.52
1.63
6
.82*
(.22)
-.17*
(.07)
.00
(.01)
-.21
(.11)
-.29*
(.12)
.50
1.52
7
.79*
(.20)
-.12
(.07)
-.00
(.01)
-.22
(.13)
-.27*
(.11)
.48
1.45
8
.71*
(.17)
-.10
(.07)
-.01
(.01)
-.26
(.13)
-.27*
(.11)
.44
1.41
12
.38*
(.15)
-.00
(.07)
-.01
(.02)
-.33
(.21)
-.26
(.15)
.29
1.25
16
.27
(.19)
.05
(.04)
-.01
(.01)
-.36
(.27)
-.29*
(.08)
.29
1.02
20
.16
(.09)
.09
(.05)
.01
(.01)
-.54*
(.09)
-.32*
(.05)
.44
0.77
Notes: y, is real output of quarter t SPREAD, equals the ten-year Treasury bond rate minus the three-month Treasury bill rate.
RFF, is the real federal funds rate (nominal minus expected inflation). All interest rates are annualized quarterly averages. GLI, is
the annualized growth in the index of leading indicators from the first month of quarter t-1 to the first month of quarter t LDEP,.k
=(400/k)0og y, - log yt4) is a lagged dependent variable. Jt,.M is the annualized rate of inflation of the GNP deflator from quarter
t-k through quarter t Newey-West (1987) corrected standard errors are in parentheses.
*Significant at the 5 percent level.
357
Table 6
Survey Forecasts versus Term Structure Forecasts
Yt+k = (400/k)(log yt+k - log yt) = cto + a, SPREAD, + p\ SURVEYFa + e,
k (Quarters
Ahead)
1
Sample
Period
February 1970
Chi-Squared(2)
«o
O,
.56
P<
.67*
R2
.08
SEE
4.08
(.59)
(.10)
1.08*
(.45)
1.30*
(.27)
.19
3.83
.46
1.13*
(.28)
.19
3.84
.15
3.11
to
«x„=0,p>l)
16.5*
[000]
April 1988
(.54)
2
March 1970
-.32
to
.26*
(.09)
.88*
(.74)
(.17)
.96*
(.36)
1.42*
(.22)
.37
2.68
.21
1.28*
(.24)
.38
2.67
-.04
(1.25)
3.01
.39
2.33
.41
2.26
5.83
[.054]
April 1988
(.93)
3
January 1982
.22
2.28
to
.27
(.26)
(3.44)
0.51
[.777]
April 1988
-.13
(.82)
1.59*
(.28)
-3.88
(2.58)
1.72*
.26
1.04
(.82)
Notes: Yt+k is the annualized cumulative growth rate of real GNP from quarter t to quarter t+k. SPREADt is the difference between
the ten-year Treasury bond and three-month Treasury bill yield. SURVEYFa is the ASA/NBER forecast of YHk. Numbers in
parentheses are Newey-West (1987) corrected standard errors. The last column presents a test for unbiasedness of the predictions
SURVEYFa. Numbers in brackets are significance levels.
•Significant at the 5 percent level.
358
D. The Yield Curve as a Predictor of Inflation
We now examine the ability of the slope of the yield curve to predict changes in the rate of
inflation. We regress the change in the inflation rate over a forecast horizon of k quarters ahead on
the level of the spread and on the first lag of the dependent variable. The basic equation is:
n u + k = a + p SPREADt + y nt.M,
where Fl^+k is 7Ckt - n\ml, T&t is the k quarter inflation rate starting in quarter t, and SPREAD is defined
as before.
Table 7 presents the results. The spread has predictive power, but only for forecast horizons of
two years or longer. Like the relationship between real activity and the spread, the association
between inflation and the spread is positive, but the significance appears at longer forecast horizons
and the statistical fit is not as close.
in. Conclusion
We studied the relationship of the yield curve to one policy instrument, the federal funds rate,
and two policy targets, real GNP and inflation. In all cases, we found historical relationships that may
be useful additions to the armamentarium of policy makers as they try to assess the economic outlook.
First, although the historical relationship between the federal funds rate and the slope of the
yield curve has varied over time, our evidence indicated that the slope has been sensitive to policy
actions and that the monetary authority has had a significant influence over the slope. Second, we
found that the yield curve has provided information about future real activity and inflation that may
have been useful to both private observers and policy makers. Clearly, since the level of the shortterm rate11 has been significant in explaining the slope, part of the information in the slope is related
to the stance of monetary policy. But there is also a component of predictability that is not related to
current monetary policy and may reflect expectations of future developments in the real sector of the
economy (shifts in the IS curve). This last component may be of interest to policy makers.
While the intertemporal relationships between the slope of the yield curve and real GNP or
inflation are clear and statistically significant, they cannot be easily interpreted in a causal manner.
This limitation suggests that the yield curve does not have a role as a formal intermediate target of
monetary policy. The relationships simply indicate that the yield curve slope has historically been one
of a series of variables that both the public and the private sector could have utilized in the
The funds rate, whether real or nominal, and the Treasury bill rate.
Table 7
Predicting Future Change in the Rate of Inflation with
the Slope of the Yield Curve
Sample: 1955-1988
n,,^ = oto + a, SPREAD, + aj IT,^ + e,
K (Quarters
Ahead)
1
Independent Variable
SPREAD
LDEP
Rf
SEE.
-.05
(.08)
-.44*
-.44*
(.07)
.18
1.74
.03
(.09)
-.29*
(.06)
.07
1.68
.00
(.08)
-.21*
(.07)
.03
1.69
.03
(.08)
-.07
(.06)
-.01
1.70
.05
(.06)
-.05
(.04)
-.01
1.71
.05
(.08)
-.11
(.07)
.00
1.75
.08
(.09)
-.14
(.08)
.02
1.79
.22*
(.08)
-.02
(.09)
.01
1.83
12
.36*
(.12)
-.12
(.09)
.07
1.93
16
.60*
(.11)
-.00
(.08)
.10
2.01
20
.70*
(.14)
.02
(.05)
.11
2.12
Notes: The dependent variable is defined as the annualized rate of inflation of the GNP price deflator from current quarter t to
future quarter t+k minus the annualized rate of inflation from quarter t-1 to current quarter t. LDEP is a lagged dependent
variable. SPREAD is defined as the current forward rate implicit in the term structure minus the three-month Treasury bill rate.
All interest rates are annualized quarterly averages.
* Significant at the 5 percent level.
360
formulation of their plans and policies. The statistical relations of the past, however, may well
deteriorate in the future, especially if policy makers attempt to make use of the slope of the yield
curve as a predictor of future macroeconomic developments.12
References
Doan, Thomas, Robert Litterman, and Christopher A. Sims. 1984. "Forecasting and Conditional
Projection Using Realistic Prior Distributions." Econometric Review.
Estrella, Arturo, and Gikas A. Hardouvelis. [1989] 1991. "The Term Structure as a Predictor of Real
Economic Activity." Federal Reserve Bank of New York, Research Paper no. 8907, May; also
forthcoming in Journal of Finance, 1991.
Fama, Eugene F. 1988. "Term Structure Forecasts of Interest Rates, Inflation and Real Returns."
University of Chicago Graduate School of Business, January, mimeo.
Granger, C. W. J., and Paul Newbold. 1977. Forecasting Economic Time Series.
Hansen, Lars P., and Robert J. Hodrick. 1980. "Forward Exchange Rates as Optimal Predictors of
Future Spot Rates: An Econometric Analysis." Journal of Political Economy, vol. 88
(October 1980), pp. 829-53.
Harvey, A. C. 1981. Time Series Models.
Johnson, Manuel H. 1988. "Current Perspectives on Monetary Policy." Cato Journal, vol. 8
(Fall 1988), pp. 253-60.
Laurent, Robert D. 1988. "An Interest Rate-Based Indicator of Monetary Policy." Federal Reserve
Bank of Chicago Economic Perspectives, vol. 12 (January-February), pp. 3-14.
. 1989. "Testing the Spread." Federal Reserve Bank of Chicago Economic Perspectives,
July-August, pp. 22-34.
Lucas, Robert E., Jr. 1976. "Econometric Policy Evaluation: A Critique." In Karl Brunner and Allan
H. Meltzer, eds., The Phillips Curve and Labor Markets. Carnegie-Rochester Conference Series in
Public Policy, no.l. Amsterdam: North Holland.
Mishkin, Frederic S. 1987. "What Does the Term Structure Tell Us about Future Inflation?" Columbia
University Graduate School of Business, mimeo.
12
For a detailed description of similar arguments, see Robert Lucas (1976).
361
Meulendyke, Ann-Marie. 1990. U.S. Monetary Policy and Financial Markets, Federal Reserve Bank of
New York.
Newey, Whitney K., and Kenneth D. West. 1987. "A Simple Positive Semi-Definite,
Heteroskedasticity and Autocorrelation Consistent Covariance Matrix." Econometrica, vol. 55
(May), pp. 703-8.
Shiller, Robert J., John Y. Campbell, and Kermit L. Schoenholtz. 1983. "Forward Rates and Future
Policy: Interpreting the Term Structure of Interest Rates." Brookings Papers on Economic Activity,
1, pp. 173-217.
Sims, Christopher A. 1972. "Money, Income, and Causality." American Economic Review, vol. 62,
no. 4.
Steindel, Charles, and Carl Palash. 1986. "The Shifting Term Structure of Interest Rates." Federal
Reserve Bank of New York, Research Paper no. 8613.
Turnovsky, Stephen J. 1989. "The Term Structure of Interest Rates and the Effects of Macroeconomic
Policy." Journal of Money, Credit, and Banking, vol. 21 (August), pp. 321-47.
362
THE USE OF DOLLAR EXCHANGE RATES AS TARGETS OR INDICATORS
FOR U.S. MONETARY POLICY
Charles Pigott and Christopher Rude
In the last decade, financial deregulation, innovation, and other economic changes have
significantly reduced the usefulness of the targets and indicators traditionally used to guide U.S.
monetary policy. As documented in other contributions to this volume, the medium-term relation
between the money and credit aggregates and nominal GNP, real GNP, and inflation has become more
variable and less predictable, while the economic impacts as well as information content of interest
rates have continued to be uncertain. The apparent deterioration of conventional monetary policy
guides has spurred renewed discussion of the utility of alternatives. This paper examines one of the
more novel of these alternatives, namely, dollar exchange rates.
In keeping with the other papers in this volume, we review the available evidence to determine
how exchange rates might be used in U.S. monetary policy as traditionally defined, where price
stability and other domestic goals are the primary concerns. While extensive, most of the literature on
exchange rates and monetary policy presupposes a somewhat different context, one in which external
objectives such as current account balance or maintenance of a fixed exchange rate are explicit goals
of policy. Accordingly, our task in this paper is to examine what the literature suggests about how
and to what extent exchange rates might aid U.S. monetary policy as it is presently defined.
We begin in the next section with a discussion of several characteristics of exchange rate
behavior that differ importantly from those of more traditional monetary policy targets and that are
likely to condition its use in monetary policy. This discussion draws on the short sketch of current
views of exchange rate determinants and their effects given in Appendix 1. The remaining sections
evaluate specific ways in which exchange rates might be employed to assist U.S. monetary policy.
Section I considers the use of the dollar as a primary intermediate target (largely) in place of domestic
interest rates or money/credit aggregates. Section II categorizes and analyzes ad hoc uses of exchange
rate considerations as "side constraints" or "modifiers" on the path of more traditional operating
instruments or intermediate targets. Section III examines evidence on the utility of exchange rates as
indicators of the stance of monetary policy and the course of the economy.
363
Given the complexity of exchange rate behavior, it should hardly be surprising that the
evidence does not provide much in the way of specific prescriptions for the use of exchange rates as
guides for U.S. monetary policy. Our review does lead, however, to three more general conclusions
that say much about both the limits and the potential of exchange rates in monetary policy
formulation. The first is that the use of exchange rates in monetary policy raises broader issues than
those associated with the use of more traditional domestic target variables such as the money
aggregates and interest rates. The influences on exchange rates are significantly more varied than
those affecting domestic interest rates and other traditional monetary policy variables, and the present
understanding of these influences is considerably less. Equally important, the systematic use of
exchange rates in U.S. monetary policy formulation would almost inevitably involve considerations of
international economic policy interdependence to a much greater degree than does the use of interest
rates or other more obviously domestic variables.
The second conclusion is that use of exchange rates as primary intermediate targets of U.S.
monetary policy, even if those targets are fairly flexible, would not improve the performance of policy
in maintaining macroeconomic stability and could easily add to instability under the circumstances that
prevailed through much of the last decade. Use of exchange rate targets would be most appropriate if
shifts in the demand for money balances were the primary macroeconomic disturbances to currency
values and the economy at large, a proposition very strongly rejected by the evidence of both the
literature and historical experience.
Nonetheless, despite this negative verdict, the literature does suggest (our third conclusion) that
exchange rates may have certain more limited policy uses. The two most important are the ad hoc use
of exchange rate considerations as policy guides when policy objectives or the efficacy of policy
instruments are impaired by foreign exchange market conditions, and the use of exchange rates as
indicators of the state of the economy and the policy stance. The difficulty is that present knowledge
is inadequate to define how such exchange rate considerations might be more reliably and
systematically used by policy makers than is presently possible.
Accordingly, our review suggests neither that U.S. monetary policy procedures should be
significantly modified to take a more systematic account of exchange rates, nor that exchange rates
should in any sense be ignored by policy. Future research seems unlikely to alter either of these
conclusions-unless the basic priorities of U.S. monetary policy are substantially altered. Nevertheless,
future research may-and, with luck, will-make the eclectic weighing of exchange rate considerations
in policy formulation a somewhat more systematic and informed task.
364
I. Basic Issues Concerning the Use of Exchange Rates in Monetary Policy
The literature has traditionally characterized the tools of monetary policy in terms of policy
instruments and operational targets (the variables controlled most directly and closely by authorities),
intermediate targets, and indicators.1 Collectively, these tools are used to achieve the fundamental
objectives of monetary policy, which in the United States are the essentially internal goals of price
stability (or an acceptably low rate of inflation) in the long run and stabilization of real growth near
potential in the medium term.2 Within this context, exchange rates plainly cannot be viewed as
objectives. Nor, for reasons that will become obvious shortly, is it likely that exchange rates can be
controlled sufficiently closely to be used as instruments or operational targets, at least without a radical
reorientation of U.S. monetary policy operations.
Accordingly, our discussion in the remainder of this paper is confined to the various ways in
which exchange rates might be employed as intermediate targets or as systematic monetary policy
indicators. As we will see, the issues that must be considered include not only those discussed in the
standard literature on domestic targets and indicators, such as the "stability" of a variable's relation to
the operating targets and goal variables and the information content of potential indicators, but also
questions arising from the special characteristics of exchange rates as macroeconomic variables.
We emphasize that our discussion concerns uses of exchange rates in addition to, or beyond,
the normal assessment of the likely economic effects of the dollar along with interest rates and other
key domestic variables that authorities now, of necessity, regularly undertake in formulating policy.
As is increasingly obvious, the United States is an open economy and monetary authorities here, like
their counterparts abroad, simply cannot ignore exchange rates in evaluating the condition of the
economy and the effects of their policy actions. The question considered in this paper is whether
exchange rate considerations should be elevated beyond this very important role to a more central and
formal place in monetary policy formulation.
Although the role of exchange rate considerations in monetary policy has been a major theme
1
See the Introduction to this volume by Richard Davis for a more precise and extensive
discussion of these tools and their functional role in monetary policy.
2
As defined in the Federal Reserve System Purpose and Functions (Board of Governors of the
Federal Reserve, October 1984 edition), "The basic goal of monetary policy . . . is to ensure that, over
time, expansion in money and credit will be adequate for the long-run needs of a growing economy at
reasonably stable prices. Over the shorter run, monetary policy is also conducted so as to combat
cyclical inflationary or deflationary pressures" (p. 13).
365
of the international economics literature for many years, most of that literature has been concerned
with the use of policy to achieve (at least in part) external objectives, such as a balanced current
account or the maintenance of a fixed exchange rate or some other exchange rate regime. Little, in
contrast, has been written about the use of exchange rates as targets or indicators in a domestically
oriented monetary policy of the sort used in the United States. This does not mean that the literature
is in any sense uninformative about the issues to be considered here, but it does mean that conclusions
must largely be based on inferences from the broader literature on exchange rates and their
macroeconomic effects.
Partly for this reason, we begin our discussion with a summary of several key features of
exchange rate behavior that are particularly relevant to the use of exchange rates in U.S. monetary
policy. Our review includes considerations likely to condition or limit the ways in which exchange
rates could feasibly be used in U.S. monetary policy and characteristics that make exchange rates
"different" in an important sense from more traditional monetary targets and indicators. The
discussion here draws in part on a summary of current views of exchange rate behavior and
determinants given in Appendix 1.
A. Considerations Shaping the Use of Exchange Rates in Monetary Policy
Exchange rate behavior has proved a notoriously complex and controversial subject and it
might seem therefore that one could make few definite claims about the use of exchange rates in
monetary policy. Such a conclusion would be premature, however. As explained in Appendix 1,
there is considerable agreement about the general framework to apply to the analysis of exchange
rates: the rates are regarded essentially as asset prices determined by a wide range of monetary,
financial, and real factors. Moreover, the major controversies about exchange rate behavior largely
concern the relative empirical importance of various potential influences rather than the fundamental
channels through which these influences operate. In this sense, significantly more is actually known
about exchange rate behavior than is often appreciated, and enough, at the least, to suggest several
general considerations about the way in which exchange rates might be used in U.S. monetary policy.
The foremost, and certainly most obvious, feature distinguishing exchange rates from more
traditional monetary policy tools is their international character. Admittedly, domestic interest rates
and other conditions are increasingly influenced by conditions abroad, but dollar exchange rates are
inherently international variables in that they are determined by foreign as much as domestic
conditions. Plainly, dollar exchange rates cannot be independently targeted by the U.S. monetary
366
authorities or their foreign counterparts; the dollar value of the yen, for example, is jointly influenced
by U.S. and Japanese monetary policy. Furthermore, movements in the dollar have important impacts
on the incomes, price levels, and trade balances of our partners as well as ourselves. Consequently, a
path of exchange rates that is desirable for the U.S. economy can easily be less desirable, or even
undesirable, for foreigners.
One obvious consequence of these interdependences is that any policy use of exchange rates
involves consideration of both foreign and domestic economic conditions. More important, such use
almost inevitably raises issues of international policy coordination and potential policy conflicts. The
reason is that any systematic attempt by the United States (or foreigners) to use exchange rates as
policy targets is likely to affect the policy trade-offs faced by foreigners and could restrict their policy
options. Thus, although U.S. authorities might in theory be able to use exchange rates as intermediate
targets while maintaining their traditional monetary autonomy and priorities, in practical terms, as we
will see in the next section, this arrangement is unlikely to be viable without some significant degree
of coordination with our partners. More generally, the use of exchange rates in monetary policy
normally cannot be separated from broader issues of international macroeconomic policy coordination
nearly as easily as the use of money aggregates or interest rates.
A second key feature of exchange rates is the exceptionally wide range of their determinants.3
Admittedly, the difference in this respect between exchange rates and other domestic financial prices,
which are also subject to many influences, is one of degree, but the difference nonetheless has
important policy implications. As explained in Appendix 1, exchange rates are influenced by
determinants of domestic and foreign interest rates, home and foreign price levels, and the relative
prices of traded and nontraded products~a list which includes virtually all macroeconomic
"fundamentals."4 Moreover, as asset prices, exchange rates are potentially influenced by expectations
3
Exchange rates, in fact, differ from other asset prices in that they are in an important sense
determined by home and foreign asset markets collectively rather than in a market for any particular
asset. The dollar value of the German mark does not represent the price of a particular dollar or mark
asset but rather a rate at which dollar and mark assets generally trade for one another. Thus the
effective "market" influencing exchange rates is very large and broad. Partly for this reason, the
management of exchange rates if used as intermediate targets is likely to involve considerably looser
control than the management of short-term interest rates or even the money aggregates under more
traditional operating procedures.
4
Exchange rates in effect represent the collective outcome of economic developments at home and
abroad and therefore cannot, except under very special circumstances, be viewed in terms of the
(continued...)
367
about all these variables (even those that are only "long-run" determinants of exchange rate
equilibrium) as well as their actual outcomes.5 Furthermore, and of particular importance here, the
evidence and historical experience amply demonstrate that all the major theoretical determinants of
exchange rates—inflation, interest rates, and real factors—have played important roles empirically in
exchange rate determination, and that the relative importance of each has varied considerably over
time and across countries.
The implication is that observed currency movements and their interactions with other
economic variables are likely to be exceptionally complex and variable. Indeed, theoretical models
have shown that any particular disturbance is likely to induce complex "dynamics" in exchange rates
in conjunction with interest rates, prices, and incomes. The pattern of these interactions depends upon
the formation of expectations (for example, "rational," adaptive, or other types of expectation) and
short- and long-term adjustment lags in various markets. Moreover, the currency movements that are
actually observed will normally be a reflection of the current and lagged effects of not one but many
such disturbances.
These observations have three important consequences for the use of exchange rates in
monetary policy formulation. First, the interpretation of exchange rate movements, taken alone or in
conjunction with other variables, is inherently at least as difficult, and generally more difficult, than
4
(...continued)
simple supply and demand frameworks often used to analyze the demand for money or short-term
interest rates. The simple monetary models introduced in the mid-1970s (for example, Frenkel 1976,
Mussa 1976, and Bilson 1978a), which were based on Harry Johnson's monetary approach to the
balance of payments (1978a, 1978b), did depict exchange rates as determined by relative demands and
supplies for home and foreign money. However, the simple monetary model requires the virtually
continuous maintenance of purchasing power parity, that is, essentially fixed real exchange rates, a
proposition likely to hold only when economic disturbances are nearly exclusively monetary in nature.
In practice, real exchange rates are observed to vary virtually as much as nominal rates, so the
empirical practicality of simple monetary models for all but fairly long-run horizons (or
hyperinflations) has proved quite limited.
5
Indeed, the fact that exchange rate changes statistically are nearly random suggests that their
fluctuations are virtually dominated by revisions in market expectations, or "news" (see Appendix 1
for further discussion). There is also ample evidence that expectations about long-run trends have
substantial influences on current exchange rates. The notion that anticipations of future inflation
influence current exchange rates is part of the consensus lore in this area. Work by Campbell and
Clarida (1986) implies that fluctuations in real exchange rates are dominated by revisions in the rates'
anticipated long-run values. Hutchinson and Throop (1985) provide evidence that market projections
of future budget deficits have played an important role in the evolution of the dollar during the 1980s.
368
the interpretation of other domestic financial variables. Extracting the "information" about the state of
the economy theoretically provided by exchange rate movements requires in practice considerable
knowledge about the economy's structure and the disturbances affecting it. Likewise, relations
between exchange rates and other monetary policy tools and objectives almost inevitably will be more
complex and less stable than those typically found for more traditional variables. Second, such
relations are likely to vary over time as the economic forces affecting the economy change. As a
result, any rules for using exchange rates as indicators or targets, beyond the most general, will
probably need to be altered periodically if they are to remain effective. In this sense, exchange rates
as guides for monetary policy are apt to be even more prone to the obsolescence that increasingly
seems to afflict more traditional monetary policy regimes. Third, exchange rates, even if used as
intermediate targets, are unlikely to be controlled as closely as more traditional target variables.
Stringent control of the dollar exchange rate would be difficult to achieve and, more importantly, most
likely undesirable.
B. Two Additional Observations
Beyond these general conclusions, two more specific considerations also condition the way in
which exchange rates might be used in U.S. monetary policy operations. The first concerns the extent
to which official purchases of dollar-denominated securities for securities denominated in foreign
currencies—sterilized foreign exchange market intervention—can have a lasting effect on exchange rates
independent of standard open market operations; this issue is largely, although not entirely, a matter of
the substitutability of dollar for foreign currency assets in investor portfolios. If sterilized intervention
has such effects, policy might, in principle, be able to influence the path of exchange rates in addition
to pursuing a more conventional intermediate target for an interest rate or money aggregate.
Overall, however, the evidence does not provide any adequate basis for regarding sterilized
intervention as a viable independent tool of basic domestic monetary policy operations. Admittedly,
the evidence on the substitutability of assets denominated in different currencies is inconclusive: the
literature does suggest that assets denominated in different currencies are imperfect substitutes to at
least some degree, and recent studies indicate that sterilized intervention may at times affect exchange
rates by influencing ("signaling") market perceptions and expectations.6 At the same time, however,
6
For a review of evidence on the effectiveness of sterilized intervention, see Edison (1990).
Dominguez (1988, 1989) and Dominguez and Frankel (1989) offer evidence that intervention can
(continued...)
369
despite years of research, attempts to systematically relate exchange rate movements to changes in the
currency composition of market portfolios have been unsuccessful. For this reason, the pursuit of
exchange rates as intermediate targets is likely to involve essentially the same tools—the discount rate,
open market operations, and so on-as those now used under more conventional regimes.
Finally, the size of the U.S. economy is an important consideration in evaluating how
exchange rates may appropriately be used as monetary policy guides. In smaller and more open
economies, exchange rate movements tend to have fairly direct and comparatively large impacts on
domestic prices, incomes, and interest rates; these impacts can significantly affect the trade-offs and
relations on which monetary policy is based. Large and unexpected exchange rate movements may
then require authorities to respond by altering the settings of their normal instruments simply to meet
previously defined objectives. In this sense, exchange rates almost inevitably become important guides
for monetary policy for smaller and more open economies. In the United States, however, exchange
rate movements normally have considerably smaller (although at times significant) impacts on the
basic trade-offs facing monetary authorities. Hence the case for making exchange rate considerations a
key formal element in policy formulation (beyond the normal assessment mentioned earlier) is almost
inevitably weaker. This point will arise in more concrete terms at several points in our discussion in
the following sections.
Our preliminary considerations can be summarized as follows. Exchange rates, viewed as
monetary policy targets or indicators, differ significantly from more traditional variables used for these
purposes, and their use in monetary policy formulation is apt to raise broader and more complex
issues. Exchange rates, if used as intermediate targets, would not be controlled as tightly as more
traditional variables, although any such control is apt to rely primarily on the same set of policy
instruments that is used now. Equally important, whether exchange rates are appropriately used in
U.S. monetary policy formulation, as well as how they are to be used, is apt to be especially sensitive
to the particular pattern of disturbances and other basic conditions of the economy. Consequently, the
efficacy of using exchange rates as targets or indicators turns more on their robustness in varying
circumstances than on their utility in hypothetical situations.
6
(...continued)
affect exchange rates through both signaling and portfolio channels. Even so, the duration and
reliability of their impacts are highly uncertain.
370
n. The Dollar as Intermediate Target
We now consider the most extensive use of the dollar in U.S. monetary policy that has been
seriously proposed, namely, as an intermediate target. If the dollar exchange rate were used in this
way, the U.S. monetary authorities would follow an explicit exchange rate target to maintain their
primary policy objective of domestic output and price stability. Under such a policy regime, targeting
the dollar would be the primary monetary policy rule.7
Detailed proposals for the use of dollar exchange rates in monetary policy formulation have
been advanced by Ronald McKinnon, John Williamson, and others.8 But, as explained in Appendix
2, their recommendations are largely based on objectives that lie outside our concerns in this paper.
McKinnon and Williamson are concerned in the first instance with increasing the degree of
international policy coordination, and thus they see the increased use of the exchange rate in monetary
policy as but one element in the general reform of the international economy. Here we are asking
whether it would be beneficial for the United States alone to use the dollar as an intermediate target to
achieve its traditional domestic objectives, assuming (even if unrealistically) that this could be done
without any formal and systematic coordination with our partners.
The operational use of the dollar as an intermediate target could take a variety of forms. The
target variable would almost certainly be a fairly broad index of the dollar's value against many
currencies rather than any one exchange rate, and most likely a moving average of those values over
one or more quarters. As with other targeting regimes, moreover, the authorities would vary their
instruments to counter deviations from a preassigned path or "zone" in pursuing an intermediate target
for the dollar. Given the doubts about the effectiveness of sterilized intervention to influence more
than short-term exchange rate movements, the authorities would normally be required to use their
standard instruments—open market operations or the discount rate—in targeting the dollar. Thus if the
U.S. authorities chose to use their instruments to target the dollar, they almost certainly could not
7
It probably would be necessary to announce or explain publicly the main dimensions of the
target and the rule. Beyond these general considerations, however, authorities most likely would not
publicly commit to any particular path, or even specific zones, for the dollar.
8
See McKinnon (1984, 1988), Williamson (1985, 1987, 1988, 1989), Williamson and Miller
(1987), and Edison, Miller, and Williamson (1987).
371
independently control interest rates or a money aggregate at the same time.9
Given the wide range of domestic and international influences on exchange rates, it is equally
clear that targeting the dollar would involve considerably looser control than has typically been the
case under interest rate or money aggregate targeting regimes.10 The target would have to be defined
in terms of a fairly wide zone around some path allowing for variation over time. The authorities
would not attempt to control the dollar closely or suppress the wide range of factors that now affect it,
but rather would try to influence its course by moderating pressures contrary to the target's direction.
Thus the use of the dollar as an intermediate target would not imply a commitment to a fixed
exchange rate. Instead, any exchange rate targeting rule adopted by the U.S. authorities would have to
be a loose one: the exchange rate would at most function as a medium-term "feedback rule" telling
the authorities if their stance is too tight or too loose relative to longer term goals.
To be consistent with policy objectives, a dollar target would clearly have to be based on
projections of foreign economic conditions-inflation at the least-in addition to the more standard
considerations now used in formulating interest rate and money intermediate targets. Partly as a
consequence, any targets for the dollar would probably be quite provisional and particularly subject to
revision. Finally, as emphasized earlier, the use of the dollar as an intermediate target would almost
certainly entail consultation with monetary authorities abroad, even if more explicit coordination were
not necessary.
A. Stabilizing Properties of Dollar Targeting
The key issue about the appropriateness of the dollar, or any other variable, as intermediate
target concerns the extent to which its use would be stabilizing in the sense that the exchange rate
"feedback rule" would offset rather than reinforce the shocks disturbing the economy.11 The question
is one of relative merits, not only whether dollar targeting is stabilizing on balance, but more
importantly, whether it is likely to be superior in this respect to alternative target variables. To resolve
this issue, two key questions, one conceptual and the other empirical, must be answered: First, under
9
The authorities could pursue a hybrid target that was some combination of individual variables,
but this approach would still normally require the use of the traditional policy instruments.
10
Lesser control would also be necessary if foreign "acquiescence" were to be at all realistic.
11
Thus the appropriateness of using the dollar as a monetary policy target must be evaluated in the
same way that Poole (1970) evaluated interest rates and the money aggregates in his classic
contribution to the stabilization policy literature.
372
what theoretical conditions would targeting the dollar be stabilizing or destabilizing? And second,
what conditions does the U.S. economy actually face?
To help answer these questions, the table on the following page shows the most probable nearterm responses of the U.S. economy to various fundamental disturbances (normalized so that the dollar
appreciates in each instance), assuming no change in monetary policy operating instruments. In each
case, exchange rate targeting can be said to be stabilizing if monetary expansion to counter the dollar's
rise would partially offset the original disturbance's effect on real output and inflation. We emphasize
that the responses in the table are those that the literature suggests for a large, comparatively closed
economy subject to a fairly high degree of international capital mobility. As indicated below, these
responses are quite ambiguous in some cases, depending upon the exact behavioral parameters
assumed. Accordingly, the table is intended as a heuristic guide to the discussion rather than an exact
or definitive summary.12
When exchange rate targeting is stabilizing
As the table indicates, there are situations in which the use of the dollar as an intermediate
target is probably beneficial, situations in which dollar targeting is probably destabilizing, and
situations in which the effectiveness of dollar is simply not clear. At one extreme, the table shows
that exchange rate targeting is most clearly stabilizing if the U.S. economy is subject to an unexpected
change in home money demand (an "LM shock"). An increase in home money demand lowers real
output and the pressure on prices by raising real interest rates and the value of the dollar. Limiting the
dollar's rise by increasing the domestic money supply would accommodate the increase in money
demand, suppressing its transmission to the general economy.
The fact that money demand in the United States has become unstable in recent years13 might
appear to suggest that the use of the dollar as an intermediate target would be beneficial—if, indeed,
this were the primary disturbance affecting the U.S. economy. But this is in general the only situation
in which exchange rate targeting is clearly beneficial. Moreover, as is well known, this is precisely
12
Disturbances that have straightforward impacts on closed economies are not infrequently found
to have ambiguous effects on open economies once the effects of exchange rate variations and the
transmission of the shocks to and from foreign economies are taken into account. See Boyer (1978),
Cooper (1976), Henderson (1979a), and Mundell (1962, 1963, 1968). Dornbusch (1980a) gives an
updated version of the "standard" model for an open economy.
13
See "Monetary Aggregates as Intermediate Targets" by John Wenninger in this volume.
373
When Exchange Rate Targeting is Stabilizing:
Probable Near-Term Response of U.S. Domestic Variables to Disturbance
When Exchange Rate Is Not Controlled
(U.S. Monetary Policy Instruments Held Constant)
2
Disturbance
(Normalized So That Dollar Rises)
Inflation
Nominal
Interest Rates
Real
Interest Rates
or o
- or o
- or o
- oro
Increase in foreign inflation
o
o
oor +
o
Increase in foreign money supply
?
?
- oro
oro
Oil or other terms-of-trade improvement
+
o or-
+
Output
Exchange rate targeting probably stabilizing
Increase in U.S. money demand
(domestic LM shock)
Decrease in foreign domestic demand
(foreign IS shock)
Exchange rate targeting probably destabilizing
Increase in U.S. domestic demand
(domestic IS shock)
Effects of exchange rate targeting unclear
Increase in demand for U.S. bonds
(away from foreign bonds)
Notes: o indicates effect is negligible. The response to the terms-of-trade case assumes that the economy is a net importer.
the situation in which interest rate targeting is appropriate.14 If the United States were subject to
uncertainty only in the domestic money market, therefore, it is unlikely that targeting the dollar would
have an obvious advantage over another, more traditional target variable-short-term interest rates.
Admittedly, countering exchange rate fluctuations caused by fluctuations in foreign expenditure
demand (a foreign "IS" shock) may also be stabilizing, but any benefits are likely to be significantly
smaller and less certain than in the prior case. A fall in foreign demand arising from, say, a
contraction in fiscal policy directly reduces the demand for U.S. exports and thus U.S. output more
generally, but as shown in case two in the table, it may also lead to downward pressure on U.S.
interest rates, a development which would tend to stimulate the U.S. economy. Unless the interest rate
decline is fairly substantial, the net effect on U.S. output and prices most likely would be
contractionary, although in most instances fairly small given the relatively large size and comparatively
closed nature of this country's economy.15 Thus, countering the dollar's movement would be
stabilizing in this case, but the effects are likely to be too small and uncertain to justify exchange rate
targeting on this ground alone. (Note that interest rate targeting would again be stabilizing under the
same circumstances.)
When exchange rate targeting is destabilizing
At the other extreme, the table shows that exchange rate targeting is likely to be destabilizing
to the U.S. economy if the authorities respond to unanticipated shifts in foreign inflation or foreign
monetary policy, if the United States' terms of trade change unexpectedly, or if the U.S. economy
experiences real demand disturbances.
Limiting the dollar's movements in the face of an unexpected change in foreign inflation
would tend to transmit the foreign price pressures to the United States.16 For example, if the U.S.
14
The argument that interest rate targeting in a closed economy context is stabilizing in the face of
monetary disturbances (and destabilizing in the face of real disturbances) was, of course, made famous
much earlier by Poole (1970).
15
This conclusion is consistent with the results reported by Bryant, Henderson, Holtham, Hooper,
and Symansky (1988) in simulations of eleven large macroeconomic models. The direct effect on U.S.
interest rates is likely to be substantial only if the foreign demand fluctuation is large enough to have
an appreciable effect on longer term world interest rates.
16
If allowed to move freely, the dollar would tend to appreciate in proportion to any increase in
foreign prices, thus effectively insulating the U.S. economy from any increase in foreign inflation.
(continued...)
375
authorities countered a dollar appreciation arising from increasing (but unperceived) foreign
inflationary pressures, they would only succeed in importing the foreign inflation: the dollar prices of
imported goods would rise with the foreign price increases, and monetary expansion would contribute
to an increase in domestic prices more generally.17
Targeting the dollar can be destabilizing if the U.S. authorities are uncertain about the course
of foreign monetary policy, because the required U.S. policy response to an unexpected change in
foreign monetary policy would mirror the foreign policy change: at the least, the U.S. authorities
would have to adjust their own stance in the same direction as the change in the foreign authorities'
stance to limit the dollar's movements. Expansionary foreign monetary policy, for example,
simultaneously increases foreign demand and the demand for U.S. exports and, by lowering foreign
interest rates, may also reduce dollar interest rates. But since an increase in the foreign money supply
also increases the value of the dollar, the net effect on U.S. output and prices is uncertain.18 On the
other hand, domestic monetary expansion to limit the dollar's appreciation would lower dollar interest
rates still further, stimulate domestic income, and thus prove destabilizing.
Using monetary policy to limit the dollar's movements in the face of an unexpected change in
the United States' terms of trade, such as a change in oil prices, can also be destabilizing. Such a
disturbance involves a change in the relative price of the good in question; preventing the exchange
rate from moving means, in effect, that domestic prices must vary to effect the relative price changewith problematic consequences for output. For example, a rise in world oil prices tends to lower U.S.
output while raising domestic prices; in the event of a rise, the dollar would probably fall on average
16
(...continued)
Long-term nominal dollar interest rates might rise somewhat, but the effects on output and inflation
would tend to be small or negligible.
17
Obvious shifts in the trend in foreign inflation could be offset by adjusting the target zone or
path for the dollar. In practice, however, whether the exchange rate is responding to such pressures or
to other factors is likely to be unclear. Moreover, too frequent shifts in the exchange rate targets,
however loose, could undermine the credibility of the authorities' policy-or at least make it confusing
to the public.
18
Mundell (1963) finds that monetary policy is transmitted negatively internationally, that is, that
foreign monetary expansion would have contractionary effects on the U.S. economy. Under his
assumption of perfect capital mobility, the exchange rate effect is dominant. In most larger
macroeconomic models (see Bryant, Henderson, et al. 1988), foreign monetary expansion tends to
increase U.S. output but to decrease U.S. consumer prices.
376
since the United States is a net importer of oil.19 Though monetary contraction to limit the dollar's
fall would reduce the upward pressure on prices, it would add to the downward pressure on output.
Thus decreasing the money supply following an unexpected rise in world oil prices would not offset
but only transform the effects of the disturbance.
Exchange rate targeting is apt to be destabilizing in the case of a shift in the demand for U.S.
products (an "IS shock") because the change in the policy instrument required to maintain the
exchange rate target would aggravate the effects of the original underlying disturbance on real output
and/or prices. For example, an unexpected (and unperceived) increase in private spending increases
domestic output and, by raising domestic nominal and real interest rates, pushes the dollar up.
Monetary expansion to limit the dollar's rise would reinforce the original disturbance by stimulating
output further and ultimately adding to inflationary pressures.20
The U.S. historical experience of the early to mid-1980s illustrates the risks of exchange rate
targeting under these circumstances. During this period, economic growth in Europe was slow, and by
comparison the United States was growing rapidly. As both nominal and real interest rates in the
United States increased and then held steady, dollar-denominated assets became an attractive
investment, and between January 1980 and February 1985, the dollar rose 94 percent on a tradeweighted basis. U.S. monetary policy thus would have been more accommodating during this period
if it had been directed toward limiting the dollar's rise. But this would have stimulated domestic
demand still further and would have required that the U.S. monetary authorities reverse their efforts to
contain the inflationary pressures that were the legacy from the late 1970s.21
19
Admittedly, the mark and the yen-currencies of countries even more dependent on oil imports
than the United States-would probably fall against the dollar, while the pound and the Canadian dollar
would rise. Nominal dollar interest rates might increase somewhat in this case in keeping with a rise
in expected inflation rates; real interest rates probably would move lower following a rise in world oil
prices as U.S. output moved lower.
20
The destabilizing effects of exchange rate targeting in the face of real disturbances are well
documented. See Mundell (1962, 1963, 1968), Fleming (1962), Dornbusch (1988), Henderson
(1979a), Feldstein (1989), and Obstfeld (1985). Clearly, exchange rate targeting is also equivalent to
interest rate targeting in this case-and both are destabilizing.
21
See Sachs (1985). The combination of fiscal stimulus and monetary contraction helped the
United States both to combat inflation and to avoid the recessionary consequences of doing so. In
fact, the dollar's rise during the period in itself contributed to the reduction in the domestic inflation
rate by putting downward pressure on all U.S. tradable goods and reducing the costs of imported raw
materials and intermediate products. Admittedly, this policy mix led to large external deficits.
(continued...)
377
When the effectiveness of exchange rate targeting is uncertain
The effectiveness of targeting the dollar when a variety of international financial market
disturbances alter market preferences among bonds of different currencies is not clear because the
impact of these shocks on the underlying economy is ambiguous. Because a decrease in the demand
for dollar bonds relative to nondollar bonds would raise long-term dollar interest rates and depreciate
the dollar, for example, the consequences for the economy depend on how the disturbance is split
between interest rates and exchange rates and on the exact response of the domestic economy to
changes in interest rates and the dollar exchange rates.
If the interest rate effect is dominant, and output and the pressure on prices move lower, then
monetary contraction to limit the dollar's decline would aggravate the consequences for real output
and/or prices. Even when the opposite is the case and the feedback relationship between the behavior
of the exchange rate and output and prices is in the right direction, exchange rate targeting would not
necessarily be stabilizing. Only when it is fairly certain that the initial impact of the disturbance will
be felt on the exchange rate alone can one have confidence that using monetary policy to limit the
dollar's rise is warranted.
Thus there is no reason in general to believe that targeting the dollar would help to guarantee
the stability of the United States' income and price performance in the face of the increased integration
of the world's financial markets in the 1980s. On the other hand, exchange rate targeting is apt to
increase the international transmission of the purely financial market disturbances themselves.22
B. U.S. Exchange Rate Targeting and International Economic Cooperation
As emphasized earlier, any target range for the dollar would have to be based on projections of
foreign as well as domestic economic conditions, and for this reason alone would require consultation
21
(...continued)
However, the evidence suggests that expansionary monetary policy could not have prevented the
deficits because its effect on the trade balance is at best transitory and small.
22
For example, a shift in demand away from dollar bonds to nondollar bonds wound tend to raise
dollar interest rates, lower nondollar interest rates, and depreciate the dollar; the measures that the U.S.
authorities would take to limit the dollar's decline would raise dollar interest rates still further and thus
reinforce the rise in dollar interest rates relative to nondollar interest rates. The exaggerated dollar
interest rate movements might be instantaneous if the U.S. policy makers' likely responses were
incorporated into market expectations.
378
with monetary authorities abroad. Other considerations, however, strongly suggest that the
implementation of a policy of targeting the dollar would probably have to go beyond consultation to
more active coordination.
In particular, unexpected foreign policy changes are quite likely to complicate the pursuit of
chosen target ranges for the dollar. Obvious shifts in foreign monetary and fiscal policy could be
offset by adjusting the target zone or path for the exchange rate. In practice, however, whether the
exchange rate is responding to such pressures is likely to be far from clear. Moreover, the U.S.
authorities would have to be concerned as well about the effects thatfrequentadjustments in their
targets might have on the credibility of their commitment to exchange rate targeting as well as their
long-term objectives. Thus U.S. authorities could easily find themselves adjusting their stance to
counter movements in the exchange rate stemming from unexpected changes in foreign policy. As we
have seen, such adjustments could well be destabilizing.
Using the dollar as an intermediate target, therefore, at a minimum requires close consultation
with the monetary authorities abroad simply to reduce the need forfrequentadjustments to the U.S.
dollar targets and thus to protect the credibility of U.S. monetary policy. Increased international
cooperation, even if unnecessary, probably would ease and enhance U.S. exchange rate targeting. But
increased cooperation would require the U.S. authorities to give up some degree of their autonomy.
Given the size of the U.S. economy in the world economy, moreover, foreign policy makers
are unlikely to be indifferent to any effort by U.S. monetary authorities to influence the dollar's
medium-term course. If, for example, foreign authorities were attempting to counter an internal (IS)
disturbance stimulating domestic demand, efforts by the United States to limit the downward pressure
on the dollar could complicate the foreign authorities' task. More generally, foreign authorities would
probably raise questions about dollar targets that they viewed as inconsistent with their own policy
objectives. Even when the dollar target was for the moment consistent with their own policy goals,
there would always be the possibility that the targets chosen by the United States in the future might
limit their maneuvering room.
Thus it is unrealistic to assume that foreign authorities would be willing to accept without
question a move on the part of the United States toward exchange rate targeting. Instead, foreign
monetary authorities, especially in the other major industrial nations, would very likely accept U.S.
exchange rate targeting only if they had some say in the exchange rate policies pursued. The more
stringent the dollar target, moreover, the more likely it is that they would expect their views to be
considered.
379
Foreign authorities would have additional concerns. They could question how a move toward
exchange rate targeting would affect the way in which U.S. economic disturbances are transmitted to
the rest of the world. U.S. exchange rate targeting, for example, would tend to increase the effects on
foreign output and prices of U.S. fiscal expansion: since U.S. efforts to limit the dollar's rise under
these circumstances would add monetary expansion to fiscal expansion, whatever price pressures U.S.
fiscal expansion placed on the U.S. economy would be transmitted more quickly and forcibly to the
rest of the world. Thus, with exchange rate targeting, foreigners would be concerned not only with
U.S. monetary policy but also with U.S. fiscal policy.
More generally, attempts by the United States to use the dollar as an intermediate monetary
policy variable inevitably confront the fact that in a world of "N" nations, there can be only "N-l"
independent exchange rate paths and hence only "N-l" exchange rate policies. Since an increase in
one country's exchange rate is a decline in that of its partners, exchange rate policies are inherently
interdependent. This observation would not be particularly consequential if the United States were a
small economy and dollar assets were not so widely traded internationally. Given the position of the
United States in the world economy, however, the use of the dollar as a monetary policy target could
easily affect the currencies of our partners and have significant consequences for their economic
conditions and policy choices. Under these circumstances, international policy cooperation and
coordination are likely to be necessary to avoid serious conflict with economic partners.23
We must conclude that the dollar exchange rate is not a particularly robust monetary target
variable, even if used in a fairly flexible way, and that the use of the dollar as a primary intermediate
target of U.S. monetary policy could easily have added to instability under the circumstances that
prevailed through much of the last decade. Using the dollar as a monetary policy "feedback rule" is
very unlikely to improve the stability of U.S. output and prices in the face of unexpected changes in
foreign inflation, foreign monetary policy, real domestic demand, and commodity prices or the United
States' terms of trade. Furthermore, dollar targeting may or may not stabilize the domestic economy
in the face of international financial market disturbances. Using the dollar as a monetary policy target
is potentially stabilizing in the face of unexpected changes in foreign demand, but the benefits of
23
Since a rise in one currency is simultaneously a decline in another currency, one can argue that
a policy of targeting the exchange rates will inevitably lead to policy conflicts. Dollar depreciation,
for example, stimulates the U.S. economy but dampens economic growth abroad. International
economic cooperation is needed both to ease the resolution of inevitable conflicts and to prevent
"beggar-thy-neighbor" policies from being adopted.
380
doing so may not be very large. The dollar is most clearly an appropriate monetary policy target in
the face of one variety of major economic disturbance—instability in the domestic money market—but
in this case, short-term domestic interest rates would more than likely be the preferred target.
Any move toward unilateral U.S. exchange rate targeting, moreover, is unlikely to be feasible.
From the U.S. point of view, increased international economic cooperation would ease the targeting
process but would entail some loss of independence. From the point of view of foreign authorities,
increased international cooperation might be seen as a prerequisite to exchange rate targeting. At the
least, foreigners would expect their views to be taken into consideration in the exchange rate targets
chosen by the United States.
m . The Dollar as Side Constraint
Although the dollar exchange rate should probably not be used as a primary intermediate target
of U.S. monetary policy, it might play a significant role in an ad hoc and temporary way as a side
constraint on the normal functioning of U.S. monetary policy when the monetary authorities are using
another variable as their primary intermediate target.24 The dollar might usefully function as a side
constraint on those occasions when there appear to be quite seriousfinancialrisks associated with
exchange market conditions or when exchange market developments appear to alter temporarily the
response of thefinancialmarkets or the underlying economy to the Federal Reserve's policy
instruments and targets to a substantial degree. The exchange rate could also function as a side
constraint when exchange rate movements appear to alter fundamentally the output-inflation trade-offs
that the policy makers originally assumed in calculating the time-paths for their instruments and
targets.
Admittedly, the relationship between the exchange rate and any primary target chosen by the
U.S. monetary authorities, both in the formulation and the execution of policy, will nearly always be a
close one. When the U.S. monetary authorities set and then implement their policy targets, they must
take account of the exchange rate responses and their economic impacts, and in this sense the
exchange rate normally acts as a factor conditioning or constraining monetary policy. But the active
24
The evidence suggests not only that the dollar has a role to play as a side constraint, but also
that the dollar may in fact have played this role in U.S. monetary policy in recent years. Indeed, the
argument presented in this section is based, in part, on both the "Record of Policy Actions of the
Federal Open Market Committee" that appears in the Federal Reserve Bulletin (see Federal Open
Market Committee) and the Federal Reserve Bank of New York's quarterly reports on "Treasury and
Federal Reserve Foreign Exchange Operations" (see Federal Reserve Bank of New York).
381
use of the dollar as a side constraint involves something more. The dollar would cease to be simply
one among a number of factors that must be taken into account and would become for the moment a
primary focus of attention; it would no longer be used simply as one of several indicator variables but
would become, for a short time, an object of major policy concern and possibly a temporary target.
Thus, as stated earlier, the question considered here is whether exchange rate considerations should be
elevated to a more central and formal place in monetary policy formulation beyond the normal
assessment of the likely economic effects of the dollar.
Since the United States is a large and, at this point, still a comparatively closed economy,
exchange market developments are likely to impinge upon the normal operation of U.S. monetary
policy only enough to justify the use of the dollar as a side constraint relatively infrequently; even on
such occasions, the dollar would supplement, but not replace, the primary intermediate target. That
is, use of the dollar as a side constraint is likely to be most appropriate when exchange market
developments alter quite substantially the balance of near-term risks faced by the authorities or the
trade-offs determining the speed at which operational instruments or intermediate targets are moved to
their desired paths. More frequent use of the dollar as a side constraint could risk undermining the
clarity and even the credibility of the authorities' normal objectives, to the extent that such use could
convey to the public the impression that the dollar was a secondary target or one part of a dual
target.25
In principle, the dollar can modify the application of the basic monetary policy rule in three
situations. First, when there appear to be significant financial market risks associated with exchange
rate developments, the exchange rate can temporarily alter the near-term weights given to alternative
ultimate policy objectives. Second, when exchange market developments modify the linkages between
the operational and intermediate targets and between intermediate targets and the underlying economy,
the exchange rate can temporarily affect how monetary policy is transmitted to the economy. Third,
when exchange rate movements significantly change the output-inflation trade-off that can be achieved
in a particular situation, those movements may temporarily affect the time-path used to achieve a given
output-inflation objective.
25
Furthermore, on those occasions when exchange market developments do become an important
policy consideration, authorities may need to explain their actions carefully and to emphasize that their
ultimate targets and goals remain valid.
382
A. Modifying the Trade-offs between Policy Objectives
Exchange market developments can modify the trade-offs between fundamental policy
objectives—and not simply the path taken to achieve a given policy objective—on those occasions when
financial market stability becomes a pressing concern temporarily outweighing goals for domestic
output and price stability. Thus, for example, the monetary authorities may be reluctant to ease
monetary policy during a period in which the dollar is especially vulnerable to sharp downward
pressures lest easing precipitate instability in the foreign exchange market that could spill over into the
other financial markets.
The dollar seems to have functioned as a side constraint on U.S. monetary policy in this way
on a number of occasions. In the spring of 1987, for example, the dollar came under strong selling
pressures, especially against the yen, and concern about the stability of the dollar spread to other
financial markets. As investors sought alternatives to dollar-denominated assets, the dollar's
depreciation precipitated sharp declines in prices of U.S. bonds and equities and contributed to price
increases in precious metals and bonds denominated in other currencies. In view of the fragility of the
dollar in the foreign exchange market, the U.S. authorities were very cautious in their provision of
reserves during this period.26 At the same time, the exchange rate was moved to the first position in
the order of the variables in the policy directive in the March 1987 Federal Open Market Committee
(FOMC) meeting and to second position in the directive in the May 1987 FOMC meeting.27
B. Altering the Response to the Authorities' Policy Instruments and Targets
Exchange market developments can temporarily alter the response of the financial markets or
the underlying economy to the Federal Reserve's policy instruments and intermediate targets in a
number of ways. In an open economy, for example, monetary policy affects domestic interest rates
both directly through the domestic financial markets and indirectly through its effects on the exchange
rate and exchange rate expectations. The direct domestic linkages are fairly straightforward: a
credible move toward monetary restraint will raise short-term nominal interest rates and probably both
short-term and long-term real interest rates. The indirect foreign exchange market linkages, however,
26
Federal Reserve Bank of New York, Open Market Analysis Division (1988, p. 56).
27
Furlong (1989). Heller (1988) argues that the order in which the various economic variables are
mentioned in the FOMCs policy directives is generally consistent with the relative importance placed
on these variables in monetary policy considerations. Furlong provides an updated listing of the order
of the variables in the policy directives.
383
are more subtle: a credible move toward monetary restraint may lead to expectations of dollar
appreciation and possibly reduce the risk premium on dollar assets as well, thus lowering medium- and
long-term dollar interest rates. The magnitude of any reduction is contingent on exactly how the
policy move affects exchange rate expectations and the perceptions of risk. Extensive international
capital mobility can complicate the effect of a change in monetary policy on domestic interest rates
because the indirect exchange market channel can alter the response of domestic interest rates in ways
that cannot always be easily predicted.
Of course, monetary authorities will normally give the exchange market channel linking
monetary policy and the domestic financial system careful consideration in the formulation of their
monetary policy rules. Indeed, any estimate of the effects of a change in monetary policy on shortterm and long-term interest rates is implicit in the policy makers' estimates of the time-paths linking a
change in their intermediate target and the financial system.28 Nonetheless, the exchange rate may
need to function as a side constraint on some occasions because the strength of the exchange rate
channel varies with market conditions and is contingent on the effects of monetary policy on market
expectations.
If the authorities are using a medium-term interest rate target, for example, they may choose to
modify their standard operating procedures when certain conditions prevail in the foreign exchange
market. During periods when the dollar is susceptible to heavy selling pressure, monetary expansion
may have smaller effects on interest rates by increasing the expected rate of dollar depreciation. More
importantly, the effect of a given change in monetary expansion on interest rates is likely to be less
predictable since it would depend upon the effect of the expansion on market expectations, which
under the circumstances are volatile. There could be serious unintended effects, therefore, if the
authorities were to continue to follow their standard operating procedures in those presumably
infrequent situations in which the normal linkages between instruments and targets and the economy
have been seriously distorted by the volatility of market expectations.
The spring 1987 episode can be interpreted from this point of view as well. During this
period, the dollar was under considerable selling pressure, and dollar interest rates were rising as
28
The studies collected in Research Papers on International Integration of Financial Markets and
U.S. Monetary Policy (Federal Reserve Bank of New York, 1987) investigate how increased
international capital mobility has changed the way in which monetary policy is transmitted to the U.S.
economy. The essays by Radecki and Reinhart (1987) and Kasman and Pigott (1987a, 1987b) focus
in particular on the changing relationship between monetary policy and the financial sector.
384
investors, motivated in part by their concern for the dollar's future stability, were selling dollar for
nondollar assets. Monetary expansion under these circumstances could well have precipitated a sharp
fall in the value of the dollar, thus increasing the market's anxieties about the dollar and dollardenominated assets; there was a risk that, instead of lowering dollar interest rates, expansion might
have precipitated a further rise. The more cautious policy stance actually taken, on the other hand,
helped to increase the market's confidence in the dollar and dollar assets and thus helped lead to a
reduction in dollar interest rates. Once the episode was over and market confidence was restored, the
more normal, less counterintuitive relationship between monetary policy and interest rates reemerged.
C. Modifying the Output-Inflation Trade-off
The exchange rate can influence the output-inflation trade-off in a given economic situation in
part because of the effect which the exchange rate itself has on these two variables: a particular timepath for short-term interest rates will yield lower inflation more rapidly for a given level of economic
activity when the dollar is rising and reduce inflation less rapidly for the same level of activity during
a period when the dollar is falling. Depending on the duration and the strength of the upward or
downward pressure on the dollar, therefore, monetary authorities might at times choose to alter the
speed at which monetary policy is changed to achieve a given output-inflation objective—for example,
postponing contraction if the dollar is under sustained upward pressure and speeding up monetary
contraction if the opposite is the case.29
The exchange rate, however, is likely to be an appropriate constraint only when exchange rate
movements so substantially alter the output-inflation trade-offs originally assumed by policy makers in
calculating the time-paths for their instruments and targets that these paths need to be changed or when
the trade-offs that the policy makers normally assume to exist are no longer viable. On those
occasions when the dollar functions as a side constraint in this way, moreover, the dollar would
probably be used more as an informational variable than as a temporary intermediate target. Thus the
authorities might postpone contraction if the dollar were under sustained upward pressure, because
contraction might accelerate the dollar's upward movement. A rapidly rising dollar might bring about
a lower domestic inflation rate more quickly and at a lower level of real activity than was desired. In
29
Again, the authorities will normally consider how the exchange rate is likely to affect the
output-inflation trade-off in their regular deliberations about setting their instruments and targets. But
the use of the dollar in this way is largely informational; that is, the dollar serves as an indicator of
higher or lower inflation and not as a side constraint.
385
certain circumstances, the authorities might also want to alter the behavior of the dollar so that the
output-inflation trade-offs normally assumed would return. For example, they might be reluctant to
tighten or might be inclined to ease somewhat until the dollar came down to a more acceptable level.
In the September 1985 Plaza Agreement among the Group of Five (G-5) industrial nations,
concern focused on what appeared to be an overvalued dollar and a major loss of U.S.
competitiveness, reflected in a decline in activity in U.S. manufacturing due to a shift to overseas
producers, as well as a large payments deficit fmanced by a massive buildup in foreign debt. The G-5
authorities agreed to intensify individual and cooperative efforts to achieve sustained noninflationary
growth and resolved that "some further orderly appreciation of the main non-dollar currencies against
the dollar [was] desirable."30 Consistent with these policy objectives, the U.S. discount rate
underwent a series of reductions and the federal funds rate declined more or less steadily in 1986.31
During this period as well, the position of the exchange rate in the order of economic variables in the
FOMC's policy directives suggests that greater importance was given to the dollar in monetary policy
considerations: the exchange rate was raised to the third position in the meetings of August 1985
through April 1986 from last place in the meetings of March through July 1985.32
One must emphasize again that the use of the dollar as a side constraint in this way is
somewhat exceptional. The U.S. economy is such a large and relatively closed economy that issues of
international competitiveness only infrequently alter the inflation-output trade-offs enough to call for a
significant alteration in the thrust of policy. One can also argue that U.S. monetary authorities used
the dollar's level in 1985 less as a side constraint than as an indicator variable. Moreover, even on
this occasion, when the use of the dollar as a side constraint might have been warranted, there was no
fundamental change in the basic paths for the authorities' actual operational targets, intermediate
targets, and final goals.
30
Group of Five Communique, September 1985.
31
Furlong (1989), noting that unemployment was above 7 percent and inflation below 3 percent in
1986, suggests that because this policy stance was consistent with a goal of stimulating the domestic
economy, one cannot definitely conclude that a desire to lower the dollar was the reason for these
policy initiatives. But this is the point: the dollar became an object of policy concern precisely
because of its effect on the domestic policy trade-offs.
32
Furlong (1989, p. 12).
386
IV. Exchange Rates as Monetary Policy Indicators
Our final topic concerns the use of dollar exchange rates as formal indicators for U.S.
monetary policy. We mean in this context the systematic use of exchange rates as indicator variables
(rather than targets) providing information about fairly specific economic conditions. This use goes
beyond the general assessment of exchange rates that authorities regularly undertake in the course of
formulating and adjusting policy.33 As we have emphasized, the behavior of exchange rates, their
likely reactions to policy actions, and their effects on the economy must always be taken into account
in determining the settings of monetary policy operating instruments and intermediate targetsregardless of whether they are used as formal indicators.
Interest in the explicit, regular use of exchange rates as indicators has grown in recent years as
the behavior of the more traditional information variables-interest rates and the money and credit
aggregates—has seemingly become more difficult to interpret. Since the mid-1980s, for example, U.K.
authorities have used the value of the pound (particularly relative to the German mark) as one
indicator of the tightness of monetary policy, as well as a supplementary target on certain
occasions.34 In the United States, a number of government officials as well as academics have
expressed interest in developing exchange rate measures in conjunction with more standard indicators
to improve the formulation of monetary policy.35
33
Admittedly, the distinction between simply "taking account" of exchange rate behavior and the
more formal use of exchange rates as specific indicators is far from precise. Furthermore, exchange
rates could in principle serve as both indicators and intermediate targets. Indeed, as Richard Davis
notes in the introduction to this volume, there are strong conceptual similarities between targets and
indicators.
34
See Copeley and Ryding (1988) as well as the Financial Statement and Budget Report of the
Chancellor of the Exchequer (1982). The report states that
the behavior of the exchange rate can help in the interpretation of monetary conditions,
particularly when the different aggregates are known to be distorted. . . . External or domestic
developments that change the relationship between the domestic money supply and the
exchange rate may therefore disturb the link between money and prices, at least for a time.
Such changes cannot readily be taken into account in setting monetary targets. But they are a
reason why the Government considers it appropriate to look at the exchange rate in monitoring
domestic monetary conditions and in taking decisions about policy, (p. 14)
35
See, for example, Federal Reserve Vice Chairman Johnson's argument (Johnson 1988) for using
exchange rates in conjunction with the interest-rate term structure and commodity price measures. He
suggests that a combination of a steepening of the term structure, accelerating commodity prices, and a
(continued...)
387
In principle, an economic indicator could be any variable that provides some useful
information about the economy, including its current state or future course, its underlying structure or
the nature of the disturbances affecting it. In practice, formal monetary policy indicators have
typically been applied somewhat more narrowly as indicators of underlying inflation, real growth
trends, and actual or potential strains in financial markets, or as measures of the tightness or ease
(stance) of monetary policy.36 Exchange rates, in particular, are most often viewed either as "leading
indicators" of inflation pressures or as indicators of the policy stance.
A. Some General Considerations
In assessing the literature on exchange rate indicators, much of which is theoretical in nature, it
is useful to bear in mind several important considerations that, while applying to policy indicators
generally, are especially relevant in the case of exchange rates. The first and perhaps most obvious is
that exchange rates, because of the relative complexity of their behavior and determinants, are likely to
be useful as indicators only when employed in conjunction with other variables. The possible
circumstances that can lead to any given exchange rate movement are so numerous that exchange rate
fluctuations by themselves are apt to be hopelessly ambiguous indicators of underlying economic
conditions. Most obviously, since the dollar's value effectively reflects U.S. conditions relative to
those abroad, any sensible judgment of what a given exchange rate movement implies about the U.S.
economy requires an assessment of foreign economic data. In addition to such data, interpretation of
exchange rate movements will nearly always require examination of domestic interest rates, inflation
trends, and other closely related developments. Because exchange rates must be considered in
connection with these other factors, they appear most likely to be useful as supplementary indicators
for monetary policy rather than primary indicators. Thus a key question is how much information
exchange rates add to that available from other measures.
A related second consideration is that an indicator, to be practically informative, must be
interpretable on the basis of the knowledge that policy makers can reasonably be expected to possess.
The theoretical literature excels in identifying circumstances under which the exchange rate or some
35
(...continued)
falling dollar (presumably relative to trend) could be taken as a sign of increasing inflation pressures
and the need to tighten monetary policy.
36
See the introduction by Richard Davis, especially page 6.
388
other variable supplies critical information—provided policy makers have fairly detailed knowledge of
the economy's behavioral parameters and are confronted by a limited array of disturbances whose
general statistical distribution is also known.37 In practice, however, policy makers almost always
have to interpret available indicators with highly incomplete and imprecise information about the
economy's structure, which in any case tends to vary unpredictably over time. An indicator whose
interpretation depends critically on knowledge of parameters only barely known, or on circumstances
that tend to change and are difficult to predict or identify, is unlikely to be very informative in
practical terms, whatever its theoretical information content. This consideration again applies
especially to exchange rate indicators, in large part because exchange rate behavior is so heavily
dependent on the structure of disturbances and on the policy reactions of authorities, both of which are
subject to fairly frequent changes.
Third, any given indicator or set of indicators is almost inevitably less informative (and the
information required for interpretation is greater) when there are many disturbances to the economy
than when there are only a few shocks, as is most often the case in theoretical models. For example,
the work on exchange rates as indicators of the policy stance (see Subsection C) is in large part based
on the observation that exchange rates in principle can help distinguish an increase in nominal interest
rates due to higher expected inflation from, say, an increase in excess money demand: in the first
case, the dollar is likely to fall (ceteris paribus), whereas in the second, the resulting increase in real
interest rates is likely to result in dollar appreciation. On this basis, exchange rates could be quite
helpful in interpreting interest rate movements when the economy is responding to only one shock at a
time (or when one disturbance dominates).
In practice, however, many disturbances affecting exchange rates, interest rates, and the
economy generally occur simultaneously. In most circumstances, a rise in the dollar coinciding with
an increase in domestic interest rates can reflect a wide combination of shifts in expected inflation,
foreign inflation, domestic money demand, and many other changes. Compared with hypothetical
situations in which disturbances occur one at a time, the observable information in this case changes
This literature is very extensive. For discussions motivated by practical monetary policy
considerations see, for example, Truman et al. (1981), Glick and Hutchison (1989), and Bergstrand
(1985a, 1985b).
389
little but the possible explanations increase greatly.38 Further complications are presented by the lag
in economic responses, which means that the observed behavior of a given set of indicators at any
time reflects the delayed influence of past disturbances in addition to any present shocks, as well as
the reactions of authorities to those disturbances and the market perceptions of all these factors. As a
result, analysts inevitably must confront questions of the following nature: is the current dollar
appreciation reflective of a new increase in U.S. relative to foreign real interest rates or simply a
response to past overly expansionary monetary policy whose earlier effects in lowering interest rates
are now receding?
A recent paper by Fuhrer and Moore (1989) graphically illustrates the problems of using asset
indicators when the economy is responding to a variety of current and past circumstances. Their
analysis shows that the covariation of asset prices, real GNP, inflation, and other fundamentals
depends quite importantly upon parameters such as the response of output and factor prices to
aggregate demand and the feedback rule followed by authorities. Indeed, their simulations indicate
that a rise in the dollar may signal a tightening of monetary policy under one targeting rule but a
loosening of policy under another rule.39
Neither this example nor the larger complications it illustrates mean that exchange rates lack
all value as monetary policy indicators. They do mean, however, that the use of such indicators will
be an inherently rough and inexact business most suitable to fairly modest objectives. This lesson is
38
This is an illustration of a more general problem. Like other asset prices, exchange rates can be
viewed as composed of a collection of unobservable "latent" variables, such as real interest rates,
expected inflation, the projected long-term real exchange rate, and so on. The problem is that
typically the latent variables greatly exceed the observable indicators so that the parameters of the
statistical distribution of the former (which are needed to infer their values) are much greater in
number than those (observable) parameters of the observables' distribution. Unfortunately, economic
theory and evidence provide little if any insight into the majority of the parameters in question,
namely, the correlations among the unobservable variables. For one illustration of the problems
involved, see Pigott (1983).
39
The statistical answer to these complications (which amounts essentially to running regressions)
is simple in theory but hardly so in practice. In principle, given a particular fixed statistical
distribution of the underlying disturbances and the economy's structural linkages, there is a "best"
predictor (the conditional expectation or ideal "regression") of the variables the authorities are trying to
infer based on the indicators available. The problem is that the predictor depends on all the current
and lagged relations among the key economic variables and on the variances and correlations among
the underlying disturbances. These would be difficult enough to determine even if they remained
constant indefinitely-which is certainly not the case.
390
confirmed by the following review of the empirical evidence.
B. Empirical Evidence on Exchange Rates as Inflation Indicators
The widespread view that exchange rate movements are fairly good "leading indicators" of
fluctuations in inflation is not based on theory alone.40 Several historical experiences, such as the
period in the late 1970s when sharp depreciation of the dollar preceded the peak in U.S. inflation,
appear to support this view as do numerous empirical studies linking exchange rates, import prices,
and the general price level for many countries. In the U.S. case, empirical analyses have fairly
consistently found that a 10 percent decline in the dollar leads ultimately to an average increase of
about 7 percent in import prices and a rise of about 1 to 2 percent in the consumer price index (CPI),
with the effects spread over eighteen months to two years. (Of course, the specific estimates in the
studies fall into a significant range centered about these means.41)
However, the interpretation of these apparently close relations between exchange rate
movements and subsequent U.S. inflation is less clear than might seem apparent for at least two
reasons. First, exchange rates, through their effect on import prices, are a direct input into domestic
costs and prices. Hence exchange rate movements directly affect inflation developments in the near
and medium term, in much the same manner as other cost components such as food or gasoline prices.
The usual equations relating the CPI or GNP deflator to exchange rate movements and other costs
must be interpreted as largely reflecting these direct effects. These relations do not generally tell us
whether exchange rate movements are good indicators of more persistent underlying inflation pressures
that are of particular concern to policy makers and also relatively more difficult to measure. A second
potential problem with these relations is that in many cases they represent reduced forms or partial
reduced forms of a larger set of structural relations relating the underlying determinants of inflation to
actual price movements. As with all reduced forms, the parameters relating exchange rate movements
to future inflation are apt therefore to vary as the structure of underlying disturbances and the
responses of policy change over time.42
40
See, for example, the comments in Willett (1985).
41
For surveys of such studies and their methodology, see Hooper and Lowrey (1979) and
Goldstein and Khan (1985).
42
For a discussion of these and other problems of interpreting standard exchange rate and inflation
relations, see Pigott, Rutledge, and Willett (1985).
391
The results of the relatively few (and recent) studies assessing the value of exchange rates as
indicators of medium-term to longer term inflation pressures have been mixed. These studies provide
evidence that movements in the dollar have "led" fluctuations in inflation in some cases and that, in
certain circumstances, the dollar has added significantly to the information about future inflation
provided by the term structure and commodity prices.43 However, as Lown (1989) shows, the
accuracy of dollar movements in predicting inflation has varied considerably; in particular, the sharp
dollar depreciation from 1985 through early 1986 was followed by a (further) decline in U.S. inflation
rather than an increase.
More generally, the use of exchange rates as inflation indicators is subject to a number of
conceptual pitfalls in addition to the ones mentioned above. First and most obviously, exchange rates
reflect relative conditions in the broad array of domestic compared with foreign product and asset
markets. Thus signals given by dollar movements about U.S. inflation necessarily involve an
assessment of foreign inflationary prospects.44 For the same reason, exchange rate movements are
unlikely to be the best indicators of developing worldwide inflationary pressures. Second, in periods
when real disturbances are important (as they have been for much of this decade), exchange rate
movements are liable to give misleading signals about underlying inflation trends. For example, the
dollar's fall beginning in 1985 is widely attributed to real factors (specifically, a shift in actual and
expected relations between national saving and investment) and perhaps for this reason proved not to
be a good indicator of future domestic inflation.
None of these caveats should be taken to mean that exchange rates cannot be helpful in
assessing inflation pressures in some circumstances. On balance, however, the utility of exchange
rates as indicators of underlying inflation pressures, as distinct from those pressures directly but
temporarily generated by their own movements, has probably been exaggerated and indeed seems, on
present evidence, to be fairly limited. Given their nature, exchange rates seem unlikely to be superior
to interest rates, commodity prices, or other equally available data as indicators of worldwide
inflationary pressures. Moreover, the more obvious historical cases in which exchange rate movements
have proved to be good leading indicators of inflationary pressures are ones in which the signals
43
Many of these studies have appeared in Federal Reserve publications. See Bergstrand (1985a,
1985b), Whitt et al. (1986), and Lown (1989).
44
Most studies relating dollar movements to future U.S. inflation in fact have not accounted
explicitly for foreign inflation, although Lown (1989) is an exception.
392
provided by interest rates and other domestic indicators appear in retrospect to be equally clear and
timely. What exchange rates can add to assessments of underlying inflationary pressures on a regular
basis, therefore, remains to be documented.
C. Exchange Rates as Indicators of the Policy Stance
A second set of empirical studies involves the use of exchange rate indicators to distinguish
movements in real interest rates from changes in the inflation premia~a step in assessing the stance of
macroeconomic policies. The handful of analyses in this area are most useful for what they reveal
about the potential information gains from using exchange rates in conjunction with the level and term
structure of interest rates~as well as the severe problems that will have be addressed before these
gains can be realized in practice.
Policy makers have often used real interest rate measures to assess the stance of monetary
policy and its likely effects on real activity and inflation. Short-term real interest rates can be gauged
relatively easily, though roughly, because anticipated near-term inflation is likely to be closely related
to recent past inflation developments (that is, because inflation tends to exhibit considerable short-term
inertia). Medium-term and longer term real interest rates are, however, generally of much greater
value in assessing the stance of policy and the future course of the economy, yet their measurement
(that is, of medium-term and longer term expected inflation) is much more difficult.45
As indicated earlier, exchange rates react quite differently to changes in real interest rates than
to changes in expected inflation; for this reason, exchange rates in principle can be informative about
movements in these two components of nominal interest rates. This observation is the basis, in fact,
for presumptions that increases in the dollar together with U.S. interest rates signal "tight" domestic
macro policies, that is, a sustained rise in real interest rates. Of particular importance in the present
context is that exchange rates, by their nature, are potentially informative about longer term real
interest rates and, for this reason, may also help in interpreting the term structures of nominal and real
interest rates.46
45
Admittedly, surveys of medium-term U.S. inflation expectations are available (in particular the
Hoey survey) but their accuracy as measures of general market anticipations is uncertain.
46
Recall that current exchange rates can be viewed as equal to their expected long-term levels
discounted by the home-foreign long-term interest differential adjusted for risk (associated with the
currency denomination). Accordingly, the "information content" of exchange rates concerning longer
(continued...)
393
Several studies of the U.S. experience with money aggregate targeting in the early 1980s
demonstrated that exchange rates could provide useful information about real interest rate movements
and the stance of policy, at least under some circumstances.47 During this period, reported increases
in Ml above announced FOMC targets tended to lead to virtually immediate increases in both shortterm and longer term interest rates. The studies used the reaction of dollar exchange rates to
discriminate between the two possible explanations of this pattern: that the monetary overshoot was
interpreted by the market as an indication of an acceleration in future money growth and hence as an
increase in inflation premia; or that the market expected the overshoot to be corrected in subsequent
months through an increase in real interest rates. The second hypothesis appeared confirmed by the
fact that dollar appreciation typically accompanied the rise in U.S. interest rates following
announcements of unexpectedly high money increases (dollar depreciation, or at least no change,
would have been expected if the first hypothesis were correct). The apparent success of this use of
exchange rate indicators must, however, be tempered by its relatively narrow application to short-term
movements in interest rates under a very specific monetary policy regime. Whether analogous
approaches can provide equally useful information, particularly about broader movements under
interest rate targeting rules, remains to be seen.
A subsequent study by Pigott (1984) attempted to apply the general principles underlying the
above studies to estimate movements in long-term real interest rates using observable data on nominal
interest rates and real exchange rates. The approach was to use certain identifying assumptions to
develop a prediction equation relating the observable interest rates and exchange rates to their
underlying components, in particular long-term real interest rates. The broad movements in the
estimated U.S. real interest rate obtained in this fashion seemed broadly plausible. Whether the
approach could be practically useful and extended to other periods is uncertain for at least two reasons.
First, the estimates almost inevitably depend on a number of fairly arbitrary identifying restrictions,
which are supported by very little independent evidence. Second, the parameters of the prediction
equation depend critically on the structure of the underlying disturbances to interest rates and exchange
rates and thus would need to be periodically (if not continuously) updated.
^(...continued)
term interest rates will generally decline as the volatility of expected long-run currency levels and the
risk factor increase.
47
See Hardouvelis (1984) and Cornell (1982).
394
In recent years, virtually no additional formal empirical analyses of the information content of
covariations in interest rates and exchange rates have appeared in the literature. There has been much
more work, of increasing sophistication, attempting to extract information about real interest rates and
expected inflation from the term structure of domestic interest rates.48 Given the possibilities raised
by the earlier studies cited above, renewed research examining the indicator properties of exchange
rates in conjunction with the level and term structure of interest rates might hold significant promise.
As yet, however, these possibilities remain largely unexamined.
V. Conclusions
This paper has reviewed evidence from the literature on the use of exchange rates in the
conduct of U.S. monetary policy. As we have emphasized, U.S. authorities in formulating policy
routinely consider the likely behavior of dollar exchange rates and its effects on the economy. In this
sense, because the U.S. is an increasingly open economy, authorities can hardly afford to ignore
exchange rate considerations in their policy deliberations. Our task here, however, has been to
consider whether the dollar should be used in a more active and systematic fashion in monetary policy
formulation, either as an intermediate target or as a formal indicator. In keeping with the other studies
in this volume, we have deliberately confined our analysis to the traditional context of U.S. monetary
policy, in which domestic goals of an acceptable inflation rate and real economic stability have been
pursued without any explicit coordination with policies abroad. The literature on exchange rates and
monetary policy, while quite extensive, has been primarily concerned with issues beyond this narrower
context, in particular the pursuit of external objectives and the maintenance of international monetary
regimes; thus our analysis of necessity has been based largely on inferences from the larger literature.
The increased attention over the last decade to the possibility of using exchange rates in the
formulation of monetary policy owes much to the breakdown of relations between the traditional
targets and indicators, the money and credit aggregates and interest rates, and nominal and real
activity. This breakdown, which has occurred in many other countries as well, has greatly increased
interest in alternative variables that might be used to guide policy. In the United States particularly,
interest in using dollar exchange rates in this context has also been spurred by growing awareness of
the degree to which the economy is influenced by conditions abroad, whose effects, at least at first,
tend most often to be transmitted through exchange markets.
See, for example, Mishkin (1989).
395
As we have argued throughout, however, the literature clearly shows that very important
differences, some fundamental, separate exchange rates from the other variables that have been
considered for use in monetary policy formulation. We have emphasized two of these differences in
particular. First, exchange rates are determined by a significantly wider range of real and financial
variables than are other asset prices, with the possible exception of stock prices; the list of exchange
rate influences includes virtually the entire spectrum of real and financial macroeconomic fundamentals
in the domestic economy and abroad. Exchange rates, no less than other asset prices, are critically
dependent upon market expectations, but these expectations generally concern a greater variety of
factors than are normally the focus in short-term money markets, bond markets, or commodity
markets. Second, exchange rates are inherently international variables that are directly affected by and
have direct effects on macroeconomic developments in both foreign and domestic economies; thus
their use in policy by any one country almost inevitably involves questions of international
macroeconomic interdependence and policy coordination.
At the least, these differences mean that use of exchange rates in monetary policy is likely to
be an inherently more complex process, involving broader issues, than the use of more traditional
variables—a conclusion underscored perhaps most clearly by our analysis of proposals to use the dollar
as an intermediate target. Loosening or tightening U.S. monetary policy as the dollar rises or falls
about some predetermined trend is likely to be beneficial for the U.S. economy only if disturbances are
dominated by shifts in money demand, and will generally be destabilizing under a wide range of
alternative, both nominal and real, shocks. The evidence of the last two decades suggests
overwhelmingly, however, that exchange rates have typically been subject to a very wide range of
nominal and real influences whose relative importance varies over time. If anything, the shocks under
which exchange rate targeting rules would be destabilizing have tended to predominate, especially
during the 1980s. Thus, countering the rise in the dollar in the first half of the 1980s through
monetary policy almost certainly would have seriously impeded the effort to reduce U.S. inflation. In
sum, the evidence in this area is fairly clear: the performance of U.S. monetary policy, judged simply
in terms of its success in achieving traditional domestic goals, would not be improved, and most likely
would be harmed, by use of the dollar as an intermediate target. Accordingly, the case for using the
dollar as an intermediate target and related proposals must be made on other grounds, in particular in
the context of international macroeconomic policy reforms in which external objectives and the
possibility of coordination of domestic and foreign policies become central. (Indeed, given the key
importance of dollar exchange rates to foreign economies, these considerations could hardly be
396
avoided if the United States, for whatever reasons, were to seriously pursue exchange rate targeting
over any sustained period.)
While it is reasonably clear that exchange rate targeting by U.S. monetary policy would be
inadvisable, the literature suggests that other, less ambitious and narrowly defined uses of exchange
rates may hold more promise. Indeed, the cumulative experience with floating exchange rates shows
that U.S. policy makers must take account of exchange rate movements, if only in gauging the effects
of their actions on the economy. Beyond this, the historical record includes a number of episodes in
which exceptional conditions in foreign exchange markets, alone or in conjunction with other financial
markets, appear either to have posed serious risks to key medium-term or longer term monetary policy
objectives or to have significantly if temporarily altered the responses to the operating instruments.
Under these circumstances, the use of exchange rates as temporary side constraints on near-term policy
actions may well be necessary simply to achieve domestic objectives or to restore the efficacy of the
authorities' instruments. The use of the dollar as a temporary side constraint seems, in fact, the
primary context in which exchange rate considerations (beyond their use as indicators) actually have
been given weight in U.S. monetary policy decisions during the 1980s. The literature, however, has
virtually nothing to say about the potential benefits or risks from such uses, a fact which is perhaps
not surprising given the unavoidably irregular and ad hoc nature of these uses. However, the
challenge for analysts is to define more clearly the circumstances under which exchange rate
considerations should override normal monetary policy procedures and to determine how exchange
rates should be used in such circumstances.
Historical experience and the formal literature, while mildly encouraging about the potential
utility of exchange rates as formal monetary policy indicators, provide only very general suggestive
guidelines as to how they might be so used. Most clear is that, given the complexity of their
determinants, any information provided by exchange rate movements themselves is hopelessly
ambiguous unless gauged in conjunction with interest rates and other indicators. If anything, the
literature, which has concentrated on identifying hypothetical circumstances in which the authorities
are faced with a very limited array of disturbances to gauge, understates the practical difficulties of
using exchange rate indicators. Given the complexity of exchange rates' interactions with other
economic variables, analysts must know a great deal about the disturbances hitting the economy and
other features of its structure if they are to extract significant additional information from currency
movements. Needless to say, this task is very difficult because these economic features tend to change
unpredictably over time. Thus it remains to be seen that exchange rates can in practice provide useful
397
regular information to guide policy in other than fairly extreme and infrequent circumstances. At the
same time, the literature has not thoroughly explored the possible ways in which exchange rate
indicators might prove useful (indeed, the task has barely begun). In particular, the extension of recent
work on the information content of the term structure of interest rates to include exchange rates may
hold some promise, although its value has yet to be tested.
Overall, therefore, the evidence does not provide any thoroughly compelling case for
substituting exchange rates, either as targets or indicators, for more traditional monetary policy guides.
In the future as in the past, policy makers will need to examine exchange rate movements in assessing
the course of the economy and the effects of their policy actions. Beyond this, the evidence does not
provide any firm basis for the use of exchange rates as targets except in fairly extreme and irregularly
occurring circumstances. Nor is there any compelling case as yet for the systematic use of exchange
rates as more than general indicators, or as supplements to more traditional indicator variables whose
properties are better understood. Thus the use of exchange rates in U.S. monetary policy is likely to,
and probably should, remain an "art" for the foreseeable future. The challenge to analysts is to
determine through further research whether that art can be refined and made more systematic and
reliable than it presently is.
Appendix 1: Exchange Rate Determinants and Effects: A Short Primer
Exchange rates, their determinants and their effects, are the subject of an enormous literature
and much ongoing controversy. Nonetheless, there is considerable continuity in the way economists
have viewed exchange rates over the years and more agreement presently than is often appreciated
about the general framework that needs to be applied to their analysis. Below we sketch the general
features and major controversies concerning the "modern" theory of exchange rates relevant to our
topic; this discussion is primarily for the benefit of those readers not familiar with these issues and is
not intended as a definitive survey.49
49
There are many excellent surveys of exchange rate models and evidence; good examples are
Dornbusch (1980a, 1980b, 1987a), Isard (1987), Obstfeld (1985), and Shafer and Loopesko (1983).
Arndt and Pigott (1985) trace the postwar development of exchange rate models and their evolving
representation of key policy questions.
398
The Economics of Exchange Rates
The common theme of exchange rate analyses for (at least) the last forty years is that currency
values must be consistent with balance of payments equilibrium.50 This equilibrium condition simply
says that the balance of a nation's current account transactions (CA) must equal, or be financed by, its
net inflows of capital (NCI):
(1) CA = NCI
Note that net capital inflows include all private as well as official inflows into the country.
During the 1950s and 1960s, the current account side of this equilibrium relation was given
primacy in explaining exchange rates. This emphasis largely reflects the fact that international capital
flows were sharply constrained by official and institutional barriers (capital mobility was quite low)
and economists' limited understanding of how exchange rate movements could affect financial market
demands and supplies. Exchange rates were typically viewed as varying either to maintain the relative
prices of traded goods needed for current account balance or, alternatively, to offset movements in
relative national price levels in accordance with the much older doctrine of purchasing power parity
(PPP).51 In either case, real factors determining equilibrium relative prices in goods markets were
viewed as critical to exchange rate determination.
The dramatic increase in the size and mobility of international capital beginning in the mid1960s and improvements in the analytical technology for analyzing financial market equilibrium led to
increasing emphasis on the capital account's role in exchange rate determination. In the highly
influential Mundell-Fleming model introduced in the 1960s, exchange rates were jointly determined by
the current and capital accounts, with the latter's importance an increasing function of the degree of
capital mobility.52 The present framework used by virtually all economists to analyze exchange rates,
known as the "asset market approach," is essentially the limiting case of the Mundell-Fleming model
50
This view was well established even before the advent of more formal treatments. For an early
example applying this view systematically, see Meade (1951).
51
Purchasing power parity dates back to the nineteenth-century writings of Casel (and earlier in
the view of some historians). Models based on current account equilibrium were essentially Keynesian
in nature, emphasizing the effects of exchange rate changes on relative prices on traded goods and
their repercussions for domestic incomes; good examples are Meade (1951) and Alexander (1959).
Subsequent extensions, notably Machlup (1956) and Laurson and Metzler (1956), emphasized broader
macroeconomic effects on wealth and asset markets.
52
Mundell 1962, 1963, 1968; Fleming 1962.
399
when capital mobility is highly, if not perfectly, mobile.53
The asset market approach views exchange rates as proximately determined in financial
markets.54 This framework is based on three assumptions: first, that exchange rates are free to vary
essentially instantaneously in response to market conditions; second, that asset markets themselves are
nearly continuously in stock equilibrium; and third, that potential, ex ante international capital flows
are much larger than fluctuations in trade flows. The first and second assumptions are in contrast to
what is commonly assumed about product markets, where prices are often constrained by contracts or
other institutional "rigidities" and where (partly as a result) full market clearing can take considerable
time; the third assumption indicates that the asset approach is most appropriate in an environment of
relatively high international financial integration and capital mobility.
The asset approach can be summarized by saying that the current value of the exchange rate
(S(t)) is the discounted value of its expected future level "n" periods in the future (Sc(t;n)), where the
"discount" factor is the risk-adjusted interest differential between the home and foreign country
(expressed as the return over the n period horizon):
(2) S(t) = S c (t;n)/[(l+i^^
where i and i* are the home and foreign interest rates and r is the differential risk premium.55 The
risk premium effectively represents the extra risk (to the market as a whole) of investing in home
assets as against assets denominated in foreign currencies. The relation simply says, equivalently, that
the return on foreign assets is equal, on a risk-adjusted basis, to the return on (comparable) foreign
assets after accounting for the expected depreciation of the home currency over the holding period.
As this relation suggests, the asset approach effectively describes exchange rate movements in
53
The asset approach developed as an outgrowth of the Mundell-Fleming framework beginning in
the early 1970s. Its development was spurred in part by work of Branson (1972), Officer and Willett
(1979), and others on stock-flow distinctions in the determination of international capital flows. Work
by Tobin (1969), Foley and Sidrauski (1970), and others on domestic portfolio balance models also
contributed. Johnson's seminal monetary approach to the balance of payments (1978a, 1978b) was at
least a precursor of the asset approach. Other relatively early examples are Kouri (1976), Kouri and
Porter (1974), and Branson (1979).
54
Informative discussions of the underpinnings of the asset approach can be found in Home
(1980) as well as in Allen and Kenen's treatise (1980).
55
This standard formulation is exploited by Isard (1980) to develop a very useful accounting
framework for classifying factors affecting exchange rates. Note that this relation assumes that legal
or other barriers to international capital flows are essentially negligible. Where such barriers are
important, the term r() includes their effects in addition to the effects of risk.
400
terms of changes in their expected long-run equilibria andfluctuationsabout those equilibria due to
variations inrisk-adjusteddomestic relative to foreign interest rates. Indeed, as is often pointed out,
the asset approach itself describes only the relation between current exchange rates and their future
values, and is incomplete without a theory of the long-run determinants of currency values.56
Economists' views about these long-run determinants remain quite similar to those incorporated in
traditional approaches. That is, long-run equilibrium nominal exchange rates (S) can be viewed as the
product of the long-term equilibrium levels of a) the real exchange rate (X), which summarizes
equilibrium relative prices of home versus foreign goods, and b) relative national price levels (P/P*):
(3) S' = X'(P/P*)' ,
where ' denotes long-run values.
The implication of the asset approach as summarized by relations 2 and 3 is that market
expectations are likely to be a major, even dominant, influence on currency values.57 Furthermore,
the "dynamics" of exchange rate adjustments to their long-run values are likely to be closely related to
the dynamics of domestic and foreign interest rate movements. Note also that, although exchange
rates are proximately determined in financial markets, real factors nonetheless can affect currency
values in the short run as well as the long run through their influence on expected future equilibrium
exchange rates.58 Indeed, any factor influencing exchange interest rates, inflation rates, or relative
prices—which is to say virtually any macroeconomic variable—is a potential exchange rate determinant.
Controversies about particular models of exchange rates amount in large part to questions
about the relative importance of various subsets of the broad range of factors that, in theory, can
56
The notion that exchange rates are heavily influenced by new information, normally referred to
as "news," has led to an extensive literature, some of which is relevant to our discussion of exchange
rate indicators in Section IV, and is now one of the generally accepted "stylized facts" about exchange
rate behavior. See, for example, the surveys by Mussa (1977) and Dombusch (1980a).
57
See, for example, Isard (1978, 1980) and Dooley and Isard (1982, 1983).
58
This view is highly traditional, as indicated by the discussion in Yeager (1976). It is also
completely consistent with the modem approach, as the treatises of Allen and Kenen (1980),
Dombusch (1980b), and Niehans (1977) make abundantly clear. The primary difference between older
and more recent models in this respect is that older approaches tend to view the current account as the
major direct channel transmitting real influences to exchange rates, whereas modem asset models focus
on expectations (about future currency values and their determinants) as the dominant channel.
401
influence currency values.59 The available models can be divided into three basic classes: monetary
models emphasizing variations in money demands and supplies; portfolio balance models emphasizing
the currency composition of assets (money and bonds) available to the public; and current account
models incorporating considerations of (long-run) current account equilibrium. These models are
consistent with a variety of complex exchange rate behaviors but do not, even collectively, consider
the full range of potential exchange rate determinants; real factors particularly tend to be ignored or (as
in the last group) treated only in a very limited and generally ad hoc fashion.
Monetary models are based (naturally) on the presumption that nominal disturbances are
dominant and are closely linked (because of money neutrality) to purchasing power parity (PPP). The
simplest models assume that output prices are perfectly flexible so that PPP holds continuously (that
is, real exchange rates remain absolutely constant). The model has proved to be most useful in
understanding broad exchange rate movements over long periods or during hyperinflations, situations
in which the relative importance of monetary disturbances are likely to be greatest.60 However, the
monetary models' utility over more normal medium-term intervals has been at best very limited, since
real exchange rates are observed to vary considerably—indeed as much as nominal rates.61
Dornbusch's extension of the simple monetary model, based on "sticky" price adjustment to excess
demand, does allow for (transient) real exchange rate movements in parallel with changes in real
interest rates in response to money shocks, but predicts (if monetary disturbances are dominant)
considerably stronger empirical interest-rate and exchange rate relations than are actually observed.62
59
The realization that this is the main issue, rather than timeless "structural" propositions about
exchange rate behavior, has emerged only gradually. Even a decade ago, empirical surveys and
evaluations of various exchange rate models tended to be couched in terms of key "hypotheses" about
exchange rate behavior in general; examples (and excellent and informative surveys) are Home
(1980), Frenkel (1980), Isard (1978), and Schadler (1977). These are usefully compared with more
recent surveys conditioned by sobering empirical experiences with various models: Dornbusch (1980a,
1987a), Frenkel (1980, 1981), and Obstfeld (1985). Arndt and Pigott (1985) also emphasize the
relative importance of disturbances as the key distinction among the various models.
60
61
See Johnson (1978a, 1978b), Bilson (1978a, 1978b), and Frenkel (1976) for seminal examples.
Another stylized fact cited in Mussa's (1977) survey.
62
Dornbusch's (1976) overshooting model of exchange rate determination spawned numerous
theoretical extensions and applications. The price/interest rate dynamics of the model represented
nothing beyond those of standard domestic macroeconomic texts and empirical models, but the
implications for exchange rate behavior were not well understood until its advent. Although initial
(continued...)
402
While very helpful in clarifying monetary policy transmission to exchange markets (and in helping to
explain why, for example, U.S. interest rates and the dollar both rose sharply during the early 1980s),
the Dornbusch model has not proved empirically robust across periods or countries.63
Portfolio balance models, essentially international versions of the domestic models of Tobin
(1969) and others, seek in large part to examine the influence on exchange rates and interest rates of
the currency composition of (outside) assets; the degree of substitutability of assets in different
currencies and the closely related question of the determination and empirical importance of the
currency "risk" premium (r) are the primary issues in these models.64 The most modern versions are
the international capital asset pricing (ICAPM) frameworks based on finance principles of risk
diversification.65 Unfortunately, given the importance of the issues addressed, portfolio balance
models have met with little success empirically, whether in reliably measuring asset substitutability
and the effects of changes in asset composition, in identifying or explaining the behavior of the risk
premium (if any), or even in verifying basic assumptions underlying the framework.66
Exchange rate models incorporating current account or related considerations can be viewed as
attempts to account for shifts in long-run equilibrium relative prices or for permanent departures from
a given PPP equilibrium.67 While there is growing evidence that such shifts may be important
62
(...continued)
empirical applications appeared promising (in particular, Frankel 1979a, 1979b), subsequent analyses
raised considerable doubts about its utility as a general explanation. See (among very many examples)
Brittain (1977), Willett and Forte (1969), and Pigott and Sweeney (1985).
63
See Dornbusch (1980a, 1987a) and Obstfeld (1987).
64
The models developed out of work by Branson (1977), Henderson (1979a), and Allen and
Kenan (1980). Early empirical analyses include Branson, Haltunnen, and Mason (1977) and Artus
(1976).
65
For example, Lewis (1988) and Engle, Frankel, and Rodrigues (1989).
66
In particular, studies have so far failed to detect a statistically significant or reliable effect of
currency composition or more general portfolio balance effects on exchange risk premia. See Frankel
(1982) as well as Engle, Frankel, and Rodrigues (1989). Portfolio balance models often do outperform
the monetary models, however (Home, 1980).
67
In attempting to explain dollar movements in the 1970s and early 1980s, Hooper and Morton
(1980) add a proxy for movements in current account equilibrium to a standard portfolio balance
model. In similar spirit, Hutchinson and Throop (1985) use the expected U.S. budget deficit in their
exchange rate model, while Keran and Zeldes (1980) examine the effect of oil prices on the dollar-yen
rate.
403
sources of currency fluctuations, their empirical modeling is still at a very early stage. Existing
models have generally relied on rough proxies and/or have been tailored to very specific conditions,
such as the effects of oil price increases; these models, like their competitors, have not been successful
empirically beyond limited episodes.
Overall, the various exchange rate models, while seemingly useful in explaining certain
episodes or aspects of exchange behavior, have proved to be empirically unstable and unable to
explain or predict currency values to any economically significant degree. Indeed, the one established
empirical proposition in this area is that exchange rate changes are nearly random.68 Studies by
Meese and Rogoff (1983a, 1983b, 1988) showed that a random walk predictor outperforms the formal
exchange rate models, even when realized values of the explanatory variables are used in place of
expectations variables. These "failures," while amply justifying pessimism about economists* ability to
predict exchange rates or estimate their "equilibrium" values with any precision, are nonetheless
informative in that they strongly suggest that exchange rates in any given period are affected by a very
wide range of disturbances, rather than dominated by any particular set incorporated in a given model.
Exchange Rate Effects on the Economy
The economic effects of exchange rates and exchange rate linkages to other macroeconomic
variables have also been the subject of a copious and longstanding literature, one which has reached
somewhat more definite conclusions than the literature on exchange rate determinants. The real sector
impacts of exchange rates are the most familiar and probably least controversial of the linkages. For
example, a fall in the dollar itself tends to raise the absolute and relative prices of foreign as against
U.S. traded goods, leading after a lag to a fall in U.S. imports and a rise in U.S. exports, developments
which in turn stimulate domestic production and income. The rise in import prices also tends to raise
the domestic price level, both directly and by influencing prices of domestically produced substitutes
and wages. Foreign countries, of course, experience qualitatively opposite effects on their own prices,
68
Simple statistical tests nearly always fail to reject the hypothesis that exchange rate changes are
nearly random. See, for example, Logue, Sweeney, and Willett (1978). Studies have also shown,
however, that future exchange rate changes can be predicted to some degree by certain (generally
nonlinear) rules based on past exchange rate behavior; see Dooley and Shafer (1983) and Logue,
Sweeney, and Willett (1978). The apparent paradox is probably explainable by the growing evidence
that the variance of exchange rate changes (volatility) evolves over time in the manner suggested by
the ARCH model of Engle (1987). That is, a relatively "large" change in the exchange rate at a given
time raises the probability of further changes that are "large" in magnitude; however, the probabilities
of a change in either direction are not affected.
404
trade balances, and output.
Empirical knowledge about these impacts and their timing, while far from exact or definitive,
is nonetheless sufficient to gauge, at least roughly, the effects of given exchange rate movements on
the real economy. It is also considerably superior to empirical knowledge about other aspects of
exchange rates. Goldstein and Khan (1985) provide a good summary of the effects of exchange rate
changes on U.S. foreign trade and prices. Their work suggests that a 10 percent depreciation of the
dollar should lead to a 7 to 10 percent decrease (increase) in the relative price of U.S. exports
(imports) over two years and that the estimated long-run price elasticities of U.S. exports and imports
are on average 1.19 and 1.23 percent, respectively. Thus the consensus view seems to be that a 10
percent decrease in the dollar should over time raise real U.S. exports and lower real U.S. imports by
roughly 8 to 12 percent. Moreover, since a 10 percent increase in import prices would raise U.S.
consumer prices by roughly 1.5 percent, the Goldstein and Khan survey suggests that a 10 percent
decline in the value of the dollar should raise U.S. consumer prices by 1 to 1.5 percent over time.
Bryant, Henderson, et al. (1988), using eleven large macroeconomic models to simulate the
behavior of the U.S. economy and the rest of the world's economy over six years, provide some useful
estimates of the likely effects of an exogenous change in the exchange rate and a change in the
exchange rate induced by monetary policy. On average, the eleven models predict that a gradual
exogenous (that is, not due to any policy shift) decline in the value of the dollar by 25 percent over
five years will raise real U.S. GNP by as much as 0.7 percent (by the third year), increase U.S.
consumer prices by as much as 5 percent ultimately, and improve the U.S. current account balance by
$20 billion (after four years)-all relative to baseline. The estimated effects of using U.S. monetary
policy to lower the value of the dollar by the same amount are, not unexpectedly, generally much
larger.69 On average, the models suggest that the increase in the monetary aggregates required to
lower the dollar by 25 percent over five years would raise U.S. GNP to a peak of nearly 5 percent
above baseline by the third year, after which the effect would diminish to about 2 percent above
baseline by the sixth year. Consumer prices would increase by 8 percent (more or less permanently)
as a result of the money-induced dollar depreciation. Owing to a J-curve effect, the models predict on
average that the U.S. current account balance would at first decline by more than $11 billion in the
second year and then gradually increase by nearly $4 billion relative to baseline by the sixth year.
69
The models suggest that the U.S. monetary aggregates would have to be increased by a little
more that 4 percent on average to bring about a 25 percent decline in the value of the dollar by the
fifth year.
405
Note, however, that these estimates reflect the average results of the eleven different simulation
exercises and are somewhat misleading because the individual models often differ considerably.70
Much less is known about the equally important interactions of exchange rates with asset
markets generally. Numerous experiences in recent years strongly suggest that developments in the
foreign exchange markets can spill over to domestic financial markets and that concerns about
exchange rate volatility or prospective movements in currency values can affect domestic interest rates
and even the stock markets. Indeed, the possibility that dollar fluctuations might undermine
international investors' confidence in dollar investments has at times been of significant concern to
authorities here and abroad. However, the literature has yet to provide a basis for systematically
assessing the effects of currency fluctuations on interest rates and other asset prices in particular
instances.
Appendix 2: McKinnon's and Williamson's International Monetary Reform Proposals
Ronald McKinnon's proposal for an "international standard for monetary stabilization" and
John Williamson's "target zone" proposal have serious implications for how the dollar might be used
in U.S. monetary policy.71 But their approach to the use of the dollar differs from the approach
70
Although some of the models predict that an exogenous depreciation of the dollar will have
stimulative effects on U.S. GNP, significantly larger than the average, others suggest that U.S. output
might be actually reduced instead. All the models predict that U.S. consumer prices will increase, but
the estimates range between 1 and 13 percentage points. Most of the models predict that the U.S.
current account balance will improve following an exogenous decline in the dollar, but a few predict
just the opposite. All of the models predict that U.S. monetary expansion to lower the dollar will
initially raise U.S. GNP relative to baseline, but the stimulative effects tend to weaken after a few
years and even become negative in a few cases. Most of the models predict that U.S. consumer prices
will increase following the required money supply increase, but the range of the estimates is from
minus 2 to plus 18 percentage points. Finally, although most of the models predict that the U.S.
current account balance will decline in the first few years following monetary expansion, the results
for the longer run are quite diverse. One model predicts that monetary expansion designed to lower
the dollar by 25 percent will reduce the U.S. current account balance as much as $80 billion after five
years; at the other extreme, another model predicts that the current account balance will increase by
the same amount.
71
See McKinnon (1984, 1988), Williamson (1985, 1987, 1988, 1989), Williamson and Miller
(1987), and Edison, Miller and Williamson (1987).
406
taken in the this paper in at least two fundamental ways: both McKinnon and Williamson assume a
high degree of international economic cooperation and both see a move toward exchange rate targeting
as a key element in a reform of the international economy.
At times, McKinnon's arguments about the usefulness of the dollar as a target variable seem to
be motivated by a similar concern, namely, the declining usefulness of the traditional intermediate
target variables-the money aggregates and interest rates. This generally seems to be the case with
McKinnon, for McKinnon's proposal seems to be motivated in part by a concern for the deterioration
of the traditional monetary policy targets, and its validity rests to a large degree on the argument that
the observed instability of money demand in the United States is due to the variability of exchange
rates.
But McKinnon's plan quickly goes beyond anything envisioned here. Arguing that the
observed instability of money demand not only in the United States but in the other major industrial
nations would be resolved at the global level, McKinnon proposes that the monetary authorities of the
three major industrial nations—the United States, West Germany, and Japan-cooperate to control
inflation under a system of stabilized exchange rates maintained by symmetrical unsterilized
intervention. For McKinnon, therefore, exchange rate targeting becomes a way to ensure the
symmetrical operation of monetary policy at the global level: in response to a rise in the dollar
against the yen, for example, the U.S. monetary authorities would ease their stance and the Japanese
authorities would tighten their stance in an equal and offsetting manner so that the stance of money
policy at the global level would be unaffected.
McKinnon's earlier proposals have a very monetarist flavor. Contending that the observed
instability of money demand at the national level was due to currency substitution, McKinnon (1984)
proposed that monetary authorities of the three major industrial nations cooperate to realize a growth
target for their combined money supplies under a system of fixed exchange rates by using national
targets for domestic credit expansion and symmetrical unsterilized intervention. Prompted by
persistent criticism of currency substitution and by subsequent experience (see Dornbusch 1983 for a
good critique of currency substitution), in 1988 McKinnon shifted to the view that monetary policy at
the global level should be directed toward maintaining global price stability, rather than monetary
growth, but continued to argue that exchange rate targeting was a way to deal with the observed
instability of domestic money demand. He also retained the view that stabilizing exchange rates
through unsterilized intervention would ensure symmetrical monetary policy but permitted nominal
exchange rate fluctuations within 10 percent of par to give each monetary authority greater flexibility.
407
John Williamson's "target zone" proposal is another monetary reform plan that has also been
associated with exchange rate targeting. But Williamson's proposal has very little to do with the uses
of the dollar examined in this paper, for his proposal stems primarily from dissatisfaction with the
regime of floating exchange rates. Arguing that floating exchange rates have failed to insulate the
U.S. economy from foreign economic disturbances because changes in nominal exchange rates have
led to significant changes in real exchange rates, Williamson has proposed that the Group of Seven
(G-7) industrial nations use monetary policy to prevent their real effective exchange rates from
departing significantly from equilibrium levels implied by current account objectives.
Under Williamson's earliest plans for international monetary reform, the Group of Seven (G-7)
industrial nations would use monetary policy to prevent their real effective exchange rates from
deviating by more than 10 percent from equilibrium levels implied by the mutually agreed-upon
current account objectives. When economists countered that this proposal would inevitably lead to
inflation,72 Williamson recommended that the G-7 nations a) use their individual fiscal policies to
realize national nominal income targets, b) adjust the average level of their real short-term interest
rates to stabilize the aggregate level of nominal income, and c) adjust their real short-term interest rate
differentials with supplemental intervention operations to keep their real exchange rates within the
target zones to maintain external balance. Since expected and actual domestic inflation rates would
both presumably be determined by the national fiscal and global monetary policies, the targeting of
real exchange rates through real interest rate adjustments would presumably imply adjusting nominal
interest to keep nominal exchange rates within certain variable ranges in practice.
Bibiography
Adams, Charles, and Daniel Gros. 1986. "The Consequences of Real Exchange Rates for Inflation."
IMF Staff Papers, vol. 33, no. 3, pp. 439-76.
Adams, Donald B., and Dale W. Henderson. 1983. Definition and Measurement of Exchange Market
Intervention. Board of Governors of the Federal Reserve System, Staff Study no. 126.
72
See Adams and Gros (1986).
408
Alexander, Sidney. 1959. "The Effects of Devaluation: A Simplified Synthesis of the Elasticities and
the Absorption Approach." American Economic Review, vol. 44, no. 2, pp. 22-42.
Alexander, Don, and Lee R. Thomas. 1987. "Monetary/Asset Models of Exchange Rate Determination:
How Well Have They Performed in the 1980s?" International Journal of Forecasting, vol. 3, pp. 5363.
Allen, Polly R., and Peter Kenen. 1980. Asset Markets, Exchange Rates, and Economic Integration.
New York: Cambridge University Press.
Arndt. Svendt W., and Charles Pigott. 1985. "Exchange Rate Models: Evolution and Policy
Implications." In Svendt W. Amdt, Richard J. Sweeney, and Thomas D. Willett, eds., Exchange
Rates, Trade, and the U.S. Economy. Cambridge, Mass.: Ballinger Press.
Artus, Jacques R. 1976. "Exchange Rate Stability and Managed Floating: The Experience of the
Federal Republic of Germany." IMF Staff Papers, vol. 23, pp. 312-34.
Board of Governors of the Federal Reserve System. 1984. Federal Reserve System Purpose and
Functions, October 1984 edition.
Bergstrand, Jeffery. 1985a. "Exchange Rate Variation and Monetary Policy." Federal Reserve Bank of
Boston New England Economic Review, May-June, pp. 5-18.
. 1985b. "Money, Interest Rates, and Exchange Rates as Indicators for Monetary Policy."
Federal Reserve Bank of Boston New England Economic Review, November-December, pp. 3-13.
Bhandari, Jagdeep S., Robert Driskill, and Jacob A. Frenkel. 1984. "Capital Mobility, and Exchange
Rate Overshooting." European Economic Review, vol. 24, pp. 309-20.
Bilson, John F. O. 1978a. "The Monetary Approach to the Exchange Rate: Some Empirical
Evidence." IMF Staff Papers, vol. 25, no. 1, pp. 48-75.
. 1978b. "Rational Expectations and the Exchange Rate." In Jacob A. Frenkel and Harry G.
Johnson, eds., The Economics of the Exchange Rate, pp. 75-96. Reading, Mass.: Addison-Wesley
Press.
Boughton, James M. 1983. "Conditions for an Active Exchange Rate Policy with a Predetermined
Monetary Target." IMF Staff Papers, vol. 30, pp. 461-90.
. 1987. "Test of the Performance of Reduced-Form Exchange Rate Models." Journal of
International Economics, vol. 23, pp. 41-56.
Boyer, Russell S. 1987. "Financial Polices in Open Economics." Econometrica, vol. 45, pp. 39-57.
Branson, William H. 1972. "Stocks and Rows in International Monetary Analysis." In Albert Ando,
Richard Herring, and Richard Marston, eds., International Aspects of Stabilization Policies,
pp. 27-54. Federal Reserve Bank of Boston.
409
. 1977. "Asset Markets and Relative Prices in Exchange Rate Determination."
Sozialwissenschaftliche Annalen, 1, pp. 69-89.
. 1979. "Exchange Rate Dynamics and Monetary Policy." In A. Linbeck, ed., Inflation and
Employment in Open Economies. Amsterdam: North-Holland Press.
Branson, William H., and Dale Henderson. 1985. "Specification and Influence of Asset Markets." In
Ronald W. Jones and Peter B. Kenen, eds., Handbook for International Economics, vol. 2,
pp. 749-806. Elsevier Science Publishers.
Branson, William H., Hannu Haltunnen, and Paul Mason. 1977. "Exchange Rates in the Short-run: The
Dollar-Deutsche Mark Rate." European Economic Review, vol. 10, pp. 303-24.
Brittain, Bruce. 1977. "Tests of Theories of Exchange Rate Determination." Journal of Finance,
vol. 32, no. 2, pp. 519-29.
Bryant, Ralph C , Dale W. Henderson, Gerald Holtham, Peter Hooper, and Steven A. Sysmansky, eds.
1988. Empiricial Models for Interdependent Economies. Washington, D. C: Brookings Institute.
Bryant, Ralph C , Gerald Holtham, and Peter Hooper. 1988. External Deficits and the Dollar: the Pit
and the Pendulum. Washington, D.C.: Brookings Institution.
Campbell, John, and Richard Clarida. 1986. "The Dollar and Real Interest Rates." Paper prepared for
Carnegie-Rochester Conference on Public Policy, November 21-22, 1986.
Chancellor of the Exchequer of the United Kingdom. 1982. Financial Statement and Budget Report.
Coleby, Anthony, and John Ryding. 1988. "Exchange Market Intervention and Monetary Policy in the
United Kingdom." In Exchange Market Intervention and Monetary Policy. Basle: Bank for
International Settlements.
Cornell, Bradford. 1977. "Spot Rates, Forward Rates, and Market Efficiency." Journal of Finance,
vol. 5, pp. 55-65.
. 1982. "Money Supply Announcements, Interest Rates and Foreign Exchange." Journal of
International Money and Finance, pp. 201-08.
Cooper, Richard N. 1976. "Monetary and Fiscal Policy in an Open Economy." Scandinavian Journal of
Economics, vol. 78, pp. 146-63.
Dankar, Deborah J., Richard A. Hass, Dale W. Henderson, Steven A. Symanski, and Ralph W. Tryon.
1983. Small Empirical Models of Exchange Market Intervention: Applications to Canada, Germany,
and Japan. Board of Governors of the Federal Reserve System, Staff Study no. 135.
Dominguez, Kathryn. 1986. "Are Foreign Exchange Forecasts Rational? New Evidence from Survey
Data." Economics Letters, vol. 21, pp. 277-81.
410
. 1988. "The Informational Role of Foreign Exchange Intervention: An Empirical
Investigation." Harvard University, mimeo.
. 1989. "Market Responses to Coordinated Central Bank Intervention." Harvard University,
mimeo.
Dominguez, Kathryn M., and Jeffrey A. Frankel. 1989. "Some New Tests of Foreign Exchange
Intervention: the Portfolio and Expectations Effects." Harvard University, Kennedy School of
Government.
Dooley, Michael F., and Peter Isard. 1982. "A Portfolio-Balance Rational Expectations Model of the
Dollar-Mark Exchange Rate." Journal of International Economics, pp. 257-76.
. 1983. "The Portfolio Balance Approach and Some Estimates of the Risk Premium in the
Foreign Exchange Market." IMF Staff Papers, vol. 30, pp. 683-702.
Dooley, Michael P., and Jeffrey Shafer. 1983. "Analysis of Short-run Exchange Rate Behavior:
March, 1973 to November, 1981." In David Bigman and Teizo Taya, eds., Exchange Rate and Trade
Stability. Cambridge, Mass.: Ballinger Press.
Dornbusch, Rudiger. 1976. "Expectations and Exchange Rate Dynamics." Journal of Political
Economy, vol. 84, pp. 1161-76.
. 1980a. "Exchange Rate Economics: Where do We Stand?" Brookings Papers on Economic
Activity. 1, pp. 145-85.
. 1980b. Open Economy Macroeconomics. New York: Basic Books.
. 1983. "Flexible Exchange Rates and Interdependence." IMF Staff Papers, vol. 30, no. 1.
. 1987a. "Exchange Rate Economics: 1986." Economic Journal, vol. 97, pp. 1-18.
. 1987b. "Exchange Rates and Prices." American Economic Review, vol. 77, pp. 93-106.
. 1988. "Doubts about the McKinnon Standard." Journal of Economic Perspectives, vol. 2,
no. 1, pp. 105-13.
Dornbusch, Rudiger, and Stanley Fischer. "Exchange Rates and the Current Account." American
Economic Review, vol. 70, pp. 960-71.
Driskill, Robert. 1981. "Exchange Rate Dynamics: An Empirical Investigation." Journal of Political
Economy, vol. 89, pp. 357-71.
Edison, Hali. 1990. "Foreign Currency Operations: An Annotated Bibliography." Board of Governors
of the Federal Reserve System, International Finance Discussion Papers, no. 380.
411
Edison, Hali, Marcus Miller, and John Williamson. 1987. "On Evaluating and Extending the Target
Zone Proposal." Journal of Policy Modeling, vol. 9, pp. 119-224.
Engel, Charles M. 1987. "Multivariate ARCH with Factor Structures-Cointegration in Variances."
University of California, San Diego.
Engel, Charles M., and Jeffrey A. Frankel. 1984a. "Why Interest Rates React to Monetary
Announcements: An Explanation from the Foreign Exchange Market." Journal of International
Money and Finance, pp. 31-39.
. 1984b. "Do Asset-Demand Functions Optimize over the Mean and Variance of Real Returns?
A Six Currency Test." Journal of International Economics, vol. 17, pp. 309-23.
Engel, Charles M., and Anthony Rodrigues. 1989. "Tests of International CAPM with Time-Varying
Covariances." Journal of Applied Econometrics, vol. 4, pp. 119-38.
Federal Reserve Bank of New York. 1987. Research Papers on International Integration of Financial
Markets and U.S. Monetary Policy.
Federal Reserve Bank of New York, Foreign Exchange Department. "Treasury and Federal Reserve
Foreign Exchange Operations." Quarterly Report, various issues.
Federal Reserve Bank of New York, Open Market Analysis Division. 1988. "Monetary Policy and
Open Market Operations during 1987." Quarterly Review, vol. 13, no. 1 (Spring).
Federal Open Market Committee. "Record of Policy Action of the Federal Open Market Committee."
Federal Reserve Bulletin, Board of Governors of the Federal Reserve System, various issues.
Feldstein, Martin. 1989. "The Case Against Trying to Stabilize the Dollar." American Economic
Review, vol. 79, no. 2, pp. 36-40.
Fleming, J. Marcus. 1962. "Domestic Financial Policies Under Fixed and Floating Exchange Rates."
IMF Staff Papers, vol. 9, pp. 369-80.
Foley, Duncan, and Miguel Sidrauski. 1970. "Portfolio Choice, Investment and Growth." American
Economic Review, vol. 60, pp. 44-63.
Frankel, Jeffrey A., 1979a. "The Diversification of Exchange Rate Risk." Journal of International
Economics, pp. 379-92.
. 1979b. "On The Mark: A Theory of Floating Exchange Rates Based on Real Interest Rate
Differentials." American Economic Review, vol. 69, pp. 610-22.
. 1982. "In Search of the Exchange Risk Premium: A Six Currency Test Assuming MeanVariance Optimization." Journal of International Money and Finance, vol. 1, pp. 255-74.
412
. 1983. "Monetary and Portfolio Balance Models of Exchange Rate Determination." In
J.S. Bhandari and B.H. Putman, eds., Economic Interdependence and Flexible Exchange Rates.
Cambridge: MIT Press.
. 1985. "The Dazzling Dollar." Brookings Papers on Economic Activity, pp. 199-217.
Frankel, Jeffrey A., and Kenneth A. Froot. 1987. "Using Survey Data to Test Standard Assumptions
Regarding Exchange Rate Expectations." American Economic Review, vol. 77, pp. 133-53.
Frenkel, Jacob. 1976. "A Monetary Approach to the Exchange Rate: Doctrinal Aspects and Empirical
Evidence." Scandinavian Journal of Economics, vol. 78, pp. 200-24.
. 1980. "The Collapse of Purchasing Power Parity in the 1970s." European Economic
Review, vol. 16, pp. 145-65.
. 1981. "Flexible Exchange Rates, Prices, and the Role of 'News': Lessons from the
1970's." Journal of Political Economy, vol. 89, pp. 665-709.
Frenkel, Jacob, and Morris Goldstein. 1986. "A Guide to Target Zones." IMF Staff Papers, vol. 33,
pp. 633-73.
Frenkel, Jacob, and Assaf Razin. 1987. "The Mundell-Fleming Model: A Quarter Century Later."
National Bureau of Economic Research, Working Paper no. 2321.
Fuhrer, Jeffrey, and George Moore. 1989. "Monetary Policy Rules and the Indicator Properties of
Asset Prices." Board of Governors of the Federal Reserve System, Finance and Economics
Discussion Papers, no. 89.
Funabashi, Yoichi. 1988. Managing the Dollar: From the Plaza to the Louvre. Washington, D.C.:
Institute for International Economics.
Furlong, Frederick T. 1989. "International Dimensions of U.S. Economic Policy in the 1980s." Federal
Reserve Bank of San Francisco Economic Review, no. 2 (Spring), pp. 3-16.
Girton, Lance, and Dale Henderson. 1977. "Central Bank Operations in Foreign and Domestic Assets
under Fixed and Flexible Exchange Rates." In Peter B. Clark, Dennis Logue, and Richard Sweeney,
eds., The Effects of Exchange Rate Adjustments. Washington, D.C.: Government Printing Office.
Glick, Reuven, and Michael Hutchinson. 1989. "Exchange Rates and Monetary Policy." Federal
Reserve Bank of San Francisco Economic Review, Spring, pp. 17-29.
Goldstein, Morris, and Moshin Khan. 1985. "Income and Price Effects in Foreign Trade." In
Ronald W. Jones and Peter B. Kenen, eds., Handbook of International Economics, vol. 2,
pp. 1041-105. Elsevier Science Publishers.
Greene, Margaret L. 1983a. "U.S. Experience with Exchange Market Intervention: January-March
1975." Board of Governors of the Federal Reserve System, Staff Study no. 127.
413
. 1983b. "U.S. Experience with Exchange Market Intervention: September 1977-December
1979." Board of Governors of the Federal Reserve System, Staff Study.
. 1983c. "U.S. Experience with Exchange Market Intervention: October 1980-October
1981." Board of Governors of the Federal Reserve System, Staff Study no. 129.
Group of Five Industrial Countries. 1985. "Communique." September.
Hakkio, Craig S. 1984. "A Re-examination of Purchasing Power Parity." Journal of International
Economics, vol. 17, pp. 265-77.
Hakkio, Craig, and Douglas Pearce. 1985. "The Reaction of Exchange Rates to Economic News."
Federal Reserve Bank of Kansas City, Working Paper no. 8501.
Hardouvelis, Gikas. 1984. "Market Perceptions of Federal Reserve Policy and the Weekly Monetary
Announcements." Journal of Monetary Economics, pp. 225-40.
. 1988. "Economic News, Exchange Rates, and Interest Rates." Journal of International
Money and Finance, March, pp. 23-36.
Heller, H. Robert. 1988. "The Making of U.S. Monetary Policy." Paper delivered at the Seminar for
Economic Policy, the University of Cologne, West Germany.
Helpman, Ehanan, and Assaf Razin. 1982. "A Comparison of Exchange Rate Regimes in the Presence
of Imperfect Capital Markets." International Economic Review, vol. 23, pp. 365-88.
Henderson, Dale. 1979a. "Financial Policies in Open Economics." Board of Governors of the Federal
Reserve System, International Finance Discussion Papers, no. 133.
. 1979b. "The Dynamics Effects of Exchange Market Intervention Policy: Two Extreme
Views and a Synthesis." Board of Governors of the Federal Reserve System, International Finance
Discussion Papers, no. 142.
. 1984. "Exchange Market Intervention Operations: Their Role in Financial Policy and
their Effects." In John Bilson and Richard Marston, eds., Exchange Rate Theory and Practice,
National Bureau of Economic Research Conference Volume. Chicago: University of Chicago Press.
Hooper, Peter. 1984. "Macroeconomic Effects of Exchange Rate Changes: Some Quantitative
Measures." Board of Governors of the Federal Reserve System.
. 1988. "Exchange Rates and U.S. External Adjustment in the Short and Long Run."
Brookings Discussion Papers in International Economics, no. 65. Washington, D.C.: Brookings
Institute.
Hooper, Peter, and Barbara Lowrey. 1979. "Impact of the Dollar Depreciation on the U.S. Price Level:
An Analytical Survey of Empirical Estimates." Board of Governors of the Federal Reserve System,
International Finance Discussion Papers, no. 128.
414
Hooper, Peter, and Catherine Mann. 1988. "The Emergence and Persistence of the U.S. External
Balance: 1980-1987." Princeton Studies in International Finance.
Hooper, Peter, and John Morton. 1980. "Fluctuations in the Dollar: A Model of Nominal and Real
Exchange Rate Determination." Federal Reserve Board of Governors, International Finance
Discussion Papers, no. 168.
Home, Jocelyn. 1980. "The Asset Model for the Balance of Payments and the Exchange Rate: A
Survey of the Empirical Evidence." Journal of International Money and Finance, vol. 2, no. 2
(August), pp. 89-110.
Hsieh, David A. 1988. "The Statistical Properties of Daily Foreign Exchange Rates: 1974-1983."
Journal of International Economics, vol. 24, pp. 129-45.
Huang, Roger D. 1981. "The Monetary Approach to Exchange Rates in an Efficient Market: Tests
Based on Volatility." Journal of Finance, vol. 36, pp. 31-41.
Huizinga, John, and Frederick S. Mishkin. 1986. "Monetary Policy Regime Shifts and the Unusual
Behavior of Real Interest Rates." Carnegie Rochester Series on Public Policy, no. 24, pp. 231-74.
Hutchinson, Michael and Adrian Throop. 1985. "U.S. Budget Deficit and the Real Value of the
Dollar." Federal Reserve Bank of San Francisco Economic Review, no. 4, pp. 26-43.
Isard, Peter. 1977. "How Far Can We Push the Law of One Price?" American Economic Review,
vol. 67, pp. 942-48.
. 1978. "Exchange Rate Determination: A Survey of Popular Views and Recent Models."
Princeton Studies in International Finance, no. 42.
. 1980. "Factors Determining Exchange Rates: The Role of Prices, Interest Rates, the Current
Account and Risk." Board of Governors of the Federal Reserve System, International Finance
Discussion Papers, no. 171.
. 1987. "Lessons from Empirical Models of Exchange Rates." IMF Staff Papers, vol. 34,
no. 1, pp. 1-28.
Johnson, Harry. 1978a. "Towards a General Theory of the Balance of Payments." In Jeffrey A. Frankel
and Harry Johnson, eds., The Monetary Approach to the Balance of Payments, pp. 46-63. London:
George Allen and Unwin.
. 1978b. "The Monetary Approach to Balance of Payments Theory." In Jacob A. Frenkel
and Harry Johnson, eds., The Monetary Approach to the Balance of Payments, pp. 147-67. London:
George Allen and Unwin.
Johnson, Manuel. 1988. "Current Perspectives on Monetary Policy." Cato Journal, Fall, pp. 253-60.
415
Kasman, Bruce, and Charles Pigott. 1987a. "U.S. Interest Rates and Exchange Rates: Analysis and
Interpretation." In Research Papers on International Integration of Financial Markets and U.S.
Monetary Policy. Federal Reserve Bank of New York.
. 1987b. "An Examination of Interest Rate Divergences among Major Industrial Countries."
In Research Papers on International Integration of Financial Markets and U.S. Monetary Policy,
Federal Reserve Bank of New York.
Kenen, Peter B. 1985. "Macroeconomic Theory and Policy: How the Closed Economy was Opened."
In Ronald W. Jones and Peter Kenen, eds., Handbook of International Economics, vol. 2, pp. 625-77.
Amsterdam: North-Holland Press.
. 1987. "Exchange Rate Management: What Role for Intervention?" American Economic
Review, vol. 72, no. 2, pp. 194-99.
Keran, Michael, and Steven Zeldes, 1980. "Effects of Monetary Disturbances on Exchange Rates,
Inflation and Interest Rates." Federal Reserve Bank of San Francisco Economic Review, pp. 7-29.
Kiguel, Miguel A., 1987. "The Non-dynamic Equivalence of Monetary Policy and Exchange Rate
Rules under Perfect International Capital Mobility and Rational Expectations." Journal of
International Money and Finance, vol. 6, pp. 207-14.
Kimbrough, Kent. 1983. "The Information Content of the Exchange Rate and the Stability of Real
Output under Alternative Exchange Rate Regimes." Journal of International Money and Finance,
April.
Koch, Paul D., Jeffrey A. Rosenweig, and Joseph A. Whitt, Jr. 1988. "The Dynamic Relationship
between the Dollar and U.S. Prices: An Empirical Relationship." Journal of International Money and
Finance, vol. 7, pp. 181-204.
Kouri, Pentti J. K. 1976. "The Exchange Rate and the Balance of Payments on the Short Run and the
Long Run." Scandinavian Journal of Economics, vol. 78, pp. 280-304.
Kouri, Pentti J. K, and Michael G. Porter. 1974. "International Capital Flows and Portfolio
Equilibrium." Journal of Political Economy, vol. 82, pp. 443-67.
Laursen, Sven, and Lloyd Metzler. 1950. "Flexible Exchange Rates and the Theory of Employment."
Review of Economics and Statistics, vol. 32, pp. 281-99.
Levich, Richard M. 1985. "Empirical Studies of Exchange Rates." In Ronald W. Jones and Peter B.
Kenen, eds., Handbook for International Economics, vol. 2, pp. 979-1040. Elsevier Science
Publishers.
Lewis, Karen. 1988. "Inflation and Asset Market Disturbances: The Mean-Variance Model Revisited."
Journal of International Money and Finance, pp. 273-88.
Logue, Dennis E., and Richard J. Sweeney. 1977. "White Noise in Imperfect Markets: The Case of the
Franc/Dollar Exchange Rate." Journal of Finance, vol. 31, pp. 761-68.
416
Logue, Dennis E., Richard J. Sweeney, and Thomas D. Willett. 1978. "Speculative Behavior of
Foreign Exchange Rates during the Float." Journal of Business Research, vol. 6, pp. 159-74.
Loopesko, Bonnie E. 1983. "Relationship among Exchange Rates, Intervention, and Interest Rates."
Board of Governors of the Federal Reserve System, Staff Study no. 133.
Lown, Cara. 1989. "Interest Rate Spreads, Commodity Prices, and the Dollar: A New Strategy for
Monetary Policy." Federal Reserve Bank of Dallas Economic Review, pp. 13-26.
Machlup, Fritz. 1956. "The Terms of Trade Effects of Devaluation upon Real Income and the Balance
of Trade." Kvklos, vol. 9, pp. 417-52.
Marquez, Jaime. 1988. "Income and Price Elasticities of Foreign Trade Flows: Econometric
Estimation and Analysis of the U.S. Trade Deficit." Board of Governors of the Federal Reserve
System, International Finance Discussion Papers, no. 324.
Marston, Richard C. 1985. "Stabilization Policies in Open Economies." In Ronald W. Jones and
Peter B. Kenen, eds., Handbook of International Economics, vol. 2, pp. 859-916. Amsterdam:
North-Holland Publishers.
McCallum, Bennett. 1988. "Targets, Indicators and Instruments of Monetary Policy." American
Institute Conference on Monetary Policy in an Era of Change, October.
McKibbon, Warwick, and Jeffrey Sachs. 1988. "Comparing the Global Performance of Alternative
Exchange Arrangements." Journal of International Money and Finance, vol. 7, pp. 387-410.
McKinnon, Ronald. 1984. "An International Standard for Monetary Stabilization." Institute for
International Economics, Policy Analysis in International Economics, no. 8.
. 1988. "Monetary and Exchange Rate Policies for International Financial Stability."
Journal of Economic Perspectives, vol. 2, no. 1, pp. 83-104.
Meade, Ellen E., 1988, "Exchange Rates, Adjustment, and the J-Curve." Federal Reserve Bulletin,
vol. 74, pp. 633-44.
Meade, James E. 1951. The Theory of International Economic Policy. New York: Oxford University
Press.
Meese, Richard A., and Kenneth Rogoff. 1983a. "Empirical Exchange Rate Models of the Seventies:
Do They Fit Out of Sample?" Journal of International Economics, vol. 14, pp. 3-24.
. 1983b. "The Out-of-Sample Failure of Empirical Exchange Rate Models: Sampling Error or
Misspecification?" In Jacob A. Frenkel, ed., Exchange Rates and International Macroeconomics.
Chicago: University of Chicago Press.
Mishkin, Frederic. 1989. "The Information in the Longer Maturity Term Structure about Future
Inflation." National Bureau of Economic Research, Working Paper no. 3126.
417
Mundell, Robert A. 1962. "The Appropriate Use of Monetary and Fiscal Policy for Internal and
External Stability." IMF Staff Papers, vol. 9, pp. 770-79.
. 1963. "Capital Mobility and Stabilization Policy under Fixed and Flexible Exchange
Rates." Canadian Journal of Economic and Political Science, pp. 475-85.
. 1968. International Economics. New York: MacMillan Company.
Mussa, Michael. 1977. "Empirical Regularities in the Behavior of Exchange Rates and Theories of the
Exchange Market." In K. Brenner and A. H. Meltzer, eds., Policies for Employment, Prices, and
Exchange Rates, pp. 9-57. Amsterdam: North-Holland Press.
. 1982. "A Model of Exchange Rate Dynamics." Journal of Political Economy, vol. 90,
pp. 74-104.
Niehans, Jurg. 1977. "Exchange Rate Dynamics with Stock/How Interaction." Journal of Political
Economy, vol. 85, pp. 1245-57.
Obstfeld, Maurice. 1980. "Imperfect Asset Substitutability and Monetary Policy Under Fixed Exchange
Rates." Journal of International Economics, pp. 177-200.
. 1985. "Floating Exchange Rates: Experience and Prospects." Brookings Papers on Economic
Activity, vol. 2, pp. 369-450.
. 1988. "The Effectiveness of Foreign-Exchange Intervention: Recent Experience." National
Bureau of Economic Research, Working Paper no. 2796.
Obstfeld, Maurice, and Alan C. Stockman. 1985. "Exchange-rate Dynamics." In Ronald W. Jones and
Peter B. Kenen, eds., Handbook for International Economics, vol. 2, pp. 917-77. Elsevier Science
Publishers.
Officer, Lawrence H., and Thomas D. Willett. 1979. "The Covered-arbitrage Schedule: A Critical
Survey of Recent Developments." Journal of Money, Credit, and Banking, vol. 2, pp. 229-48.
Papell, David. 1984. "Anticipated and Unanticipated Disturbances: The Dynamics of the Exchange
Rate and the Current Account." Journal of International Money and Finance, vol. 3, no. 2,
pp. 179-94.
. 1986. "Exchange Rate and Current Account Dynamics Under Rational Expectations: An
Econometric Analysis." International Economic Review, vol. 27, pp. 583-600.
Pigott, Charles. 1984. "Indicators of Long-Term Real Interest Rates." Federal Reserve Bank of San
Francisco Economic Review, no. 1, pp. 45-63.
Pigott, Charles, and Richard J. Sweeney. 1985. "Testing the Exchange Rate Implications of Two
Popular Monetary Models." In S.W. Arndt, Richard J. Sweeney, and Thomas D. Willett, eds.,
Exchange Rates, Trade and the U.S. Economy, pp. 91-106. Washington, D.C.: American Enterprise
Institute.
418
Pigott, Charles, John Rutledge, Thomas Willett. 1985. "Estimating the Inflationary Effects of Exchange
Rate Changes." In S.W. Arndt, Richard J. Sweeney, and Thomas D. Willett, eds., Exchange Rates,
Trade and the U.S. Economy, pp. 245-65. Washington, D.C.: American Enterprise Institute.
Poole, William. 1970. "Optimal Choice of Monetary Policy Instruments in a Simple Stochastic Macro
Model." Quarterly Journal of Economics, vol. 84, pp. 197-216.
Radecki, Lawrence, and Vincent Reinhart. 1987. "Capital Mobility and Short-Term Interest Rate
Control." In Research Papers on International Integration of Financial Markets and U.S. Monetary
Policy, Federal Reserve Bank of New York.
Rogoff, Kenneth. 1983. "Time-Series Studies of the Relationship Between Exchange Rates and
Intervention: A Review of the Literature." Board of Governors of the Federal Reserve System, Staff
Study no. 132.
. 1984. "On the Effects of Sterilized Intervention. An Analysis of Weekly Data." Journal of
Monetary Economics.
Sachs, Jeffrey D. 1985. "The Dollar and Policy Mix: 1985." Brookings Papers on Economic Activity,
pp. 117-97.
Schadler, Susan. 1977. "Sources of Exchange Rate Variability: Theory and Empirical Evidence." IMF
Staff Papers, vol. 24, no. 2, pp. 253-96.
Shafer, Jeff R., and Bonnie E. Loopesko. 1983. "Floating Exchange Rates after Ten Years." Brooking
Papers on Economic Activity, 1.
Shafer, Jeff, and Edwin Truman. 1981. "International Portfolio Disturbances and Domestic Monetary
Policy." Board of Governors of the Federal Reserve System, International Finance Discussion Papers,
no. 175.
Smith, Ralph, and Brian Madigan. "Exchange Market Management and Monetary Policy in the United
States." In Exchange Market Intervention and Monetary Policy. Basle: Bank for International
Settlements.
Taylor, Mark P. 1988. "An Empirical Examination of Long-run Purchasing Power Parity Using
Cointegration Techniques." Applied Economic Journal, vol. 20, pp. 1369-81.
Tobin, James. 1969. "A General Equilibrium Approach to Monetary Theory." In Essays, chap. 18.
Truman, Edwin. 1989. "International Economic Policy: The Role of Exchange Rates." Board of
Governors of the Federal Reserve System, International Finance Discussion Papers, no. 356.
Truman, Edwin, et al. 1981. "The New Federal Reserve Operating Procedures: An External
Perspective." Staff Review of Monetary Control Procedures, vol. 2, Board of Governors of the
Federal Reserve System, February.
419
Tryon, Ralph W. 1983. "Small Empirical Models of Exchange Market Intervention: A Review of the
Literature." Board of Governors of the Federal Reserve System, Staff Study no. 134.
Urich, Thomas, and Paul Wachtel. 1981. "The Informational Content of Weekly Money Supply
Announcements in the 1970s." Journal of Finance, vol. 36, pp. 1063-72.
Whitt, Joseph A., Paul D. Koch, and Jeffrey Rosensweig. 1986. "The Dollar and Prices: An Empirical
Analysis." Federal Reserve Bank of Atlanta Economic Review, pp. 4-18.
Williamson, John. 1985. "The Exchange Rate System." Institute for International Economics, Policy
Analysis on International Economics, no. S.
. 1987. "Exchange Rate Management: The Role of Target Zones." American Economic
Review, vol. 77, no. 2, pp. 200-04.
. 1988. "On McKinnon's Monetary Rule." Journal of Economic Perspectives, vol. 2, no. 1,
pp. 114-20.
. 1989. "The Case for Roughly Stabilizing the Real Value of the Dollar." American
Economic Review, vol. 79, no. 2, pp. 41-54.
Williamson, John, and Marcus H. Miller. 1987. "Targets and Indicators: A Blueprint for the
International Coordination of Economic Policy." Institute for International Economics, Policy
Analysis on International Economics, no. 22.
Willett, Thomas. 1985. "The Dollar Exchange Rate as a Leading Indicator for American Monetary
Policy: Comments." Proceedings of the Federal Reserve Bank of San Francisco Academic
Conference, Fall 1985.
Willett, Thomas, and Francesco Forte. 1969. "Interest Rate Policy and External Balance." Quarterly
Journal of Economics, pp. 242-62.
Woodward, Robert S. 1988. "The Effect of Relative Monetary Surprises on Spot Exchange Rates."
Applied Economics, vol. 20, pp. 1129-36.
Working Group on Exchange Market Intervention. 1983. "Report of the Working Group on Exchange
Market Intervention," Philippe Jorgensen, chairman.
Yeager, Leland B. 1976. International Monetary Relations: Theory, History, and Policy. New York:
Harper and Row.
420
OPTIMAL MONETARY POLICY DESIGN: RULES VS. DISCRETION AGAIN
A. Steven Englander
This paper reviews some recent academic literature on "optimal" monetary policy design. Over
the last fifteen years, the entire direction of the debate on optimal monetary policy has been reversed.
Earlier literature assumed that an optimal monetary policy could be devised by solving a "dynamic
optimization" problem. It assumed that given a set of policy objectives and a model of the economy,
the optimal path of inflation and GNP could be obtained.1 The current literature argues that it is
precisely the public's recognition that policy makers engage in such optimization that leads the public
to expect positive inflation in equilibrium, even when both the public and government view any
inflation as undesirable. According to this argument, the public will perceive that the authorities are
willing to exploit a trade-off between inflation and output, and it will adjust its expectations
accordingly.
The main thrust of this literature is to explain how a positive inflation rate can emerge on
average, even when all parties view this as an inferior outcome that produces no extra output. The
phenomenon that the authors are trying to explain is readily apparent: inflation has averaged above
zero in all OECD countries in the postwar period, but few policy makers or economists believe that
these inflation rates have contributed to economic well-being. The persistence of inflation at above
desired levels in most OECD countries has led analysts to propose economic models yielding inflation
as an equilibrium phenomenon explained by optimizing behavior on the part of the public and policy
makers rather than happenstance.
The proposed explanation is that the public views policy making as opportunistic: policy
makers are willing to exploit a short-run inflation/output trade-off even if a long run trade-off neither
exists nor is thought to exist. This explanation also yields a strong policy conclusion. If the public
expects positive inflation because it believes that policy makers are trying to exploit this trade-off, the
key to lowering actual and expected inflation is to guarantee that no such exploitation will occur. The
mechanism by which this can be accomplished is to propose a readily visible rule that eliminates
1
For example, Kalchbrenner and Tinsley (1976).
421
policy makers' discretion to inflate opportunistically. In large part this literature argues that the mere
ability of policy makers to use discretion, even if the discretion is not actually exercised, will lead the
public to expect positive inflation. Hence, the new monetary policy literature examines the old
question of "rules versus discretion" from a new perspective.
The argument that the structure of the monetary policy making process (that is the presence of
discretion) rather than the conduct of monetary policy is the source of inflationary bias also points, as
the authors see it, to the solution. Changing the structure of policy making to one guided by formal
rules, they contend, might yield lower inflation on average with a relatively small loss of output.
Implicit in such a recommendation is the assumption that only small losses will arise because the rule
prevents policy makers from responding to shocks or disturbances. Advocates of such rules generally
argue that feedback mechanisms can be incorporated into the rules to offset shocks and that the
magnitude of such shocks might be lower if a consistent noninflationary policy rule were installed.2
The alternative explanation of the prevalence of inflation in recent decades is that it was caused
by a combination of mistaken policies and adverse shocks, and subsequently compounded by the
unwillingness of policy makers to accept the output costs of disinflation through much of the 1970s.
Changing the structure of policy making, in this view, would not accomplish much if the public and
policy makers were unwilling to accept the costs of policies aimed at lowering inflation.
Such considerations have a direct connection with the issues surrounding the use of
"intermediate targets" for monetary policy. It can be argued that if policy makers do not have a
reputation for maintaining low inflation, they may find it necessary to pursue an intermediate target
rule that can be monitored easily and on a timely basis by the public. This course may involve some
loss of output or inflation control if the intermediate target is imperfectly linked to the final objectives.
Nevertheless, the visible pursuit of a nominal intermediate target may provide sufficient offsetting
benefits in the form of improved credibility and lowered inflation expectations to offset the imperfect
linkage. In one sense, the rules-versus-discretion question involves comparing the losses from the
imperfect linkages of intermediate targets to final objectives under a rules mechanism with the losses
due to the inflation bias alleged to arise from discretion.
The focus of this paper, however, is the interaction between the policy makers' goals and the
public's expectations and behavior in response to these goals. Thus, intermediate targets will be
2
Most of the literature is theoretical and does not make any effort to calculate the benefits or
losses from adhering to a rule. In a series of articles, McCallum proposes a specific feedback rule and
attempts to estimate the loss from it. See, for example, McCallum 1987.
422
discussed again only as a potential means for improving credibility. Much of the discussion below
will assume that policy can successfully hit not only intermediate targets but also ultimate goal
variables, such as inflation or nominal income growth. More specifically, the discussion will assume
that policy makers can achieve their long-run inflation target and hit an output target temporarily by
exploiting a short-term inflation/output trade-off. Over the long term it is assumed that output growth
is at trend and is independent of policy.3
Recent literature has also examined the question of optimal monetary policy under conditions of
considerable uncertainty about the structure of the economy and the policy makers' ability to hit
targets on a period-by-period basis. In the face of such uncertainty, some results are weakened
because the public, as might be expected, finds it more difficult to distinguish policy moves from
random shocks—and to distinguish policy makers who are inflation prone from those who are not.
Uncertainty about the structure of the economy also generally makes strict adherence to rules
undesirable because it is difficult to design rules suitable under a broad range of conditions; in forming
inflation expectations during periods of uncertainty, the public will usually place more weight on the
policy makers' past inflation record.
In recent years several other reviews of the literature on optimal monetary policy design have
appeared.4 This paper seeks to emphasize the intuition underlying the models (Sections I, IIA, and
IE) and to evaluate some of the suggested mechanisms for achieving credible policies (Section IIC).
After examining the lessons that can be drawn for policy, the paper argues that the policy relevance of
this literature has been overstated. The theoretical insights emerging from this literature differ little
from those of the earlier literature, and are achieved only at the cost of analytical assumptions that are
difficult to sustain empirically (Section IIB). Moreover, outside of a few extraordinary episodes, it is
very difficult to find any concrete illustrations of the recent literature's key policy prediction that a
credible disinflation can be relatively costless (Section IV).
3
This assumption can be identified with the Lucas supply curve, which is common in the
literature. See Lucas (1972), and Barro and Gordon (1983b).
4
Other nontechnical discussions can be found in Barro (1985), Blinder (1987), and Fischer (1989).
Somewhat more technical but largely readable are the surveys in Blackburn and Christensen (1989),
Persson (1988), McCallum (1984), and Rogoff (1987). Of these discussions, Barro's and McCallum's
are most sympathetic to the policy thrust of the literature, and Blinder's the least.
423
I. Some Terminology
The recent literature on optimal monetary policy is difficult for nonspecialists to read, in part
because the terminology is difficult. This section reviews the terminology and puts it in the context of
the issues to be discussed in greater detail subsequently.3
In common language, a "consistent" policy is one that follows a well-defined set of rules over
time. It would normally be viewed as superior to an "inconsistent" policy. Because the new literature
on monetary policy emerged out of the earlier optimal control literature, the common usage has been
altered. A time consistent policy is one that results from solving a long-term dynamic optimization
problem without incorporating the effect of current policy actions on the public's expectations of the
future.6 The "consistency" which emerges in solving such problems is that the optimal policy in all
future periods is the same as was determined in the initial period, provided that there are no
unexpected occurrences or shocks to the economy. Put yet another way, in the absence of shocks, the
optimal policy path laid out in time period 0 continues to appear optimal in period 1, period 2, and so
on.7 In no future period do policy makers have any reason to alter the policy path that they devised
in period 0, again assuming that no shocks have occurred. If such shocks do occur, the time
consistent policy path has the property that no currently anticipated developments would lead policy
makers to anticipate changing their program in the future.
This type of consistency does not necessarily mean that the resultant policy path is desirable,
only that policy makers see themselves as unable to do better. Whether the outcome is desirable in
fact depends on how the public formulates its expectations. The assumption made by policy makers
following a time consistent policy is that the public's behavior in each period depends on past policy
decisions only. If the public's expectations are rational, however, so that on average the public
correctly anticipates and reacts to future policy actions, the policy makers' decisions and the public's
5
The terminology and literature began with Kydland and Prescott (1977) and Calvo (1978).
Buiter (1981) discusses the relationship between the new literature on monetary policy design and the
older optimal control literature.
6
Such models are often referred to as "causal" models since current behavior can be traced
directly to past events. By contrast, "noncausal" models allow expectations of future events to affect
current behavior.
7
Buiter (1981), citing Kydland and Prescott (1977) and Bellman (1957), states that "a sequence of
policy actions is time consistent if, for each time period, the policy action in that period maximizes the
objective function, taking as given all previous policy actions and private agents' decisions and as
given that all future policy actions will be similarly determined."
424
actions could be based on different views of the impact of the policy decisions. The public could
correctly (on average) anticipate future policy moves because it recognizes the incentives faced by
policy makers and incorporates these expectations into its current behavior, while policy makers
assume that the public's decisions are independent of their future actions. In this situation, policy
makers are aware of the public's current expectations, but ignorant of how those expectations respond
to policy actions. In this respect, the assumption of rational expectations on the part of the public
provides it with an informational advantage over the policy makers.
The equilibrium that emerges is the outcome consistent with both views of the public's inflation
expectations; it maximizes the policy makers' objective function, contingent on the public's current
expectations. It is not necessarily the best outcome by any means. The public may base its
expectations on worst case assumptions, and policy makers may find that the "optimal" policy in this
case has the effect of validating these assumptions.
An example may illustrate this point. Assume that the public correctly believes that policy
makers wish to lower the unemployment rate as much as possible provided that inflation does not
exceed some critical threshold. For policy makers, the time consistent policy is to remain
expansionary as long as inflation is below this critical value. The public, knowing that this is the
policy makers' rule, will quickly adjust its inflation expectations to the critical level, since it knows
that government policy will quickly bring inflation there. Hence, the time consistent outcome is that
inflation expectations and actual inflation adjust upward to the critical level, leaving the authorities
little room in fact to implement the expansionary policy—that is, to lower the unemployment rate
below some "natural rate."
In this example, the time consistent outcome has the following properties:
a) Policy makers always follow their perceived optimal rule of expanding output until inflation
hits a critical level.
b) The public is not fooled in that it correctly predicts policy makers' action.
c) The outcome, characterized by a rapid jump in inflation expectations to the equilibrium level
(the policy makers' threshold level), is unlikely to produce the output gains sought by policy
makers.
d) At the equilibrium inflation rate, policy makers have no incentive to alter their policy.
By contrast, a time inconsistent policy path, which may in fact represent the optimal long-term
425
Payoff Matrix from the Policy Makers' Viewpoint
Policy Decision
Public's
Expectation
High Inflation
No Inflation
High inflation
2 (A)
1 (B)
No inflation
4 (C)
3 (D)
Higher numbers represent preferred outcomes.
policy path, does not necessarily appear optimal to policy authorities on a period-by-period basis.8 As
a result, each period policy makers would be tempted to renounce the initial time inconsistent policy
path and substitute a new one. In the example above, the time inconsistent policy is to resist the
temptation to lower the unemployment rate below the natural rate, even when the public's expectation
is for zero inflation and expansionary policy would appear desirable from the policy makers'
viewpoint.
The distinction between time consistent and time inconsistent policies can be illustrated further
in the context of game theory. Consider the policy makers' payoff matrix, which specifies the value
of a given outcome under a variety of circumstances and which is assumed to be known by the public
(see the table).9 From the policy makers' viewpoint, the best option is to inflate when the public
expects no inflation, thereby gaining the benefits of faster growth (outcome C).10 The worst option is
to disinflate when inflation expectations are high, thereby entailing a loss of output (outcome B).
In between are equilibrium outcomes. When the policy and the public's expectations are
noninflationary (outcome D), the outcome is slightly worse than when the inflation takes the public by
surprise, but better than when both public expectations and policy are inflationary (outcome A).
The key point is that the public recognizes that the government has an incentive to generate
8
To quote Blanchard and Fischer's (1989) definition, "A policy is dynamically inconsistent when
a future policy decision that forms part of an optimal plan formulated at an initial date is no longer
optimal from the viewpoint of a later date, even though no relevant new information has appeared in
the meantime.11
9
Blackburn (1987) provides a comprehensive review of the game theory elements of this literature.
10
The policy makers' objective function will be discussed below in greater detail.
426
inflation whether the public expects high or low inflation. Outcome A is preferred to outcome B and
outcome C is preferred to outcome D~that is, the high inflation strategy dominates.
The time consistent outcome is A: policy and expectations match, creating an equilibrium, and
the authorities can do no better given the public's expectations. Outcome D, however, is the time
inconsistent equilibrium and is clearly superior to A, but this outcome may be unsustainable. Once the
public's expectations are decided, policy makers can do better by inflating. The public will also
recognize that if inflation expectations are low, the authorities will choose C. Hence, the public will
never expect the low inflation equilibrium because policy makers' optimizing behavior consistent with
that expectation yields high inflation.
Thus, in some instances adhering to a time inconsistent policy path is superior to following a
time consistent path, provided that the public can be made to believe that policy makers are sincere in
their pursuit of a policy that forgoes short-run optimization. In the game theory example, the superior
time inconsistent outcome D could be achieved if policy makers could guarantee that they would not
try to achieve C, their true optimum. Much of the policy makers' problem consists of convincing the
public of their resolve to follow the time inconsistent path, when the public realizes the temptation to
reoptimize. The problem resembles that of the Prisoner's Dilemma in that the outcome without
cooperation between the players (in this case, the policy makers and the public) is likely to be inferior
to that with cooperation. What prevents the superior outcome is the absence of a mechanism to
guarantee the cooperative solution, when cheating promises a better result for policy makers acting on
their own.
A key element of the coordination problem is that the public is assumed to arrive at its
expectation of current period policy before policy makers make their decision. If the authorities
moved first, the coordination problem would be mitigated because there would be no opportunity to
fool the public. Paradoxically, the time consistency problem would be resolved because there would
be no incentive to deviate from preannounced plans. Many of the proposed solutions to the time
consistency problem amount to removing "surprise" as a policy tool. In the context of the table, they
amount to forcing policy makers to choose between the no inflation equilibrium (D) and high inflation
equilibrium (A).
To resolve the coordination problems that arise if the time inconsistent policy path is superior,
policy makers may wish to commit or precommit themselves to the time inconsistent policy, which
they know to be superior in the long term, and renounce the possibility of reoptimization. By
committing themselves to the time inconsistent policy, they may hope to convince the public that they
427
will not inflate, even when it would be advantageous to do so. A further difficulty may arise,
however. If policy makers face no sanctions for violating their commitment or if the public cannot
monitor on a timely basis policy makers' commitment, any commitment may lack credibility. Both
the public and the policy makers may agree that the committed policy is best, but the public will not
believe that the policy makers will follow through because of the period-by-period temptation to
renege.
In practice, it may often be difficult to determine whether policy makers are adhering to the
precommitted policy. Targets can be missed either because of random shocks to the economy or
because policy makers are reneging on their commitments. Because of this ambiguity, advocates of
precommitted policies often argue that following fixed rules makes it easier for the public to observe
adherence to announced policies.11 The rules can be very simple (for example, constant money
growth rules) or more complicated, but they have to be understandable, and compliance has to be
readily visible.
The requirement of ready visibility may make rules with no feedback (open loop rules) at times
superior to rules in which policy actions are contingent on actual events. The public may lose
confidence in its ability to monitor adherence to a rule if the rule permits action in response to events
not readily observable. For example, assume that a particular monetary aggregate deviates from its
precommitted path. The central bank may claim that it is merely accommodating a money demand
shock. But the public, having no way to ascertain that such a shock has occurred, may assume that
the deviation represents a policy easing and may therefore adjust inflation expectations upward.
To sum up, the long-run optimal policy may be time inconsistent if the public can understand
and predict future policy responses (that is, the public has rational expectations). It may be preferable
for policy makers not to optimize on the basis of expectations which they view as fixed, but rather to
anticipate the negative effect that such optimization will have on expectations of future policy actions.
More concretely, in the monetary policy case, policy makers who are expected to take advantage of
low inflation expectations in order to pursue expansionary (and inflationary) policies may find that
expectations are extremely sticky at undesirable levels in subsequent periods. Recognizing this, the
policy makers may wish to commit themselves to a series of policy actions that may not be optimal on
a period-by-period basis, but that is consistent with low inflation expectations in the long run. To
11
The other papers in this volume provide extensive references to the literature on monetary policy
rules.
428
succeed in the long run, such a commitment must be credible, and credibility in turn may depend on
adherence to readily visible fixed rules. Fixed rules with no feedback make it easiest for the public to
observe that policy is following its precommitted path.
II. Is There an Inflationary Bias to Monetary Policy?
This section considers the conditions under which positive inflation may emerge as an
equilibrium, even when both the public and the policy makers view the outcome as inferior to one of
zero inflation.12 It examines the circumstances under which dynamic optimization by policy makers
will produce an inferior result to a policy following relatively fixed rules. Following the presentation
of the basic model, a critical discussion of the assumptions needed to yield the equilibrium inflation
result is presented. The section concludes with possible approaches to mitigating the alleged
inflationary bias of policy.
A. How Do Inflationary Biases Emerge?
The basic structure of the models under discussion is very simple.13 Policy makers try to
achieve inflation and output goals that are inconsistent. The desired output level is greater than could
be achieved at stable inflation.14 Policy makers face the choice between maintaining stable inflation
at an output level lower than they would otherwise try to achieve or achieving desired output levels at
the cost of ever-increasing inflation. Higher inflation emerges in the second case because the only
mechanism available to policy makers by which output can be increased to desired levels is a positive
inflation surprise. In a multiperiod context, inflation surprises would be needed each period to
12
For convenience of exposition and in accord with the rest of the literature, this paper will treat
zero as the inflation target. In practice, measurement problems or nominal wage and price stickiness
may make a positive but low level of inflation preferable. What is essential for the analysis is that the
public view policy makers as willing to inflate above the target, whether it is zero or a positive value.
13
Barro and Gordon (1983a), Barro (1983), and Blanchard and Fischer (1989) provide clear
descriptions of the analytical model underlying this section.
14
Economists usually attribute this to some distortion that lowers output below its potential. The
most common example is the distortionary tax that lowers supplies of labor and capital. Alternatively,
political considerations may lead to a greater emphasis on short-term output gains near elections. See
Alesina (1988), Alesina and Sachs (1988), and Nordhaus (1989). In theory, the government's
objective function could be strictly rising with output, but this would imply a willingness to trade
leisure for output that would not be consistent with utility maximization by the public.
429
maintain desired output; hence, spiraling inflation would emerge.
It is assumed that policy makers are less willing to tolerate additional inflation when inflation
rates are already high. For example, going from 0 to 2 percent inflation will cause policy makers
some discomfort, which may be offset by the temporary output gain. Each successive increment of
inflation causes additional discomfiture, ultimately reaching a point where policy makers are unwilling
to accept the higher inflation levels, even if output can thereby be maintained above the level
corresponding to the natural rate. Thus, under these assumptions, there is a strict upper limit to the
inflation rate policy makers would engineer, even if the public's inflation expectations were set
naively. In many cases this upper limit will also be the public's equilibrium inflation expectation
since the public knows that policy makers would not intentionally raise inflation any further.
It is also assumed that the public cannot be systematically fooled or surprised by inflation. The
public knows policy makers' preferences and the structure of the economy, and knows that policy
makers have an incentive to try to produce surprise inflation. The public also knows the increasing
discomfiture of policy makers at high inflation rates. (The public's preference function is often
assumed to be the same as the policy makers'--that is, to eliminate conflicting preferences as an
underlying cause of equilibrium inflation.)
The public tries to predict the inflation rate by evaluating how policy makers are likely to act.
The public recognizes that if policy makers observe low inflation expectations, they will have an
incentive to create surprise inflation in order to reap output gains from the surprise. But the public
also knows that at sufficiently high expected inflation levels, policy makers, by their own choice,
would never inflate further, even by surprise, and might even choose to disinflate because of the
perceived costs of a high level of inflation.
Using this knowledge, the public forms its expectations. It will never expect inflation to be so
low that policy makers will have an incentive to create surprise inflation. Nor will the public expect
the government to produce an inflation rate that is so high that the government would subsequently be
tempted to engineer a recession (that is, create surprise disinflation) to reduce inflation to more
acceptable levels. The rational expectation is thus for an inflation rate just high enough to eliminate
the incentives for policy makers to surprise inflate and low enough to remove the incentive to surprise
deflate. From the viewpoint of policy makers, losses from additional surprise inflation at this inflation
rate just balance the perceived benefit of the additional output. The government, facing this
expectation, has no incentive to produce any surprise. This outcome is characterized by inflation that
is positive in equilibrium and output that is at the natural rate (but below the government's desired
430
level). Nothing is gained on the output side from the additional inflation, but a welfare loss is
incurred because of inflation. Thus, the outcome that emerges is inferior to the one that could be
attained at zero inflation.
Although the extent of the knowledge attributed to the public by these models strains credulity,
many of the specific assumptions are analytically convenient without being essential. What is essential
is the public's assumption that policy makers are willing to use surprise inflation as a tool to generate
higher employment. It is not necessary that the public know the exact form of the policy makers'
preferences or the exact structure of the economy.
B. Underlying Assumptions
The basic ingredients creating a conflict between short- and long-term policy making are
(i) irreconcilable output and inflation goals, (ii) forward-looking or rational expectations on the part of
the public (but not the policy makers), and (iii) a perceived ability on the part of policy makers to
"surprise" the public with unexpected inflation.
Although these assumptions seem technical in nature, assessing their realism will clarify the
realism of the entire analysis. In particular, the sensitivity of the analysis and the results to changes in
the assumptions will help us to evaluate the claim that the structure of policy making is the source of
persistent inflation in recent times. Indeed, one of the major contentions of this paper is that the
implications of the time consistency literature are virtually indistinguishable from those of a standard
backward-looking adaptive expectations framework. The additional theoretical elegance of the time
consistency models is achieved only at the cost of assumptions whose empirical robustness is dubious.
Incompatible targets
The assumption of incompatible goals is essential.15 In particular, policy makers are assumed
to strive for an unemployment rate that is inconsistent with the natural rate. By assumption, the
natural rate is the only unemployment rate at which inflation is stabilized; hence, policy makers must
balance approaching their targeted unemployment rate against the extra inflation generated in getting
there. There is no conflict between time consistent and time inconsistent policies if policy makers
have only a single goal or multiple goals that are mutually supportive. If policy makers aim only at
15
This assumption dates back to Kydland and Prescott (1977). It is used in Barro and Gordon
(1983a, 1983b), Cukierman and Meltzer (1986), and virtually every other paper on the subject.
431
stabilizing inflation (at zero or any other value) or at stabilizing the unemployment rate at the natural
rate (that is, the rate consistent with stable inflation), the time consistent policy path produced by
dynamic optimization is fully consistent with the time inconsistent policy path toward the equilibrium
of zero inflation (or any desired rate). Hence, the structure of policy making is irrelevant if policy
makers are perceived as pursuing only a zero inflation target or a sustainable output target. It is only
when the public views policy makers as regarding favorably the prospect of trading additional inflation
for additional output that the inflationary bias emerges.
The reason that time consistency problems do not emerge when the output target is the natural
rate is that the public has no reason to question the willingness or ability of policy makers to achieve
their inflation and output goals. Because there is no conflict among goals, there is no question of
commitment or credibility and no policy trade-off to exploit.16
As to the policy makers' objective function, the theoretical elegance of time consistency models
appears greatly oversold. Undesirably high inflation as an equilibrium is derived at the cost of
assuming that policy makers pursue targets that they know to be inconsistent. Often the pursuit is
justified as a necessary consequence of the political process or as a way of offsetting other outputreducing distortions in the economy. In general, however, scant attention is paid to motivating the
policy makers' assumed objective function empirically or theoretically.
Rational expectations
The other requirement for time consistency problems to emerge is rational expectations by the
public. That is, the public knows enough about the preferences of policy makers and the structure of
the economy to forecast policy accurately on average.17 Under rational expectations, policy makers
cannot systematically fool the public and so cannot gain the extra output that is sought, even
temporarily. There is an asymmetry here in that while both the public and the policy makers know the
structure of the economy and the policy makers' preference, only the public optimizes on the basis of
future events. Indeed, in the earliest models that developed the time consistency problem, it was
16
Malcomson and Hillier (1984) argue that the essence of the time consistency problem is that the
policy makers have two targets, inflation and output, but only one instrument. Surprise inflation
becomes a second instrument that the policy makers are attempting to utilize.
17
Rational expectations are not strictly required. As long as the public's behavior responds
somewhat to its expectation of future policy, a time consistency problem can emerge. However,
virtually all of the literature assumes rational expectations.
432
explicit that the public reacted to both past and future policies, while policy makers optimized only on
the basis of past events.18 Such myopia on the part of policy makers is often attributed to their
susceptibility to short-term political influences. Policy makers do not recognize that the public
discerns and reacts to their incentives. If policy makers recognized that the public cannot be fooled,
they would not make the effort to do so. Furthermore, in many cases, even if policy makers assumed
(incorrectly) that the public had backward-looking expectations, they would nevertheless be deterred
from inflating opportunistically as long as their discount rate was not too high and they viewed the
public's expectations as responding reasonably promptly to actual inflation.19 By implication, in
those models where high inflation equilibria emerge, policy makers believe that they can fool most of
the people for a long time.
The assumption that the public holds rational expectations can also be challenged on empirical
grounds. Most empirical tests of the rational expectations hypothesis reject it. In particular, inflation
expectations appear to be more backward- than forward-looking and inflation "surprises" can last for a
long time.20 If such is the case, the premise that adherence to a credible policy rule will produce
costless disinflation may prove to be far off the mark. In practice, policy makers may find it risky to
adopt a policy path whose success depends crucially on the assumption that the public will both
anticipate correctly and react immediately to the effects of future policies.
In considering the robustness of policy conclusions to be drawn from the models under review,
it is important to recognize that backward-looking (for example, adaptive) expectations on the part of
the public can yield many of the same results produced by rational expectations in these models.
Adaptive or backward-looking expectations in a multiperiod context would not be strictly "rational,"
but in regimes of moderate or low inflation the results would not diverge greatly from rational
expectations. As long as expectations eventually catch up to actual inflation, any systematic inflation
surprise can only be transitory. During this transition, policy makers could temporarily generate
18
For example, see Kydland and Prescott (1977). In equilibrium, expectations are fulfilled, so
both the policy makers* and the public's expectations are rational ex post.
19
For example, the low inflation reputational equilibria discussed in Barro and Gordon (1983a)
can be interpreted as emerging because policy makers recognize that inflation expectations respond
quickly to actual policies. Also see Chari and Kehoe (1988) and Grossman (1990). Backus and
DriffiU (1986) find that the response of expectations even with Fischer-Taylor-type overlapping wage
contracts is sufficiently quick to avoid the bulk of the costs associated with time inconsistent policies.
20
See Englander and Stone (1989) and the references therein.
433
higher output (a course not open to policy makers if strictly rational expectations are assumed), but
long-run output growth would be unaffected as long as policy makers were unwilling to accept everincreasing inflation. Equilibrium inflation would be higher and output temporarily higher.
With such backward-looking expectations, however, it makes no difference if policy makers are
credible, and there is no conflict between the time consistent and time inconsistent solutions. From the
policy makers' point of view, they are obtaining the best solution given their preferences and the
structure of the economy. That is, they may feel that if inflation is running at very low levels, the
short-run increase in output can justify a small, but long-run, increase in inflation. In practice,
however, if inflation expectations react quickly to increases in inflation, the willingness to inflate is
likely to be extremely curtailed.
The key point is that in the absence of rational expectations, policy makers, perhaps reflecting
the tastes of the public, have preferences that lead them to exploit the inflation/output trade-off and
that make them unwilling to accept the output losses required for a return to zero inflation. In this
case, it is probably better to choose better policies or better policy makers than to impose a structure
of rules that may respond inflexibly, and thus suboptimally, to economic shocks or changes in
priorities.
Surprise inflation as a policy tool
The final critical assumption of these models is that policy makers can generate surprise
inflation to exploit an inflation/output trade-off temporarily. While this assumption is commonplace in
the literature, the process by which the inflation/output trade-off is exploited in practice is not clearly
described. Indeed, it seems to rest on two assumptions: 1) that anticipated policy moves (such as an
expected easing in monetary policy) should have little or no effect on output, and 2) that policy
makers can manipulate the surprise component of inflation to alter the path of output temporarily.
Surprise inflation is not a tool directly at the disposal of policy makers. Some other
instrument-interest rates, money growth, reserve requirements—must be used to implement policy. By
common consensus, however, long and variable lags separate movements in these potential instruments
from changes in inflation or output. It is doubtful whether the degree of surprise experienced by the
public when inflation rates change is any greater than that experienced by policy makers or whether
434
economic behavior is greatly affected because of ignorance of the price level.21 Hence, it is unlikely
that mistaken beliefs as to the level of real wages or relative prices can generate significant output
fluctuations.
The question of the mechanism by which surprise inflation affects aggregate output, while
apparently arcane, is important for determining whether the structure of policy making is the key
factor inducing persistently high inflation expectations. If policies that have been previously
announced, or for some other reason are already expected, nevertheless can have an effect on real
output, the structure of the problem assumed in the time consistency literature is altered
fundamentally.22 The reason is that policy makers can achieve output gains, at least in the short run,
without resorting to policy moves that fool the public. Policy makers would optimize subject to their
knowledge that unsustainable expansionary policies lead to inflation. Depending on the policy makers*
objective function, they might tend to choose inflationary or noninflationary policies, but the source of
the inflation would be the policy makers' actions rather than the structure of policy making or
expectations conditioned on future policies.
The public might revise its inflation expectation upward when it observed expansionary policy
being implemented, but it would not do so in the absence of such policy. Again, the conduct of policy
making, rather than its structure, appears to be the underlying determinant of inflation.
Recognizing that ignorance of the level of prices or real wages is unlikely to produce major
output effects, some analysts have argued that the effects of inflationary policy moves are seen
immediately in asset values and capital accumulation decisions (but before the inflationary effects
show up in actual prices). Hence, the policy surprise operates through wealth rather than inflation.
The empirical consequences of such redistributions of wealth, however, are difficult to pin down.
Some authors contend that inflation leads to higher output because the lower real value of government
debt allows the government to engage in further spending. In contrast, others argue that price inflation
21
As discussed below, Barro and Gordon (1983) and Kydland (1989) argue that the effects of
surprise inflation on nominal asset values and capital accumulation are of greater empirical
significance than the effects of wage or relative price surprises on output.
22
See Mishkin (1983), for example. Both his original work and his reworking of Barro and Rush
(1980) suggest that, if anything, anticipated policy moves have more impact on output than
unanticipated policy.
435
may actually lead to a reduction in output by lowering the incentives to accumulate capital.
While the issue appears abstract, the considerable uncertainty attending the effects of surprise
asset inflation makes it unlikely that such surprises are the mechanism by which an inflation/output
trade-off is consciously exploited by policy makers. Yet the structure of such models and the policy
conclusions that they yield presuppose that surprise inflation is the only means by which policy can
affect outcomes. If this assumption is false, it is hard to make the argument that the mere presence of
discretionary policy making yields an inflationary bias. Again the time consistency problem seems
less important than systematic policy errors or preferences in generating inflation.
Credibility
If the zero inflation outcome is preferable to the equilibrium outcome in the eyes of both
parties, why do they not agree to maintain the preferred alternative? The time consistency literature
argues that the answer to this question involves the credibility issue. The public recognizes that policy
makers have every incentive to assert that they will maintain low inflation, but it also recognizes that
policy makers have a greater incentive to renege if the public accepts the assertion at face value.
According to this view, the public in general will not believe that low inflation will be maintained
unless policy makers are viewed as strong adherents of low inflation or policy makers can provide
evidence that they are following a policy rule that will yield low inflation. It is in this latter context
that adhering to an intermediate target path believed consistent with low inflation, for example, may
reduce inflation expectations.
This is where credibility issues become important. A commitment can be credible either
because policy makers have a reputation for backing their commitments or because a way of enforcing
the commitment exists. Among the suggested strategies for achieving commitment are
i) requiring commitment through legislation
ii) ensuring that any breaches are obvious
iii) choosing policy makers whose sole objectiv
@FedFRASER