The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
Working Paper Series
The Identification of Monetary
Policy Disturbances: Explaining
the Liquidity Puzzle*
Steven Strongin
LIBR A R Y
J A N 2 6 1993
FEDERAL RESERVE
BANK OF CHICAGO
W o rk in g P a p e rs S e rie s
M a c ro e c o n o m ic Issu e s
R e s e a rc h D e p a rtm e n t
F e d e ra l R e s e rv e B a n k o f C h ic a g o
N o v e m b e r 1 9 9 2 (W P -9 2 -2 7 )
*A substantially different version of the p a p e r
w as released u nder the s a m e title d esig nated
W P -9 1 -2 4 .
FEDERAL RESERVE BANK
OF CHICAGO
T h e Id en tifica tio n o f M on etary P olicy D istu rb an ces:
E x p la in in g th e L iq u id ity P u zzle
Steven Strongin*
Federal Reserve Bank of Chicago
11-14-92
*The author acknowledges a debt to Martin Eichenbaum for his helpful comments and
suggestions on numerous drafts of this paper. He would also like to thank Anne
Marie Gonczy and Mark Watson for their comments on earlier drafts, as well as the
participants at the NBER’s Economic Fluctuations and Monetary Economics meetings,
the macro workshop at the Federal Reserve Bank of Chicago, Francesca Eugeni and
Monika Merz for their research assistance in the early stages of this project and
Carolyn McMullen for editorial assistance. The opinions expressed in this paper are
not necessarily those of the Federal Reserve Bank of Chicago or the Federal Reserve
System.
Abstract
This paper examines the recent work on the identification of exogenous monetary
policy disturbances by Sims (1992) and Eichenbaum (1992), as well as Sims (1980)
and Litterman and Weiss (1985). The main finding is that the anomalies documented
by these authors reflect their failure to properly take account of the Federal Reserve’s
policy of accommodating short run reserve demand disturbances. This, in turn, leads
them to misidentify demand shocks as supply shocks. A new method of identifying
the supply shocks, which avoids this confusion, is proposed and estimated. Once this
is done, a nonborrowed reserves based measure of monetary policy is used to
successfully address the difficulties which normally afflict attempts to measure policy
innovations, such as the liquidity puzzle and Sims’ price puzzle. Specifically, using
the proposed specification, a positive innovation in nonborrowed reserves adjusted for
the accommodation of reserve demand shocks has a strong and persistent negative
effect on interest rates regardless of sub-sample and conditioning variables. Second,
contrary to Sims (1992), but in line with Eichenbaum (1992), there is no persistent
and significant positive price impact from a contractionary policy disturbance. (There
is, however, a positive persistent price effect from the Federal Reserve’s
accommodation of reserve demand shocks.) Third, using the proposed specification,
"money" Granger causes output even in the presence of interest rates and explains
approximately 49% of the variance in output over a two year horizon. Fourth, the
explanatory power of interest rates for industrial production drops to approximately
zero once the proposed policy measure is taken account of. Finally, the paper finds
strong evidence that the liquidity effects of policy on interest rates are persistent and
that the persistence is directly related to the persistence of nonborrowed reserves’
response to an innovation in policy.
This suggests that even anticipated
accommodative policy actions have a substantial negative impact on interest rates.
L Introduction
Recently, there has been renewed interest in the identification of exogenous monetary
policy disturbances. The recent search for an appropriate way to measure the impact
of monetary policy has followed two well-worn paths: interest rates and monetary
aggregates. When monetary aggregates are used to measure exogenous policy
disturbances, three problems consistently arise. First and most troublesome is that
innovations in the standard monetary aggregates seem to be associated with rising
rather than falling interest rates. This "liquidity puzzle" has been surveyed by
Reichenstein (1987) and was recently re-documented in detail by Leeper and Gordon
(1992). The Leeper and Gordon paper shows that what relationship there is between
innovations in the monetary base (as well as broader monetary aggregates) and
interest rates is highly uncertain, varies across time and usually has the opposite sign
than that predicted by the liquidity hypothesis. The second problem is that while
monetary aggregates Granger cause output in VAR specifications that exclude interest
rates, once interest rates are included in the specification, monetary aggregates no
longer Granger cause output (see Sims (1980) and Litterman and Weiss (1985)). A
third, (and related puzzle also documented in Sims(1980) and Litterman and
Weiss(1985)) is that in VARs which include interest rates and some form of money,
innovations in money explain a very small proportion of the variance of output, often
1% or less. In sharp contrast, innovations in interest rates typically explain a very
high percentage of the variance of output, often exceeding 40% over a two year
horizon, apparently indicating that innovations in policy (as measured by money) have
little explanatory power.
These problems with money based measures of monetary policy have led Sims (1992)
and Bemanke and Blinder (1992) among others to identify policy directly with
innovations in interest rates. This identification scheme is reasonably successful in
producing results consistent with a priori expectations about the effects of monetary
policy in that both output and money fall in response to a contractionary monetary
policy disturbance, e.g. a positive innovation in interest rates. However, a number of
conceptual and empirical problems remain. First and most important, without any
demonstrated empirical linkage between Federal Reserve actions and interest rate
movements, it is unclear how innovations in interest rates can reasonably be attributed
to monetary policy. Second, the validity of any such identification hinges critically
on the specification’s ability encompass all other significant determinants of the
interest rate except policy. This suggests that inferences about policy could be very
sensitive to specification and omitted variable problems. Only if monetary policy far
outweighs all excluded variables in terms of explaining interest rates would inferences
be robust, a proposition hard to argue from theory or previous empirical work. Sims
(1992) documents just how severe this fragility can be. In specifications that follow
Litterman and Wiess (1985) or Sims (1986) a positive innovation in policy, an easing
action, is associated with a strong and persistent drop in the price level, exactly the
opposite of what would be expected to happen in response to a expansionary monetary
1
policy. Sims is able to reverse this counter-intuitive response by including exchange
rates and commodity prices in the specification, variables whose direct explanatory
power is fairly small both for interest rates and output. Beyond these problems of
fragile inference, this approach suffers from a more general failure in that when policy
is identified purely by its consequences, i.e. the residual in an equilibrium condition,
rather than directly by Federal Reserve action, there is no way to disentangle the
policy response function, thus eliminating a useful diagnostic on the reasonableness
of the specification, nor can any questions about policy regimes be meaningfully
entertained.
Thus, to this point neither of the aforementioned approaches to identifying monetary
policy disturbances seem satisfactory. Eichenbaum (1992) argues that Sims (1992)
was rash in rejecting aggregate based measures of monetary policy and suggests that
the innovation in nonborrowed reserves is the correct measure of monetary policy.
This suggestion is highly successful on a number of dimensions. Positive
nonborrowed reserve innovations are associated with increased production and
declining interest rates. Eichenbaum’s solution, however, is less than fully satisfying.
First, nonborrowed reserve innovations produce the same counter-intuitive price
response that interest rate innovation do, i.e. contractionary policies generate persistent
increases in the price level, although in the case of the nonborrowed reserve
innovations the price response is statistically insignificant Second, the explanatory
power for real output is very small; only about 1% of the variance of industrial
production can be explained by innovations in nonborrowed reserves. Interest rate
innovations still account for the bulk of the explanatory power, nearly 40%. Third,
nonborrowed reserves do not Granger cause output in the presence of an interest rate.
And finally, Eichenbaum does not explain why his measure of monetary policy
disturbances should solve either the liquidity or price puzzle. There is no conceptual
explanation for why an innovation in nonborrowed reserves is the right measure of
monetary policy, relative to previous attempts. His justification for the use of
nonborrowed reserves is much like those normally used for the monetary base (a
failed measure by Eichenbaum’s analysis), i.e., that the Federal Reserve can control
nonborrowed reserves more easily than a broader aggregate.
This paper argues that the main source of the difficulty in identifying monetary policy
from the type of reserve data used by Eichenbaum (or any monetary aggregate) is that
a significant proportion of the variance in the reserves data is due to the Federal
Reserve’s accommodation of innovations in the demand for credit and money rather
than policy induced supply innovations. This leads to a confounding of supply and
demand innovations. This paper utilizes a linear representation of the Federal
Reserve’s operating procedures, which includes both the level of total reserves and the
mix of borrowed and nonborrowed reserves supplied by the Federal Reserve, to
identify the exogenous disturbances to monetary policy net of accommodation. By
using two measures of reserves with different responses to supply and demand
innovations it becomes possible to distinguish between changes in reserves which
2
result from Federal Reserve policy innovations and changes in reserves which result
from the Federal Reserve’s accommodation of demand innovations. The key
assumptions are that the level of total reserves is largely determined by the Federal
Reserve’s short run accommodation of the demand for reserves1 and that policy
innovations are reflected in the mix of borrowed and nonborrowed reserves used to
meet that demand2. Specifically, policy innovations are measured by using the
innovation in total reserves to extract changes in the reserve mix between borrowed
and nonborrowed reserves that are due to the Federal Reserve’s accommodation of
reserve demand shocks, leaving only those changes in the mix which are true policy
innovations. This generates a well identified measure of exogenous monetary policy
disturbances which can be used to address the puzzles set forth at the beginning of
the paper.
The main empirical results can be summarized as follows. First and most important,
in sharp contrast to previous work interest rates fall in response to expansionary
monetary policy disturbances. This liquidity effect holds in all sub-periods, is
precisely estimated, and its persistence relates directly to the persistence of the own
response of the nonborrowed reserve policy measure, indicating that anticipated
monetary policy as well as innovations affect interest rates. Second, the proposed
measure of monetary policy disturbances has substantially more explanatory power
for interest rates and real output than a pure nonborrowed reserves measure or any
other single monetary aggregate based measure, accounting for 49% of the variance
in output at the end of two years. Third, the marginal explanatory power of interest
rates is substantially reduced in the presence of the proposed measure of monetary
policy, accounting for less than 2% of the variance of output at the two year horizon.
In this specification nonborrowed reserves Granger cause output while interest rates
do not Fourth, unlike interest rate or pure nonborrowed reserves based measures of
monetary policy, accommodative policy leads to a permanent and statistically
significant increase in the price level. Fifth, these results are largely insensitive to the
details of the contemporaneous modeling of the VARs error structure. The key is to
fully span the reserve space in term of both level and mix of reserves, thus allowing
for the systematic accommodation of demand innovations. This implies that the key
point in identifying monetary policy is using the mix between borrowed and
nonborrowed reserves and not the details of the contemporaneous adjustment
'This contrasts sharply with earlier work where it was assumed that the Federal Reserve did not
accommodate demand shocks and that innovations in the aggregates or the monetary base could be used
as a direct measure of monetary policy disturbances. For examples of empirical research which follow
this interpretation see Sims (1972), Barro (1978), Mishkin (1981,1982), and Rush (1986).
2
In the nomenclature used by the Federal Reserve only the part of reserve demand meet by
nonborrowed reserves is usually referred to as accommodation.
3
process.3
The core of the paper is in three sections. Section n discusses the relationship
between Federal Reserve operating procedures and the identification of monetary
policy innovations. It also develops a new identification scheme for identifying
monetary policy innovations in the context of a VAR. Section HI reviews the history
of Federal Reserve operating procedures and, based on this history, predicts what
differences in behavior should correspond to different sub-periods of the data. Section
IV presents sub-sample and full sample analysis similar in form to Leeper and Gordon
(1992), Sims (1992) and Eichenbaum (1992) using the VAR identification scheme
developed in Section II on the data sub-samples outlined in Section in. The appendix
analyzes a more general model in which less restrictive identification schemes can be
discussed.
II. Federal Reserve operating procedures and the identification of policy
The standard assumption4 in the identification of monetary policy is that the Federal
Reserve controls the supply of total reserves5 and therefore any innovations in total
reserves can be thought of as policy innovations. As a matter of actual practice, the
Federal Reserve accommodates innovations in the banking system’s demand for
reserves. Failing to take account of this policy of accommodation can lead to
misleading results. As Sims (1986) put it, "we forgo separating bank behavior from
Federal Reserve behavior and thereby also forgo properly separating demand from
supply behavior." However, once the analysis takes account of the Federal Reserve’s
operating procedures, policy disturbances can be identified in a fairly simple manner
by using two reserve measures which respond to demand and supply innovations
differently. The key assumptions are that 1 .) in n o v a tio n s in th e l e v e l o f t o ta l
reserves
are
la r g e l y
th e
r e s u lt o f th e
F e d e r a l R e s e r v e ’s
a c c o m m o d a tio n
of
in n o v a tio n s in t h e d e m a n d f o r t o t a l r e s e r v e s a n d 2 .) th e F e d e r a l R e s e r v e e x e r ts its
i n f lu e n c e o n t h e r e s e r v e m a r k e ts b y a lte r in g th e m ix o f b o r r o w e d a n d n o n b o r r o w e d
r e s e r v e s i t s u p p l ie s to m e e t c u r r e n t r e s e r v e d e m a n d .6*
3The robustness of the qualitative results is not a matter of theory, it is due to the fact that for
industrial production virtually all of the explanatory power, not contained in the own shock, is contained
in the linear combination of reserves that the paper identifies as the policy shock. As a result, other
contemporaneous identification schemes only reallocate the same basic impulse across 1 to 3 shocks
without effecting any of the qualitative results.
4See footnote 1.
sUsually this is applied to even broader aggregates, such as the monetary base, Ml or M2. For a
recent and thorough review of empirical results using this approach see Christiano and Eichenbaum
(1992a).
6In the Federal Open Market Committee’s policy directive, this mix is referred to as " the degree of
pressure on reserve positions."
4
Using assumptions 1 and 2 above, policy innovations can be identified as those
changes in the mix of borrowed and nonborrowed reserves which are not the result
of the Federal Reserve’s accommodation of demand innovations. This can be
accomplished by modeling the forecast innovations of both borrowed and
nonborrowed reserves as follows. Let vs be the policy shock, where an increase in
nonborrowed reserves is a positive shock (i.e. an easing action) and where there is an
exactly offsetting reduction in borrowed reserves. The assumption that a positive
innovation in nonborrowed reserves is exactly offset by a reduction in borrowed
reserves guarantees that only innovations in the mix of reserves will be identified as
policy innovations.7 Let vd be the reserve demand shock, where vs and vd are
independent Abstracting from other variables, it follows that the innovation in
nonborrowed reserves can be written
a)
and that the innovation in borrowed reserves can be written
“br= -vs+ (l~ W v<i >
(2)
where <{>is the operating procedure determined split in the accommodation of reserve
demand innovations between borrowed reserves and nonborrowed reserves.8*
Equations 1 and 2 imply that the innovation in total reserves is a pure demand shock.
This see this, note that adding equations 1 and 2, produces
If this specification is correct, it is easy to see why attempts to use the monetary base
or broader aggregates to measure monetary policy might produce misleading results.
The monetary base sums the borrowed reserves and nonborrowed reserves shocks as
well as the innovation in currency, canceling out vs, the policy innovation, while
leaving vd. the accommodation of the reserve demand shock. The only policy signal
7In the appendix a more general model, which encompasses the current specification, is described
and analyzed. The general specification allows for feedback both from policy innovations to total
reserves and from interest rate innovations to nonborrowed reserves. Thus allowing both for policy to
contemporaneously affect total reserves and for the Federal Reserve to follow interest rate smoothing
policies. Analysis of the general specification leads back to the specification described below and the
key liquidity results are largely invariant to these assumptions, once the reserve mix is identified as the
key policy measure.
8
Of course this is only a linear approximation; operating procedures could specify a non-linear
splitting rule. The non-linearities maybe especially important when borrowings levels are very low and
the non-negativity constraint on borrowing is binding. This specification also abstracts from the role of
the discount rate.
5
that would be left is determined by how much vs affects currency demand
contemporaneously.
In this structure interest rates would then be determined by both the supply and
demand shocks where
u.I—it v.+Xv
+v_
d
s
ff
where X is the liquidity effect of policy actions on interest rates and 7t is the response
of interest rates to reserve demand innovations and vff is the interest rate innovation.9
To understand why these identifying assumptions described above are reasonable it
is helpful to understand some of the perceived rigidities in the process, and also to
understand how altering the reserve mix between borrowed and nonborrowed reserves
has real effects on interest rates.
First and foremost among these rigidities is the belief that there is very little that
either the Federal Reserve or the banks can do to affect the short run level of total
reserves. In the most recent Federal Reserve Bank of New York guide to the
implementation of monetary policy, U.S. Monetary Policy and Financial Markets.
Ann-Marie Meulendyke describes the short run response to a shortfall in reserves that
might arise from a contractionary open market operations as follows:
...the banks adjustment options to a reserve excess or shortage are in
actuality quite limited. Banks confronting a shortage would have
several possible options, but most of the options would be impractical
to cany out in such a short time period.... When nonborrowed reserves
within a reserve maintenance period are insufficient to meet the
demand, the banking system as a whole has no practical alternative to
borrowing more reserves at the discount window.... As banks come to
believe that the reduced reserve availability is a deliberate policy move
they will begin (emphasis added) to make basic adjustments to the
pricing of loans and deposits.10
9
It is useful to note that two different corrections are taking place in this specification. The first is
that the specification adjusts for contemporaneous accommodation of total reserve shocks. The second is
that this correction shifts the emphasis from the level of total reserves to the mix of nonborrowed to
total reserves. While both corrections appear to be useful in identifying monetary policy, it turns out
that empirically the shift of focus to the mix of nonborrowed to total reserves dominates the specifics of
the contemporaneous identification. How these two effects can be separated will be discussed in detail
in the estimation section.
10Meulendyke (1989), p.137-138
6
Meulendyke’s assertion that reserve short falls generate increased borrowing from the
discount window rather than an actual reduction in the level of total reserves follows
from two separate institutional arguments; first, she argues that the demand for
required reserves is largely predetermined by current and past deposit levels, and
second, she argues that the demand for excess reserves is largely independent of
policy. The predetermination of required reserves follows from the belief that banks
have extremely limited short run control over the level of deposits held on account
and that any operating procedure that attempted to force banks to significantly alter
their short run deposit base and hence their short run demand for required reserves
would generate unacceptable levels of interest rate volatility. Put in more common
terms, the demand for money, which largely determines the demand for reserves, is
slow to adjust and has a relatively low short run interest rate elasticity."
The argument that excess reserve demand is largely independent of policy hinges on
the nature of excess reserve demand. There is no direct demand for excess reserves
only a derived demand due to the costs of managing reserve positions more closely.
For large banks the demand for excess reserves is effectively zero. These banks
expend considerable resources on managing their reserve accounts and rarely actually
hold excess reserves above their carryover allowance12. For smaller banks daily
reserve flows can be very large relative to the amount of the reserves they actually
need to hold for reserve requirements and the costs of closely managing such flows
may be deemed excessive relative to the potential gain. Such banks do hold excess
reserves and could as a consequence economize on their excess reserves holding in
response to a policy action. As noted above, the Federal Reserve views this potential
as unimportant empirically. If a bank holds excess reserves only because it is too
expensive to control them relative to average interest rates, it is unlikely that is would
pay that same bank to manage them in response to 25 basis point policy actions.
Further, since the Federal Reserve has purposely attempted to avoid large swings in
the Federal Funds market at the end of reserve maintenance periods both by direct
accommodation through open market operations and by easier administration of the
discount window at the end of reserve maintenance periods, it has minimized any
incentives for banks to develop ways of systematically managing excess reserves on
a short run basis.
Taken together, the predetermination of required reserve demand and the inelastic
demand for excess reserves generate a substantial amount of short run rigidity in the
"See Judd and Scadding (1982) for a survey of money demand articles documenting these
phenomena across a wide variety of specifications.
12The carry over allowance is the amount of excess reserves banks may carry from one reserve
maintenance period to the next, currently 2% of required reserves. Thus, while these reserves appear in
the reserve accounts as excess reserves, they are not in fact excess in that they are counted against next
period’s reserve requirements. There is no direct measure of true excess reserves.
7
demand for total reserves which should not be underestimated. During the last week
of December banks typically do not want to be seen borrowing from the discount
window. Discount window borrowings in the final week of the year appear in yearend financial statements and are viewed as signaling financial weakness.13 This
increased reluctance to borrow, in turn, interferes with the normal functioning of the
discount window as a safety valve for accommodating unexpected shifts in the
demand for reserves. During this week the Federal funds rate has spiked to as high
as 100% for individual trades, despite attempts by the Federal Reserve to avoid such
spikes. The size of these spikes in interest rates also go a long way toward explaining
the Federal Reserve’s reluctance to attempt short run control of reserves as well as
providing evidence that banks have a very inelastic short run demand for reserves.
Given the perceived inability to control the short run supply of total reserves and the
Federal Reserve’s "policy" of accommodating reserve shortfalls through the discount
window, it is important to understand how altering the mix of reserves between
borrowed and nonborrowed reserves has real effects on interest rates. After all,
borrowed and nonborrowed reserves both satisfy reserve requirements equally well.
The reserve mix matters because banks are reluctant to borrow from the discount
window.14 As a result, if the Federal Reserve fails to provide the necessary reserves
to meet reserve demand through open market operations, banks will first try to meet
their reserve requirements through the Federal funds markets, thereby bidding up the
Federal funds rate. And only when the Federal funds rate has risen sufficiently to
overcome banks’ reluctance to borrow from the discount window will banks borrow
the necessary reserves. This rise in the Federal funds rate will then lead to lower
output and to banks and individuals reducing deposit levels in the future. This is
sometimes referred to as demand management
III. A Short History of Federal Reserve Operating Procedures
13This has been especially true in the last few years.
14The source of this reluctance to borrow from the discount window is not immediately apparent,
especially since reserves acquired through the discount window typically carry a lower explicit interest
rate than reserves borrowed through the Federal funds market (i.e. the discount rate is usually below the
Federal funds rate). The basic reason for the discrepancy is that individual banks do not have unlimited
borrowing privileges and that borrowing through the discount window involves exercising a nontransferable option to borrow again in the near future. Beyond this, the banking system’s ability to
bypass the restrictions on individual banks’ borrowing privileges is severely constrained by the fact that
individual banks are not allowed to lend in the Federal funds market and borrow from the discount
window at the same time. Thus, a single bank can use the discount window only to make up its own
short fall and its gain from doing so is limited the size of its own reserve short-fall times the difference
between the current Federal funds rate and the discount rate. See Goodfriend (1983) for a detailed
examination of these issues.
8
Examination of equations 1 and 2 suggests that <|> should change if the Federal
Reserve changes its operating procedures. It should be near 0 under strict
nonborrowed reserves targeting regimes and near 1 under strict borrowed reserves
targeting regimes. The value of <|) under mixed regimes will depend on the nature of
the mix. In reality, all of the operating procedures covered in the data sample are
mixed procedures, though most have been close to strict borrowed reserves targeting.
This section describes the history of Federal Reserve operating procedures and
develops a set of dates for sub-sample testing in order to examine potential problems
with parameter instability in § and other regime shift induced changes. The section
also attempts to give the reader a feel for the evolution of the monetary policy process
through time and for some of the subtleties of how the procedures were implemented
over time. This will provide some useful benchmarks for the interpretations of the
sub-sample results presented in Section IV of the paper. For those uninterested in this
topic, the last paragraph in this section provides a quick summary.
Broadly speaking, the period from 1959 to the present can be split into five periods.
An attempt has been made to define reasonably consistent and homogeneous sub
periods. A lot of judgment is involved in this process. Where dates have attained a
place in the literature, such as the "vanishing liquidity" effect that Melvin (1983)
suggests began at the end of 1972, the literature date is used, as long as it roughly
corresponds to actual operating regime history.15 A more complete survey of these
issues can be found in Meulendyke (1989). The five periods used in this article are
as follows:
1959-1966, Free reserves targeting before the modem Federal funds market
1966-1972, Free reserves targeting and the bank credit proviso
1972-1979, Money growth / Federal funds targeting
1979-1982, Nonborrowed reserves targeting
1982-present, Borrowed Reserves/Federal funds targeting
1 9 5 9 -1 9 6 6
Though the Federal funds market actually dates back to the early 1950s, it was not
until the mid-1960s that the Federal funds market and reserve management began to
resemble their modem equivalents. Meulendyke (1989) describes this as follows:
The interbank market was not very broad as the 1960s
began, but activity was expanding. Until the mid-
15In order to make sure that none of the empirical results presented later are dependent on specific
dates, a grid search of 12 months on each side of every proposed date was performed for all dates
except the 1979 and 1982 dates. Only dates after October 1979 and before October 1982 were checked.
No sensitivity to dates over these ranges was uncovered.
9
1960s, the Federal funds rate never traded above the
discount rate. During "tight money periods," when the
desk was fostering significant net borrowed reserve
positions, funds generally traded at the discount rate,
and the funds rate was not considered a useful indicator
of money market conditions.... There was considerable
surprise when the funds rate first rose above the
discount rate briefly, in October 1964 and more
persistently in 1965. As large banks became more
active managers of the liability side of their balance
sheets, they borrowed funds in the market in a sustained
way.16
As a result, the pre-1966 period may be different than subsequent periods. The
reserves market simply worked in a different fashion than it does today. Basically,
this period was characterized by free reserves targeting. Free reserves are the negative
of borrowed reserves plus excess reserves (-BR+ER). Free reserves targeting is a
variant of borrowed reserves targeting with explicit accommodation of excess reserves
(<f) should be approximately equal to 1). In the modem form of borrowed reserves
targeting the accommodation of excess reserves is implicit. The Federal Reserve had
no quantitative operating targets during this period and the desk in New York simply
sought to stabilize general money market conditions17, in line with the FOMC’s
directives.
1 9 6 6 -1 9 7 2
The modem era begins in 1966. The period from 1966 through 1972 was still
characterized by a free reserves operating procedure, but the Federal funds market had
begun to perform the same function it does today as a major source of bank liquidity.
This period was also marked by the introduction of the "proviso" on bank credit into
the FOMC’s policy directive. The proviso stated that if bank credit growth deviated
significantly from target then the desk could adjust the Free Reserve target
"modestly." Brimmer (1971) argues that this shift to a quantitative goal for the
operation of monetary policy was a major landmark for Federal Reserve operating
procedures, which he labels "The Reform of 1966." As mentioned earlier, before this
shift, there were no quantitative objectives. Brimmer also argues that this signaled
a substantial increase in the role of monetarist thought within the FOMC. This period
ended in the early 1970s as monetary aggregates slowly replaced bank credit as the
16Meulendyke (1989), p. 36-36
17
Money market conditions can be thought of as a very loose combination of interest rates and
reserve pressures thought to relate to the general availability of credit
10
main focus of long run policy and the Federal funds rate began to replace free
reserves as the main operating focus. In general, the Federal Reserve was still
stabilizing general money market conditions and was still accommodating reserve
demand shocks through nonborrowed reserves. Again, $ should be close to 1 during
this period.
1 9 7 2 -1 9 7 9
It is interesting to note that the shift from the free reserves targeting of the previous
period to the Federal funds targeting of the 1972-1979 period was not viewed as a
major event Borrowings targets and Federal funds rate targets are quite similar in
practice, though the dynamics are not quite identical. Stable borrowing levels usually
imply stable interest rates. The difference in procedures only becomes evident when
there is a shift in the borrowings function. In a free reserves targeting procedure a
shift in the borrowings function will cause interest rates to change. In a Federal funds
rate targeting procedure, the reserve mix will be adjusted to exactly offset the shift
in the borrowings function and keep the funds rate steady. Thus, <{>should be less
than 1, but closer to 1 than 0 during this period.
This period is dated as beginning at the end of 1972, following Melvin (1983).18
The Federal funds rate targeting procedure was not actually made public in the
FOMC’s policy directive until 1974, though it could be argued that the switch
occurred earlier. Brimmer (1971), for example, dates the shift to September of 1970.
Meulendyke (1989) suggests that while the change began in the early 1970s, its
implementation was gradual, lasting until the mid to late 1970s. The ambiguity arises
from the fact that the period was characterized by progressively tighter targeting of
the Federal funds rate rather than an abrupt shift to strict Federal funds targeting. It
was still very much a borrowing reserves targeting type of procedure, except that the
operating procedures automatically adjusted for shifts in the borrowings function by
stabilizing the Federal funds rates. These operating procedures were in use until 1979
when nonborrowed reserves targeting was introduced. In general, much of the
motivation for the 1979 change resulted from a conviction that the preoccupation with
keeping interest rates stable that had developed in the 1972-1979 period had created
an inflationary bias in the application of policy and that only by allowing substantially
more interest volatility could inflation be tamed. Similarly, monetary growth was
regularly exceeding target growth ranges due to an unwillingness to raise interest rates
enough.
1 9 7 9 -1 9 8 2
During this period the Federal Reserve adopted a nonborrowed reserves operating
18Melvin (1983) actually cites the collapse of Bretton Woods as the precipitating event
11
procedure in order to assert greater control over the money stock. Nonborrowed
reserves targeting was the most complicated of the reserve operating procedures that
the Federal Reserve has ever used and it lasted the shortest length of time. This
operating procedure, in fact, targeted neither nonborrowed reserves or any other
reserve aggregate in any straight forward way. Considerable debate within the Federal
Reserve system about how these procedures actually worked is still going on. Despite
the avowed intention of controlling money, the Federal Reserve missed its money
targets more in this period than in any other before or since. On the other hand,
numerous studies, such as Spindt and Tarhan (1987) and Strongin and Tarhan (1990),
show that the desire to control money played a large role in the conduct of policy.
Financial deregulation and a general increase in economic volatility make it very
difficult to sort out exactly what happened during this period. What is clear is that
Federal Reserve operating procedures were quite different during this period than in
any other, and that little if any effort went into stabilizing either borrowings or
interest rates. Therefore, § should clearly be less than 1 during this period.
The basics of the procedures were quite simple. Short run money growth targets were
established at each FOMC meeting. Nonborrowed reserves targets consistent with
these money targets were then derived. The nonborrowed reserves targets were then
revised each week in order to bring money back to path. If money growth was above
target, the nonborrowed reserves path was lowered. This meant that short run
fluctuations in money and the resulting total reserves movements were being used by
the Federal Reserve to determine future nonborrowed reserves targets (i.e.
nonborrowed reserves targets changed each week in response to last week’s reserve
demand shock) and the market used this same information to forecast future Federal
Reserve actions and interest rates [see Strongin and Tarhan (1990)]. Thus, total
reserves movements contained information on future Federal Reserve actions that had
contemporaneous effects on interest rates due to Federal Reserve actions that had not
yet been taken. This interaction between reserve demand shocks and policy could
potentially cause the total reserve variable to absorb more of the policy signal over
this period than any other. There are a number other potential problems with this
period. First, there is very little data; the operating procedures lasted only 3 years.
Second, the first six months of this period are suspect, in that it took a while to
develop and implement the new operating procedures. And third, the last six months
are equally suspect, because dissatisfaction with the procedures led the FOMC to
accept larger and larger deviations from its short run money targets during 1982 until
borrowed reserves targeting was officially adopted in October of 1982.
1 9 8 2 -p re se n t
After October of 1982, the Federal Reserve adopted a borrowed reserves targeting
procedure, where the borrowing target changed only by policy action rather than in
response to reserve demand shocks. During this period <j>should be close to 1. The
operating procedures in this period are very similar to the procedures in force in the
12
early 1970s, except that the Federal Reserve has shown much more willingness to
change the Federal Funds rate.
A note on interest rate volatility
As an additional note, something should be said about the Federal Reserve’s attitudes
toward interest rate volatility across these periods. This is not an easy task in the
sense that the Federal Reserve typically does not have a specific policy toward interest
rate volatility per se. Nevertheless, the Federal Reserve’s attitudes toward interest rate
volatility have undergone some large shifts over the years. Beginning in the early
1970s, the Federal Reserve began to look upon interest rate stabilization as an end in
itself, and the target ranges for the Federal funds rate became progressively smaller.
Earlier periods had emphasized the stability of overall money market conditions. The
1979 shift was in large part a rejection of the interest rate smoothing goal and interest
rate volatility was no longer avoided. As the monetary control experiment ended in
1982, dissatisfaction with the large amount of interest rate volatility that nonborrowed
reserves targeting had created was one of the strongest factors in the rejection of
continued nonborrowed reserves targeting. The 1982-1991 period has once again seen
increasingly narrow ranges for Federal funds rates, though the target ranges are moved
more often than in the 1972-1979 period. Thus, 1972-1979 and 1982-1991 periods
stand out as having interest rate smoothing as a relatively important Federal Reserve
operating goal, especially the 1972-1979 period. The 1979-1982 period stands out as
being by far the least concerned with interest rate smoothing. The implications of
these changes in attitude are not obvious for reserve management, but it is likely that
if a significant percentage of reserve actions were aimed at stabilizing interest rates,
it might dampen observed interest rate effects of policy.
By and large one would expect to see differences between periods, but only the 19791982 period represents a major shift in operating procedures. All of the other
operating regimes are variants of borrowed reserves targeting and Federal funds rate
targeting, which in the course of Federal Reserve history have tended to blur into each
other and are both characterized by <j>s much closer to 1 than 0, though 1972 to 1979
should have the lowest <)) of these periods due the heavy emphasis on interest rate
targeting. Within the literature, Melvin (1983) suggests that 1973-1979 has a
diminished liquidity effect, and Cochrane (1989) finds it reappeared in 1979.
However, most work finds little or no liquidity effect in any of these periods [see
Reichenstein (1987) for a survey of this work]. Further, it is also clear the 1972-1979
is characterized by the greatest desire to smooth interest rates, while 1979-1982 is
characterized by the least desire to smooth interest rates.
IV. Estimation and results
Estimating the specification outlined in Section II is actually very easy. Noting that
the forecast innovation in total reserves is vd, vs can be identified simply by having
13
total reserves immediately precede nonborrowed reserves in a standard Choleski
decomposition. Once this is done vs can be identified as the orthogonalized error in
the nonborrowed reserves equation.19 Normalizing total reserves and nonborrowed
raises a question since the normal method of using log levels would violate the linear
identification developed in Section EL In the results that follow total reserves and
nonborrowed reserves are both normalized by the level of total reserves in the prior
month, i.e. both are divided by the lag of total reserves. This preserves the strict
linear identification formulated in Section EL Also, by explicitly including a reserve
mix variable ( i.e. nonborrowed reserves divided by the lag of total reserves) in the
specification, it is easier to examine issues involving the persistence of policy shocks
and the effects of anticipated policy actions.20 More importantly by explicitly
using a measure of the mix of nonborrowed to total reserves (NBRX) it is
possible to separate the contemporaneous modeling of reserve accommodation
from the observation that the Federal Reserve influences interest rates through
controlling the mix rather than the level of reserves. This turns out to be
important in analyzing the impact of policy innovations on output since the results
presented hinge critically on the reserve mix, but are fairly robust to specification of
the contemporaneous error structure. In fact, as we will see, eliminating the total
reserve shock entirely has only a second order impact on the system as long as the
nonborrowed to total reserve mix is used to identify policy rather than rather than a
variable which relates to the level of reserves. A direct implementation of the
specification from Section II in levels would simultaneously implement both aspects
of the identification scheme making it impossible to assess which was more important.
The proposed specification has the addition technical benefits of allowing easier
analysis of policy persistence.
The results are presented in two subsections, Sub-sample analysis and Full sample
analysis. The Sub-sample analysis section analyzes the liquidity effect in each of the
sub-samples enumerated earlier. Two sets of VARs are presented for each sub
sample. The first are three variable VARs containing TR (total reserves), NBRX (the
nonborrowed reserves mix)21* and FF (the level of the Federal funds rate) in that
19
The essential point is thatby projecting nonborrowed reserves on total reserves, the policy
disturbances can be extracted from the Federal Reserve’s accommodation of reserve demand shocks.
This is easily done in the context of sequential OLS estimations or structural VARs.
20A number of other normalizations and linear approximations were also examined, including
simply using logs for both reserve series and a variety of demand and mix variables, where a mix
variable is some variable that measures the mix of reserves as opposed to the demand for total reserves.
The general notion is that a reserve demand proxy must precede a reserve mix variable. Variable choice
made very littledifference on the qualitative results analyzing the impact of the decomposed errors.
21The nonborrowed reserve series is adjusted for extended credit and so-called special borrowing.
These are borrowing programs which supply reserves to banks on an extended basis and do not involve
the using up the borrowing privilege discussed earlier. These reserves are implicitly treated by the open
14
order. Data for TR and NBRX are seasonally adjusted and adjusted for reserve
requirement changes. TR and NBRX as noted above are divided by the level of total
reserves in the prior month. The data is all monthly and the data sample begins in
January 1959 and ends in February of 1992. The second group of monthly VARs
presented are 5 variable VARs. These VARs contain IP (the log of industrial
production) and CPI (the log of the consumer price index) preceding TR, NBRX and
the Federal funds rate in the Wold causal ordering.22 The ordering of TR, NBRX,
and the Federal funds rate is dictated by the identification scheme developed in
Section EL IP and CPI are placed first, so that policy can respond to on going
events.23 This ordering is also consistent with Sims (1992), Eichenbaum (1992) and
Christiano and Eichenbaum (1992a and 1992b). All sub-sample VARs were run with
6 lags of data. (Longer lags were also estimated in those cases where there was a
sufficient number of observation to perform a reasonable estimation with little
difference in results.) All innovations discussed are the orthogonalized innovations
from the ordering described above. The first set of VARs correspond most closely
to the liquidity puzzle literature and most of the important results about the liquidity
effect can be seen in this first set of VARs. The second set is presented to
demonstrate the robustness of the results to more complicated systems and to provide
a bridge to the third set of VARs presented in the Full sample subsection. The Full
sample analysis subsection uses the same data as the 5 variable VARs, but uses 12
lags of data. This subsection emphasizes the identification of monetaiy policy and the
assessment of its impact on the economy.
Sub-sample analysis
Figures 1 and 2 show the impulse responses of TR, NBRX, and FF to TR and NBRX
innovations. The error bands are plus and minus 2 sigma bands and are generated by
Monte Carlo techniques. Each column shows one impulse response function for each
of the five time periods. Figure 1 contains the responses to TR innovations and
Figure 2 the responses to NBRX innovations, the proposed policy measure. The
reserve to reserve impulse response functions are the response to a one percent
innovation in reserves. This is done so that initial values are interpretable as <{>. All
other impulse response functions are the response to a one standard deviation shock.
Table 1 shows the estimates of <|) and its standard error and the concurrent period
market desk as nonborrowed reserves. This has no impact prior to the 1980s and has only marginal
impact since. The adjustments simply eliminate outliers that result from episodes such as the failure of
Continental Bank.
22There are a number of structural VARs that could easily be estimated, based on Sims (1986).
Such models were estimated and provide very similar results. The Wold ordering results are presented
for simplicity and for comparability to previous results. Differencing IP and CPI also generates similar
results. The general question of identification isaddressed in the appendix.
23The placement of CPI and IP in the Wold ordering is not important.
15
estimate of the liquidity effect for a one standard deviation shock in NBRX on the FF
and its standard error for each sub-period. Table 2 shows the decompositions of
variance for the effect of NBRX innovations on TR and the Federal funds rate and
the decomposition of variance of TR innovations on the Federal funds rate for the end
of year 1 and 2 (standard errors are in parentheses).
The key results are that for each and every sub-sample there is a clear liquidity effect
and it is always negative and highly significant. Further, the persistence of NBRX
shocks varies considerably across sub-samples and so does the persistence of the
liquidity effect on interest rates (more on this later). The TR or reserve demand shock
has the opposite sign on interest rates and thus reflects the "normal" perverse result
in the literature, suggesting that previous specifications have often confused reserve
demand shocks with policy disturbances. It also shows why the use of broader
aggregates often leads to the absence of an estimated liquidity effect.
Also of interest is the response of TR to NBRX innovations and of NBRX to TR
innovations. First, there is a strong contemporaneous impact of total reserves on
NBRX implying that (J) is clearly not zero and that the reserve demand correction is
statistically and quantitatively important Second, other than this contemporaneous
effect, NBRX and TR are not strongly related. This suggests that the ability to pick
up policy innovations from TR or the monetary base (to say nothing of the broad
aggregates) is likely to prove extremely difficult. In addition, it is likely that even for
relatively long sampling periods, such as quarterly or annually, NBRX is likely to
remain a significantly better measure of monetaiy policy than a broad aggregate, even
when used in isolation.24
Another very important implication of these impulse response functions taken as a
group concerns how the persistence in NBRX’s response to policy disturbances relates
to persistence in the estimated liquidity effect The more persistent NBRX’s response
is, the more persistent the liquidity effect is, suggesting that anticipated movements
in NBRX are quite important. In fact, careful examination of the "Effect of NBRX
on NBRX" graphs in comparison with the "Effect of NBRX on FF" graphs shows a
remarkable similarity, sub-period by sub-period. If NBRX’s response to an NBRX
innovation damps, FF’s response to an innovation in NBRX damps. If NBRX’s
response to an innovation in NBRX changes sign n periods in the future, FF’s
response also changes sign approximately n periods in the future. This is true for
every sub-period with shapes matching in each case. This suggests that Federal
Reserve actions have significant effects on interest rates, regardless of the policy
regime and the degree to which Federal Reserve actions are anticipated.
24In alternate specifications which include money itisclear that total reserve shocks are largely
technical arising from instabilities in the reserve multiplier. This follows from the fact that while NBRX
isuseful for forecasting money, TR has only limited ability to forecast future money growth. Impulse
response functions for a typical six variable VAR with money are included in the appendix.
16
Table 1 indicates that <j) varies across time in line with the expectations outlined in
Section HI. For the borrowed reserves targeting periods, 1959-1966, 1966-1972 and
1982-1991 the values are in the .95 range and are statistically indistinguishable from
1. In the Federal funds targeting period, 1972-1979, and in the NBRX targeting
period, 1979-1982, <|) is .656 (.167) and .546 (.296), respectively, indicating much
more mixed operating procedures, though in no case would the hypothesis of a pure
borrowed reserves targeting operating procedure be rejected. Table 1 also shows that
in sharp contrast to Leeper and Gordon (1992) and most of the previous literature
there is a strong negative impact on FF from a positive innovation in monetary policy
as measured by the innovation in NBRX in each and every sub-period. This
estimated liquidity effect is strong, immediate and statistically significant in all sub
periods. Consistent with previous findings25, the liquidity effect is especially strong
in the 1979-1982 sub-period, nearly 5 times greater than for any other sub-period;
however, given the very short sample period, one should not be make too much of
this result.
Table 2 provides some evidence that the Federal Reserve’s attitude toward interest
rate smoothing is important. Specifically, Table 2 shows that in the two periods
where interest rate smoothing was of the greatest interest -- 1972-1979 and 1982-1991
— NBRX innovations had the least impact on interest rates. Similarly, the largest
explanatory power was in the 1979-1982 period, when the least attention is being paid
to interest rate smoothing. Overall these results show a very strong correspondence
to the history of operating procedures covered in Section III.
Table 3 shows the estimate for (J) and its standard error and the concurrent period
estimate of the liquidity effect for the NBRX innovations on the Federal funds rate
and its standard error for each sub-period for the five variable system. The estimates
of <J>are similar to those shown in Table 1 and are still within 2 standard deviations
of 1. The estimated liquidity effects are slightly smaller in magnitude in all 5 cases,
only in the 1979-1982 period is the change significant. Nevertheless, the estimated
liquidity effects are still negative and highly significant in all sub-periods.
Table 4 shows for the five variable VARs the decompositions of variances reported
in Table 2. These results are generally consistent with the results shown in Table 2,
though in most cases the explanatory power is lower. The greater uncertainty
associated with point estimates of the decompositions of variance suggest that none
of these differences are significant. Figure 3 shows the impact of TR and NBRX
innovations on the Federal funds rate for all the sub-periods as well as the NBRX own
effect, so that we can examine the persistence result shown in the three variable case.
Once again it is clear that the persistence of NBRX’s response to a policy shock is
directly related to the persistence of the liquidity effect, indicating that expected
^Cochrane (1989) and Strongin and Tarhan (1990)
17
policy actions matter.
The Full sample analysis
The full sample five variable VAR reinforces these general results. Figure 4 shows
the complete impulse response functions to one standard deviation shocks for the five
variable system estimated over the full sample in line with Sims (1992) and
Eichenbaum (1992). The overall results in Figure 4 are consistent with the previous
sub-sample analysis. They also meet Sims’ and Eichenbaum’s criteria for a good
measure of monetary policy. Real output effects are positive, statistically significant
and persistent in levels, but not in changes. Total reserves rise in response to an
easing action. Early price effects are highly uncertain and near 0, though the point
estimates are negative; however, the long run price effect are positive and
significant.26 TR innovations have a strong positive and persistent impact on the
price level indicating that to the extent that money is allowed to grow, inflation rises.
This suggests that passive behavior on the part of the Federal Reserve may have
important implications for long run inflation, which is quite consistent with the
monetarist position that inflation is not the result of transient policy actions, but the
result of persistent biases in the application of the efforts to smooth interest rates and
output Again, it should be noted that TR innovations have positive, but imprecise
effects on the Federal funds rate. These effects are consistent, not with supply
innovations, but with reserve demand shocks.
The decompositions of variance for the full sample VAR are presented in Table 5.
NBRX innovations, the proposed monetary policy measure, account for 49% (std. dev.
10.04) of the variance of IP at the end of two years. This compares with only 0.4%
using Eichenbaum’s (1992) specification that uses only the log of NBRX without the
reserve demand correction (The main results for Eichenbaum’s specification are
shown in Table 6. DNBR is the log level of nonborrowed reserves with the extended
credit and special borrowings corrections.) NBRX innovations also account for 26%
(std. dev. 6.76) of the variation in the Federal funds rate at the end of 1 year, as
opposed to Eichenbaum’s 11% (std. dev. 5.63). More importantly, the residual
explanatory power of interest rates to explain IP variance at the end of two years falls
from 36% (std. dev. 11.8) in Eichenbaum’s specification to only 0.23% (std. dev.
1.91) in the proposed specification.
On the other hand, NBRX has virtually no explanatory power for either TR or CPI,
which as stated before is consistent either with the notion that most policy actions at
the monthly frequency do not represent innovations in long run policy goals for either
prices or the monetary aggregates or that the impact of policy on inflation has long
26The graphs do not show enough time span for the price effects to become significantly positive,
which happens around a 5 year horizon.
18
lags. However, positive innovations in TR generate a significant and persistent
positive impact on the price level, suggesting that the Federal Reserve’s passive
accommodation of reserve demand has a more immediate impact on inflation,
although the decompositions of variance suggest that this has not been a very
important factor, only explaining about 11.76% (std. dev. 8.15) of the variance of CPI
at the two year horizon. Perhaps what is most striking about the results in Table 5
is the general lack of explanatory power in interest rates innovations once TR and
NBRX are taken account of.
There is an additional important point about alternative identification schemes that is
evident in Table 5, but is even more clear in Table 6. Many questions about the very
strong assumptions made in the identification scheme proposed in this paper can be
reduced down to a very simple question with careful examination of the
decomposition of variance. The three variables in this system that can be reasonably
assumed to have some explanatory power for output based on monetary policy are
TR, NBRX and FF. In fact, as noted earlier, each of these variables has been
suggested as the one true measure of policy. If the question of VAR identification
was redefined along the line of canonical correlations, we could ask what linear
combination of these variables has the most explanatory power for industrial
production. In general such a combination would not have any economic
interpretation. In the present case, examination of either Table 5 or 6 shows that this
best linear combination would be the NBRX-<(>TR, the exact combination suggested
as a measure of policy. This correspondence follows from the fact that the
explanatory power of the FF and TR shocks are effectively 0. The very low
explanatory power of TR and FF (always less than 2% regardless of horizon) also
indicates that there is only one shock of any significance for IP contained in the linear
space defined by TR, NBRX and FF. The clear implication of this is that all any of
various possible alternate identification schemes can do is redistribute the explanatory
power of this one shock. Thus, the only real identification question is what should
NBRX-(J)TR be called. Any other shock constructed from these innovations would
either be uncorrelated with output or just relabeled NBRX-<|>TR. As the earlier part
of the paper argues, there are strong institutional reasons to argue that monetary policy
is the correct label for NBRX-<|)TR.
Table 6 shows the way explanatory power for IP can be reassigned by various Wold
orderings. It also shows two alternate 4 variable VAR systems. Alternate model 3
which shows a NBRX specification without TR and Alternate model 4 which shows
Eichenbaum’s suggested specification. As can be seen there is only a small loss in
explanatory power from dropping TR. However, there is a much larger loss when the
change in nonborrowed reserves (DNBR) is used. Clearly the mix is the key concept
As a point of comparison to previous work, Granger F-tests show that NBRX Granger
causes output (F-statistic=2.73, P-value=.001), while interest rates do not (F-
19
statistic=.95, P-value=.49).27 This is in contrast to Eichenbaum (1992) or Litterman
and Weiss (1985), where interest rates Granger cause output, but an uncorrected
NBRX measure or any other aggregate does not For instance, in the Eichenbaum
specification, which uses the log of nonborrowed reserves and does not include total
reserves, the F-test for nonborrowed reserves is 1.45 (P-value=.14) and the F-test for
interest rates is 2.90 (P-value=.009).
V. Conclusion
This paper examined recent work on the identification of exogenous monetary policy
disturbances by Sims (1992) and Eichenbaum (1992), as well as Sims (1980) and
Litterman and Weiss (1985). The main finding is that the anomalies documented by
these authors reflected their failure to properly take account of the Federal Reserve’s
policy of accommodating short run reserve demand disturbances. This, in turn, leads
them to misidentify demand shocks as supply shocks. A new method of identifying
the actual supply shocks is proposed and estimated. The new measure of monetary
policy disturbances is used to successfully address all of the difficulties normally
encountered in attempts to measure policy disturbances that were listed in the
introduction. Specifically, a positive innovation in nonborrowed reserves adjusted for
reserve demand shocks has a strong and persistent negative effect on interest rates
regardless of sub-sample and conditioning variables. Second, unlike Sims (1992) but
in line with Eichenbaum (1992), there is no persistent and significant positive price
impact from a contractionary policy disturbance. There is, however, a positive
persistent price effect from the Federal Reserve’s accommodation of reserve demand
shocks. Third, using the proposed specification, NBRX Granger causes output even
in the presence of interest rates and innovations in NBRX explain approximately 49%
of the variance in output over a 2 year horizon. Fourth, the proposed measure of
policy contains all of the explanatory power for industrial production contained in
total reserves nonborrowed reserves and interest rates. Finally, the paper finds strong
evidence that the liquidity effects are persistent and that the persistence is directly
related to the persistence of nonborrowed reserves’ response to an innovation in
policy. This suggests that even anticipated accommodative policy actions have a
substantial negative impact on interest rates.
27
Using the levels specification can cause problems in Granger causality tests; however, ifIP and
CPI are differenced the results are even stronger. The F-statistic for NBRX Granger causing output is
3.45 with a P-value of .017, while the F-statistic for FF Granger causing output is 1.48 with a P-value
of .22. Sub-sample results are inconclusive due to a lack of precision, but they maintain the qualitative
resultthat NBRX is more important than FF.
20
Appendix
The key identifying assumption used in the paper is that policy innovations in
nonborrowed reserves are exactly offset by changes in borrowed reserves. This is
equivalent to assuming that policy has no short run effects on the supply of total
reserves. To more fully understand what happens if this assumption is relaxed, we
need a more general form of the specification in the paper. Also it is generally
argued that the Federal Reserve seeks to smooth interest rates in the short run. To
incorporate these two features, let the forecast innovation in total reserves be
represented as
and the forecast innovation in nonborrowed reserves be represented as
M„6r= < K +V 0 V
where y is the degree to which policy actions in nonborrowed reserves are reflected
contemporaneously in total reserves, 0 is the interest rate smoothing parameter and
<j) is the operating procedure determined split in the accommodation between borrowed
reserves and nonborrowed reserves in response to a reserve demand shock. The
relevant range for y is from 0 to 1. As long as y is greater than 0, a reserve drain (a
negative policy impulse) will cause total reserves to fall contemporaneously. Negative
values would indicate that reserve drains lead to increases in the supply of reserves.
Values above 1 would indicate that reserve injections cause banks to borrow more.
Neither of these possibilities are sensible. Setting y and 0 to 0 reproduces the model
used in the paper. Setting y » 0 would recreate the normal model in the literature
where innovations in supply are fully reflected in total reserves. Interest rate
smoothing requires that 0, the interest rate smoothing parameter, be positive. A
negative value for this coefficient would imply that the Federal Reserve was seeking
to generate volatility in interest rates.
Clearly, this more general model is under-identified. Nevertheless, there is a
reasonable estimation strategy, at least with respect to y. If we take as given the
notion that we are identifying supply and demand shocks and that positive demand
shocks cause interest rates to rise and positive supply shock cause rates to fall, then
we can set y such that the contemporaneous effect of an innovation in supply has as
the maximum negative impact on FF. This adds an identifying assumption of a type
not normally useable in a VAR context If the specification suggested in the paper
is correct then within the constrained space the minimum should occur at y=0. If, as
is normally assumed, total reserves are exogenous then the minimum should occur for
y significantly greater than 0. The interest rate smoothing parameter, on the other
hand, is not subject to a similar method of identification and must be treated as a
conditioning parameter, though as it turns out results are not particularly sensitive to
21
reasonable choices for this parameter.
Figure 5 shows the result of parameterizing y and 0 over the relevant ranges specified
above for the key liquidity effect parameter, A,28. This was accomplished by solving
the quadratic form linking the sample covariance and the structural system. This is
substantively different than the standard structural VAR approach in which the
restrictions are placed on the relationship of forecast errors to each other. Here, the
restrictions are on the parameters which link structural errors to the forecast errors
(i.e. restrictions on Aq’1 in the standard nomenclature rather than on \ ) . As can be
seen X is everywhere negative, ranging from a value of -15 to -35. It can also be
seen that X is minimized when y equals 029, as predicted by the model. It can also
be seen that assuming moderate amounts of interest rate smoothing would only
increase the impact of policy on interest rates. Thus, unless interest rate smoothing
were to be viewed as the major focus of Federal Reserve policy the presented results
would hold up almost exactly as stated, perhaps with a slight increase in the
estimated liquidity effects. Further, we can infer from the canonical correlation
interpretation of NBRX-<|)TR that interest rate smoothing cannot be a first order issue;
however, to fully explore this issue it is useful to take a look at a model which takes
interest smoothing as the key goal of monetary policy. This would be similar in spirit
to Bemanke and Blinder(1992).
Specifically, "Alternate Model 2" from Table 6, where FF precedes the reserve series,
can be viewed as the extreme interest rate smoothing model. In such a system the
innovation in FF would be policy (as in Bemanke and Blinder (1992)) and the
innovation in NBRX would due to shifts in the borrowings function. Such a model
runs into serious contradictions seeking to explain why the remaining innovation in
NBRX still has such large explanatory power for IP, 35% at a two year horizon,
which strongly exceeds FF at 11.5%. Further, it is difficult to understand why shifts
in the borrowings function would produce impulse response functions of the type
shown in Figure 7, which look just like the original NBRX impulse response
functions, especially the interest rate response which still shows lagged effects on
interest rates. This, as stated above, is not at all surprising given the canonical
correlation interpretation of NBRX-<j>TR, which strongly indicates that interest rate
smoothing can only be of second order importance.
A final point on the impact of alternate specification on impulse response functions
28Hgure 5 shows the entire constrained manifold in which the original specification isembedded.
Values outside of this range are either rejected by theory as specified or cannot produce the sample
covariance regardless of how the other coefficients are set The solution method used isbased on
ongoing work with Tom Gittings and Mark Watson. At the fringes of the manifold, parameter values
become complex.
29The global minimum for X occurs when y = -.0883 and 8=.003 in which case A=-39.2.
22
is in order. As would be suspected from the analysis of decomposition of variances
in the paper, alternate specifications do not have much impact on the NBRX impulse
response functions, they simply transfer approximately proportional variants of
NBRX’s impulse response functions to FF and TR and add noise to the NBRX shock.
This is, in fact, the case, as can be seen in Figures 6-8 which show the impulse
response functions for the alternate NBRX models in Table 6, in order. Figure 9
shows a slightly larger system in which M l is included. The only point to this
expanded system is to show how little impact the inclusion of money has, the lack of
a strong short run relationship with TR, and that NBRX innovations generate
permanent increases in the money supply.
23
References
Barro, Robert J. (1978), "Unanticipated Money, Output, and the Price Level in the
United States," Journal o f Political Economy 86, August, 549-80.
Barro, R. J. and M. Rush (1980), "Unanticipated Money and Economic Activity,"
in: Fischer, Stanley, ed., Rational Expectations and Economic Policy (Chicago:
University of Chicago Press), 23-74.
Bemanke, B. and A. Blinder (1992) "The Federal Funds Rate and The Channels of
Monetary Transmission," American Economic Review, Vol. 82 September 1992, pp.
901-921.
Brimmer, Andrew F. (1971), "Evolution and Impact of Monetarism in the Federal
Reserve System," The Political Economy o f Money, December, 1971.
Bryant, Ralph C. (1983), Controlling Money: The Federal Reserve and its Critics
(Washington, D.C.: The Brookings Institution).
Christiano, L. and M. Eichenbaum (1992a), "Identification and the Liquidity Effect
of a Monetary Shock", in Political Economy, Growth and Business Cycles, A.
Cukierman, L.Z. Hercowitz, and L. Liederman (eds.), MIT Press, forthcoming, 1992.
_____________________________ , (1992b) "Liquidity Effects, Monetary Policy and
the Business Cycle," NBER Working Paper No. 4129, Northwestern University.
Cochrane, John H. (1989), "The Return of the Liquidity Effect: A Study of the
Short-run Relation Between Money Growth and Interest Rates," Journal o f Business
and Economic Statistics 7, 75-83.
Eichenbaum, M artin (1991), "Comments on ’Interpreting the Macroeconomic Time
Series Facts: The Effects of Monetary Policy’", European Economic Review, Vol. 36,
June 1992, 1001-1012
Gibson, W.E., (1970), "Interest Rates and Monetary Policy." Journal o f Political
Economy, May/June 1970, 431-55.
Goodfriend, Marvin (1983), "Discount Window Borrowing, Monetary Policy, and
the Post-October 1979 Federal Reserve Operating Procedures," Journal o f Monetary
Economics, 12, 345-356.
__________________ (1990), "Interest Rates and the Conduct of Monetary Policy,"
Working Paper 90-6, Federal Reserve Bank of Richmond, August
24
Judd, John P. and John L. Scadding (1982), Journal o f E conom ic Literature,
September 1982, 995-1002.
King, Robert G. (1990), "Money and Business Cycles," Manuscript, University of
Rochester, October.
Leeper, Eric M. and David B. Gordon (1992), "In Search of the Liquidity Effect,"
Journal o f Monetary Economics, June 1992, 341-369.
Litterman, R. B. and L. Weiss (1985), "Money, Real Interest Rates, and Output: A
Reinterpretation of Postwar U.S. Data," Econometrica 53, January, 129-56.
Melvin, Michael (1983), The Vanishing Liquidity Effect of Money on Interest:
Analysis and Implications for Policy," Economic Inquiry 21, 188-202.
Meulendyke, Ann-Marie (1989),U.S. Monetary Policy and Financial Markets,
Federal Reserve Bank of New York
Mishkin, Frederic S. (1981), "Monetary Policy and Long-Term Interest Rates: An
Efficient Markets Approach." Journal o f Monetary Economics, January 1981,29-55.
____________. (1982), "Monetary Policy and Short-Term Interest Rates: An Efficient
Markets-Rational Expectations Approach." Journal o f Finance, March 1982, 63-72.
____________. (1983), A Rational Expectations Approach to Macroeconometrics
(Chicago, IL: University of Chicago Press).
Reichenstein, W., (1987) "The Impact of Money on Short Term Interest Rates,"
Economic Inquiry, Vol. XXV, No. 1, 67-82.
Romer, C. D. and D. H. Romer (1989), "Does Monetary Policy Matter? A New Test
in the Spirit of Friedman and Schwartz," in: Blanchard, O.J. and S. Fischer, eds.,
NBER Macroeconomics Annual 1989 (Cambridge, Mass.: MIT Press), 121-170.
Rush, Mark, "Unexpected Money and Unemployment, 1920-1983," (1986), Journal
o f Money, Credit and Banking.
Sims, Christopher A. (1980), "Comparison of Interwar and Postwar Business Cycles:
Monetarism Reconsidered," American Economic Review Papers and Proceedings 70,
May, 250-57.
__________________. (1986), "Are Forecasting Models Usable for Policy Analysis?"
Federal Reserve Bank of Minneapolis Quarterly Review, Winter, 2-16.
25
__________________. (1988), "Identifying Policy Effects," in: Bryant, R.C., D.W.
Henderson, G. Holtham, P. Hooper, S.A. Symansky, eds., E m pirical M acroecon om ics
f o r Interdependent E conom ies (Washington, D.C.: The Brookings Institution), 305-21.
__________________. (1992), "Interpreting the Macroeconomic Time Series Facts:
The Effects of Monetary Policy", European Economic Review, Vol. 36, June 1992,
975-1000
Spindt, Paul and Vefa Tarhan, (1987), "The Federal Reserve’s New Operating
Procedures: A Post-Mortem", Journal o f Monetary Economics, Vol. 19, No. 1.
Strongin, Steven and Vefa Tarhan (1990), "Money Supply Announcements and the
Market’s Perception of Federal Reserve Policy," Journal o f Money, Credit and
Banking, May.
26
TABLE 1
3-VARIABLE VARS BY PERIOD
4
Contemporaneous
EffectofNBR on FF
1959-1966
0.920
(0.103)
-0.073
(0.027)
1966-1972
0.951
(0.108)
-0.143
(0.033)
1972-1979
0.656
(0.167)
-0.180
(0.043)
1979-1982
0.546
(0.296)
-0.590
(0.097)
1982-1991
0.909
(0.054)
-0.090
(0.041)
NOTE: Numbers in parenthesis are standard errors.
TABLE2
DECOMPOSITION OF VARIANCE USING 3-VARIABLE VARS
1959-1966
NBR-»TR
TR->FF
NBR-^FF
End of year 1
12.59
(6.20)
1.79
(5.62)
37.22
(14.22)
End of year 2
12.47
(6.46)
1.50
(5.68)
31.35
(15.13)
End of year 1
6.42
(4.86)
3.79
(6.92)
48.51
(18.07)
End of year 2
6.69
(5.64)
3.91
(738)
46.26
(17.41)
End of year 1
7.62
(5.43)
29.33
(15.16)
27.32
(15.12)
End of year2
7.96
(6.22)
30.42
(16.32)
18.19
(16.97)
End of year 1
22.91
(10.55)
33.32
(12.72)
54.17
(13.17)
End of year 2
27.49
(12.84)
34.13
(13.31)
53.50
(13.30)
End of year 1
6.65
(5-52)
0.57
(6.27)
25.88
(16.76)
End of year 2
6.64
(6.74)
0.40
(9.72)
22.10
(18.81)
1966-1972
1972-1979
1979-1982
1982-1991
NOTE: Numbers in parenthesis are standard errors.
TABLE 3
5-VARIABLE VARS BY PERIOD
£
Contemporaneous
Effect ofNBR on FF
1959-1966
0.922
(0.094)
-0.072
(0.021)
1966-1972
1.113
(0.132)
-0.082
(0.024)
1972-1979
0.714
(0.140)
-0.129
(0.031)
1979-1982
0.671
(0.184)
-0.343
(0.046)
1982-1991
0.888
(0.065)
-0.051
(0.025)
NOTE: Numbers in parenthesis are standard errors.
TABLE4
DECOMPOSITION OF VARIANCE USING 5-VARIABLE VARS
NBR->TR
TR-»FF
NBR-»FF
End of year 1
17.25
(5.39)
3.06
(4.63)
43.20
(11.77)
End of year 2
1733
(5.54)
2.67
(4.64)
46.18
(12.16)
End of year 1
4.79
(4.53)
2.67
(5.95)
19.06
(11.28)
End of year 2
5.09
(5.63)
2.60
(6.45)
26.63
(13.85)
End of year 1
6.01
(4.60)
13.20
(11.82)
7.92
(6.64)
End of year 2
6.15
(5.12)
10.89
(10.99)
8.09
(6.72)
End of year 1
8.98
(4.02)
60.68
(13.68)
15.11
(6.48)
End of year 2
8.65
(4.39)
52.54
(14.70)
9.89
(5.12)
End of year 1
5.49
(4.06)
6.17
(8.69)
4.96
(4.95)
End of year 2
7.57
(5.81)
12.51
(12.72)
11.58
(8.66)
1959-1966
1966-1972
1972-1979
19.79rl9.S2
1982-1991
NOTE: Numbers in parenthesis are standard errors.
TABLE 5
DECOMPOSITION OF VARIANCE FOR ALL VARIABLES
USING 5-VARIABLE VAR (1959-1991)
End of year 1
End of year 2
Equation IP
Variance explained hv
IP
CPI
TR
NBR
FF
83.84
0.30
0.87
14.77
0.22
(6.66)
(1.28)
(2.35)
(5.96)
(1-36)
48.96
3.81
0.64
46.36
0.23
(10.04)
(4.15)
(2.89)
(10.03)
(1.91)
IP
CPI
TR
NBR
FF
18.98
65.26
8.87
4.06
2.83
Equation CPI
(7.44)
25.87
59.42
(8.89)
(5.40)
11.76
(3.94)
1.87
1.08
(2.83)
(10.28)
(11.57)
(8.15)
(3.28)
(2.64)
IP
CPI
TR
NBR
FF
2.91
5.02
83.58
2.54
5.95
Eqg&iQn TR
3.21
(1.67)
6.28
(2.25)
80.60
(3.45)
3.23
(1.58)
6.68
(2.23)
(1.75)
(2.58)
(3.96)
(1.76)
(2.46)
IP
CPI
TR
NB R
FF
27.44
4.37
12.69
53.77
1.73
Equation NBR
31.26
(7.31)
4.22
(3.20)
(3.91)
12.65
(6.89)
49.75
(1.61)
2.13
(7.66)
(3.13)
(4.75)
(7.05)
(2.56)
Equation FF
IP
CPI
TR
NBR
FF
40.29
2.50
8.48
25.95
22.78
(8.36)
(2.82)
(5.72)
(6.76)
(5.69)
NOTE: Numbers in parenthesis are standard errors.
45.73
4.48
11.66
18.24
19.90
(9.40)
(4.91)
(8.44)
(5.91)
(6.90)
TABLE 6
Variables listed
in Wold causal
order
End of
Year 1
Decomposition of Variance at
End of
Year 2
End of
Year 3
Proposed 5
Variable VAR
IP
CPI
TR
NBRX
FF
83.8(6.7)
0.3(1.3)
0.9(2.3)
14.8(6.0)
0.2Q.3)
49.0(10.0)
3.8(4.2)
0.6(2.9)
46.4(10.0)
0.2(1.9)
34.4(9.3)
5.6(6.3)
1.2(3.9)
58.6(10.4)
0.2(2.4)
83.8(6.4)
0.3(1.3)
12.7(6.0)
3.0(3.7)
0.2(1.1)
49.0(9.6)
3.8(4.7)
30.3(10.6)
16.7(8.9)
0.3(1.9)
34.4(8.6)
5.6(6.7)
33.3(11.5)
26.5(11.5)
0.2(2.5)
83.8(6.6)
0.3(1.3)
3.8(3.5)
1.5(2.3)
10.6(5.4)
49.0(9.6)
3.8(4.5)
11.5(7.1)
1.1(3.0)
34.5(9.6)
34.4(8.9)
5.6(6.6)
16.3(9.2)
1.1(3.1)
42.6(10.5)
83.6(6.6)
0.5(1.3)
15.3(6.3)
0.7(1.6)
48.4(10.1)
6.2(5.6)
42.9(9.9)
2.5(3.4)
34.2(9.3)
9.5(8.8)
51.2(10.7)
5.1(5.3)
90.4(5.6)
0.3(1.4)
3.8(3.8)
5.5(4.1)
70.8(10.8)
6.3(6.1)
9.1(7.7)
13.8(8.1)
54.8(12.9)
15.5(10.3)
8.2(7.8)
21.4(10.9)
Alternative Model 1
IP
CPI
NBRX
TR
FF
Alternative Model 2
ip
CPI
FF
TR
NBRX
Alternative Model 3
IP
CPI
NBRX
FF
Alternative Model 4
ip
CPI
DNBR
FF
NOTE: Numbers inparenthesisarestandarderrors.
F IG U R E 1.
Effect of TR on TR
Effect of TR on NBRX
Effect of TR on FF
58:8 to 65:1?
58:8 to 65:12
58:8 to 65:12
1966:1 TO 1972:12
1966:1 TO 1972:12
1966:1 TO 1972:12
1973:1 TO 1979:10
1973:1 TO 1979:10
1973:1 TO 1979:10
1
7
13
1
19
13
19
1982:11 TO 1992:2
1.251
I
1982:11 TO 1992:2
0.75-
0.60
\
0.20
0.00
-0.20
^v
------------
\V / ,
1 1 v.
— — -------------V
--------------1
7
13
19
MOTE: Error bands are + 2 sigma.
7
0.500.25I v /a I "
ftft.
u.uu
ft.4OC.
S v V
. “U
CO*
^
— -------
-0.50*
13
13
19
1979:11 TO 1982:10
1 .00- 1 .
0.80
0.40
1
1979:11 TO 1982:10
1979:11 TO 1982:10
1.20
1.00
7
19
0.40i
0.300.20 0.10 n ftft.
u.uu
-.10 -.20 -.30-.40-.50-
1982:11 TO 1992:2
•»»***
s
\
1
s~
\ ___.
7
13
19
F IG U R E 2 .
Effect of NBRX on TR
Effect of NBRX onNBRX
Effect of NBRX on FF
58:8 to 65:12
58:8 to 65:12
58:8 to 65:12
1966:1 T O 1972:12
1966:1 TO 1972:12
1966:1 T O 1972:12
1973:1 T O 1979:10
1973:1 T O 1979:10
1973:1 T O 1979:10
1979:11 T O 1982:10
1979:11 T O 1982:10
1979:11 T O 1982:10
1982:11 T O 1992:2
1982:11 T O 1992:2
1982:11 T O 1992:2
F IG U R E 3.
Effect of TR on FF
Effect of NBRXon FF
58:8 to 65:12
Effect of NBRXon NBRX
58:8 to 65:12
FROM 1966:1 T O 1972:12
FROM 1966:1 T O 1972:12
FROM 1966:1 T O 1972:12
FROM 1973:1 T O 1979:10
FROM 1973:1 T O 1979:10
FROM 1973:1 T O 1979:10
FROM 1979:11 T O 1982:10
FROM 1979:11 T O 1982:10
FROM 1979:11 T O 1982:10
FROM 1982:11 T O 1992:2
FROM 1982:11 T O 1992:2
FROM 1982:11 T O 1992:2
F IG U R E 4.
Effect of CPI on IP
EffectofIPon IP
Effect of TR on IP
0.0064
0.0048
0.0032
0.0016
0.0100 v
0.0075
0.0050 ' \ y ^ ___
0.0025
0.0000
\
”
-.0025
\
-.0050
—
-.0075
8
30
0.0000
-.0016
-.0032
-.0048
-.0064
8
30
8
30.
Effect of CPI on TR
Effect of IP on TR
30
Effect of TR on TR
8
. Effect of NBRX on TR
0.00160
0.0080
0.00120
0.0060
0.00080
0.00040
0.0040
0.00000
-.00120
-.00160
-.00200
NOTE: Error bands are
_ ___
1/
0.0000 ,'m1Hr— —
1
%
\
-.0050
+ 2 sigma.
30
33
Effect of NBRX on NBRX
Effect of CPI on NBRX
0.0025
8
Effect of FF on TR
-.00040
-.00080
9
-.0025
30
0.0100
0.0020
0.0000
-.0020
Effect of IP on NBRX
30
Effect of FF on CPI
Effect of CPI on CPI
8
8
8
30
Effect of FF on NBRX
F IG U R E 6.
Effect of IP on IP
Effect of CPI on IP
• Effect of TRon IP
0.0125
0.0032]
-"
0.0016 •*/i\
0.0100 f \ \
0.0000
0.0075 h \ \
-----* -.0016
0.0050
v\.
0.0025
-.0032 ^\V--- 1^-\
-.0048
0.0000 —
\
— •
-.0025
-.0064
\
^*-.0050
-.0080
-.0096
-.0075
8
30
8
30
tv
8
30
Effect of IP on CPI
0.0016] #A
0.0000 h
\VnN\
-.0016 \\
^
^
-.0032 V\
-.0048 \ \
-.0064
\
-.0080
\\
V.
-.0096
N__
-.0112
8
30
Effect of TR on CPI
0.0060
8
30
Effect of FF on CPI
0.0040
0.0020
0.0000
-.0020
-.0040
7
Effect of IP on NBRX
8
28
30
Effect of NBRX on NBRX
. Effect of CPI on NBRX
0.0025
0.0020
0.0010
0.0000
-.0010
-.0020
A11
0.0000 /‘'ViJbMr - ---- ~
-.0030
-.0040
-.0050
-.0060
-.0070
a
-.0025
-.0050
8
Effect of IP on TR
8
30
. Effect of CPI on TR
0.0015
0.0010
0.0005
0.0000
-.0005
-.0010
-.0015
-.0020
8
30
Effect of NBRX on TR
0.0060
0.0050
0.0040
0.0030
30
Effect of TR on TR
0.0020
0.0010
0.0000
-.0010
-.0020
Effect of CPI on FF
26
NOTE: Error bands are
30
Effect of FF on TR
0.0020]
0.0015 J.
0.0010 ;{)
0.0005 ItlyWV/r W_ _
0.0000 _flilJWi ft .j-- -.0005
-.0010
'
-.0015 i f
..noon
UvtU
8
30
. Effect of FF on FF
7
+ 2 sigma.
8
26
7
26
F IG U R E 7.
0.0125
Effect of IP on IP
Effect of CPI on IP
f\
7\ *
0.0075 f ' \ \
-------0.0050
0.0025
\ \ ^
n
<YWYV
w.UUw
\
~
-.0025
-.0050
-.0075
8
30
0.0100
Effect of FF on IP
0.0025
0.0000
-.0025
-.0050
-.0075
-.0100
0.0032
8
Effect of FF on CPI
0.0016
0.0020
-.0032
7
28
Effect of TR on CPI
0.0080
0.0040
-.0016
30
30
0.0060
0.0000
8
. _ Effect of NBRXon IP
-.0048
0.0000
-.0064
-.0020
7
26
Effect of IP on TR
8
30
Effect of IP on NBRX
30
. Effect of FF on NBRX
8
NOTE: Error bands are + 2 sigma.
8
30
8
30
Effect of TR on NBRX
8
30
Effect of NBRX on NBR>
F IG U R E S.
Effect of CPI on IP
5
22
Effect of NBRXon IP
39
Effect of FF on CPI
Effect of IP on CPI
5
Effect of IP on NBRX
21
37
5
22
39
Effect of FF on NBRX
Effect of CPI on NBRX
5
Effect of IP on FF
Effect of FF on IP
Effect of CPI on FF
22
Effect of NBRX on FF
39
Effect of FF on FF
F IG U R E 9.
Effect of M1 on IP
Effect of IP on
0.0120
0.0100
Effect of NBRXon IP
Effect of TRon IP
0.0120
0.0100
0.0080
0.0060
' 0.0040
0.0080
0.0060
0.0040
0.0020
0.0000
. -.0020
-.0040*
0.0020
0.0000
-.0020
12
Effept of IP on
0.0080
0.0070
0.0060
0.0050
0.0040
0.0030
Effect of TR on CPI
Effect of NBRX on CPI
0.0020
0.0010
0.0000
-.0010
-.0020
0.0020
0.0010
0.0000
-.0010
12
Effect of CPI on LM1
Effect of IP on LM1
Effect of M1 on LM1
0.0035
0.0060
0.0025
0.0040
✓
0.0030
0.0030
✓
0.0048
/
✓
s
0.0025 1
0.0020
0.0036 v V \
r
0
.0020
0.0000 }>*/
0.0010
0.0024
0.0015
0.0000 \ f y
0.0012
-.0010
-.0020 V ' A -
-.0030
-.0040
-.0050
Effect of FF on IP
0.0140
\ /
12
Effect of IP on TR
0.0000
-.0012
-.0025 \
'-S. x '
nrrcn
-.uuou
12
Effect of CPI on TR
0.0015 *
0.0010
0.0005
0.0000
-.0024
_nnqe
-.V/U
OO
0.0010
t ' ' -------
0.0005
0.0000 i
-.0005
. nnin
*.U
U1u
12
Effect of M1 on
-.0010
n
-.0015 1
12
12
Effect of CPI on NBRX. Effect of M1onNBRX
0.0032r
0.0024
0.0016
0I0008
0.0070
0.0060
0.0050
0.0040
0.0030
0.0000-.0008
-.0016
-.0024
-.0032^
9
33
33
-.0030
-.0040 s r
-.0050 V
-.0060^
/
/
12
Effect of FF on TR
0.0015
0.00080
0.00040
0.00000
0.0000
-.00040
-.00080
-.0010
-.00160
-.0020
12
Effect of TRonNBRX
0.0020
0.0010
0.0000
-.0010
-.0020J
12
-.0020
0.00200
0.00160
0.00120
0.0020
0.0010
0.0005
-.0005
-.0015
Effect of NBRXon NBRX Effect of FFonNBRX
12
Effect of NBRX on FF
NOTE: Error bends are + 2 sigma.
0.0020
/ •
0.0010
r
\
*
0.0000 \ y
. nnm
-.W
1u
12
-.00120
-.0020
Effect of IPonNBRX
Effect of NBRX on LM1 Effect of FF on LM1
0.0060
0.0020
s'
✓
s
0.0010
0.0050
/
/
/
0.0000
/
0.0040
S'
^
^
V
-.0010 1 / /
0.0030
Effect of NBRX on TR
•i
]/AvV-^
-.0005
12
-.0030
-.0040
-.0050
-.0060
-.0070
@FedFRASER