The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
M A Y / J U N E
1 9 9 5
•
V O LU M E
7 7 ,
N UM BER
3
Channels off M o n e ta ry Policy
P roceedings o f the N ineteenth A nnual Econom ic Policy C onference
o f the F ed era l R eserve Bank o f St. Louis
Theoretical Issues of
Liquidity Effects
Resolving the
Liquidity Effect
Is There a "C re d it
C h an nel" for
M o n e ta ry Policy?
Distinguishing
Theories of the M o n e ta ry
Transmission Mechanism
Inform ation, Sticky Prices
and Macroeconomic
Foundations
A C on ference Panel D iscussion:
W ha t Do We Know About
H ow M o n e ta ry Policy
Affects the Economy?
P resident
Th om a s C. M e lz e r
D irecto r o f R esearch
W illia m G . D e w a ld
A s s o c ia te D irecto r o f R esearch
Cletus C. C o u g h lin
R esea rch C o o rd in a to r a n d
Review E d ito r
W illia m T. G a v in
B a n k in g
R. A lto n G ilb e rt
D a v id C. W h e e lo c k
In tern ation al
C h risto p h er J . N e e ly
M ic h a e l R. P a k k o
P atricia S. P o lla rd
M a cro econ om ics
D o n a ld S. A lle n
Richard G . A n de rso n
Ja m e s B. B u lla rd
M ich a e l J . D u e k e r
Joseph A . R itter
D a n ie l L. T h o rn to n
P eter Yoo
R eg ion al
M ich e lle A . C lark
K e v in L. K liesen
A d a m M . Z a re ts k y
D irecto r o f E d ito ria l S erv ices
D a n ie l P. B re n nan
M an agin g E d itor
Charles B . H enderson
G r a p h ic D esigners
B ria n D . E bert
Inocencio P. Boc
R eview is published six tim es p er year b y the Research
D ep artm en t o f th e F e d e ral R eserve B an k o f S t. L ou is.
S in g le -c o p y s u b sc rip tio n s are availab le free o f ch arge.
S en d re q u e sts to: F e d e ra l R e serv e B a n k o f S t. L o u is,
P u b lic A ffa irs D e p a rtm e n t, R O . B o x 4 4 2 , S t. L o u is,
M issou ri 6 3 1 6 6 - 0 4 4 2 .
T h e view s expressed are those o f the individual auth ors
and do n o t n e cessa rily re fle c t o ffic ia l p o sitio n s o f the
Fed eral Reserve B an k o f St. Lou is, the Fed eral Reserve
S y ste m , o r th e Board o f G o v e rn o rs.
A rtic le s m ay be
reprinted or reproduced if the sou rce is credited. Please
provide the Pu b lic Affairs D epartm ent w ith a copy o f the
reprinted m aterials.
83 Distinguishing Theories of
the M o n e ta ry Transmission
Mechanism
Stephen G. Cecchetti
V o lu m e 7 7 , N u m b e r 3
98
C o m m e n ta ry
Mark Gertler
Hi
C o n trib u tin g A u th o rs
v
P re sid e nt's M essage
101 Inform ation, Sticky Prices
and Macroeconomic
Foundations
Thom as C. M elzer
v ii
E d ito rs ' Introdu ction
Daniel L. Thornton
David C. W heelock
Allan H. Meltzer
119
C o m m e n ta ry
Randall Wright
125
R e p ly to W rig h t's C o m m e n ta ry
Allan H. Meltzer
3 Theoretical Issues of
Liquidity Effects
A Conference Panel
Lee E. Ohanian and
Alan C. Stockman
26
Discussion: W h a t Do We
K now A bout H ow M o n e ta ry
C o m m e n ta ry
Policy Affects the Economy?
Kevin D. Hoover
33 Resolving the
Liquidity Effect
Adrian R. Pagan and
Joh n C. Robertson
55
C o m m e n ta ry
Lawrence J. Christiano
63 Is There a "C re d it C hannel"
fo r M o n e ta ry Policy?
R. Glenn Hubbard
78
C o m m e n ta ry
Bruce D. Smith
127
Ben S. Bernanke
131
Thomas E Cooley
138
Manfred J.M . Neumann
HfVlfM
MAY/JUNE
1995
Contributing Authors
Ben S. Bernanke
Princeton University
Woodrow W ilson School of Public
and International Affairs
Princeton, NJ 08544-1013
M anfred J . M . Neum ann
Universitat Bonn
Institut fur Internationale
W irtschaftspolitik
Lennestrasse 37
D 53113 Bonn Germany
Stephen G . Cecchetti
Boston College
Department of Economics
Chestnut Hill, MA 02167
Lee E. O hanian
University of Pennsylvania
Department of Economics
Philadelphia, PA 19104
Lawrence J . Christiano
Northwestern University
Department of Economics
2003 Sheridan Road
Evanston, IL 60208
Thomas F. Cooley
University of Rochester
Graduate School of Management
Rochester, NY 14627
M a rk G e rtle r
Department of Economics
New York University
269 Mercer Street, 7th Floor
New York, NY 10003
K evin D. Hoover
University of California-Davis
Department of Economics
Davis, CA 95616
R. G lenn Hubbard
Columbia University
Graduate School of Business
Uris Hall
New York, NY 10027
A lla n H. M eltzer
Carnegie-Mellon University
GSIA
Pittsburgh, PA 15213
A d ria n R. Pagan
The Australian National University
Econom ics Program
RSSS
Canberra Act 0200
Australia
John C. Robertson
The Australian National University
Econom ics Program
RSSS
Canberra Act 0200
Australia
Bruce D . Smith
Cornell University
Department of Econom ics
Ithaca, NY 14853-7601
A lan C. Stockman
University o f Rochester
Department of Economics
Rochester, NY 1 4 627-0156
Randall W rig h t
University o f Pennsylvania
Department of Economics
3 7 1 8 Locust Walk
Philadelphia, PA 19104-6297
President's Message
The view that, in the long run, monetary
policy can affect only nominal variables is
now widely held, both within the Federal
Reserve and among economists generally.
Moreover, there is a growing consensus that
to achieve maximum achievable sustainable
economic growth, the main objective of mon
etary policy should be price stability. On the
other hand, there has never been agreement
about how monetary policy actions are trans
mitted through the economy to prices.
Certainly, if we could reach a consensus about
this process, we might better agree on how
best to achieve our long-run goal.
Many believe that monetary policy
works through interest rates. By influencing
short-term real interest rates, they contend,
policy actions have an effect on long-term
real rates and, in turn, spending, output and,
eventually, prices. The extent to which interest
rates respond to policy actions, however, is
unclear, as is the degree to which interest rate
changes affect spending. 1 confess that I am
skeptical about our ability to consistently and
predictably influence longer-term real interest
rates. For this and other reasons, I am uneasy
about trying to achieve price stability by
adjusting nominal short-term rates.
Some researchers have argued that mone
tary policy actions affect the real economy by
influencing the availability of credit. Indeed,
some contend that the sharp decline in bank
lending at the beginning of this decade— the
so-called credit crunch— was evidence that
monetary policy was too restrictive. All firms
do not have equal access to credit markets.
Thus, monetary policy actions may affect
some more than others. W hat this tells us
about the formulation and implementation
o f monetary policy is unclear to me, however.
Nevertheless, I believe readers of this issue of
Review will be intrigued about the discussion
on the credit channel and its implications for
the conduct of monetary policy.
O f course, the money channel is not
without its problems. In the past, setting
monetary target ranges and reducing them
over time “operationalized” the Fed’s resolve
to contain inflationary pressures and move
toward price stability. The apparent break
down in the short-run relationship between
nominal spending and the monetary aggre
gates— first M l and, more recently, M2—
has made the aggregates less useful in guiding
policy, at least for now. What’s more, a reduced
emphasis on the aggregates may undermine
the credibility of the Federal Open Market
Committee’s commitment to price stability.
Perhaps it is time to follow the lead of
other central banks and quantify our long
term objective through an explicit inflation
target. That, however, is a topic for another
time, except that in debating channels of
monetary policy, we cannot lose sight of what
a central bank ultimately influences: prices.
Nor can we lose sight of what a central bank
does, which is influence reserve availability.
W hat we need to learn more about is how
these actions are transmitted through the
economy to prices. A clearer understanding
of this transmission mechanism can enhance
policy credibility as well.
Thomas C. Melzer
President and Chief Executive Officer
Federal Reserve Bank of St. Louis
review
MAY/JUNK
199
Editors' Introduction
The conventional wisdom once held
that money doesn’t matter. Now there is wide
agreement that monetary policy can signifi
cantly affect real economic activity in the short
run, though only the price level in the long
run. Considerable debate remains, however,
about how monetary policy affects the real
economy and prices. The Nineteenth Annual
Econom ic Policy Conference was devoted to
assessing and deepening our understanding
of how monetary policy works. This volume
of the Review contains the proceedings of
that conference. We believe that the careful
and wide-ranging analyses presented here
will focus the debate and perhaps further the
profession’s understanding of the monetary
transmission mechanism.
The conference focused on what many
consider to be the two principal channels of
monetary policy: 1) its effect on interest rates;
and 2) its effect on the availability o f credit.
These channels are sometimes referred to as
the money (or interest rate) channel and the
credit channel, although the conference made
clear the difficulties associated with these
labels. For example, some participants noted
that monetary policy could affect the real
economy even if it had little or no effect on
real interest rates. Consequently, they objected
to equating the money channel with the
interest rate channel. Furthermore, one form
of what is generically referred to as the credit
channel requires that monetary policy actions
significantly affect the real interest rate. This
form of the credit channel differs from the
interest rate channel in the way that changes
in the real interest rate affect output. Conse
quently, calling one the interest rate channel
and the other the credit channel is potentially
misleading.
expansion of bank reserves or a monetary
aggregate and short-run reductions in short
term interest rates.” Ohanian and Stockman
then explore the liquidity effect in generalequilibrium, representative-agent models.
Ohanian and Stockman’s analysis shows
that in a general-equilibrium environment,
exogenous changes in money can, in principle,
affect real output, prices or the interest rate.
If money is neutral and prices adjust instan
taneously, monetary policy changes the price
level, but not output or the real interest rate.
If prices do not adjust instantaneously, a liq
uidity effect occurs— the real interest rate
declines in response to a monetary expansion.
The failure of the price level to adjust imme
diately to its new long-run equilibrium, how
ever, also produces expectations of inflation.
From the Fisher relation, the nominal interest
rate may either rise or fall, depending on the
relative size of the liquidity and price expec
tations effects. If the liquidity effect is domi
nant, both the nominal and real rates fall. If
the price expectations effect is dominant, the
nominal rate rises.
Ohanian and Stockman show that within
this class of models, variants differ in both
the mechanisms that produce sluggish price
adjustments and in how monetary policy
actions influence the real economy. Their
review considers a wide variety of equilibrium
models: one- and two-sector sticky-price
models, two-country models, limited-participation models (with and without representa
tive agents), and even models where the only
role for money is to reduce intermediation
costs. In the last case, the liquidity effect is
perverse: An increase in the money supply
causes the real interest rate to rise because a
monetary innovation represents a technolog
ical change. Many of these models include a
cash-in-advance constraint and all require a
mechanism that causes prices to adjust slowly
to their equilibrium level.
In his discussion of Ohanian and
Stockman’s paper, Kevin Hoover re-interprets
TH E L IQ U ID IT Y EFFECT
The first article, by Lee Ohanian and
Alan Stockman, defines the liquidity effect
as “the purported statistical relation between
vii
REVIEW
MAY/JUNK
parameters in VARs, that is, that the covariance
matrix is diagonal and that the structure is
recursive are, a priori, no more or less credible
than the identifying restrictions adopted by
the Cowles Commission approach. Since the
Cowles Commission approach frequently
results in over-identified systems, and the
Wold casual ordering exactly identifies the
structural parameters of the system, Pagan
and Robertson state that “One m igh t.. .cate
gorize the difference as simply amounting to
whether one wants to work with an exactly
identified system or not.”
Pagan and Robertson also point out
the similarity in the approaches to selecting
among what are essentially observationally
equivalent structures. VAR practitioners
frequently select from alternative Wold
causal orderings by choosing the one whose
impulse response functions are most consis
tent with their prior beliefs. Researchers in
the Cowles Commission tradition generally
perform dynamic simulations of alternative
models, choosing the one whose dynamic
responses most closely correspond to their
prior beliefs.
Pagan and Robertson go on to confirm
what previous empirical work suggests,
namely, that finding a statistically significant
liquidity effect depends critically on the defi
nition of money used. A statistically signifi
cant liquidity effect is generally found only
with nonborrowed reserves or the ratio of
nonborrowed to total reserves. No statistically
significant liquidity effect is found using total
reserves, the monetary base or M l.
The authors investigate the robustness
of the estimated liquidity effect to alternative
specifications of the system by including first
commodity prices, and then exchange rates
and foreign interest rates. W hile these vari
ables affect the magnitude and persistence
of the liquidity effect, the overall conclusion
remains; that is, a statistically significant
liquidity effect is obtained only when non
borrowed reserves is used.
Pagan and Robertson also find that the
magnitude of the estimated liquidity effect
depends on the sample period. Specifically,
the liquidity effect essentially vanishes when
the VAR is estimated over the period 1982:12
to 1993:12. Christiano investigates whether
many of these models within a standard IS-LM
aggregate-supply/aggregate-demand frame
work. Hoover questions whether the research
agenda on which Ohanian and Stockman
report, namely the modeling of monetary
policy within a representative agent, cash-inadvance constraint framework, is useful for
understanding the liquidity effect. Specifically,
he questions whether the liquidity effect in
such models truly reflects what most of the
profession thinks of as the liquidity effect.
First, he notes that in general-equilibrium
models all of the endogenous variables are
determined simultaneously and, hence, the
interest rate cannot be “causally efficacious,”
as in most discussions of the liquidity effect.
Second, Hoover argues that the liquidity
effect is a feature of financial markets and
that the financial sectors of these models are
simply not sufficiently rich to capture the
liquidity effect adequately. In models of the
type presented by Ohanian and Stockman,
the interest rate is solely determined by the
shadow prices associated with consumption,
leisure and saving choices. In the end,
Hoover concludes “that we still are a long
way from understanding the liquidity effect.”
TH E EM PIR ICAL EVIDENCE
W hile the theoretical foundations for
the liquidity effect remain controversial, the
article by Adrian Pagan and Joh n Robertson,
and commentary by Lawrence Christiano,
narrow the disagreement about the empirical
relevance of the liquidity effect. Pagan and
Robertson thoroughly review the empirical
literature on the liquidity effect, differentiating
between single-equation and systems-modeling
approaches. Arguing for a systems approach,
they focus their attention on the estimation
of vector autoregressions (VARs), which is
the most promising tool for identifying a sta
tistically significant and empirically relevant
liquidity effect.
Pagan and Robertson start with an inter
esting discussion of some basic differences
between Sim’s VAR approach and the older
Cowles Commission methodology, which
involves estimation of a simultaneous-equation
structural model. The authors point out that
the assumptions used to identify the structural
199
viii
REVIEW
MAY/JUNE
this marked change is due to the small sample
size or to a fundamental change in the variance-covariance matrix and concludes: “The
primary reason for the shift in the impulse
response function appears to lie in a shift in
the variance-covariance matrix of the VAR
disturbances.” He notes, however, that the
presence of autoregressive conditional heteroskedasticity (ARCH) in the covariance
structure during this period means that this
result could be a statistical artifact, rather than
a true shift in the structure. Nevertheless, noting
that the Fed’s operating procedure changed in
late 1982, Christiano speculates whether the
observed change in the liquidity effect might
be due to a change in policy regime.
Even ignoring the apparent disappear
ance of the liquidity effect recently, Pagan
and Robertson find the liquidity effect to
have been a relatively unimportant determi
nant of the behavior of the federal funds rate
in the past. They find that a 1 percentage point
increase in the level of nonborrowed reserves
reduces the funds rate by about 13 basis points.
Since the average absolute monthly change in
nonborrowed reserves is about 0.9 percentage
point, the immediate effect of policy actions
on the funds rate seems modest. Moreover,
even when they allow the effect to accumulate
until the impulse response function turns
positive, they find it was rare for the sum to
be smaller than -60 basis points. Hence, they
conclude that “most of the factors historically
driving the federal funds rate do not seem to
be due to the Fed... .”
TH E CREDIT C H A N N EL FOR
M O N E T A R Y P O LIC Y
Articles by Stephen Cecchetti and
R. Glenn Hubbard survey the credit channel
for monetary policy. Although these papers
discuss much of the same literature and evi
dence, the confluence of their approaches
provides a richer understanding of the issues
than either article alone. Both make clear
that there are two possible credit channels
for monetary policy, and that both require
asymmetry in the access of “small” and “large”
firms to credit. The bank credit channel
operates directly on the ability of depository
institutions to make loans through the effect
199
of monetary policy actions (open market
operations) on bank reserves. For example,
restrictive monetary policy actions reduce
reserves and, thereby, loans. Unable to obtain
bank or other external finance, bank depen
dent firms curtail planned spending.
The second credit channel, which goes
by various names (such as the financial accel
erator, excess sensitivity or the broad lending
view), works through the effect of a policyinduced change in interest rates on the balance
sheets of borrowers. For example, by reducing
their real net worth, a policy-induced increase
in the real interest rate makes it difficult for
some, typically smaller, firms to attract capital.
Unable to attract funds, these firms curtail
planned spending. This view differs from the
traditional analysis, whereby a policy-induced
increase in interest rates makes marginal
investment opportunities unprofitable.
Both Cecchetti and Hubbard evaluate
the state of the macroeconomic and crosssectional evidence on the credit channel.
Cecchetti focuses on the aggregate evidence
and frames his analysis in an interesting
discussion of the difficulties associated with
identifying changes in monetary policy, an
analysis of several commonly used indicators
of policy, and a discussion of how to differen
tiate alternative views using both aggregate
time-series and cross-sectional data.
Hubbard, on the other hand, focuses
on cross-sectional evidence. He concludes
that there is considerable evidence that “the
spending decisions of a significant group of
borrowers are influenced by their balance
sheet condition in the ways described by
financial accelerator models.”
The discussions of these papers by Mark
Gertler and Bruce Smith follow very different
lines. Agreeing with essentially all of what
Cecchetti said, Gertler emphasizes the com
plementarity between the credit and tradi
tional views of the effect of policy-induced
changes in interest rates on spending. He
illustrates how what he calls the financial
propagation mechanism can magnify the
traditional effect of policy-induced changes
in interest rates. Gertler’s emphasis on the
propagation mechanism makes clear that this
effect comes into play whatever the source
of the impulse to interest rates. Hence, this
REVIEW
MAY/JUNE
propagation mechanism may have important
implications for output and interest rates
even if Pagan and Robertson are correct that
the influence of monetary policy on interest
rates is small.
Smith’s analysis, on the other hand,
undercuts the significance of the empirical
research that Hubbard finds most supportive
of the broad credit view. Smith shows that a
general-equilibrium, modified-neoclassical
growth model capturing several key features
o f models associated with the broad credit
view of monetary policy, has considerably
different implications. In contrast with most
models generating a credit channel, Smith’s
model has multiple equilibria, with both a
low capital stock, low-income equilibrium,
and a high capital stock, high-income equi
librium. In the latter, a low marginal effi
ciency of capital and high income provide
firms with significant amounts of internal
finance. In this equilibrium, an expansionary
monetary policy reduces output, the capital
stock and credit. Hence, in one equilibrium,
credit market imperfections magnify the
effect of monetary policy in a way consistent
with the broad credit view. In the other, the
outcome is inconsistent with the credit view.
Moreover, even when monetary policy pro
duces results consistent with the credit view,
it is not for the reason given by proponents
of the credit view.
rather than real prices; and 3) that many
prices change slowly over time. W hile not
rejecting a role for menu prices, imperfect
competition, relative and absolute price
confusion and aggregation in explaining
sticky-price behavior, Meltzer argues that
these alternative explanations are not consis
tent with one or more features o f price data.
Instead, he argues that the cost of acquiring
information and the inability o f individuals
to fully distinguish permanent from transitory
shocks provide better micro-foundations for
the sluggish adjustment of nominal prices
observed in the data.
Meltzer argues that the now widely
adopted approach to providing m icro-foun
dation to macroeconomics, which features
representative agents and complete ArrowDebreu markets, have not, and will not, prove
useful. He contends that this framework
provides no role for monetary disturbances.
Hence, he concludes that “it is not the appro
priate micro-foundation for macroeconomics.
No amount o f squeezing, cutting and pasting
will make it so.”
In his discussion, Randall Wright
focuses on Meltzer’s remarks about the state
of macroeconomics. W right defends the use
of general-equilibrium modeling in macro
econom ics, arguing that this methodology
has produced great strides in the professions’
understanding of business cycles, labor
markets and econom ic growth. Moreover,
he argues that the use of overlapping-generations models has produced significant contri
butions to our understanding of the properties
of monetary economies and the monetary
policy debate, as well as about economics
generally. W hile conceding that overlapping-generations models have not captured
the medium of exchange function of money,
he points out that there are other generalequilibrium models that explicitly capture
this function of money and the private infor
mation problem. Finally, W right concedes
that these models “sometimes take the pricing
aspect of the Arrow-Debreu paradigm too
seriously” He notes, however, that the effect
of sticky prices has been addressed by such
models and states that Meltzer’s article has
not convinced him of the value of explaining
endogenously sticky prices.
M IC R O -F O U N D A T IO N S ,
E N D O G E N O U S PRICE
STICK IN ESS A N D
M O N E T A R Y P O LIC Y
The last paper of the conference, by
Allan Meltzer, revisits a theme of the first,
namely, that sluggish price adjustment is
necessary for monetary policy to have real
effects. Meltzer’s purpose is to provide
microeconomic foundations for price setting
and the gradual adjustment of prices to new
information.
Meltzer argues that differences in infor
mation and costs associated with acquiring
information explain three facts about price
setting behavior: 1) that many prices are set;
2) that price setters choose to set nominal
1995
X
REVIEW
MAY/JUNE
PANEL D IS C U SS IO N
A panel discussion provided a capstone to
the conference. The panelists, Ben Bemanke,
Thomas Cooley and Manfred Neumann, each
took a different approach to summarizing the
profession’s understanding o f the effects of
monetary policy Bemanke argues that the
semi-structural VAR approach is a fruitful
method for investigating how monetary policy
actions are transmitted through the economy
He also finds limited-participation models to
be a realistic approach but, sounding a theme
reminiscent of Hoover’s comments, argues that
the cash-in-advance constraint is implausible.
He suggests that a more promising avenue
would be to combine the limited-participation
and sticky-price assumptions.
Bernanke acknowledges recent criticisms
of the bank lending channel of monetary policy,
and points to the need to differentiate between
the bank lending channel and the balance
sheet channel. He argues that continued work
on the credit channel is desirable because of
quantitative problems with the other leading
models o f the transmission mechanism, and
because ongoing institutional changes will
likely affect both the potency of policy and the
interpretation of monetary policy indicators.
Cooley argues that the papers presented
at the conference seem to take as given that
monetary policy can affect the real economy
at cyclical frequencies. He argues that the
theoretical evidence that the Fed can moderate
cyclical fluctuations in econom ic activity is
weak and that the empirical evidence for this
proposition is “extremely fragile.” He argues
that the evidence based on VARs or structur
al VARs is sensitive to the set of conditioning
variables, the sample period and the identifying
restrictions. Moreover, he asserts that models
that treat money as exogenous are simply
“meaningless.”
Cooley argues that a more interesting and
fruitful approach is to investigate the growth
and welfare consequences of monetary policy
shifts by modeling artificial economies and
examining them using calibration methods.
He argues that this approach permits explicit
modeling of essential features of the hypothe
sized transmission mechanism and broadens
the scope of inquiry from output effects at
1995
business cycle frequencies to growth and
welfare.
Neumann examines the structure of
what he terms the new money-credit view of
monetary policy. Comparing this view with
the monetarist view o f the transmission
mechanism of monetary policy, he concludes
that the monetarist approach is the “more
comprehensive.” He points out that the
monetarist approach assumes that all assets,
financial and real, are imperfect substitutes.
A change in base money sets in motion a
broad process of portfolio substitution over
a full array of real and financial assets, and
over a broad array of financial institutions
and firms. W ith this as background,
Neumann points out how the monetarist
approach o f Brunner and Meltzer encom
passes the traditional IS/LM analysis and the
new money-credit view.
Neumann points out that the timing of
the effect of monetary policy actions on bank
holdings of government securities and loans,
documented recently by Bemanke and Blinder
(1992), is a direct implication of the monetarist
theory of relative prices that Brunner (1 970)
had pointed out some time earlier. He also
questions the need to find evidence in support
of the broad credit channel because differences
in financial assets and the existence o f infor
mation cost are unassailable.
Finally, Neumann points out that the
fact that monetary policy has distributional
effects, impacting more on smaller, financially
weaker firms, is not surprising. This has
no implications for the conduct of monetary
policy, he argues, except to reinforce the
monetarist advice to avoid large swings in the
creation of reserves or the monetary base.
The conference opened new areas of dis
cussion and revisited others. It appears that
the profession is progressing slowly toward a
consensus view o f the monetary transmis
sion mechanism.
The theoretical foundation for the com
monly accepted liquidity effect is disputed.
Moreover, the empirical evidence indicates
that the liquidity effect is relatively weak and
short-lived. If monetary policy exerts rela
tively modest influence over the federal funds
rate and if this influence has weakened sig
nificantly recently, for whatever reason, the
effect of monetary policy through the interest
rate channel or broad credit channel is dubi
ous. The support for the narrowly focused
bank lending channel is also weak, being
open to considerable criticism and lacking
empirical support. Although the profession
has yet to agree on how monetary policy
impulses are ultimately transmitted to the
price level, we hope that the empirical and
theoretical scrutiny of alternative monetary
transmission mechanisms reported in these
proceedings will stimulate new research
into the channels of monetary policy.
Finally, a nod to the analysts in the
Research Department of the Federal Reserve
Bank of St. Louis who helped to review the
text and data for each of the articles: Jerram
Betts, Kelly Morris, Tom Pollmann, Steve
Stohs, Rich Taylor and Chris Williams.
Daniel L. Thornton and David C. W heelock
St. Louis, Missouri
May 15, 1995
REVIEW
MAY/JUNI
1995
Lee E. Ohanian is an assistant professor at the University of Pennsylvania. Alan C. Stockman is chairman of the Department of Economics at
the University of Rochester, and serves as a consultant to the Federal Reserve Bank of Richmond and the Federal Reserve Bank of Cleveland.
Theoretical
Issues of
Liquidity Effects
the nominal interest rate. It is difficult to
interpret these claims without a coherent
model of nominal liquidity effects.
The monetary policies that the Federal
Reserve claim s that it follows require the
existence of liquidity effects. Many central
bank operating procedures that involve use
of the federal funds rate (or any other interest
rate) as a target, instrument, or operating
variable of monetary policy require a liquidity
effect. The current operating procedure of
the Federal Reserve is predicated on the exis
tence of a liquidity effect in the sense that
the Fed uses the federal funds rate as its
proximate instrument of policy and contracts
quantities of reserves and monetary aggregates
by raising the funds rate (and vice versa).
W hen the Fed raises the federal funds rate,
it reduces reserves by the amount sufficient
to achieve the desired increase. The smaller
the required reduction in reserves, the larger
the implied nominal liquidity effect. Of
course, a central bank operating procedure
that attempts only to tie down the nominal
interest rate (which ties down the real inter
est rate plus the expected rate of change of
the price level) may lack a nominal anchor
to tie down the level of prices. If, however,
the operating procedure also includes a pro
vision to revert to control over the level of
monetary aggregates if inflation exceeds some
critical level, then the price level may be
anchored at least within a certain range.
Attempts to isolate liquidity effects
empirically are often subject to a unique
problem: If the central bank operating proce
dure involves direct targeting of a short-term
interest rate, statistical work and econom ic
models that treat a monetary aggregate as
exogenous and the nominal interest rate as
endogenous may be misleading. This has
led many economists to question the exis
tence o f liquidity effects. Although we do
not attempt to resolve that issue in this article,
we note that other kinds of evidence (that do
not involve regressions o f interest rates on
allegedly exogenous monetary aggregates)
suggest important liquidity effects in the
Lee E. O h a n ia n and
A la n C. Stockm an
ne o f the most pervasive real effects
long-claimed for monetary policy is its
ability to affect interest rates in the short
run through channels other than the standardexpected inflation effect. The alleged short
term inverse relationship between interest
rates and monetary policy is often called the
“liquidity effect” of monetary policy. We use
the term liquidity effect to refer to the pur
ported statistical relation between expansion
of bank reserves or monetary aggregates (or
perhaps only surprise expansions of these
aggregates) and short-run reductions in
short-term interest rates. The liquidity effect
can also refer to the common interpretation
of this purported statistical relation: that the
same central bank action that changes bank
reserves or monetary aggregates also changes
short-term interest rates. This definition
corresponds to early use of the term, for
example, by Friedman in 1968.1
We distinguish between a nominal
liquidity effect (the aforementioned relation
with a nominal interest rate) and a real
liquidity effect (the aforementioned relation
with a real interest rate). Either may occur
without the other. For many purposes, real
liquidity effects are more interesting because
they indicate real effects o f monetary policy.
On the other hand, central banks around the
world claim that their operating procedures
directly target or control nominal interest
rates— that they reduce reserves of the banking
system (perhaps through open market sales)
to raise the nominal interest rate or raise
reserves of the banking system (perhaps
through open market purchases) to reduce
O
3
1 Some economists use the term dif
ferently, viz. to refer to a particular
class of theoretical models attempting
to explain the purported relation.
RIVKN
MAY/JUNE
curve), which is consistent with empirical
estimates of income elasticities, and assume
that real income rises by less than 10 percent in
response to the exogenous 10 percent increase
in nominal money. In this case, the nominal
interest rate would fall to equilibrate money
supply and money demand, thus generating
the liquidity effect. In m ost neoclassical
flexible-price models, however, the price level
would rise sufficiently in response to a per
manent increase in the money supply so that
the real money supply, real income and nom
inal interest rates would be unchanged.2
data. Cook and Hahn (1 9 8 9 ), for example,
interpret their results as showing that changes
in Fed targets for the federal funds rate have
large, immediate effects on three-month, sixmonth and 12-month Treasury bill rates,
w ithout any apparent reverse effects of
Treasury bill rates on the funds rate.
This article assumes the existence of real
and nominal liquidity effects in the data and
discusses the main explanations for liquidity
effects that have been advanced in the litera
ture. The theoretical issues associated with
liquidity effects are important because different
models imply different welfare effects of
monetary policies and different effects on
interest rates and other variables. Also, dif
ferent models of liquidity effects have different
implications for optimal monetary policies.
They also provide different interpretations
of the data. Finally, differing implications
of various models suggest potential tests of
those models.
A One-Sector, Sticky-Price RationalExpectations Model
Though it is not difficult to generate a
liquidity effect in an IS-LM model with sticky
prices, further research has shown that this
is not a generic feature of sticky-price models.
To see why, we consider a simple neoclassical-growth model with money and exogenous
price stickiness, as in recent work by Cho
and Cooley (1 9 9 0 ) and King (1 9 9 1 ). A rep
resentative household maximizes discounted
expected utility, with preferences defined
over consumption of a single physical good,
c(, and leisure, lt:
L IQ U ID IT Y EFFECTS IN
STICK Y-PR ICE M O D ELS
Traditional Sticky-Price Models
The liquidity effect is a characteristic of
traditional sticky-price (Keynesian) models.
Consider a model with a conventional
money-demand function, and a price level
that is perfectly sticky,
f
(1)
log
m
1995
(2)
M a x E 0^ j8 'u ( c „ l t).
1=0
The household faces a period budget
constraint:
\
= a 0+ a 1logCy)
(3)
- a 2log(l + i) + £m,
w,nt + (rt + ( l - £ ) ) k , + -^£- i ^ L
V
V,
> c + fe , + ^l±L
1 ,Vt+ l n
P,
—
2 Of course, there can be non
superneutralities of money in flexh
ble-price equilibrium models, such
as in cash-in-advance models, in
which the inflation tax reduces
inputs of labor or capital.
where md is nominal money demand, p is the
pre-determined price level, y is real income, i
is the nominal interest rate, and £ is a meanzero disturbance to money demand so that
E(em) = 0. Given the double-log specification,
the parameters a 1 and a 2 are income and
interest elasticities, respectively.
Suppose that the money supply, m,
increases permanently by 10 percent. Because
the price level is perfectly sticky, and money
demand equals money supply in equilibrium,
the real money supply also rises by 10 percent.
Assume that 0 <
^ 1 (a relatively flat LM
The household’s wealth (measured in
units of the consumption good) consists of
wage income, wt n(, capital incom e and
undepreciated capital stock, (rt + (1 —S ))kt,
and the real value of money, including lump
sum monetary transfers from the govern
ment, (m( + r()/p(. (In this economy, the
price level is simply the dollar price of the
single good). The household uses its wealth
to purchase consumption, and acquire new
capital and new money. We assume that
consumption purchases are subject to a cashNK or ST. LOUIS
4
MAT/JUNE
in-advance constraint; consumption can only
be acquired with existing cash:
(4)
(1 3 ) InM , = lnM ,^
+/i(ln(M t_1) - ln(M,_2)) + £mt.
mt+ T t > p tcr
Now, consider the one-period interest
rate on a (nominally) risk-free bond between
today and tomorrow. Although this asset will
not be traded in this representative-agent
economy, it is straightforward to compute
the equilibrium asset price. The equilibrium
interest rate implies that the representative
household has, at the margin, no incentive to
trade this security. The interest rate on this
one-period bond is given by the relation:
A competitive firm produces the single
good using a stochastic constant-returns-toscale production technology, zt f (K(, N(), that
takes labor (N) and capital (K) as inputs. The
term z is an exogenous productivity distur
bance with the following autoregressive law
of motion:
(5)
1995
Z, = (l-</>) + 0Z(_1 + £a .
The firm maximizes profits, treating factor
prices parametrically:
<1 4 >
+
= Pt
ft
P t+1
( 6)
M ax z J ( K t,N t) - w tN t - r lK l.
Equilibrium with
Pre-Determined Prices
Because the technology is constant returns to
scale, maximum profits are zero. Profit maxi
mizing input choices by the firms yield the
following functions for factor prices (where
subscripts indicate partial derivatives):
(7)
wt = z J N(K t,N t)
(8)
rt = Ztf K(K t,N l).
Suppose prices are set one period in
advance at the expected market-clearing price.
(The commodity for which the pre-determined
price is an equilibrium is expected consump
tion conditional on information at date t—1.)
Given that the price is pre-determined, it is
necessary to specify a rule for allocations:
We first assume that output in this economy
is purely demand-determined. That is, the
representative firm sells as much output to
households as demanded at the pre-determined
price. This assumption is consistent with
recent sticky-price literature, as in Blanchard
and Kiyotaki (1987), in which monopolistically
competitive firms willingly supply extra
demand, as long as price exceeds marginal cost.
Unlike the IS-LM type model discussed
at the beginning of this section, it is ambigu
ous whether the nominal interest rate falls in
response to an unexpected increase in the
money stock in this sticky-price economy.
Assuming that the cash-in-advance con
straint binds, consumption is relatively high
today, which implies that the marginal rate of
substitution between consumption today and
tomorrow (the expected real interest rate) is
low, which tends to reduce the nominal
interest rate. If money growth is positively
serially correlated, then expected inflation is
high, which tends to increase the nominal
interest rate. It is easy to see this result if we
The resource constraint in this economy is:
(9)
z tf { K t,N t) + a - S ) K t > C t+ K t+l.
Remaining equilibrium conditions are
given by household first-order conditions
and market-clearing conditions. Efficient
household choices for consumption, labor
input, capital accumulation and money, with
subscripts indicating partial derivatives, and
A, denoting the date-t marginal value of
wealth, are:
( 10)
A, = /?Et[At+1(rt+1 + (1 - 5 ))]
( 11)
uly = Atw,
( 12)
~ = PE,
P,
ct+l
\Pm
We assume that money growth is exogenous,
and is given by the autoregressive process:
5
REVIEW
MAY/JUNI
assume that households have perfect foresight,
and that momentary utility is additively
separable:
conducted a simulation of the demanddetermined version of this model. Because
the model does not possess a closed form, we
computed an approximate equilibrium using
a version of Marcet’s (1 9 9 0 ) procedure. We
choose functional forms and parameters that
have been commonly used in the business
cycle literature. We assume that the momen
tary utility function is isoelastic, which is
consistent with steady-state growth:
i-p
(15)
u = - ---- + t>(l).
1 -p
w
Taking natural logs of the asset-pricing rela
tion under perfect foresight, we obtain the
Fisher decomposition of the nominal interest
rate into a real component and a nominal
component reflecting future inflation:
(16)
1995
*1+1 ~ - ln(/^)+ p (ln (ci+i) - ln (ci))
+ ln(p(+i ) - ln (p,)Production possibilities are assumed to
be Cobb-Douglas:
Thus, the nominal interest rate falls only
if the utility curvature parameter, p, is sufficiendy large that the decline in the real interest
rate reflecting negative-consumption growth
more than offsets the increase in inflation. This
typically implies that the curvature parameter
p must exceed 1. (That is, risk aversion of
the representative household exceeds that of
log utility).
Note that the effect of an unexpected
increase in the money stock on the nominal
interest rate in this cash-in-advance economy
depends on the allocation (rationing) rule.
Suppose instead that output is determined by
the minimum o f quantity demanded and quantity
supplied, as in Barro and Grossman (1 9 7 1 ),
rather than being determined by the quantitydemanded allocation rule. That is, house
holds will be rationed in response to a positive
money shock, and firms will be rationed in
response to a negative money shock. In this
“short-side” case, the cash-in-advance con
straint no longer binds if there is a positive
money shock, and the nominal interest rate
must fall to zero. (This extreme response of
the nominal interest rate is an artifact of the
cash-in-advance framework of this model. It
would likely disappear in a similar model in
which the interest elasticity of the demand for
money were non-zero.)
(18)
The discount factor, /3, equals 0.99, which
implies a steady-state real interest rate of about
4 percent. The preference parameter, y/, deter
mines the share of discretionary time spent
in producing market goods. We set I//= 0.37,
which implies that households work about
one-third of their discretionary time. The
curvature parameter p is set to 2. The pro
duction parameter, 6, is equal to capital’s
share of income and has averaged about 0.36
in the United States. The depreciation rate,
5, is set to 0.0 2 5 , which implies an annual
depreciation rate of 10 percent. The persis
tence parameter for the technology shock,
4>, is 0 .9 5 , which is comparable to numbers
used by Hansen (1 985) and Prescott (1986).
The innovation variance is set to 0.007, which
is the estimate used by Prescott (1 986) and
others. The serial correlation parameter for
money growth, p., is 0.5, and the innovation
variance is set to 0.009.
The experiment consists of holding the
technology shock fixed at the unconditional
mean, and letting the money supply increase
by 1 percent at date t. The increase in the
money stock is completely unanticipated.
Unexpectedly high money growth raises real
output in this model. Assuming that the
cash-in-advance constraint binds, the per
centage increase in consumption equals the
percentage increase in the money stock.
Figures 1-3 present the impulse response
Solving and Simulating the
One-Sector Sticky-Price Model
To gain some insight into this issue with
a more general form of preferences, we have
z f(K ,N ) = zK eN 1~e .
6
REVIEW
MAY/JUNI
functions of capital, consumption and labor
input to a 1 percent, unanticipated, permanent
increase in the money stock. The capital stock
increases only slightly; its increase is not
sufficient to generate persistent changes in
consumption or labor input. The response
of the nominal interest rate to the money
shock appears in Figure 4. The immediate
effect of the money shock is to increase the
nominal interest rate slightly: A Fisher decom
position shows that the real interest rate
declines, but that the increase in expected
inflation more than offsets this fall.
W hile we do not pursue a comprehen
sive analysis of this one-sector model, this
example indicates that it is not necessarily
easy to generate nominal liquidity effects
in sticky-price models with explicit intertem
poral optimization. Robert King reaches the
same conclusion in a related monetary model
which does not have unitary income elasticity
of money demand, and which includes
multi-period price setting.
199
F ig u re 1
Impulse Response of Capital Stock to
M on e y Shock
Percent deviations from steady state
F ig u re 2
Impulse Response of Consumption to
M on e y Shock
Percent deviations from steady state
Liquidity Effects in Models with
Some Sticky Prices
The model discussed in the preceding
section had the property that the price level
was sticky in response to a monetary shock.
This section analyzes the liquidity effect in
an economy in which some, but not all, prices
are sticky. The analysis in this section is
drawn from Ohanian and Stockman (1994).
The motivation behind this model is that
while there is considerable evidence suggesting
that some nominal prices change infrequently
(see Carlton, 1989), there is also abundant
evidence that many goods have prices that
change frequently, such as food, automo
biles, computers and gasoline. We consider
a model with two physical consumption
goods, X and Y, money introduced through
a cash-in-advance constraint, and complete
asset markets, with the exception of the
friction induced by the cash-in-advance
constraint. We first analyze a very simple
economy without capital. The equilibrium
in this simple economy can be calculated
very quickly, and as a result it is possible to
evaluate the properties of this economy for
a wide variety of parameter values.
F ig u re 3
Impulse Response of Labor Input to
M on ey Shock
Percent deviations from steady state
7
REVIEW
MAY/JUNE
from the government, Px and PYare nominal
prices, M is the nominal money the household
chooses as it leaves period-t asset markets and
enters period-t product markets, Vt is a vector
of other assets the household owns at the
beginning of period t, with dividend vector d
and ex-dividend price-vector q. We assume
that households have constant elasticity of
substitution preferences across the two phys
ical goods, where cr is the elasticity of substi
tution between x and y and p is a measure of
overall curvature of the utility function. And
v is a leisure-preference parameter. (1/p is
the elasticity of intertemporal substitution.)
Assume that the cash-in-advance con
straint (equation 21) binds as an equality
every period and that t = 0. It is easy to see
that the flexible-price perfect foresight equi
librium for this simple production economy
satisfies
F igu re 4
Impulse Response of Interest Rate to
M o n e y Shock
Percent deviations from steady state
A representative household maximizes
discounted expected utility:
(19)
m a x £ 0^
'
1
M
(
ax ,
199
( 22 )
'
(23)
Mst = rPxtLxt
I6 T
+ rPYt^Yn
Le
PXtA,
ax
l-pa /<7-1
fzLv
(24)
subject to the sequence of constraints
(20)
Py,A( = ^ caj
'
+ (i-a )y ;
J
-0/(7
( l - a ) L -Y
yt »
n,_j + Tt + Pxt_lL Xt_l + Pyt_iLyi(_[
- M , + v t(qt + d t) - v t+lq = 0
(2 5 )
v = P -P Xl d L V A1+1,
(2 6 )
V = p P Y t0 L e- i At+l,
and
(21)
M ,- P X(X t - P yiYt > 0 ,
where M[ is the (exogenous and constant,
because r = 0) money supply at the end of
period-t asset markets and A is the currentvalue Lagrange multiplier on the constraint
of equation 20. (Note that A = y, where y is
the current-value multiplier on the cash-inadvance constraint, because of the first-order
condition for the choice o f M,.) Moreover,
we can solve for the nominal interest rate on
a one-period nominal asset using the pricing
condition:
where equation 20 is a budget constraint for
period-t asset markets and inequality (equa
tion 21) is the cash-in-advance constraint
which applies to period-t product markets
(which immediately follow period-t asset
markets, as in Lucas, 1982). The terms x
and y refer to consumption of goods X and
Y, Lx and L y refer to the labor hours to pro
duce goods X and Y, 0 ^ 6 < 1 is a parameter
o f the production function, nl_l refers to the
household’s money holdings at the end of
period (t—1) product markets, r refers to a
lump-sum transfer of money to the household
(27)
8
1 + i:
A,
REVIEW
MAY/JUNE
Equilibrium When Some
Prices Are Sticky
Ta b le 1
Benchmark Case
We now suppose that nominal prices
in the X industry are pre-determined: Sellers
choose the nominal price Px at the end of
period t—1. The nominal price of Y, on the
other hand, adjusts to clear markets each
period. We can vary the amount of price
stickiness in the economy by varying the rel
ative sizes of the X and Y industries. The
nominal price of X is set to equate expected
quantities supplied and demanded. As in the
case of the one-sector model, we assume that
output in the X industry is determined by the
quantity demanded. An interesting feature
of this setup is that it encompasses the stan
dard Keynesian model and the flexible-price
neoclassical model as special cases.
We begin with the economy in a
nonstochastic, steady-state equilibrium with
a constant money supply, and consider an
unanticipated, permanent change in the
nominal money supply at the beginning of
period t. Real variables dated at t + 1 and later
are unaffected by this change in the money
supply, but real variables at date t change
because PXt is pre-determined. Suppose the
money supply falls permanently by 1 percent
at date £, with PXt fixed for one period. Because
the quantity of X produced is determined
by the quantity demanded, equation 25
(describing the supply of X ) does not hold
in the short run. Instead, we have equations
2 2 -24 and 26 in the four variables Lx (, Ly (,
PYt and At, (with At+1 taking its new steadystate value). Because a change in the money
supply has no steady-state effect on x, y or Lx,
equation 23 implies that it has no steady-state
effect on PXjl+1A|+1. Therefore, since the fall
in the money supply lowers PXjt+1 by 1 percent,
it necessarily raises A1+1 by 1 percent.
Tables 1-4 present the quantitative effects
of a permanent 1 percent rise in the money
supply (from 10.0 to 10.1) when a = 0.5,
6 = 0.64, v = l , f3 = 0.99 and p = 2. We choose
the elasticity of substitution between x and y
( a = 0.5) to be less than the Cobb-Douglas
case (a - 1) since it seems reasonable to assume
that the short-run substitutability between the
two categories of goods may be relatively low.
The value chosen for the production parameter,
1995
/
px
py
labor in X
labor in /
Output of X
Output of Y
GNP
Total labor
Old SS
SR
4.167
7.925
7.925
0.4869
0.4869
0.6309
0.6309
1.262
0.9738
3.701
7.925
8.012
0.4924
0.4882
0.6355
0.6320
1.267
0.9807
N ew SS
Ratio
4.167
8.004
8.004
0.4869
0.4869
0.6309
0.6309
1.262
0.9738
-0 .4 7
-0 .9 9
0.10
0.13
0.27
0.72
0.17
0.45
0.70
Ta b le 2
V e ry Small Sticky-Price Sector
i
px
py
labor in X
Labor in Y
Output of X
Output of Y
GNP
Total labor
Old SS
SR
N ew SS
Ratio
4.167
5.346
10.74
0.1044
0.7248
0.2355
0.8138
0.9939
0.8291
4.057
5.346
10.85
0.1053
0.7252
0.2368
0.8141
0.9943
0.8305
4.167
5.4
10.85
0.1044
0.7248
0.2355
0.8138
0.9939
0.8291
-0 .1 1
-0 .9 9
0.02
0.86
0.06
0.55
0.04
0.04
0.16
(0 = 0 .6 4 ), is often used in the equilibrium
business cycle literature, and is identical to the
value used in the one-sector model. We select
overall curvature of the utility function (p = 2)
that is consistent with empirical estimates, and
is also identical to the value used in the onesector model. Also, V is a leisure preference
parameter and does not play an important
role for the experiments we conduct.3
The first column shows the variables of
interest: the nominal one-period interest rate
(in percent); the nominal prices of X and Y;
labor input in each industry; output in each
industry; real GNP evaluated at the equilibrium
prices and production shares; and the total
labor input. The second column displays the
old steady-state (Old SS) levels of these vari
ables, before the change in money. The SR
9
3 We have analyzed specifications
with different preferences over
leisure, and the results ore qualita
tively similar to those reported
below. For example, if preferences
are logarithmic in leisure, the effect
of a money shock on interest rates
is about 70 percent as large as in
the case of linear preferences over
leisure.
REVIEW
MA Y/JUNK
But in the short run, with px pre-determined,
real GNP rises about 0.45 percent. There are
significant differences across sectors: Output
in the sticky-price sector rises 0 .7 2 percent,
while output in the flexible-price sector rises
0 .17 percent. The rise in money raises the
nominal price of Y, which reduces the rela
tive price of X. This raises the quantity of X
demanded and creates excess demand in the
X industry. Since output of X is determined
by the quantity demanded, output of X rises.
Output of Y rises less because consumers
substitute into purchases of X. (If the elasticity
of substitution were greater than 1, this sub
stitution would be larger and the output of Y
would fall.) Notice that the nominal price of
Y rises by about as much as if the price of X
were flexible. It overshoots its long-run
equilibrium by one-tenth of 1 percent (it would
undershoot if the elasticity of substitution
between X and Y were greater than 1).
The rise in the money supply has a
short-run “liquidity effect” on the nominal
interest rate. In Table 1, the nominal interest
rate falls 4 7 basis points, from 4 .1 7 percent
to 3.70 percent, in the short run. Because
expected inflation is positive (the CPI is
expected to rise another 0 .50 percent), this
represents a fall in the real interest rate (mea
sured in terms of the output bundle) of about
1 percentage point. Notice that the liquidity
effect occurs despite the introduction of
money through a cash-in-advance constraint,
which (when binding, as in these examples)
builds in a zero interest-elasticity of the
demand for money.
Table 2 shows that a change in the
money supply can have a substantial liquidity
effect on nominal and real interest rates in
the short run even if only a small fraction of
the economy has sluggish prices. This table
presents results with the same parameter
values as in Table 1, but with a = 0 .0 4 , so
that the sticky-price sector accounts for only
about 12 percent of employment. A perma
nent 1 percent rise in money reduces the
nominal interest rate by about 12 basis points.
Table 3 shows that with p = 2, the liquidity
effect is somewhat smaller if one-third of
labor is employed in the sticky-price sector.
If, however, the elasticity of intertemporal
substitution is one-third (p = 3) rather than
Ta b le 3
Interm ediate Case
/
px
py
Labor in X
Labor in Y
Output of X
Output of Y
GNP
Total labor
Old SS
SR
4.167
7.032
8.95
0.3191
0.6235
0.4814
0.7391
1.205
0.9426
3.861
7.032
9.045
0.3223
0.6246
0.4845
0.7399
1.208
0.9469
N ew SS
Ratio
4.167
7.103
9.04
0.3191
0.6235
0.4814
0.7391
1.205
0.9426
-0 .3 1
-0 .9 9
0.06
1.01
0.18
0.65
0.11
0.24
0.46
Ta b le 4
Sm aller Elasticity of
Intertem poral Substitution
/
px
PY
Labor in X
Labor in Y
Output of X
Output of Y
GNP
Total labor
Old SS
SR
4.167
6.137
8.103
0.3546
0.7672
0.5151
0.844
1.337
1.122
3.6
6.137
8.139
0.3582
0.7685
0.6183
0.8449
1.339
1.127
N ew SS
Ratio
4.167
6.199
8.184
0.3546
0.7672
0.5151
0.844
1.337
1.122
-0 .5 7
-1 .0 0
0.06
1.00
0.17
0.64
0.11
0.21
0.43
column shows the short-run effects of the
1 percent rise in money (while the nominal
price of X is fixed at its previous level). The
fourth column (New SS) shows the new steady
state, and the column labeled ratio displays
the percentage by which a variable falls short
of, or exceeds, its new steady-state level. For
the interest rate, this column shows the differ
ence between the short-run and steady-state
interest rates.
Table 1 shows the results when a = 0.5,
so that the sticky-price sector represents half of
the economy’s output and half o f all labor is
employed in the sticky-price sector. A perma
nent 1 percent rise in the money supply is
neutral in the long run, with a 1 percent rise in
nominal prices and no effects on real variables.
199
10
REVIEW
MAY/JUNK
one-half, the interest rate falls 57 basis points
when one-third of labor is employed in the
sticky-price sector (see Table 4). W ith p = 3,
the nominal interest rate falls 27 basis points
even if the sticky-price sector accounts for
only 15 percent of employment, and raising
p from 3 to 4 further doubles the size of the
interest rate response, holding fixed the share
of the economy with sluggish prices.
These examples demonstrate that a sig
nificant liquidity effect is consistent with a
relatively small sticky-price sector. A further
analysis of the relationship between the size
of the sticky-price sector and the response of
interest rates to a money shock is presented
in Figure 5. This plot displays the (p, a )
combinations that generate the midpoint
estimate of the liquidity effect reported by
Christiano and Eichenbaum (1992b ), and
shows that reasonable values of p combined
with a small sticky-price sector are consistent
with their estimates for U.S. data. W hile the
size of the liquidity effect depends on the
parameter p, in all other respects the responses
of the economy to an increase in the money
supply are virtually unaffected by changes in
the elasticity of intertemporal substitution.4
F ig u re 5
R ho/Alpha Pairs Y ie ld in g Estimated
(M id p o in t) Liqu idity Effects
Curvature of utility function (rho)
Relative importance of stickey price goods (alpha)
(28)
1 V CT-1)/CT
E0£ / 2 ‘
(i -
A Two-Country Model
p
)I
'
subject to the sequence of budget constraints
In previous work, such as Stockman and
Ohanian (1 9 9 4 ), we examined the effects of
money supply changes in a two-country world
in which some sectors of the economy have
nominal prices that are sticky in the short run
and other sectors have flexible prices. We
showed that money supply changes have liq
uidity effects (a fall in the money supply raises
the real and nominal interest rate) both within
and across countries, and creates a cross
country, real interest rate differential.
We discussed a two-country model in
which each country produces and consumes
two internationally tradable goods, X and Y,
using only labor as an input. There are two
monies introduced through cash-in-advance
constraints with the usual convention in which
sellers’ currencies are the medium of exchange
for all transactions. Because the two countries
are identical ex ante, we describe only the
domestic country. A representative household
in the home country maximizes
199
fv2.Q^
y
Px,i-i h x,t-i x,t-i -4"r P
+ v, (q, + d,) +
h
- v , q , - M t - e tN t = 0
and sequences of the two cash-in-advance
constraints,
(30)
M t —m in jx ,,* ,}? * ,
— m i n { y , , y , | P Yt = 0 ,
and
(31)
N t - m a x jx , —3c,,oJpX(
—m ax {y t —y ,,0 } p ; = 0 ,
where equation 29 is a budget constraint for
period-£ asset markets and equations 30 and
31 are the cash-in-advance constraints that
apply to period-t product markets. The
terms xt and y t refer to total home consump-
11
* In future work, we plan to analyze
liquidity effects in a version of this
hybrid model with capital accumula
tion and o mixture of monetary and
technology shocks.
REVIEW
MAY/JUNE
tions of excess supply, buyers purchase from
sellers in the country with the lowest price
(adjusted for the exchange rate). W hen prices
are equal in both countries, buyers purchase
first from sellers in their own country. We
assume that in situations o f excess demand
for a good in some country, buyers residing
in that country are first in line and buyers
from the other country are last in line to
buy that good.3
Necessary conditions for home-currency
and foreign-currency bonds yield expressions
for one-period nominal interest rates like 2.10
that, along with the law of one price for good
Y, Py = e P * , and interest parity imply:
Ta b le 5
Home M oney Falls from 10 to 9 .8
Old SS
SR
New SS
Percentage
Excess of SR
to N ew SS
i
1.01
1.537
1.01
0.52
if
1.01
1.48
1.01
0.46
e
1.00
0.9794
0.98
-0 .0 5
PY
7.333
7.183
7.187
-0 .0 5
p yf
7.333
7.334
7.334
Ix
0.6534
0.6427
0.6534
-1 .6 4
Ixf
0.6534
0.6532
0.6534
-0 .0 3
ly
0.6534
0.6499
0.6534
-0 .5 4
lyf
0.6534
0.6535
0.6534
5 Our previous paper discusses the
various cases involving alternative
corner solutions to the rationing
problems that can arise in this
model.
0.002
0.017
(32)
where X* is the multiplier on the foreign
representative household’s current-period
budget constraint. Equation 15 follows
directly from the usual expression of interest
parity (e'/e = ( l + i ) / ( l + i * ) ) and the standard
asset-pricing equations for riskless nominal
one-period bonds in each currency. In addi
tion, we need the separate budget constraints
for home and foreign households. The home
household can buy (or sell) one-period nominal
bonds B at the price 1/(1 + i).
Table 5 shows the effects of a permanent,
unexpected, 2 percent fall in the home country’s
money supply (from 10.0 to 9.8), starting from
a steady-state equilibrium with a constant
money supply and price level. We hold fixed
the foreign money supply in this initial exercise
and assume a = 0.5, cr = 0.5, 8 = 0.9, V = 1,
/3 = 0.99 and p = 2. This implies that half of
GDP in each country consists of output of good
Y, the relative price of Y in terms of X is initially
unity, the exchange rate is initially 1, and the
real (and nominal) interest rate is 1//3 — 1.
Since < j < 1, the two goods are relatively poor
substitutes. We also assume there is no initial
international indebtedness, so initially the
countries are identical and there is no interna
tional trade. (After a change in the money
supply in one country— or in both— B can
become non-zero and can remain non-zero
in the new steady state.)
The first column of Table 5 shows the
endogenous variables: the nominal price of Y
tion of goods X and Y, regardless of where
the goods were purchased, x t and y refer to
home production o f the two goods, M( is the
home household’s stock of home money at
the beginning of the product market, Nt is its
stock of foreign money, used for purchasing
imports ( if imports are positive), and e is the
exchange rate (in units of home money per
unit of foreign money).
We assume that assets cannot be traded
conditional on monetary transfers or taxes
(positive or negative r), so any decrease in
the home money supply is financed by lump
sum taxes (negative r) on households in the
home country only, and any decrease in the
foreign money supply directly affects only
foreign households. Assuming r = r* = 0,
where r* is the transfer or tax in the foreign
country, and kxt = kx, t = k YI = k YU = 1 for all t,
we showed that one flexible-price equilibrium
is the same as in a closed economy, with no
international trade or foreign money holding.
We assume that Px and P * (the foreigncurrency nominal price of X produced and
sold in that country) are pre-determined,
chosen one period in advance. The nominal
prices Py and P *, on the other hand, are flex
ible. Assuming flexible exchange rates and
holding constant the foreign money supply
Ns, we consider a small, unanticipated, per
manent fall in Mts (the home money supply)
starting from a nonstochastic steady-state
equilibrium with constant money. In situa-
1995
12
RfVUW
MAY/JUNE
in the home and foreign countries (py and pyj);
the nominal interest rate (in percent) in the
home and foreign countries (i and if); the
exchange rate (e); labor inputs in the x industry
in the home and foreign countries (be and be/);
and labor inputs in the y industries in the home
and foreign countries (ly and lyf). The second
column, Old SS, shows the old steady-state
levels of the variables (before the change in
money) from which the analysis begins. The SR
column shows the short-run effects of the fall
in money (while the nominal price of X is fixed
at its previous level for one period). The New SS
column shows the new steady state, and the
column labeled percent shows the percentage
by which a variable falls short of or exceeds its
new steady-state level. For the interest rate,
this column presents the difference between
interest rates in the short run and in the new
steady state.
The 2 percent fall in money leads, in the
long run (New SS colum n), to a 2 percent
fall in the nominal prices of goods X and Y,
from 7.333 to 7.187. (The new steady-state
relative price of Y in terms of X is 1, so the
new price of X is also 7.187.) The interest
rate is unaffected in the long run by the one
time change in the level of money, and the
exchange rate falls 2 percent, from 1.00 to
0.98, in the long run. Long-run levels of
employment in each industry in the home
country (be and ly) are unaffected, as are for
eign employment levels in each industry (be/
and ly f) and long-run levels of output in
each industry and in each country.
W hile the unexpected change in money
is almost neutral in the long run ( “almost”
because it redistributes wealth and so has
permanent effects), it is not neutral in the short
run. The impact effect of the unexpected fall
in home money is to raise the home-country
nominal interest rate by 53 basis points. If
one interprets this as a quarterly model (since
the discount parameter is 0.99 per period),
with one-quarter nominal price stickiness in
the X industry, then the steady-state interest
rate is 1.01 percent per quarter, or 4.04 percent
per year. Then the 2 percent fall in home
money raises the annualized home nominal
interest rate by 211 basis points, to 6.15 per
cent per year. The foreign nominal interest
rate also rises, by 4 7 basis points, on a per-
1995
Ta b le 6
Home and Foreign M oney Fall
from 10 to 9 .8
N ew SS
Percentage Excess
of SR to N ew SS
/
2.012
if
2.012
e
1.
py
Ix
7.183
- 0 .0 5
0.6426
- 1 .6 6
Ixf
0.6426
- 1 .6 6
ly
0.65
- 0 .5 2
lyf
0.65
- 0 .5 2
0
period basis, which is 188 basis points on an
annualized basis with this interpretation. The
home nominal interest rate is then 20 basis
points above the foreign rate on an annualized
basis. This is reflected also in a slight over
shooting of the exchange rate in the short run
(it falls 0.05 percent below 0.98) followed by
a small, expected (and actual) appreciation
of home currency. Employment in the home
country falls in both industries, particularly
in the X industry with sticky prices. Overall
output is unchanged in the foreign country,
though there is a small sectoral reallocation
of production from the X industry to the Y
industry.
The short-run appreciation of home cur
rency, combined with the stickiness of both
the home-money price of X sold at home and
the foreign-money price of X sold abroad,
implies that X is cheaper in the foreign country
than in the home country, creating excess
demand for X in the foreign country and excess
supply in the home country. Foreigners are
unconstrained in buying good X in their own
country and home residents, who are last in
line there, import X and buy the rest from
sellers in their own country.
Table 6 shows the case in which both
countries reduce their money supplies by the
same percentage. The result is the same in
each country as in a closed economy, and
there is no international trade in either the
short run or in the new steady state. The table
shows the effects of an unexpected, permanent,
2 percent fall in money in both countries.
This has identical effects in the two countries,
FED ER A L RESERVE B A N K OF ST. L O U I S
13
REVIEW
MAY/JUNE
199
Ta b le 7
Ta b le 8
Foreign M oney Falls to 9 .7 ,
then to 9 .8 0 0 9 8 ; Home
M oney Falls to 9 .8
Foreign M on ey Falls to 9 .8 5 ,
then to 9 .7 9 9 5 2 ; Home
M on ey Falls to 9 .8
N ew SS
if
e
3.041
3.041
1.
Ixf
7.18
0.6463
0.6317
ly
lyf
0.6466
0.6466
PY
Ix
New SS
1.503
if
0
-0 .1 0
-1 .0 8
e
1.503
1.
py
Ix
7.185
0.6407
Ixf
0.648
0.6518
-1 .0 4
ly
lyf
-1 .0 3
so we can discuss only the home country.
The fall in money reduces aggregate nominal
spending, which reduces the nominal price
o f good Y. Because Px is fixed in the short
run, this increases the relative price of X, so
consumers substitute good Y for good X, which
further reduces output of X and works against
the fall in spending on Y. If the elasticity of
substitution in consumption, cr, were 1, out
put in the Y sector would remain unchanged
and the nominal price of Y would fall by 2
percent. With cr < 1, output of Y falls along
with output of X and PY overshoots its longrun fall. (If cr > 1, output of Y rises and its
nominal price undershoots its long-run fall.)
One way for the foreign country to peg
its exchange rate (in the absence of any other
shocks) is to change its money supply in
proportion to the change in the home money
supply; in this model, there are other paths of
monetary policy that also result in a pegged
exchange rate. But these policies have vastly
different effects on real and nominal interest
rates.
Suppose the home country’s money supply
falls by 2 percent as before, and suppose the
foreign country pegs its exchange rate at unity.
Suppose also that the foreign government can
credibly commit to a future path for the money
supply. Because nominal prices are set one
period in advance, for only one period, antic
ipated future changes in money can be fully
incorporated into price-setting behavior.
Table 7 shows the results of a foreign mone
/
1
GO
CO
i
Percentage Excess
of SR to New SS
0.6517
Percentage Excess
of SR to N ew SS
0
-0 .0 3
-1 .9 4
-0 .8 3
-0 .2 6
-0 .2 6
tary policy that reduces the foreign money
supply by 3 percent from Ms*= 10 to Ms*= 9.7
in the short run (while the home money sup
ply falls from 10 to 9 .8 ), and then changes
Ms* to 9.80098 in the long run, assuming that
a = 0.5, u - 0.5, 8 - 0.9, V= 1, j8 = 0.99 and
p = 2, as in Table 5. The exchange rate remains
at exactly 1, but the rise in world interest
rates of 203 basis points exceeds the 100 basis
point rise that occurs along the baseline path.
Table 8 shows the results when the for
eign money supply falls less than the baseline
case: It falls from 10 to 9.8 5 for one period
and then permanently goes to 9 .7 9 9 5 2 (while
home money falls to 9.8 ). We continue to
assume a = 0.5, cr= 0.5, 8 = 0.9, v = 1, /3 = 0.99
and p = 2. If the fall to 9.85 were permanent,
foreign currency would depreciate and X
would be cheaper in the foreign country. This
would add to excess supply for X in the home
country and reduce excess supply of X in the
foreign country. This occurs up to the point
at which the relative price is unity, that is, at
an unchanged exchange rate. In this case,
the rise in world interest rates is smaller (49
basis points) than in the baseline case, and
similar to the rise in Table 5, even though
the size of the change in the money supply is
different. Finally, Table 9 shows the results
when the foreign money supply falls even
less in the short run— from 10 to 9.9 before
permanently going to 9 .79904. In this case,
there is no nominal liquidity effect (though the
real interest rate falls). If the home money
14
MAY/JUNE
that the utility function for the representative
household is linear in leisure.
Final goods, Y, are produced from an
isoelastic technology using intermediate inputs,
Ta b le 9
Foreign M oney Falls to 9 .9 0 ,
then to 9 .7 9 9 0 4 ; Home
M oney Falls to 9 .8
New SS
/
Percentage Excess
of SR to N ew SS
(34)
0.9979
0.9979
if
e
PY
lx
Ixf
ly
lyf
1.
7.186
0.6389
0.6534
-2 .2 3
-0 .0 0 2
0.6535
0.6534
0.006
-0 .0 0 4
1995
where m(i) is the amount of intermediate
input used in the production of final goods;
a) represents the measure of intermediate
goods-producing firms and is fixed exoge
nously. Intermediate-goods firms produce
output from an increasing returns-to-scale
technology that uses capital and labor,
0
0.0006
(35)
supply falls by only 0.5 percent in the short
run, but the exchange rate is pegged by a
commitment to future policy, then the nomi
nal interest rate actually fa lls in each country
(though, again, the real rate rises).
These tables illustrate that real and
nominal interest rates do not depend solely
on domestic monetary policy. Foreign policy
and expected future domestic and foreign
policies can create significant changes in the
responses of both real and nominal interest
rates. In particular, even the sign o f the interest
rate response to domestic monetary policy
depends on foreign monetary policy. In
addition, the response of nominal interest
rates to changes in the money supply is highly
nonlinear. That nonlinearity, illustrated by
these tables, suggests that linear statistical
analysis may miss key features of the rela
tions between money and interest rates.
D r.
(36)
It is assumed that r > 1, so that costs are
reduced by acquiring real balances. Banks
are owned by households and maximize the
present discounted value of cash flows,
A very different model with sticky prices
has been analyzed by Beaudry and Devereux
(forthcom ing). An infinitely lived represen
tative household maximizes discounted
expected utility:
(3 7 )
M a x ^ Q ,(s ,)C F (s ,)c JS ,
1=0
where Q is a state-contingent pricing func
tion and CF is the bank’s cash flow.
Final-goods producers are price takers,
but intermediate-goods producers are
monopolistic competitors. The increasing
returns parameter y plays a fundamental
M ax E 0 ^ /?' (ln (c,) - rjn, j.
t=0
Beaudry and Devereux make use of the
Rogerson-Hansen construct, which implies
n(i,t) = ZtF (K u,N 1()r,
where the degree of increasing returns is
indexed by y; z, is an exogenous technology
shock and the log of z is assumed to follow
a random walk. Money plays a very different
role in this economy relative to the other
models discussed in this article. Households
have no demand for money in this economy;
instead, cash is held by banks because it
reduces intermediation costs. Banks accept
deposits from households and lend to inter
mediate-goods producers, who must finance
capital inputs before selling their product to
final-goods producers. The representative
bank’s intermediation cost function is assumed
to be isoelastic in real balances and deposits:
Liquidity Effects, Increasing Returns
and Multiple Equilibria
(33)
Y=
IS
REVIEW
MAT/JUNE
productivity of the banking sector. Lower
intermediation costs, with strong increasing
returns, lead to the substantial increase in
output that occurs in this model.
The striking feature of this model is that
small monetary shocks lead to significant and
persistent liquidity effects, as well as large,
persistent increases in real quantities. Of
course, the very strong internal propagation
mechanisms in the model that make these
phenomena occur have not been established.
Large increasing returns are required in this
economy, which raises a number of questions.
If actual production technologies exhibit
economies of scale in this range, we would
expect to see greater temporal concentration
of production (periods of very high producdon,
followed by periods of no production). The
volatility of output, consumption, investment
and labor input in this increasing-returns
model is alm ost surely much greater than the
corresponding volatility in the data. In addi
tion, this model suggests large profits (or
high per-period fixed costs) for business
enterprises that are not obviously evident in
the data. Finally, with even a small interest
elasticity of money demand, the model would
imply a large effect on capital accumulation,
output and other variables of a change in
secular inflation, because a rise in inflation
would operate like a tax on financial inter
mediation (analogous to a negative monetary
shock in their model).
The nominal liquidity effect and inverse
real liquidity effect implied by this model
reflect the fact that an increase in the money
stock leads immediately to a sharp deflation.
O f course, the standard interpretation of liq
uidity effects is that monetary increases are
associated with lower nominal and real inter
est rates, and it is perhaps not surprising that
a model in which an increase in the nominal
money stock leads to a future increase in the
nominal value of money reduces nominal
interest rates. Though some vector autore
gressions suggest that nominal prices do not
immediately increase in response to a m one
tary shock, it is not yet an established empir
ical fact that higher money leads to a falling
price level over horizons corresponding to
business cycle frequencies, as is the case in
the Beaudry and Devereux model.
role. In models with substantial increasing
returns, there is a continuum of stationary
equilibria. This occurs because, with large
enough increasing returns, the eigenvalues
governing the policy functions are both out
side the unit circle and the model no longer
has the standard saddlepath property. Since
there are multiple equilibria, the authors
choose the equilibrium in which nominal
prices do not respond to current innovations
to money or technology. Sticky prices in this
economy have a much different implication
than in standard sticky-price setups. In con
ventional sticky-price models, nominal prices
would change in response to monetary or
technology innovations if that were possible.
In the Beaudry and Devereux economy, how
ever, there is no ex post regret in that no pro
ducer has an incentive to change his price
after shocks are revealed. Monetary and tech
nology shocks generate substantial changes
in economic activity in this model. In partic
ular, there are large and persistent increases
in output, consumption and investment in
response to either type of shock. Moreover,
an unanticipated increase in money leads to
a significant and prolonged reduction in the
nominal interest rate. W ith strong increasing
returns, this is one of the few models that
has internal propagation mechanisms capable
of generating persistent liquidity effects.
However, the model generates only a nominal
liquidity effect and not a real liquidity effect.
In fact, monetary expansion raises the real
interest rate in this model. (Because prefer
ences are separable over consumption and
leisure, marginal utility of consumption
depends only on date t consumption, so a
rising real rate is implied by the consumptionEuler equation.) So the fall in the nominal
rate is due entirely to substantial, persistent
deflation induced by the monetary shock.
The channel through which money
affects real quantities in this model differs
significantly from the monetary transmission
mechanisms in the other models discussed
in this article. In fact, an increase in the
money stock in this economy is isomorphic
to a favorable technology shock that affects
financial intermediaries. An increase in the
stock of money, combined with sticky prices,
results in higher real balances and raises the
1995
16
MAY/JUNE
IN TER EST RATES A S
PREDICTORS O F FUTURE
A C T IV IT Y
insure that the agent is willing to hold equity,
and there is no incentive for an agent to
issue debt.
For current purposes, we assume initially
that money grows deterministically:
It is reasonably well established that
short-term interest rates rise prior to reces
sions. These correlations have been inter
preted as important evidence for the exis
tence of significant liquidity effects, and for
monetary business cycle models. In this sec
tion, we consider a very simple equilibrium
model in which increases in nominal interest
rates precede econom ic downturns, but the
correlations between nominal interest rates
and future changes in output are due to an
exogenous shock. This model is used to
illustrate in a simple way that there are alter
native interpretations of these correlations
that are consistent with neoclassical economic
theory and observations.
This section is drawn from Cooley and
Ohanian (1990). Consider a representative
household with preferences given by:
(38)
(41)
\a
(42)
r
_1
M a x E T p 1- 1--------.
(l-c r f
(4 3 ) ln(R ,) = —ln(/?) —- var(A ln(c(+, ))
1- CX
+ ln(G,+1) + ( a - l)E [ A ln(ct+1)],
where Aln(cl+I) is defined as the growth rate
of consumption between today and tomorrow.
Suppose that the log o f the endowment
follows an integrated process:
mt + T, + Rtb, + z , ( q , + p,dt)
and the cash-in-advance (asset market) con
straint:
(44)
ln(c(+1) = a + ln(c,) + b(L)£t,
where b(L) is a polynomial in the lag operator
L and is square summable, and e t is an i.i.d.
random variable with E (e) = 0, E (e 2) = V2.
R,bt + m t + T, > p,ct + bt.
The budget constraint states that con
sumer wealth, which consists of nominal
money holdings (mt), a lump-sum monetary
transfer ( t() , interest and principal on oneperiod bonds (R br), and the value of equity,
Zt (q t + p d t) , must be sufficient to finance
consumption (c() new money (m,+,), new
one-period debt ( bt+I) and new equity (zt+I q,)The price level for the economy is given by pt.
The equilibrium for this model is straightfor
ward: Consumption of the representative
agent must be equal to the endowment (d),
the equilibrium prices of equity and bonds
1 = /3E,
For analytical convenience, we assume
that the endowment is generated by a lognormal distribution. This implies that the
one-period interest rate is given by:
> p tct + m t+1 + b t+1 + z (+i<Jt
(40)
m1+1 = G tMt, G, > 1.
The endowment process is stochastic,
and is the only source of uncertainty in this
model. One-period debt is specified as sure
nominal debt: One dollar today yields R dol
lars tomorrow, R S 1. Since this is a representative-agent economy, this security will be
in zero-net supply, but the asset can be priced
by using the household’s marginal condition:
Consumers maximize the expected pre
sent value o f utility subject to the budget
constraint:
(39)
1995
k
Defining dk = ^
it can be shown that
i= 0
the one-period nominal interest rate in this
economy is given by:
(4 5 )
ln^R, j = - ln ( / ? j- --------^ -u 2 + ( c T - l) a
K * - 1)
fe=0
-ln(G,+i).
17
Note that in this economy, the nominal
interest rate necessarily rises prior to a reces
sion if the risk-aversion coefficient (cr) is less
than 1. This is because if the economy is at a
cyclical peak, then the term
As long as the Fed did not (or was unable to)
completely smooth price-level fluctuations,
we would observe Fed tightening, higher
nominal interest rates, and future output
declines. The behavior of the money supply,
however, would be entirely endogenous.
This model illustrates how observations
that are often interpreted as results of the liq
uidity effect can have very different explana
tions consistent with neoclassical models.
W hile it is unlikely that real shocks account
entirely for the observed correlations between
nominal interest rates, the money stock and
output, given the strong countercyclical behav
ior of the price level, it is not at all unreason
able to expect that this mechanism is respon
sible for at least some of these associations.
<0.
6 Jovonovic (1982) presents o related
model.
Moreover, if one compares the spread between
short- and long-term debt, then it is also the
case that the yield curve necessarily inverts
prior to a recession, and this is a strong feature
of the data. The explanation for this is due
to expected inflation. If households antici
pate the endowment to fall next period, there
are two forces at work on the interest rate.
First, given the constant growth rate rule for
money, higher expected inflation tends to
push up the nominal rate. A falling endow
ment, however, implies that the real rate will
fall. If risk aversion is less than unity, then
inflation risk is more important than endow
ment risk, and the interest rate rises prior to
a downturn.
O f course, the price level in this model
is countercyclical; high price levels (and infla
tion rates) are associated with low endowment
states. But as Kydland and Prescott (1990)
and Cooley and Ohanian (1990) have pointed
out, the price level in the United States is
strongly countercyclical over the postwar
period. The predictions of this simple model
are also in line with observations reported by
Fama (1981) regarding a negative associa
tion between stock returns and inflation.
As an extension, this model could be
used to interpret an even richer set of corre
lations that has been reported (for example,
Christiano and Eichenbaum, 1992a) in which
open market sales (Federal Reserve tightening),
high interest rates and subsequent downturns
occur. This would simply require price level
smoothing on the part of the Fed. For exam
ple, suppose that individuals expect a fall in
the endowment, and a corresponding rise in
the inflation rate (in the absence of any change
in monetary policy). If the Fed is interested
in pursuing price level smoothing, then the
Fed would conduct open market sales of
securities to reduce the amount of cash in
the economy, and lower the future price level.
LIM ITE D -P A R TIC IP A TIO N
M O D ELS
Limited-participation (LP) models
refer to a class of models, originally proposed
independently by Rotemberg (1 984) and
Grossman and Weiss (1 9 8 3 ), and later devel
oped further by Lucas (1 9 9 0 ).6 These models
provide an alternative interpretation of liq
uidity effects. W hile the sticky-price models
discussed above all imply that assets can be
priced by using consumption-Euler equations,
so that the effects of a monetary disturbance
on the time path of consumption determines
whether there is a real liquidity effect (as
well as how large it is and how long it lasts),
limited-participation models provide a means
of breaking the link between consumptionEuler equations and real interest rates.
The basic econom ics of the limitedparticipation theory can be illustrated with
a modified version of the Grossman and Weiss
model with logarithmic utility. Households
are staggered in their visits to financial mar
kets. “Evens” visit financial markets in evennumbered periods and “Odds” visit in oddnumbered periods. It takes time for people
to exhaust their money balances, so most peo
ple do not participate in financial markets
continuously. At any point in time, some are
in financial markets and some are out of
financial markets. As in cash-in-advance
models, households must use cash to buy
goods, but in this model households spend
18
REVIEW
MAT/JUNE
their money over two periods rather than
one. Also as in cash-in-advance models, each
period consists of an asset market (AM) fol
lowed by a product market (PM).
A nonstochastic steady-state equilibrium
in the Grossman and Weiss model can be
described as follows. At period-one asset
markets, Odds obtain money for spending
during product markets in periods one and
two. Then, at period-one product markets,
Odds spend a fraction </>of their money on
goods, where <f) = 1/(1 + /3) and /3 is the dis
count rate, and save the remaining fraction
(1 —4>) of their money to spend during the
second period. At the same time, in periodone product markets, Evens spend all the
money they have left, which is a fraction <f>of
the money they had acquired last period (in
period-zero asset markets). This will be utility-maximizing behavior for households with
separable logarithmic utility and a constant
discount rate facing constant nominal prices
and opportunities to hold only money and
riskless one-period nominal bonds as assets.
Consider a steady state in which Odds
and Evens are equally wealthy and have the
same consumption profiles (except that they
are out of phase by a period). In a steady state
with a fixed nominal money supply, Ms, and
with constant endowments o f goods, y = 1,
equilibrium nominal prices are constant and
total nominal spending on goods each period
is (1/(2—</>))Ms, while ((1 —</>)/(2—</>))Ms
money is not spent (because it is carried over
to the next period by the households that will
not be in asset markets next period).
Starting from this steady state, an unan
ticipated open market purchase has real effects
in the short run: The increase in money must
initially be acquired by those households that
are in asset markets when it occurs. Suppose
the open market operation occurs in an odd
period, so odd households initially acquire it
all (by selling bonds for money). Because all
households spend cash slowly (over two
periods), not all the new money is spent at
first. The price level rises less-than-proportionally to the money supply. Because Even
households (who did not attend financial
markets this period) planned already to spend
all their money on goods, the increase in the
price level reduces their consumption. W ith
1995
constant endowments, equilibrium requires
that Odd households consume more this
period. However, this increase in consump
tion by Odd households is temporary, so the
anticipated growth rate of the Odd household’s
consumption falls. The consumption-Euler
equation for Odd households then implies
that the real interest rate over two periods
(from now until the Odd household again
enters assets markets) falls. Notice, however,
that the model breaks the link between real
interest rates and the consumption-Euler
equation of Even households, so it breaks
the link between real interest rates and the
path of aggregate consumption.
More precisely, there are equal numbers
of Odd and Even households. Odd households
choose consumptions, c(, and withdrawals of
money from financial markets (every other
period), M(, to maximize
(4 6 )
X r 'l n f c ? ) ,
t=i
subject to a sequence of constraints
(4 7 ) Ptc° + P(+1c”+1 = M°
for t odd
and
(48)
+ Pt_j = M" + . ^l+2 , + r, for t odd,
(1+ w )
and initial conditions on B0°, the initial level
of “bonds” held by the representative Odd
household, and P(), the period-zero price level;
r( is a lump-sum tax payment that the house
hold must pay (to balance the government
budget). Odd household own claims on the
endowment streams of firms: They are enti
tled to the dividends paid by the firm during
asset markets at odd-numbered periods (from
sales in the product market at the previous
even-numbered periods). Firms pay their
entire revenue as dividends. The term M,°
shows the money that the Odd household
acquires during asset markets at date 1 for
use in product markets at dates 1 and 2.
This money comes from dividends paid by
firms from their sales of goods at date-zero
product markets. Notice that utility maxi
mization implies that P( c ° =
and
19
REVIEW
MAY/JUNE
all its remaining money, ((1 —<£)/(2—<£))M.
Again, total nominal spending on goods is
( 1 / ( 2 Thi s sequence repeats in the
steady state. Because output is unity, the
steady-state nominal price level is (1/(2
Now consider a parametric change in
the money supply at date 1, starting from
this nonstochastic steady state. The govern
ment buys a one-period bond (from the Odd
household) with newly printed money. The
Odd household now has (1/(2—<£))M + AM
dollars and spends </>((l/(2—</>))M+AM) on
goods. The Even household still has
((1 —</>)/(2—<^>))M dollars to spend. Total
nominal spending and the price level are then
Pt+i c°t+1 = (1 —
Even households
solve an analogous maximization problem.
The government collects lump-sum taxes
and uses the proceeds as interest on its debt;
the representative household has a tax liability
equal to the present value of the total gov
ernment debt. The government may also
engage in open market operations. The
government’s budget constraint is
l + if
-+ T = B f
which says that the government finances it
debt obligations by printing money, or by
borrowing from or imposing lump-sum taxes
on households currently in asset markets. In
the steady state, this budget constraint becomes
simply r = (i/ (l+ i))B g. The initial level of
government bonds is given exogenously.
Equilibrium requires a sequence of
prices and interest rates so that households
maximize utility, and product and asset
markets clear:
(50)
(53)
(52) M“ = Pt_! + M f —Mf_,
for £ odd.
The latter conditions require that households in
financial markets acquire all the money paid by
firms as dividends that period plus any new
money printed by the government.
Consider the following sequence of events
in a steady-state equilibrium with fixed money
supply M. In each odd period £, the represen
tative Odd household acquires money
(1/(2—
at asset markets and then spends
(4>/(2—</>))M in product markets, saving
((1 —
cf)))M to spend next period. The
representative Even household spends all its
remaining money, ((1 —4 > )/ (2 —
on
goods. Total nominal spending on goods is
(1/(2—4>))M. In each even period, the repre
sentative Even household acquires money
(1/(2—<p))M at asset markets and then spends
(</>/(2—$ ))M in product markets, saving
((1 —<£)/(2—4>))M to spend next period.
The representative Odd household spends
1-0
2-0
-M + AM + -- - - - — M
M + 0AM.
W ith </>^ 1, the price level rises but
falls short of its new steady-state value of
(1/(2—<£))(M+ AM). This rise in the price
level reduces the real consumption of Even
households from 1 —<f>to ((1 —<f))/(2—<p))M/
((1/(2 —4>))M+ AM). Equilibrium real con
sumption of Odd households rises by the
amount that Even household consumption
falls.
The following period (t = 2), Even
households acquire all the money that was
spent at £ = 1 and spend a fraction <f>of it,
so they spend $ [(1/ (2—<£))M+ 4>AM],
Odd households spend their remaining
(1 —</>)(( 1/(2 —(pi) )M + AM) on goods, so
total nominal spending (and the price level)
is (1/(2-<f>))M+ (</>2+ l —</>)AM. With
V2 ^ r/>< 1, the price level rises at £ = 2 and
overshoots its new steady-state level. The
price level then falls below its steady-state
level at £ = 3 and shows damped oscillations
as it approaches its new steady state. (The
subsequent adjustment of the price level can
be described by the difference equation,
and
for £ even,
0
2-0
c '+ c f
(51) Met
199
P, =
+ ( 1 - 0 P .- 2 .)
Equilibrium real interest rates can be
computed in this model from consumptionEuler equations. Consumption by Odd
households rises at £ = 1 (when the open
market purchase occurs), then falls at £ = 2
(as Even households go to asset markets and
20
AWN
MAY/JUNE
acquire the portion of the new money that
Odd households spent at £ = 1). The twoperiod change in consumption for Odd
households is also negative, as the economy
approaches (with two-period oscillations) its
new steady state. So the two-period market
real interest rate falls at t = 1, and the implicit
one-period real interest rate from the consumption-Euler equations of Odd households
also falls. This is the liquidity effect of mon
etary expansion in the basic limited-participation model. Notice that because con
sumption of Even households falls at £ = 1,
the implicit one-period real interest rate from
their consumption-Euler equations rises at
t = 1, but this is not reflected in any market
interest rates because these households are
not currently participating in asset markets.
At t = 2, the two-period market real interest
rate rises above its steady-state level because
the two-period change in consumption of
Even households is positive. So the liquidity
effect in the limited-participation model is
necessarily of limited duration: It vanishes
(and in fact reverses itself) when the identity
of the participants changes.
The liquidity effect from the limited-par
ticipation model results from the temporary
change in consumption of the households
who have use of a disproportionate share of
newly printed money. In the simple model
discussed above, these households cannot
use this money to finance a permanent
increase in consumption. More generally,
the increase in money may raise liquidity (in
the model above, relax the two-period, cashin-advance constraint) by more than it raises
wealth, so households that obtain the addi
tional money may choose an increase in con
sumption that is (at least partly) temporary.
Although in equilibrium other households
must then experience a temporary fall in
consumption, the limited-participation
nature of the model breaks the link between
interest rates and the consumption-Euler
equations of those households.
Lucas (1 9 9 0 ) and Fuerst (1 992) developed
variations on the limited-participation model
that simplify it by using a representative
household, thereby eliminating wealth-redistribution effects. Their models go further
than the heterogenous-agent LP model dis
cussed above by eliminating the connection
between real interest rates and any consump
tion-Euler equation. The models split the
representative household into individuals
with unique tasks who later pool wealth and
consumption. One person in the household
purchases goods with money while another
participates in financial markets and receives
new money transfers. The new money, in
the hands of the latter person, cannot imme
diately reach the former person and is there
fore not available for immediate spending in
the goods market. As a result, nominal
goods prices do not immediately reflect the
new money. (W ith a binding cash-inadvance constraint in the goods market,
nominal prices do not depend at all on the
size of the current monetary transfer.) In
this way, the model generates short-run price
stickiness in response to unanticipated
increases in the money supply. The new
money introduced into the economy enters
the loans market as firms must use money to
pay inputs. Households work for money they
can use to buy goods next period. Because
they know that nominal prices will rise next
period, the nominal reservation wage rises in
proportion to that increase in prices. This
raises the nominal amount of money the firm
must borrow to pay wages. However, because
a disproportionate share of new money is
used in this factor market (rather than being
spread throughout all markets that require
money), the real interest rate falls.
Although the model is simpler than the
LP model in that there is a representative
household, its timing and household splits
add new complications. We describe here
only the setup of a basic RHLP model (read
ers are referred to the papers by Lucas and
Fuerst for discussions of the model’s solution
and im plications). The basic model has sev
eral steps. First, households start each period
with all the economy’s money, while firms
will hold all the economy’s money at the end
of each period. Initially, households divide
Representative Household LimitedParticipation (RHLP) Models
Limited-participation models are com
plicated because they involve heterogeneity.
199
21
REVIEW
MAY/JUNI
money between buying goods and lending
to financial firms: They lend D, dollars to
financial firms and keep Mt—D, dollars to
spend on goods. Second, financial firms
receive a lump-sum transfer, r , from the
government. Third, financial firms lend
their money, D, + r ,, to goods-producing
firms. Fourth, goods-producing firms use
some or all of this cash to pay w, L, for labor
services (in a perfectly competitive labor
market) because labor services are subject
to a cash-in-advance constraint. Fifth,
goods-producing firms produce f ( k t, z,L,)
goods using this labor and (previously
installed) capital; they decide how many
goods to install as capital for next period,
I , , and how many to sell to households in a
perfectly competitive environment. Sixth,
goods-producing firms sell f ( k t, z, L,) —I,
goods to households for Mt—D, dollars.
Seventh, goods-producing firms repay (with
interest) their loans from financial firms:
They pay (D, + r , ) ( l + it) to financial firms.
Eighth, goods-producing firms pay divi
dends, F I , , to households with all their
remaining money:
(54)
n, =M (- D ,- ( D ,+ T , ) ( l
+
where n t and vt are the nominal dividends
paid at the end of period t by goods-producing
firms and financial firms, Mt is beginning-ofperiod money balances, i,D, is the interest
the household earns on its deposits at finan
cial firms, w,L, is labor income, and P,c, is
nominal spending on consumption goods.
This budget constraint can be rewritten as
(57)
(58)
i1)
(59)
_< M
in ,
D..
U.
Info,
where Uc t+1 is the representative household’s
marginal utility of consumption at date t,
Pt+i is the price level at date £+1, and Infot is
the firm’s information set at date t. The fir-.
m’s production function depends on capital,
labor and a productivity shock, z t, so pro
duction is f ( k t,L t, z,)- Nominal profit equals
nominal income from sales, P( f ( k t,L r zt),
minus expenditure for investment, Pt kt+1,
minus expenditure for labor, w(L :
p t+1[l/(c,, 1 —L,) |Info, ],
Et
(6 0 )
t=0
subject to a sequence of budget constraints:
iPc
Households m ust choose labor supply and
deposits for date £ prior to the realization of
uncertainty at date t, but can choose con
sumption after the resolution of uncertainty
at that date.
The representative competitive goodsproducing firm maximizes
Finally, financial firms pay interest of (1 + i,)Dt
to households on their loans, and dividends
to households with all their remaining money,
(D, + T , ) ( l + i , ) — ( l + i,)D t. So the total
amount of money that financial firms pay
to households at this point is (1 + i,)D, +
(D, + r ,) (1 + i,). After this payment, house
holds have money balances of M, + r ,, which
come from four sources: interest on loans to
financial firms; wage income; dividends from
goods-producing firms; and dividends from
financial firms.
The representative household chooses
consumption, labor supply and deposits to
maximize expected utility,
(56)
M,+i = M t + r,
where t, is the nominal lump-sum transfer of
new money to the representative financial
firm and f ( k t,ztL t) — k l+1 is output of goods
minus investment spending by the represen
tative goods-producing firm. (This formula
tion assumes 100 percent depreciation of
capital each period.) The household is also
subject to a sequence of cash-in-advance
constraints:
+ D .+ T ,—w,Lr
(55)
1995
n, = P tf ( k t,L t,z t) - P tk t+l —w ,Lt.
The representative financial firm
acquires loans (deposits) D, from households
by paying interest it, receives a lump-sum
M (+1 - M, + I I t + u, + i,D, + wtLt - Ptc t,
22
REVIEW
MAY/JUNE
transfer r, from the government, and lends Bt
to goods-producing firms at the interest rate
i,. It chooses loans and deposits to maximize
(61)
data are not connected with asset prices in
the usual way because individual consump
tion decisions are made at finite intervals
that are longer than the measurement inter
val for asset prices (aggregate consumption
is related to asset prices indirectly and in a
different way, however). (This is reminiscent
of the Grossman-Laroque model of purchases
of durable goods, which are made infrequently
due to transactions costs). Lynch also assumes
individual heterogeneity in that decisions of
different individuals are staggered. This
assumption ensures that there is no decision
interval for which the model can be rewritten
in terms of a representative agent; hence,
aggregate consumption is not relevant for
asset prices. The staggering of decisions
makes the model similar to limited-participa
tion models: W ith two groups of agents, say,
Odds and Evens as in the earlier discussion,
Evens finance consumption in odd-number
periods out of previously held assets by selling
riskless, zero-transaction-cost assets. Finite
intervals for consumption decisions, with
staggered decision periods across households,
smooth the aggregate intertemporal marginal
rates of substitution and reduce their correla
tion with asset prices. Lynch studied the
implications of this model for the equitypremium/riskless-rate puzzle (with mixed
success). (The infrequent decisionmaking
might be thought to be due to the costs of
making decisions; Lynch calculates that the
total utility loss relative to every-period deci
sionmaking is about 1 percent of wealth.)
In Lynch’s model, consumption plans are
followed through with certainty between
decision intervals.
Though limited-participation models
of this sort appear to have met with at least
some success in asset-pricing issues more
generally, heterogeneous-agent limitedparticipation models have not been applied
quantitatively to liquidity effects. Like the
representative-household limited-participation
models, they break the simple link between
consumption and interest rates implied by
the usual consumption-Euler equation. In
contrast to the RHLP model, they do not
replace that connection with a similar rela
tion involving firms. Instead, they place
restrictions on movements in interest
Ot \lnfo,
where its nominal dividends l>, are given by
(62)
Note that the total amount a financial firm
pays to households at the end of a period
equals dividends plus interest payments,
or Tt + itB,.
The basic idea of the model above is
similar to the limited-participation models
of Rotemberg (1984) and Grossman and
Weiss (1983). In this model, all households
fully participate in financial markets, but
monetary transfers enter through credit
markets (in the sense that they go to financial
firms, which then lend the money) and
households cannot use this new money to
buy consumption goods. This breaks the
link between consumption growth and the
real interest rate. Although the separation
of product and financial markets creates a
sluggish response of the nominal price level
to a monetary shock, it is the dispersion of
markets rather than price stickiness p er se
that creates the effect of money on real and
nominal interest rates in the model.
Limited-participation models suggest
that econometricians can disregard aggregate
consumption data when examining the con
nection between consumption and interest
rates implied by the consumption-Euler
equation. The models instead impose a dif
ferent necessary condition relating real inter
est rates to different intertemporal marginal
rates of substitution. The Grossman and
Weiss model related it to the consumption
of a subset of consumers, that is, those who
are “in financial markets.” The RHLP model
relates it instead to an intertemporal margin
faced by firms.
Several other researchers have extended
these kinds o f models to deal with other
asset-pricing issues. In a recent article, for
example, Lynch (1994) develops a non-monetary model in which aggregate consumption
199
23
KEVIN)
MAY/JUNE
1995
Beaudry, Paul, and Michael Devereux. "Monetary Policy and the Real
Exchange Rate in a Price-Setting Model of Monopolistic Competition,"
Carnegie-Rochester Conference Series (forthcoming).
rates and consumption by a subset of the
population. This suggests those models
would be easier to test (using panel data)
than models in which money demand by
firms for purchases of inputs plays a major
role. It also suggests a possible common
model for explaining liquidity effects and
resolving other asset-pricing puzzles. We
believe additional research along these
lines may be useful.
Carlton, Dennis. "The Theory and the Facts of How Markets Clear: Is
Understanding Industrial Organization Valuable for Understanding
Macroeconomics?" in R. Schmalensee and R.D. Willig, eds., Handbook
of Industrial Organization. North-Holland, 1989, pp. 909-46.
C O N C LU S IO N S
Cho, Jang-Ok, and Thomas Cooley. "The Business Cycle with Nominal
Contracts," working paper (1990), University of Rochester.
Blanchard, Olivier, and Nobuhiro Kiyotaki. "Monopolistic Competition
and the Effects of Aggregate Demand," The American Economic
Review (September 1987), pp. 647-66.
Christiano, Lawrence J., and Martin Eichenbaum. "Liquidity Effects ond
the Monetary Transmission Mechanism," The American Economic
Review (May 1992a), pp. 346-53.
Most economists believe that liquidity
effects appear in the data for the U.S. economy,
though the size of the effects (if it even exists)
is a subject of controversy, due largely to
identification problems in statistical work.
The theoretical explanation for nominal or
real liquidity effects also remains controversial.
W hile many economists interpret liquidity
effects as results of sluggish nominal price
adjustments, others interpret them as reflecting
costs of complete and continuous participation
in markets that allow monetary changes to
cause redistributions or to channel spending
into certain areas (such as increased spending
by firms on factors of production). Others
suggest that liquidity effects reflect part of
the economy’s coordination on a particular
equilibrium when multiple solutions are pos
sible. Other alternative explanations may
appear in future research. Goodfriend (1995)
has recently suggested a model in which
imperfectly competitive firms face kinked
demand curves and price sluggishness emerges
endogenously, creating real effects of monetary
policy in which liquidity effects play a role.
More generally, the problem of explaining
liquidity effects theoretically is part of the
broader problem of explaining the effects of
monetary policies on a wide range of eco
nomic variables. Current explanations may
be suggestive, but no definitive model has
yet emerged.
_ _ _ _ _ _ and_ _ _ _ _ _ _ . "Liquidity Effects, Monetary Policy, and
the Business Cycle," National Bureau of Economic Research Working
Paper No. 4129 (August 1992b).
Cook, Timothy, and Thomos Hahn. "The Effect of Changes in the
Federal Funds Rate Target on Market Interest Rates in the 1970s,"
Journal of Monetary Economics (November 1989), pp. 331-51.
Cooley, Thomas, and Lee E. Ohanian. "Term Structure Inversion and
Real Activity," working paper (1990), University of Rochester.
Fama, Eugene F. "Stock Returns, Real Activity, Inflation and Money,"
The American Economic Review (September 1981), pp. 545-65.
Friedman, Milton. "Factors Affecting the Level of Interest Rates,"
Savings and Residential financing, 1968 Conference Proceedings,
U.S. Savings and Loan League, 1968, p. 7.
Fuerst, Timothy. "Liquidity, Loanable Funds, and Real Activity," Journal
of Monetary Economics (February 1992), pp. 3-24.
Goodfriend, Morvin. "Moderately Inflationary Monetary Policy," working
paper (1995), Federal Reserve Bank of Richmond.
Grossman, Sanford, and Laurence Weiss. "A Transactions-Based Model
of the Monetary Transmission Mechanism," The American Economic
Review (December 1983), pp. 871-80.
Hansen, Gary. "Indivisible Labor and the Business Cycle," Journal of
Monetary Economics (November 1985), pp. 309-27.
Jovanovic, Boyan. "Inflation and Welfare in the Steady State," Journal
of Political Economy (Februory 1982), pp. 561-77.
King, Robert G. "Money and Business Cycles," working paper (1991),
University of Rochester.
Kydland, Finn, and Edward Prescott. "Business Cycles: Real Facts and a
Monetary Myth," Federal Reserve Bonk of Minneopolis Quarterly
Review (spring 1990), pp. 3-18.
REFERENCES
Lucas, Jr., Robert E. "Liquidity and Interest Rates," Journal of Economic
Theory (April 1990), pp. 237-64.
Barro, Robert, and Herschel Grossmon. "A General Disequilibrium
Model of Income ond Employment," Jhe American Economic Review
(March 1971), pp. 82-93.
_ _ _ _ _ _ . "Interest Rates and Currency Rates in a Twotountry Model,"
Journal of Monetary Economics (November 1982), pp. 335-59.
24
Lynch, Anthony W. "Staggered Decision-Making by Individuals: Pricing
Implications and Empirical Evidence," working paper (1994),
University of Chicago.
Marcet, Albert. "Solving Nonlinear Models by Parameterizing
Expectations," working paper (1990), Federal Reserve Bank of
Minneapolis.
Ohanian, Lee E., and Alan C. Stockman. "How Much Price Stickiness is
Necessary for Reasonable Liquidity Effects?" working paper (1994).
Prescott, Edward. "Theory Ahead of Business Cycle Measurement,"
Carnegie-Rochester Conference Series on Public Policy (spring 1986),
pp. 11-44.
Rotemberg, Julio. "A Monetary Equilibrium Model with Transactions
Costs," Journal of Political Economy (February 1984), pp. 40-58.
Stockman, Alan C.( ond Lee E. Ohanian. "The Short-Run Independence
of Monetary Policy under Pegged Exchange Rates," working paper
(1994), University of Rochester.
U lS
25
REVIEW
MAT/JUNE
199 5
Kevin D. Hoover is a professor of economics at the University of California, Davis. The author thanks Kevin Salyer, Daniel Thornton and the
participants in the University of California, Davis, Macroeconomics Workshop for helpful comments.
Com m entary
jo in t and not independent. But if we keep
these restrictions in mind, we can still use the
IS-LM model to understand what some of
the issues are.
K e v in D . H o o v e r
THE L IQ U ID IT Y A N D
IN F L A TIO N EFFECTS
ee Ohanian and Alan Stockman’s paper
presents a careful, lucid survey of a series of
technically difficult models of the liquidity
effect. I would like first to clarify for the reader
(exacdy as I had to do for myself) what seem to
me to be the key issues and conclusions of
their survey and then to offer an assessment
(not of their admirable paper, but) of the
research program on which it reports.
In years, I am somewhat younger than
Alan Stockman and slightly older than Lee
Ohanian. In training, however, I am com
pletely antediluvian. I will therefore attempt
to clarify the issues using, for the most part,
the venerable (and much maligned) IS-LM
model. Ohanian and Stockman begin their
analysis of the liquidity effect with an IS-LM
model. But they treat it as if it were ju st
another model on par with the sequence of
micro-foundational models that they explore
through the rest of the paper. Actually, the
IS-LM model operates on a different plane
from the other models. It displays the rela
tionships among aggregates with only the
most im plicit commitment to particular
micro-foundations. Therefore, each of
Ohanian and Stockman’s models can be seen
as attempts to fill in the details that lie behind
the IS-LM-AS (aggregate supply) analysis,
rather than as substitutes for that analysis.
Thus, consider, for example, their first model:
the sticky-price model. Cast as a dynamic
optimization model for a representative agent,
it can be seen as particularizing the IS-LM
model: It treats the consumption function as
a life-cycle or permanent-income type; the
investment function as the most rudimentary
neoclassical type; money demand as governed
by the quantity theory; and labor supply as
derived from utility maximization; and, most
important of all, it insists that these functions
and the decisions that lie behind them are
L
1 That it is only the IS curve that is
shifted by a change in anticipated
inflation results from nominal inter
est rates being plotted on the verti
cal axis. Were the vertical axis to
measure real interest rates instead,
then the LM curve would shift and
the IS curve would not be directly
affected. See Mundell (1971,
pp. 18-19).
Consider the canonical (classical)
IS-LM-AS in Figure 1. The aggregate supply
curve is vertical, indicating market clearing
in labor and product markets. If the money
supply increases by AM, the LM curve shifts
to the right, and equivalently the AD (aggre
gate demand) curve shifts upward. At the
original price level, p, aggregate demand
exceeds aggregate supply. If such a situation
is technically infeasible, then prices must rise
from p to p', which shifts the LM curve back to
its original position, since the location of the
LM curve depends on the real money supply,
M/p. Interest rates remain unchanged. This
is the flexible price case.
Now, if, in the short-run aggregate
supply can exceed its long-run level at AS
(which might be, as Ohanian and Stockman
observe, a result of the assumption of
monopolistic competition) and if prices are
sticky, say at p, then output rises and interest
rates fall from i to i'. This is the liquidity
effect: An expansion of money results in a fall
in interest rates. It is a fall in the nominal
interest rate and, because prices are sticky, in
the real interest rate as well.
If increases in the money supply are not
simply lump-sum, but are this period’s real
ization of higher growth rates of money asso
ciated with increasing rates of inflation, then
there is another effect to consider. Investment
is a function of ex ante (or anticipated) real
rates, not of nominal rates of interest. As
anticipated rates of inflation increase, a con
stant ex ante real interest rate must be repre
sented by higher nominal rates of interest.
Thus, as the rate of inflation increases in
Figure 2 by Ap, the IS curve must shift verti
cally by Ap in order for each level of GNP to
correspond to the correct real interest rate.1
F EDERAL RESERVE B A N K OF ST. L OU I S
26
REVIEW
MAY/JUNK
F ig u re 1
If prices are sticky at p, then the nominal
interest rate rises by less than the increase in
inflation to i': Anticipated inflation stimulates
output, investment and (short-run) growth.2
Real rates therefore fall. This is the MundellTobin effect.3 If, however, prices are flexible,
prices rise until aggregate demand equals
aggregate supply (from p to p' ), shifting the
LM curve back and increasing interest rates
from i to i". Here, the nominal interest rate
fully reflects the increased rate of inflation.
This is the Fisher effect: An increase in infla
tion corresponds to a one-for-one increase in
nominal interest rates.4
There can also be a Mundell-Tobin effect
with flexible prices. If consumption depends
not only on GNP but also on the real value
of nominally denominated wealth (bonds or
money), then, as in Figure 3, there are two
effects: First, the upward shift of the IS curve
by the full amount of the anticipated inflation
and, second, a partly offsetting downward
shift of the IS curve owing to rising prices
that reduce real wealth in the consumption
function.5 The LM curve has to shift less far
back to restore equality between aggregate
demand and supply, so that interest rates rise
from i to i', which is less than the increase in
inflation. Real rates of interest therefore fall.
Because of the full employment assumption,
current output is unchanged, but investment
rises (and consumption falls) as a result of
lower real rates of interest, which increases
the rate of GNP growth.
Therefore, we have three effects: a liquidity
effect; a Fisher effect; and a Mundell-Tobin
effect. These constitute a powerful taxonomy,
and— I think— one that is clearer than that
used by Ohanian and Stockman. Ohanian
and Stockman define a liquidity effect to be
the systemic change in an interest rate as the
result of a monetary expansion. They distin
guish between a real and a nominal liquidity
effect. I think that it would better to reserve
the term liquidity effect only for those changes
in the real rate of interest induced directly by
money expansion. The only interesting liquid
ity effects are real because, without a change
in the real rate of interest, there can be no
accompanying effect on any other real vari
ables that interest us: GNP; investment; con
sumption; and employment. W hat we should
199
Nominal Interest Rate
2 The steody-state rote of growth is
rot affected.
recognize is that liquidity effects are often
(and depending upon the model, often nec
essarily) accompanied by the other two effects.
The systemic effect is the sum of several par
tial effects.
TH E M O D ELS
Ohanian and Stockman consider a series
of models. How do these effects play out in
each of them?
One-Sector, Sticky-Price, RationalExpectations Model
Some of the key features of this model
were already described above. One addition
al feature is that money is modeled using the
cash-in-advance constraint. In the IS-LM
model, this means that the LM is vertical at a
level of GNP determined by the real supply
of money (Figure 4 ). The increase in money
(the shift of LM to the right) has the usual
liquidity effect, ceteris paribus, reducing the
interest rate. In this model, we can under
stand that effect as arising from an increase
in current consumption because of the relax-
27
3 Tobin (1965) concentrates on the
increase in the steody-state capital
stock that results from the depres
sion of real interest rates due to
anticipated inflation, and the
Mundell-Tobin effect is often
thought of as a proposition about
capital deepening and the accom
panying increase in steody-state
consumption. The presentation
here, however, closely follows
Mundell (1971, chapter 2).
Mundell focuses on the failure of
the Fisher hypothesis through
exactly the some mechanism.
Mundell's treatment suits the issues
raised in the current discussion of
the liquidity effect. For an excel
lent historicol discussion of the rela
tionship of the Fisher hypothesis to
the Mundell-Tobin effect, see
Cottrell (1993).
4 See Fisher (1930).
s The initial shift of the IS curve from
IS lftp to lS l^ + A ftp would,
ceteris paribus, result in a shift in
aggregate demand from AD to AD'.
Since aggregate demand falls with
rising prices because of the wealth
effect, the final aggregate demand
curve is flatter than AD".
ffEVIEW
MAY/JUNE
1995
of intertemporal substitution between con
sumption today and tomorrow). Yet, either
way the real rate falls, which is the important
point about the liquidity effect. Christiano
(1 9 9 1 ) usefully distinguishes between a dom
inant liquidity effect (Figure 4a) and a non
dominant liquidity effect (or dominant infla
tion effecting, as shown in Figure 4b).
F igu re 2
Nominal Interest Rote
Models with Some Sticky Prices
6 This point deserves emphasis since
Ohanian and Stockman note it in a
way that suggests mild surprise or
novelty and because many econo
mists talk os if an interest-elastic
money demand implied a liquidity
effect. Although the interest elas
ticity of money demand affects the
size of the liquidity effect, it is nei
ther necessary nor sufficient for its
existence. It is not necessary
because there is a liquidity effect in
Figure 4, where ttie LM curve is
vertical. It is not sufficient because
there is no liquidity effect in Figure
1, where the LM curve is not verti
cal, unless we make the additional
assumption that prices are sticky.
In the model with perfectly flexible
prices, there is no anticipated inflation effect,
because the price level jum ps immediately to
bring aggregate supply and demand back into
equality. If some prices are flexible this period,
fewer prices will have to adjust in future
periods. Therefore, while there will be an
anticipated inflation effect, it will be smaller
than in the model with all prices sticky. It is
therefore in this case less likely that the infla
tion effect will dominate as in Figure 4b, and
more likely that both nominal and real interest
rates will fall. Ohanian and Stockm an’s con
tribution with respect to this model is to show
that for reasonable parameterizations only
small degrees of price stickiness are enough
to produce dominant liquidity effects.
ation of the cash-in-advance constraint.
Optimally, agents want to consume more in
future periods as well (an im plication of the
life-cycle/permanent-income hypothesis) and
so increase capital accumulation today, low
ering the rate of return on capital and the real
rate of interest. (Notice that the liquidity
effect here has nothing whatsoever to do with
the interest elasticity of the demand for money;
it works the same whether the LM curve is
vertical or simply upward sloping.)6
Prices are pre-determined in the current
period in this model. They can, however,
adjust in the future. Thus, an unexpected
increase in money this period is associated with
perfectly anticipated inflation. Consequently,
the IS curve shifts up at the same time that
the LM curve shifts out. There are two cases.
In Figure 4a, the liquidity effect outweighs
the inflation effect, and both real and nom i
nal interest rates fall. In Figure 4b, the infla
tion effect outweighs the liquidity effect, and
nominal interests rates rise, although since i'
is less than i", real interest rates fall. As
Ohanian and Stockman observe, which case
one obtains depends on the degree of relative
risk aversion (or equivalently on the degree
Capital Accumulation and
Technology Shocks
The possibility of Hicks-neutral technol
ogy shocks and capital accum ulation intro
duces two new effects into the analysis. A
technology shock can be represented as in
Figure 5 as shifting the AS curve to the right.
If prices were perfectly flexible, they would
fall immediately (without anticipated infla
tion) from p to p', shifting the LM curve
right and lowering the interest rate to i".
This is the pure technology-shock effect. If
prices are not perfectly flexible, however,
there will be anticipated deflation, and the IS
curve will also shift down, so that the final
equilibrium interest rate is i". This fall in the
nominal rate is greater than the pure Fisher
effect ( i—i" ') so that the real rate falls.
Unlike a neutral technology shock, capi
tal accumulation, ceteris paribus, not only
shifts the AS curve to the right, it also
reduces the marginal product of capital,
INK OF ST. L OU I S
28
review
MAY/JUNE
which can be represented by a downward
shift of the IS curve. This would simply
magnify the drop in interest rates (real and
nominal) in Figure 5. The interesting ques
tion, which can be answered in Ohanian and
Stockman’s parameterized model, but not in
this qualitative model, is what the typical
time-series behavior of interest rates would
be given repetitive shocks to both money
and technology in the face o f capital accu
mulation.
1995
F ig u r e 3
Nominal Interest Rate
The Increasing-Returns Model
The increasing-returns model is much
more difficult to cast in an IS-LM-AS frame
work, although it could no doubt be done
with some ingenuity. The model, which I
know only from Ohanian and Stockman’s
sketch, does not appear to have very attrac
tive properties with respect to the empirical
observations that motivate Ohanian and
Stockman’s survey. I therefore omit further
consideration of it.
Limited-Participation Models
Ohanian and Stockman consider two
models in which asset markets and goods
markets are separated so that household
decisions with respect to holdings o f money
and consumption must be made sequentially.
The simpler of the two for our purposes is
the one associated with Grossman and Weiss,
and Rotemberg. This model is less easily
rendered in an IS-LM version. Nevertheless,
it is not difficult to understand its essence.
I
think that it is helpful to consider an
even simpler model than Ohanian and
Stockm an’s simplification of Grossman and
Weiss’ model. I like to think of the
Grossman and Weiss model as a kind of
Hindu overlapping-generations model. The
young in the overlapping-generations model
corresponds to Grossman and Weiss’ odd
agents, and the old to their even agents.
Instead of the old dying as they do in a stan
dard overlapping-generations model, they
are reincarnated again as young, while the
young become old. Unlike Ohanian and
Stockman, or Grossman and Weiss, let us
first consider the model without a cash-in-
advance constraint. That an increase in the
money supply that is disproportionately
received by one of the generations in an
overlapping-generations model results in a
less-than-proportional increase in prices is a
standard result.7 The intuition is simple. If
the young receive an increase, they appear to
have higher wealth at current prices and
wish to consume more, now and in the
future. If total resources are fixed, then
prices rise. This has two effects: It reduces
the real value of the increased money avail
able to the young, and it reduces the real
value of the money held by the old, reducing
their consumption and freeing up resources
available to the young. Because extra resources
are available to the young, the price level need
not rise as far to reduce the real value of the
young’s holdings of money to a sustainable
level. The trick only works because the
resources available to the young can be
extracted from the old through an inflation
tax. As the money becomes more dispersed—
as it must when the young spend some of
it— both young and old come to hold the new
money balances. But since inflation can
transfer resources but not create them, even
NK OF ST.
29
LOUIS
7 See Hoover (1988, chapter 6,
section 3).
F ig u re 4
(a)
TH E PROSPECTS FOR
L IQ U ID IT Y M O D ELS
Nominal Interest Rate
Nominal Interest Rate
tually prices must rise in direct proportion
to the increased money supply. Thus, even
though prices rise less than proportionally this
period, further inflation can be anticipated.
Since money is the only asset in my sim
plified version of this model, the real rate of
interest is ju st the inverse of the rate of infla
tion, and the nominal rate of interest is zero.
The liquidity effect in this model is the fall
in the real rate due to increased inflation.
Ohanian and Stockman’s version of Grossman
and Weiss’ model is able to retain a non-zero
nominal interest rate by imposing a cash-inadvance constraint that applies to two periods
instead of ju st one for those in the asset
market, and to one period for those out of
the asset market. This means that even if
nominal interest rates rise above zero, money
will of necessity still be held.
The Lucas model goes one step further
in collapsing the two types of agents into a
single household with common budgets and
consumption, but with sequencing of finan
cial and consumption decisions or differenti
ated roles for household members. The fun
damental idea remains similar.
The critical question to ask about all of
these models is whether they really capture
the liquidity effect that we think we see in the
world or, for that matter, whether they capture
our intuitive understanding of the liquidity
effect. I think there is reason to doubt that
they do. I see two related problems.
First, in all of these models— including
even the basic IS-LM model with a vertical
LM curve— the liquidity effect seems to
operate through the wrong mechanism. The
modus operandi of the liquidity effect in the
models surveyed by Ohanian and Stockman
is to affect the marginal rates of substitution
between consumption in different periods or
between consumption and leisure. In most
of the models, although an interest rate is
determined, interest rates do not matter in
the sense that they are not causally impor
tant, but rather are priced as redundant
assets. Thus, for example, the sticky-price
model sets the interest rate according to the
following logic: If there were a bond, it
would have to yield a real rate of interest
equal to the marginal rate of substitution
between consumption today and consump
tion tomorrow, or it would be dominated in
rate of return by real capital. In the model,
that logic is perfectly correct; it picks out the
correct shadow price. My concern, however,
is precisely that this is a shadow price of a
shadow asset. Bonds do not do anything
important in the model; they are dispens
able— a fifth wheel for the econom ic car.
Yet, normally, we think the liquidity effect is
important because monetary policy affects
interest rates and interest rates are causally
effective in altering consumption and, more
particularly, investment decisions.
These models miss this feature for two
reasons. First, because they are generalequilibrium models in which all the endoge
nous variables are determined simultaneous
ly, they cannot adequately model the causal
efficacy o f interest rates. Causes operate in
particular directions, and directionality
requires recursive rather than simultaneous
structure. The second reason that these
models miss the causal efficacy of interest
REVIEW
MAY/JUNK
rates is that there is no mechanism for inter
est rates to be determined other than
through the shadow prices associated with
consumption, leisure and saving choices.
The financial sectors in these models are
simply not rich enough. Partly this is a
result of the models assuming representative
agents. Robinson Crusoe does not need a
financial system. The point is not that there
is something wrong in principle with general-equilibrium models. Rather, it is that a
model of the liquidity effect should have
important financial markets. Models in
which the same equilibria are supported
whether or not bonds are present appear to
miss the crucial point.8
Actually, Grossman and Weiss’ model is
not technically a representative-agent model
if that is taken to mean a single-agent model,
but it is close enough to make no difference
for my point. Similarly, it is true of all the
limited-participation models that bonds have
a non-redundant function. Their function,
however, is an artifact of the cash-in-advance
constraint, about which I shall have more to
say presently.
It might be useful to compare the under
standing of interest rate determination in
this model with that of Joh n Maynard
Keynes in The G eneral Theory.9 Keynes
assumed that the economy was populated
by heterogeneous people with a diversity of
opinions about what normal interest rates
were. At current rates, those who feared
capital losses would hold money (which
for Keynes included Treasury bills and other
short-term, interest-bearing assets) and those
who hoped for capital gains would hold
bonds. The market interest rate was deter
mined as the point at which those hopes and
fears were balanced. An injection of money
lowered interest rates because a lower rate
increased the proportion of the population
who feared capital loss and therefore increased
their willingness to hold the new money.
Investors in real capital then looked to the
market rate as part of their process of evalu
ating the desirability of a new investment. I
am not arguing that Keynes’ analysis is nec
essarily correct, but it does seem better than
the representative-agent models at capturing
the spirit of the liquidity effect: The finan-
199
F ig u re 5
Nominal Interest Rate
rial market matters fundamentally and inter
est rates are causally efficacious. Two further
points: Keynes highlights the importance of
the term structure, because he distinguishes
between short (or money) rates of interest
and long-term rates of interest and, at the
same time, he raises the question of what to
count as money.
What to count as money raises the second
major problem with all of the models in
Ohanian and Stockm an’s survey: They rely
on the cash-in-advance constraint. Money,
therefore, is whatever asset is a direct con
straint on current expenditure. The cash-inadvance constraint is, however, a lousy way
to model money, mainly because it is not
clear that any asset serves uniquely to limit
current expenditure. For very few purchases
is cash literally necessary in advance of pur
chase: Coins in vending machines are one of
the few obvious examples. For any number
of other purchases, cash may be preferred,
for example, because of lower transactions
costs or anonymity. Reflection on the trends
in transactions technology, however, suggests
that cash is becoming less and less essential;
credit cards do for almost everything. And,
F EDERA L RESERVE R A N K OF ST. L OU I S
31
8 McCollum (1983) mokes a related
point about money. In an accept
able monetary model, the exis
tence of money should alter the set
of possible real outcomes for the
economy.
9 See Keynes (1936, chapter 13).
REVIEW
MAY/JUNE
REFERENCES
although credit card accounts must be settled,
ultimately through the transfer of central
bank reserves between banks, the settlement
is not in advance but in arrears. And this is
not ju st the trend of the future; it is for large
parts of the economy the established practice
from ancient times. Businesses operate on
trade credit. My grandparents, who ran a
rural grocery in Georgia, extended credit to
the inhabitants of Chattanooga Valley and
were paid when their customers became more
liquid. The amusing chapter in Thackery’s
Vanity Fair on “How to Live on Nothing a Year”
is premised on the practice of 19th-century
English merchants of extending credit payable
on “quarter days.” A good model of money
is elusive, but the cash-in-advance constraint
is a weak reed on which to build it.
So where does that leave us? Ohanian
and Stockman’s efforts mean that we know a
good deal more about a particular research
program, but I am afraid what they have
convinced me is that the particular research
program— representative-agent, cash-inadvance models— is a dead end and that we
still are a long way from understanding the
liquidity effect.
1995
Christiano, Lawrence J. "Modeling the Liquidity Effect of a Money
Shock," Federal Reserve Bank of Minneapolis Quarterly Review
(winter 1991), pp. 3-34.
Cottrell, Allin. "Keynes and the Keynesians on the Fisher Effect," working
paper (October 1993), Wake Forest University.
Fisher, Irving. The Theory of Interest. Macmillan, 1930.
Hoover, Kevin D. The New Classical Macroeconomics: A Sceptical
Inquiry. Blackwell, 1988.
Keynes, John Maynard. The General Theory of Employment Interest and
Money. Macmillan, 1936.
McCallum, Bennett T. "The Role of Overlapping-Generations Models in
Monetary Economics," in Kad Brunner and Allan H. Meltzer, eds.,
The New Monetary Economics, Fiscal Issues and Unemployment.
Carnegie-Rochester Conference Series on Public Policy, vol. 23.
North Holland, 1983, pp. 13-46.
Mundell, Robert A. Monetary Theory: Inflation and Growth in the World
Economy. Goodyear, 1971.
Tobin, James. "Money and Economic Growth," Econometrica
(October 1965), pp. 671-84.
32
REVIEW
MAY/JUNE
1995
Adrian R. Pagan is a professor of economics at The Australian National University and the University of Rochester. John C. Robertson is a
lecturer in econometrics at The Australian National University. The authors would like to especially thank Dan Thornton for his help in prepar
ing this article. Bill Lastrapes, Larry Christiano, Eric Leeper and David Longworth have also given the authors valuable criticism and insight.
All computations were performed using programs written in GAUSS v3.2.4. The programs and data are available on request through e-mail:
john.robertson@anu.edu.au.
Resolving the
Liquidity Effect
that excess money balances have a powerful
direct influence on expenditures, conventional
wisdom on the transmission mechanism of
monetary policy has been that the effects are
felt via interest rates. A very stylized view of
this mechanism is available from the money
demand and supply relations, which are
either explicit or im plicit in m ost models:
A d ria n R. P a g a n an d
Jo h n C. R obertson
“Resolving: To separate into constituent or
elem entary parts”
(The Macquarie Dictionary)
m f = a 1 + a 2rt + e f
(2)
m st = /3l + P 2rt + e st
mf = m(s,
he effect on interest rates of a change in
monetary policy has long been an impor
tant topic in monetary econom ics, and
there is now a large body of literature that
has studied the existence and magnitude of
any such effect. Strong conclusions have
emerged, and yet, little is available by way of
work that attempts to account for the diver
sity of conclusions. This article aims to fill
some of this gap. As the title suggests, it does
this by separating out the basic elements of
the arguments that lead to the recorded con
clusions. In later sections, these are enumer
ated and discussed. The first section of the
article sets out the framework underlying
existing studies, followed by an examination
of whether the proper object of investigation
is a single relationship or a complete system.
We come down in favor of the systems view
point. Even then, there are many other factors
that can account for a diversity of outcomes,
and section three is devoted to a consideration
of these, ranging from issues of measurement
to the sample of data selected for the empirical
work. The fourth section explores the inter
relationship of monetary policy and the term
structure, while the final section presents
some conclusions.
where d indicates demand, s supply, mt is the
log of nominal money, r, is the nominal interest
rate, while £ f and ets are mutually uncorrelated
demand and supply shocks. In the textbook
treatment of this model, r„ responds to shifts
in the money supply, engineered by varying
f t , and the relation dr/d[3l = ( a 2—f t ) ' 1 means
that the interest rate decreases when money
supply increases, provided a 2< 0 and
f t ^ —a 2. This negative reaction of the inter
est rate to a rise in money supply is termed
the liquidity effect.
W hen there is a random variable attached
to money supply, a change in /3, can be thought
of as a movement in the expected value of
f t + £ s„ and the money supply shock might
simply be re-labeled e\', with the conceptual
experiment performed by changing the
expected value of e ffro m f t to a new value.
Since, mathematically, there is no difference
between the response to a change in £\ or a
change in the expected value of £ s', we will
henceforth concentrate upon describing the
effects of a change in £\. Such an orientation
is now standard in the literature and will be
adopted here, so that the liquidity effect will
focus upon the simulated response of inter
est rates to a money supply shock, setting all
other shocks to zero.
The above model is static and implies
that all adjustments are instantaneous. To
make it dynamic, one might augment each
relation in equations 1 and 2 with lagged
T
TH E BASIC M O D E L
Although there has been some dissent
over the years, mainly from those believing
(1)
33
review
MAY/JUNE
values in m, and r, to produce
(3)
variables are important to money demand
and supply). It is now no longer sufficient to
focus ju st upon the interest elasticity of the
demand and supply of money.
In practice, the relations in equation 5 will
also exhibit dynamics, possibly with lagged
values of all the variables appearing on the
right-hand side of each function. If we collect
the variables that are regarded as being part
of the system in an n x 1 vector z„ we could
write the supply and demand functions as
m, = <*! + a 2r, + Bdm(L)m , 4- Bdr(L)r, + e f
m , = Pi + f a + Bsm(L)m , + Bsr (L)r, + e\
with By(L) being polynomials in the lag
operator of the form bUjL + b 2ijL2+ ...... There
is now a distinction to be made between
impact effects and the responses over time.
In general, one can solve these equations to
produce a moving-average representation for
interest rates:
(4)
(8)
rt = Cd( L ) e f + C s( L ) e st
1 From now on we will identify struc
tural equation errors according to
the variable taken to appear on the
left-hand side of the equation. This
hos the advantage of freeing up the
choice of whether it is the interest
rate or money that should be the
dependent variable in a demand or
supply equation. Hence, e f is the
error in the structural equation thot
has m( on the left-hand side, and
this might be either demand or
supply, depending upon the con
text. For example, Gordon and
Leeper (1994) choose to normalize
the demand equation with money
and the supply equation with the
interest rate.
m, = a , + a 2rt + a 3p,
+ « 4 J . +
where Cj(L) = (c0j+ cljL + ....), and the impact
effect will be c0s - ( a 2—/32)'1 while the effects
over time are measured from the impulse
responses cks.
In the framework ju st described, strong
restrictions have been placed upon both the
demand and supply functions of money, as
the demand for money would also be expected
to depend, inter alia, on the level of income
(or wealth) and the price level, while the
supply of money depends upon the “reaction
function” of the authorities. In the scenario
described by equation 2, the reaction func
tion depends solely upon the current level of
the interest rate, whereas one might expect
that current developments in the price level,
exchange rates, output and so on would also
play a role. Thus, ignoring dynamics for the
moment, equations 1 and 2 might become
(5)
199
(9)
V
D
z . +
e ,5
r, = y 1+ y 2m ,+ y 3p,
+ y ^ y t+ B lk( L ) z l + e f
More generally, the whole system might be
written as
(10)
B0z,_! = Btz,_t + ... BpZl_p + £(z.
Pre-multiplying equation 10 by B^ yields the
“reduced-form” vector autoregression (VAR)
representation for zt,
(11)
= B-1B l V l + ... + B-1BpZ,_p + B-‘ <
= A iP z ^ + e *
and solving for rt, gives us a moving-average
representation of the interest rate of the form
( 12 )
m, = a x+ a 2rt + a 3p, + a 4y, + e f
rt = D z ( L X
= Dz (L )B -‘e*
m , = Pi + P 2r, + P ip, + P *y, + £ 5,.
= C m(L )e" + C -z( L ) e f,
where p, and y l are logs of the price level and
output, respectively. If one inverts the money
demand function to produce
(6)
where z, are the elements in z, excluding m„
and £™= e\ is the money supply shock.1 Note
that there are two decompositions presented
here; one involving the “reduced form” shocks
from the VAR in equation 11, and one
involving the “structural” shocks e\ from
equation 10.
Questions over the existence and magni
tude of the liquidity effect are seen to hinge
critically upon the measurement of the para
meters in the “structural relations.” In par
ticular, to isolate the money supply shock, it
is necessary that one be able to estimate both
the contemporaneous effects, a,, yr and the
rt = y l + y 2m, + y 3p , + y ,y , + e[
the immediate liquidity effect will be
(7)
dr, /de\ = y 2dm t /de\
+ y i d p , / d e s, + y , d y , / d e s,
and this depends upon more parameters
than ju st a 2 and /32 as r, could change either
directly, or indirectly through variations in p,
and y t. To evaluate the full effect, therefore,
requires us to consider the complete system
formed from m„ r„ p„ y, (and whatever other
34
REVIEW
MAY/JUNE
Subtracting equation 13 from equation 6
then yields a relation among the reduced-form
errors:
nature of the dynamic relationships. For
example, if the terms Bsr(L)rt were omitted
from equation 3, the identified supply shock
would actually be Bsr(L )rt+ e st, and so the
computed impulse responses would be
incorrect. It is no wonder then that much
of the controversy about the presence and
nature of the liquidity effect really comes
down to estimation issues.
(14 )
Single-Equation Estimation Methods
In a single-equation method, an attempt
is made to directly estimate the terms of Cm(L)
in equation 12. Early studies, summarized in
Thornton (1 9 8 8 ), absorbed C?(L)£f into the
error term, and then proceeded to measure
£™by regressing m, against lagged values of
m„ y t and p„ and so on. However, such a
regression does not produce an estimate of
e " in general, but rather the reduced-form
error e™. The two will coincide only if there
are no contemporaneous effects of any vari
ables upon money. Hence, the methodology
involves strong assumptions. A further problem
is that the error term in the regression of r, on
£,”j (j ^ 0 ), cannot be uncorrelated with e f
unless all the shocks are uncorrelated. This
assumption seems most problematic if the
system has been under-specified, either in
terms of lag length or the number of variables
taken to constitute it. Failure to account for
these effects will lead to biases in the estimated
coefficients. A different complication is the
fact that residuals replace £ ™in the estimated
relation. Because one is estimating the coef
ficients of lagged values of £™, the situation is
that analyzed in Pagan (1 9 8 4 ), where it is
shown that the estimated standard errors
are understated.
A related single-equation approach which
focuses on estimating the impact response c0s
is that of Mishkin (1981, 1982). He inverted
the money-demand equation as in equation 6
and took expectations with respect to some
assumed information set
to produce
Systems Methods
Simultaneous-equation estimation
methods address the issue of how to estimate
the parameters of a system such as those in
equation 10. However, some assumptions
have to be made about the nature of the system
if consistent estimates are to be obtained,
and a number of approaches have emerged in
this regard. Each approach is in evidence in
the literature on the liquidity effect and
involves some constraint upon the covariance
matrix of the errors £f and/or the parameters
in the matrices B0, Bh ... . Table 1 summarizes
the four main approaches in this context.
In the Cowles Commission methodology,
co v (ef) was left unrestricted, but the Bj(j > 0 )
was restricted.2 For models of monetary
phenomena, this often meant that enough
lagged values of rt, m, and so on, were omitted
from the system to identify the coefficients
attached to the endogenous variables remaining
in the demand and supply equations. Sims
(19 8 0 ) condemned such exclusion on
restrictions as “incredible,” a stance that has
been taken up by the academic community
to such an extent that one now rarely sees
the Cowles Commission approach mentioned
E(rt\nt_l ) = y 1+ y 2E(m t\nt_l )
+ 7 3E(p 1k - i) + y 4 E ( 3 'tk - i ) -
e\ = y 2e" + y3ef + y , e yt + £,r.
Effectively, one is attempting to estimate
the parameters of a money-demand function.
However, one might query whether this is a
satisfactory method for doing so. First, e™
only measures the money supply shock if
there are no contemporaneous effects of p, or
r, on money supply (a restriction explicitly
recognized by Mishkin). Second, e pt, e yt and
so on are correlated with e\ in general, since,
from equation 11, ef = Bgle f will be a func
tion of e\. Finally, it is necessary that precise
estimates of £'" be extracted, and this necessi
tates making the set of conditioning variables
large enough to completely describe the
money supply relation.
The two methods ju st described will be
referred to as single-equation procedures and
designated as SING1 and SING2, respectively.
E S TIM A T IO N M E TH O D S
(13)
199
35
2 Of course, the Cowles Commission
methodology recognized other pos
sibilities, which effectively corre
sponded to the other approaches
documented in Table 1, but these
were rarely implemented. See, for
example, Koopmons, Rubin and
Leipnik (1950).
REVIEW
MAY/JUNE
the Bj’s than were needed to exactly identify
the parameters. The assumption of a recur
sive model exactly identifies the parameters
of the system and, hence, imposes no testable
restrictions on the VAR. One might therefore
categorize the differences as simply amounting
to whether one wants to work with an exactly
identified system or not.
The Wold ordering technique seems to
be very popular in the literature on the liquidity
effect, being used by Leeper and Gordon
(1 9 9 2 ), Eichenbaum (1 9 9 2 ), Christiano
and Eichenbaum (1 9 9 2 ), Sims (1 992) and
Eichenbaum and Evans (1 9 9 2 ), inter alios.
This method will be denoted as SYS1 in what
follows. For a given set of variables, authors
utilizing the SYS1 approach often experiment
with many different orderings, and seem to
select between these observationally equivalent
structures according to some prior belief about
the signs and persistence of selected impulse
responses computed from the system. For
example, Eichenbaum criticizes the ordering
adopted by Sims (1 9 9 2 ), in which the interest
rate is taken as pre-determined, on the grounds
that a monetary expansion, brought about by
a decrease in e \, produces persistent negative
effects upon prices. Actually, this modus operandi
is quite similar to the approach taken by
researchers within the Cowles Commission
tradition, in the sense that the validity of
their estimates was often analyzed by the
simulation properties of the models, that is,
the dynamic responses of endogenous vari
ables to selected exogenous variables.
O f course, there are intermediate posi
tions. The order condition for identification
requires that the number of unknown para
meters in B0 must not exceed n(n+1)/2 —n,
and these might be distributed throughout B0
rather than being placed so as to make it tri
angular. This method is often referred to as
a structural VAR (SVAR) approach, in the
sense that while no restrictions are imposed
upon the dynamics via Bj (j 2 :1 ), non-triangular restrictions are imposed on B0. We will
designate this as the SYS2 method. In the
liquidity literature, the main representative
of an SYS2 structure is Gordon and Leeper
(1 9 9 4 ), who work with a system of seven
variables [m, r, u, y, p, rl0, cp], where u is the
unemployment rate, r10 is the 10-year bond
Restrictions on Equation 10 Used in
Different Systems Methods
B0
Cowles Commission SEM
VAR(SYSl)
VAR (SYS2)
VAR (SYS2+SYS3).
3 Sims actually found covtef) and
Bo such that$0' coviej)(i)0'Y] =
cov(ej), where the right- hand side
is the estimated covariance matrix
of reduced-form (VAR) errors.
Numerically, this decomposition can
be effected by applying a Choleski
decomposition to the right-hand
side. We feel that this description
of the estimator obscures the fact
that a simultaneous equation sys
tem has been assumed recursive, a
point emphasized by Cooley and
LeRoy (1985) in their critique of
Sims' work.
x
x
x
x
Bj (/ * 1)
x
1995
cov(e,’)
x
x
x
x
in macroeconometric work. Having decided
that no elements in Bj ( j & 1) could be restricted,
that is, all lagged values appear in every equa
tion, Sims was forced to adopt two other
assumptions to estimate B0.
First, he proposed that the structural
errors e f have a diagonal covariance matrix,
that is, they were uncorrelated, so that a
money-supply shock could be regarded as
independent of a money-demand shock.
Second, he chose to make B0 lower triangular.
Together, these assumptions produced a Wold
causal ordering, and that terminology is one
frequently used in the literature. Thus, the
ordering {m, p, y, r } means that m, is deter
mined; mt depends only on lagged values
of m„ pn y, and r,. The next variable in the
ordering depends on contemporaneous values
of the previous variables in the ordering and
lagged values of itself and the remaining
variables; for example, p, depends on m, and
lagged values of p„ y, and rt. An alternative
way of expressing the implications of these
assumptions is that the simultaneous system
in equation 10 has been transformed to one
that is recursive, making OLS the appropriate
estimator of the unknown parameters in B0.3
It is rather unclear why this set of assump
tions is viewed as any more credible than
those proposed by the Cowles Commission.
Indeed, if Sims’ assumptions are invalid,
inconsistent estimates of the contemporaneous
impact of the variables will result, ju st as they
would be obtained if the exclusion restric
tions adopted by the Cowles Commission
were incorrect.
One important difference to the Cowles
Commission framework is that the latter
generally works with over-identified systems,
that is, more restrictions were placed upon
N K OF ST . L O U I S
36
REVIEW
MAY/JUNE
—Bp). Suppose that one of the long-run
multipliers is zero, say the (i, j ) ’th. Then
[adjCBg—B, —. . . —Bp)]y=0 and this imposes
some restrictions between the parameters in
B0 and those in B 1(. . . ,Bp. To illustrate the
impact of this, consider estimating the first
equations
rate, and cp is the log of the commodity price
index. The system is taken to be recursive
except for money demand and supply which
have the form
(15)
m t = a l + a 2rt + a i p t
+ a ^ y l + Bmz(L )z, + e "
r, =
7
i + y 2mt+ y 3r10,
n
(2 0 )
+ y 4cP (+ Brz (L)z, + e rt ,
(21)
+
^ 1 2 ^ 2 1 - 2 "I" £ l r
Now, the long-run multiplier being zero
will generate a restriction that <p(bn , b ,,1, b n 2,
fc>12' , b n 2) = 0, and we should be able to write
b u ' = <f>(b12, b u \ b 12, b 122) so that the equa
tion reduces to 4
(22)
z lt = bl2z2t +
“*“ ^ 1 2 ^ 2 1 - 1
+ bnZii_2
^ 1 2 ^ 2 1 -2
® lf
This restriction frees up an instrument
for z2l among z,,.,, zlt.2, Z2,-i and z2t.2 since 4>
is known once the other parameters are given.
Consequently, provided the long-run restric
tion actually involves the parameters of interest
(which may not happen as it is [acij(B0—B, —.
. . —Bp) ]y which equals zero), one can esti
mate bn using as instruments z,M, . . . , z2t.2In the liquidity literature, the SYS3 approach
has been applied by Lastrapes and Selgin
(1 9 9 4 ), while Gali (1 9 9 2 ) uses ideas from
both the SYS2 and SYS3 approaches.
As is evident from the proceeding dis
cussion, there have been many proposals
about how to estimate the parameters of the
simultaneous system. In all instances, certain
moment conditions are used, and so the esti
mators can be given instrumental variable
(IV) interpretations, in which pre-determined
variables in the system are used as instruments.
In the Cowles approach, it is necessary that
the pre-determined variables excluded from
an equation be uncorrelated with the equa
tion’s error term while, in the recursive systems
approach, the structural equation errors need
to be uncorrelated with one another as well
as any right-hand side endogenous variables.
If this was a traditional system, b 12 and h2l
are not identifiable. However, if one imposes
the restriction that E ( e u£ 2t) = 0, one of them
is estimable. Now, let us consider the long-run
response of zlt to e 2l, which is (b l2+ b 12)/
{ ( l - b n X l - b 22) - ( b 12+ b n X b 21+ b 21) ) . U
this response is set to zero, then b 12= ~ b n
and equation 16 becomes
2 lt = b 12( z 2t —Z2t_l ) + b n z lt_1+ £u,
and so b l2 can be estimated consistently by
using z 2t.i as an instrument for Az2l. Hence,
this procedure in SVAR work is identical to
the long-recognized possibility of estimating
B0 by imposing restrictions (other than
exclusion ones) upon the parameters of a
simultaneous equations system.
The argument generalizes to a system of
the form
+ . .. + Bpz t_p + £t
in the following way. Let the long-run multi
pliers of a change in z, to £, be (B0—B, —. . .
—B p )1= adj(B 0—B 1—. . . - B p)/det(B0—B, —. . .
1=1 fe= l
z 1( = b12z 2t + buz lt_, + bnz u_2
~t~bl2Z2l- l
(17) Z2, — b 2iZu + ^21^11-1 "*■^22^2t-l+ ^2t
B0z t =
n
simplified by setting n = 2, p = 2 to get
(16) z 1( —b l2z 2, + bllz Il_i + b l2Z2,-i + £u
(19)
p
Z |, = y,bu,Z|,t+ ^ ' . ' y ,.b|i,Zia_i + c 1,,
h=2
respectively.
An alternative way for reducing the
number of unknown parameters in a SVAR is
to impose restrictions between the elements
of B0 and B( (j 3=1), a strategy we will refer
to as SYS3. These constraints arise from the
belief that certain multipliers in the system
have known long-run values. Shapiro and
Watson (1988) provide a general treatment of
re-parameterizations for studying models that
have the SYS3 nature, and they show that such
strategies free up some of the elements in
2i.j(j=® 1) to be used as instruments. To illus
trate this, consider the simple bivariate system
(18)
199
37
4 Parameters from the equation for
z2,will also appear in the restriction.
HtVIEN
MAY/JUNI
5 Some have attempted to control for
simultaneity by choosing data peri
ods and intervals in which m,can
be reasonably regarded as predeter
mined, for example, by using
weekly data in the logged reserve
accounting regime— see for exam
ple, Cochrane (1989).
W hen the number of unknown parameters
equals the number of moment conditions, as
in a recursive VAR, it is impossible to test the
validity of such restrictions, and it becomes
simply an act of faith that they are valid. If
the assumption is wrong, then it would
be expected that there will be biases in the
estimates of the parameters. For example,
observe that a liquidity effect may require
that the demand-interest elasticity be negative.
In the event that a liquidity effect is not found,
one might ask: W hat is problematic about
the implicit demand function being estimated?
Given that we are concerned with a simultaneous-equation system, the most likely
explanation would be bias due to the simul
taneity. For example, if the system is ordered
recursively as {m, p, y, r }, but m, is not pre
determined for r„ then the OLS estimator of
the contemporaneous liquidity effect will be
biased away from a true negative value and
might even produce a positive value. Hence,
it is hard to know whether any lack of evi
dence for a liquidity effect is due to the actual
state of the world or estimation/identification
difficulties.5 Accordingly, it seems that there
is always going to be an element of indeter
minacy in a study of the existence of the
liquidity effect.
Another estimation issue concerns the
usefulness o f the available instruments. In
particular, it is important that the instruments
are correlated with their respective endoge
nous variables. W hen instruments X! are in
a structural equation already, it is the correla
tion of the complete set of instruments X with
the endogenous variable, after partialling out
X u that is important. It may be that the raw
correlation is high while the partial correla
tion is very low. Studies by Staiger and Stock
(1 9 9 3 ), Pagan and Jung (1993), Kocherlakota
(1990) and Nelson and Startz (1 9 9 0 ) have
all concluded that there can be large biases
in the estimators of the parameters attached
to the endogenous variables if the partial
instrument correlation is weak, for example,
< 0.2. Thus, it is important that this quanti
ty be examined. In the simple SYS3 example
constructed above, the correlation between
the instrument and regressor is determined
by the magnitude of the autocorrelation in
Z2t- As the autoregressive root tends to unity,
199
one would get worse estimates of b u . This
problem has been studied by Sarte (1 994)
and, in the context of the liquidity effect,
Pagan and Robertson (1995).
E X A M IN IN G TH E S TU D IES
Table 2 presents a summary of some of
the evidence on the liquidity effect for stud
ies using monthly or quarterly data. Perhaps
the most striking characteristic is the fact
that early failure to detect a liquidity effect
(largely based on single-equation methods)
has been replaced by a conclusion that there
generally is a liquidity effect when inferences
are based on systems methods. Although
this is a comforting outcome, the transition
needs to be analyzed carefully, to ensure that
the observed relation is in fact robust to any
assumptions made in order to identify it.
Four concerns can be distinguished, involving
how sensitive the conclusion is to:
1. different definitions of the monetary
stance;
2. different models;
3. different estimation procedures and
restrictions; and
4. different data samples.
In what follows, we examine these
issues using monthly data. Descriptions of
the data are contained in the appendix. The
money, price and output series are measured
in logs and are seasonally adjusted. Three
sample periods have been chosen. The longest,
from 1959:1-1993:12, was fitted with a 14thorder VAR, while the shortest runs from
1 9 8 2 :1 2 -1 9 9 3 :1 2 and has a sixth-order VAR.
An intermediate period of 1 974:1-1993:12
with an eighth-order VAR was selected to
roughly coincide with the period of flexible
exchange rates. These choices also reflect
those adopted in the literature. Equationby-equation and system diagnostic tests (not
reported) indicated the absence of residual
autocorrelation, but found autoregressive
conditional heteroskedasticity (ARCH) and
some non-normality, particularly in the
money and interest rate equation residuals
estimated over longer sample periods. The
ARCH effect was less evident in models
using post-1982 data.
38
MAY/JUNK
1995
Ta b le 2
Sum m ary of Selected Studies on the Liquidity Effect
Sample
Freq
Money Variables
Interest
Rate
Mishkin (1982)
Reichenstein (1987)
Thornton (1988)
Leeper & Gordon (1992)
1959:1-1976:4
1965:01-1983:03
1958:08 1987:06
1954:07-1990:12
Q
M
M
M
AM 2,AM 1
R6-R3
Sims (1992)
Eichenbaum (1992)
Christiano & Eichenbaum (1992)
Eichenbaum & Evans (1992)
Christiano, Eichenbaum & Evans (1994)
Gali (1992)
Lastrapes & Selgin (1994}
Gordon Sleeper (1994)
1958:04-1991:02
1965:01-1990:01
1959:01-1990:03
1974:01-1990:05
1960:1-1992:3
1955:1-1987:3
1959:01-1993:12
1982:12-1992:04
M
M
M/Q
M
Q
Q
M
M
Author
Different Money Variables
a
R3
A P ,A Y
ap , a y , u
A M I, AMO, ANBR
AR3
A P ,A Y
AM2, A M I, AMO
FF
P,Y
Ml
FF
P,Y,FR,CP
M l ,M 0 ,m
FF
P,Y
M l, MO, m
FF
P,Y
FF
Y,P,RF, RER
NBR
FF
P, Y, CP, TR
AM I
R3
A P ,Y
MO, M l, M2
R3
Y ,M - P
M2, TR
R l, FF
RIO, P,
Y,U,CP
NBRX
Model lype
SING1
SING1
SING1
SINGl/2,
SYSl
SYSl
SYSl
SYSl
SYSl
SYSl
SYS2/SYS3
SYS3
SYS2
to the money-supply equation are identified
with monetary policy. For example, one
might assume an ordering such that money
is predetermined for the interest rate (and
possibly other variables as well) and use the
error from the money equation and the esti
mated dynamics to derive the impulse
responses of the interest rate. Ignoring the
dynamics, this amounts to assuming that the
supply function of money is perfectly inelas
tic with respect to the interest rate. A differ
ent strategy, employed by Sims (1 9 9 2 ), and
Bernanke and Blinder (1 9 9 2 ), is to order the
VAR such that the interest rate is predeter
mined for money and to treat shocks to the
interest rate equation as the monetary policy
indicator. This yields an interest rate or
R-rule interpretation, since, ignoring the
dynamics, this is equivalent to assuming that
the supply function is perfectly elastic with
respect to interest rates. Empirically, defin
ing money as MO or M l does not result in
a liquidity effect in a recursive VAR under
M-rule interpretations, while using NBR or
NBRX does yield a liquidity effect for either
M-rule or R-rule identification schemes. For
example, Figure 1 presents the implied interest
rate responses to a one-unit monetary expan
sion under an M-rule (an increase in £?) for
various measures of money, and two alternative
A crucial question is whether changing
the definition of money has been important.
Here, it would seem as if the answer is yes.
The consensus from Table 2 is that for sin
gle-equation and recursive models, defining
money as MO or M l does not result in a liq
uidity effect, while finer measures such as
nonborrowed reserves, NBR, or the ratio of
nonborrowed to total reserves, NBRX, do.
Nevertheless, one should dig a little deeper
into the issue of measuring monetary action.
Remember from equations 1 and 2 that we
are concerned with the response of interest
rates to a shift in the intercept of the money
supply equation, and this was measured by
computing the impulse response of interest
rates to the money supply structural errors.
Hence, if one could identify a series corre
sponding to shifts in the intercept over time,
that would constitute the basis for an appro
priate way to measure the monetary stance.
Such series have been constructed by Romer
and Romer (1989) and Boschen and Mills
(1 9 9 3). Eichenbaum and Evans (19 9 2 ) have
shown that there is a strong liquidity effect
when the first of these measures is used.
For recursive models, a money-supply
or M-rule interpretation implies that shocks
AM I
Main Other
Variables
NK OF ST. L OUIS
39
Max
Lags Finding
4
no
4
no
6 no/yes
36/18 no
14
yes
14 no/yes
14/5 no/yes
6
yes
4
yes
4
yes
14
yes
6
yes
REVIEW
MAY/JUNE
M l, since an expansionary monetary action
(in his case, an R-rule contraction in £ [) led
to a persistent fall in the price level).
Eichenbaum’s proposed solution to this was
to replace M l or M0 with NBR, and to place
P and Y prior to money and interest rates in
the ordering, so that the Federal Reserve’s
M-rule responds contemporaneously to price
and output variables, but not interest rates.
Eichenbaum reports a small positive response
to expansionary monetary policy in this case.
Earlier, Thornton (1988), in a single-equation
analysis, observed that NBR was the only
measure of money which displayed evidence
of a liquidity effect. Thornton’s conclusion
has been reiterated by Christiano and
Eichenbaum (1 992) in a systems context
(see Figure 1). Subsequently, Strongin
(1 9 9 2 ) has suggested that the ratio of NBR
to total reserves, TR, denoted NBRX, is the
best monetary measure, and Eichenbaum
and Evans (1 9 9 2 ) have adopted NBRX in
their work on exchange rates.6
Figure 2 presents the impulse responses
of P, Y and F F to monetary shocks in VARs
ordered as {Y, P, NBR, FF, TR), {Y, P, NBRX,
FF} and {Y, P, NBR, F F ), respectively. In
contrast to the finding in Eichenbaum (1992),
it is apparent that the price puzzle is still pre
sent regardless of which monetary measure is
adopted, although in all cases the estimated
responses are relatively small.7 The difference
between these and the Eichenbaum results
can be explained by noting that Eichenbaum
used a slightly different sample period
(1 9 6 5 :0 1 -1 9 9 0 :0 1 ). Computing impulse
responses from a VAR fit to this sub-sample
does produce impulse responses very similar
to those he reports. Hence, it seems as if the
estimated price-impulse responses are unstable,
at least if NBR or NBRX are used to measure
monetary actions. We examine the issues of
model stability and the precision of the point
estimates in more detail further in this article.
Perhaps the most controversial issue with
the use of nonborrowed reserves is whether
it constitutes an effective way o f measuring
monetary policy. The variable NBRX is very
highly negatively correlated with borrowed
reserves BR (-0 .8 2 over the period 1959:011 9 9 3:12), raising the question about how the
latter should be treated. Suppose that total
F ig u re 1 a , b
M -»F F , {M ,Y ,P ,FF} M od e l,
1959:01 -1 9 9 3 : 1 2
Basis points
80
60
40
20
0
-20
-4 0
-6 0 W
........... ....................................................
2 6 10 14 18 22 26 30 34 38 42 46 50 54 58
Periods
M — FF, {Y ,P ,M ,FF} M od el,
1959:01-1993:12
Basis points
Periods
6 Strongin (1992) actually used the
ratio of NBR to TR r
7 These results are also quite robust
to reversing the ordering of Yand P
at the top of the recursion.
orderings of a four-variable VAR of m, r, y and
p, where r is measured by the federal funds
rate, FF, p is measured by the log of the con
sumer price index, P, and y is measured by
the log of the industrial production index, Y.
The VAR is fit to the sample 1959:01-1993:12,
and the recursive models parallel some of those
reported in Christiano and Eichenbaum (1992).
It is not sufficient, however, to simply
concentrate upon the impulse response func
tions relating to interest rates and money, as
it is possible that a model producing a plau
sible liquidity effect also creates implausible
effects of monetary policy upon other vari
ables in the system. This was Eichenbaum’s
(1992) objection to Sims’ work. Sims pointed
out that there was a “price puzzle” generated
from a simple four-variable model based on
1995
NK OF ST. LOUIS
40
REVIEW
MAY/JUNE
reserves, TR = NBR + BR, showed no varia
tion. Then, if BR has a positive relation to
FF, NBR must be negatively related to FF A
model of this sort was constructed by Gilles
and others (1993). They effectively fix the
total demand for reserves by making it
depend upon real factors exogenous to the
monetary sector, and then add a “discount
window” function in which the supply of
borrowed reserves is a positive function of
FF Hence, they concluded that the observed
negative relation between NBR and FF sim
ply reflects the way that the Federal Reserve
has operated the discount window. The import
of this model is not entirely clear because it
makes the supply of BR a function of FF,
whereas the data indicates that the relation is
between BR and the spread between the
Federal funds and the discount rate, RD—
that is, SPRD = F F —RD (see Mishkin, 1992),
and therefore, BR is not a function of FF
alone. Indeed, statistically, it would not make
sense to relate BR solely to FF, as the latter is
best described as an integrated process while
the former is not. This is evidenced by aug
mented Dickey-Fuller (with 12 lags) tests of
— 1.88 (FF) and —3.47(B R ), as compared to
a 5 percent critical value of —2.86.
W hat is in dispute here is the degree of
substitutability of NBR and BR. W ith zero
substitutability, NBR would appear to sum
marize monetary policy quite well. But if
there was perfect substitutability, total
reserves would be a better measure, and,
with the exception of the study by Gordon
and Leeper (1 9 9 4 ), this does not seem to
result in a liquidity effect, all responses being
quite similar to those from MO or M l. An
attempt to allow for non-zero substitutability
might be to incorporate demand and supply
functions for both NBR and BR into the
analysis. A variant of this idea would be to
include both NBR and total reserves (TR) in
the VAR, and this has been done by
Christiano and others (1994). Doing so pro
duces more reasonable price and income
responses than the {Y, P, NBR, FF} model,
and broadly similar responses to those from
the {Y, P, NBRX, FF) model (Figures 2a and
2 b ), although the price effect is still negative
for a long period of time. There is also some
increase in the magnitude of the liquidity
1995
F ig u re 2 a -c
M -^ P .3 M odels,
1959:01-1993:12
Log
Periods
M -*-Y.3 M odels,
1959:01-1993:12
Log
Periods
M -^ F F .3 M odels,
1959:01-1993:12
Log
effect, and it is less persistent than for the
model {Y, P, NBR, FF) (Figure 2 c).8
The result that neither of the NBRX or
NBR/TR formulations are capable of com
pletely eliminating the price puzzle is consis
tent with the view of Sims (1 9 9 2 ) that the
main source of the price puzzle is the
41
8 To avoid the potential problems
associated with using NBRX in the
recursive VAR formulations exam
ined in later sections, we will
henceforth adopt the strategy of
including both NBR and TR in the
recursive models.
MAY/JUNE
system. Too small a set of variables implies
misspecified relations, which can affect esti
mates of both contemporaneous and dynamic
responses. Because there is a cost to making
the list of variables too large, it is imperative
that theoretical ideas and past research are
used to indicate what variables are likely to
be of major importance. For example, Sims
(1 9 9 2 ) and Christiano and others (1 994)
extend the NBR/TR formulation to include a
measure of commodity prices. In particular,
they consider the M-rule ordering (Y, P, CP,
NBR, FF, TR|, where CP is a commodity
price index. Thus, output, the general price
level and commodity prices are taken as pre
determined in setting policy. Estimating
their model using the monthly data, we find
that the Y response is initially negative, but
then persistently positive after a few months,
while the P responses are now persistently
positive (Figure 3a) and the liquidity effect
lasts approximately seven months (Figure
3b). It seems that including additional vari
ables in the policy setting rule goes some
way to eliminating the anomalous price
effects that were obtained using simpler
models.
Another possible model variation is to
allow for interaction with the foreign sector.
Open economy models, for example,
M cKibbin and Sachs (1 9 9 1 ), emphasize the
determinants of the size of the liquidity
effect in the following quotation:
Figu re 3 a , b
N B R ->Y, N B R — P. CP M od el,
1959:01-1993:12
Log
Periods
N BR -^FF, CP M od e l,
1959:01-1993:12
Basis points
Periods
absence of some pre-determined inflation
indicator variable in the Fed’s policy response
function. This implies that the model should
be extended to include variables other than
ju st money, interest rates, output and the
general price level. This line of argument is
taken up in the next sub-section, which
deals with the issue of using alternative
model formulations.
“If the effect of the exchange rate on
domestic demand is large (through the
effect on the trade balance), and if the
effect of domestic demand on money
demand is large (through the income
elasticity of demand), and if the home
currency depreciation causes a rapid rise
in domestic prices, then it can be shown
that home nominal interest rates will tend
to rise after the money expansion ... But
if one or all of these three channels are
weak, then domestic nominal interest
rates will tend to fall after the money
expansion...
Model Variation
One explanation for the range of conclu
sions regarding the liquidity effect arises
from the non-uniqueness of models. We
have already alluded to this when discussing
recursive versus non-recursive systems, and
even within a given causal framework mod
els can vary, as reflected in the ordering or
set of variables taken as constituting the
1995
Using the MSG model, their simulations
show a strong liquidity effect for the United
States but a weak one for Japan, even though
42
REVIEW
MAY/JUNE
the interest elasticity of demand in both
countries is assumed to be the same.
It is clear from such studies that there is
a need to allow for an exchange rate er offset.
Introducing an exchange rate also demands
the addition of a foreign interest rate ry , to
allow for the possibility of uncovered interest
parity, that is, e = r — . W ithin a recursive
system, rf would need to appear as the first
variable and e will appear after r. Eichenbaum
and Evans (1992) and Sims (1992) contain
results which suggest that the conclusions
reached with systems excluding er and
remain valid, although the magnitude of any
effects differ. Using the trade-weighted
exchange rate, ER, a weighted foreign interest
rate series, RF, and an ordering {RF, Y, P, CP,
NBR, FF, ER, TR], later referred to as the
exchange rate model (ER), we find that the
liquidity effect is reduced slightly from that
observed for the “commodity price” (CP)
formulation {Y, P, CP, NBR, FF, TR) (Figure
4a). There are greater qualitative differences
for the price responses. Figure 4b shows
these for the CP and ER models. Unlike
the situation for the full sample, there is a
perverse price response with the CP model
that is largely corrected by the ER model,
pointing to the fact that the long-run responses
can be very different as models change, even
though the short-run responses are similar.
In contrast, the estimated short- and longrun responses of Y are similar in both the CP
and ER models, as shown in Figure 4c.
The question of how to choose between
alternative models is a vexed one. As men
tioned previously, most analyses seem to
concentrate upon how closely multipliers
correspond to prior conceptions. This seems
to be a restrictive viewpoint. Structural rela
tions have been estimated in getting the mul
tipliers and it seems appropriate that one
should examine how plausible the estimates
of these parameters are. In particular, the
nature of the liquidity effect directs us to the
demand for money function, and we would
expect that it should feature negative interest,
positive income and (probably) positively
signed price elasticities. A full set of struc
tural coefficient estimates for the CP and ER
models is presented below. The CP model
results for the periods 1959:01-1993:12 and
1995
F ig u re 4 a -c
NBR+FF, 1 9 7 4 : 0 1 - 1 9 9 3 : 1 2
Basis points
Periods
N BR ^P, 1 9 7 4 :0 1 -1 9 9 3 :1 2
Log
Periods
N BR ^Y, 1974:01-1993:12
Log
Periods
1974:1-1993:12 are presented in equations 23
and 24, respectively, and the ER model results
for the sub-period (1 9 8 2 :1 2 -1 9 9 3 :1 2 ) are in
equation 25. Note that because money is
ordered immediately prior to the interest rate
in the CP and ER model, the initial impulse
response of the interest rate to money shocks
F EDERAL RESERVE B A N K OF ST. L OU IS
43
REVIEW
MAY /JUNE
is simply given by the magnitude of the interest
elasticity of demand for money. This follows
directly from equation 7 as the stated recursive
structure has dpt !9 e \= d y j de\ = 0. More
generally, however, it is clear that it would be
possible for the liquidity effect to “exist” and
yet for all of the parameter estimates in the
demand function to be incorrectly signed.
(23)
P
CP
NBR
FF
(24)
P
CP
NBR
FF
(25)
P = .002R F + .055Y
CP = -.0 1 2 R F + .2 1 2 Y - .1 3 0 P
NBR = —.008RF - .492Y
+ .171P + .049C P
FF = .4 0 1 R F + 1 1 .7 3 2 Y + 7 .3 3 2 P
—2.052C P -1 4 .9 5 2 N B R
TR = .0 0 5 R F + .0 3 7 Y + .4 1 9 2 P
—.006C P+.410N BR
+ .0 05F F + .012E R
1995
Perhaps the main use of the idea that
one should think of the issue in structural
terms is that it forces one to think carefully
about the complete specification of the sys
tem, and such considerations suggest that
there may be problems in modeling the data
with particular choices of the set of variables.
For example, suppose M2 is used as the mea
sure of money. Then, for such a broad mea
sure of money, one really needs to have
another interest rate in the system to capture
the fact that a large component of the assets
making up M2 are interest-bearing. If the
dependent variable in the (inverted) demand
for money function is taken to be the threemonth T-bill rate, R3, then we might take the
federal funds rate as proxying the rate of
return on M2 assets. For a VAR ordered as
(P, Y, M2, FF, R31 we find that the estimated
implied demand for money function appears
relatively stable based on a CUSUM test, and
a liquidity effect is observed. But the demand
relation is quite unstable if FF and/or its lags
are omitted from the VAR. Hence, a VAR
only in the variables {P, Y, M2, R3) would
appear to be a poor choice. More generally,
given the large body of literature that has
evolved pointing to the instability of U.S.
money demand functions, the fact that esti
mated parameters of a demand for money
function are fundamental to any conclusion
regarding the liquidity effect has to be cause
for concern. Even if the menu o f variables
seems complete, it still may be that the rela
tionship between them is unstable, or the
use of linear models inappropriate, for some
measures of money and interest rates, and
for some sample periods.
= .0186Y
= .4 3 P + .2 5 Y
= - . 0 0 7 CP —.39P - . 2 5 Y
= 8 .0 8 Y —5.42P - 2 . 6 2 CP
—12.41NBR
TR = - .0 2 Y + .3 6 C + 0 1 2 C P
+ .423N B R + .006FF
= .0 3 8 Y
= —.42P4-.40Y
= .04CP —.29P - .4 1 Y
= 11.73Y + 19.20P - 1 .7 3 C P
-1 7 .0 7 N B R
TR = .0 0 5 Y + .6 P —.009CP
+ .41N B R + .006FF
W ith the possible exception of the P
variable in the demand for money function
(FF) of the CP model estimated over the
period 1959:01-1993:12, the estimated struc
tural relations are what would be expected,
with prices responding in a procyclical way,
monetary policy (in terms of real NBR move
ments) reacting negatively to expansions in
prices and output, and a demand for money
function that has positive income and nega
tive interest rate effects. Interpretation of the
equation for TR is harder, but it is interesting
in that it shows that changes in NBR are only
partially reflected in TR, which can be inter
preted as indicating that there is substi
tutability between NBR and BR.
Different Estimation Methods
and Restrictions
How m uch do systems methods con
tribute to the analysis of the liquidity effect?
Potentially, a good deal. As previously men
tioned in the discussion on single-equation
estimation procedures, the estimates made of
the monetary stance are ideally the structural
rather than reduced-form errors, and so a
regression of r, upon a distributed lag of
these values could produce quite different
results. Only if the monetary policy variable
44
RfVltW
MAY/JUNE
1995
(26)
Table 3
m, =
+«4 y ,
Im p u lse Responses o f FF to NBR
{NBR, P, \ F F ]
+ a 2rt + a 3p t
rt
{P, \ NBR, FF]
+ B mz( L ) z t + e™
= 7 i + y 2m( + /3riot
+ y ,c p t + Brz(L )z t + e rt ,
coS
f|S
<25
<3s
<45
C5S
-1 5 .3 9
-2 5 .7 0
-2 5 .7 9
-2 4 .5 8
-2 3 .5 6
-2 5 .6 2
-1 4 .5 6
-2 3 .0 3
-2 0 .3 2
-1 8 .4 8
-1 8 .2 8
-1 9 .9 2
respectively, with E (e"e^ ) = 0. The rest of
the system is taken to be recursive, ordered
as {u, y, p, r10, cp). Because these variables
are predetermined for m, and r„ X, = {1, u„ y„
p„ rlgt, cpt, z t.j,j > 0} provide a valid set of
instrumental variables for r, in the moneydemand equation, and for m„ in the moneysupply equation. They estimate equation 26
subject to E(£™£() = 0 via FIM L, using a sixlag VAR, and monthly data from 1982:12 to
1992:04. Pagan and Robertson (1995)
extend the sample period to 1993:12, giving
T = 127 observations, and focus on the
results for m - TR and r = FF.
The existence of the liquidity effect
hinges upon the signs and magnitudes of
both the demand and the supply elasticity,
and there are a number of issues in this
regard. First, the precision of estimation of
the demand elasticity stems in part from the
use of the residual of the supply equation as
an additional instrument, and the structural
residuals are only valid instruments if
E(s'"ert ) = 0. In this instance, the assump
tion may be checked as the system is over
identified— that is, there are more instru
ments among X, than are needed to estimate
the parameters. Using the parameter esti
mates from doing IV with X, only, that is,
excluding the supply-equation residuals,
reveals that the correlation between the
demand- and supply-equation residuals is
—0.39, which is significantly different from
zero (if money is measured by M2, the corre
lation becomes —0 .7 9 ). Also, the excess
instruments in X ( contribute little to the
prediction of TR in the supply equation.
The F-test of the hypothesis that they do
not enter the first-stage TR regression yields
a value of only 1.49, compared to a 10 per
cent critical value of 2.18. The presence of
weak instruments means that the elasticity
estimates may be severely biased. Finally,
as Gordon and Leeper acknowledge, R10 is
probably not a valid instrument for FF in
the demand equation.
is determined solely by past quantities will
the two coincide. In terms of the recursive
VAR, a single-equation approach corresponds
to a case in which the monetary variable is
ordered first, whereas the systems approach
generally has money appearing later in the
ordering. However, it turns out that the
conclusions reached concerning the liquidity
effect do not differ greatly because of this
modification, as evidenced by the close cor
respondence of the distributed lag coeffi
cients from the regression of F F against 36
lags of
in Table 3, in which £"' is alterna
tively measured as the structural errors from
the two orderings {NBR, P, Y, FF) and (P, Y,
NBR, FF}. Apparently, the conclusions
reached by Thornton (1988) in his single
equation study are not changed by purging
the monetary variable of any contemporane
ous effects.9
In the discussion in the first section, it
was suggested that the estimation issues
relate to how to consistently estimate, inter
alia, the parameters of both the demand and
supply of money functions. Working with
recursive systems, we assume the interest
rate is not to enter into one of these curves,
thereby sidestepping the simultaneity issue.
If one wishes to estimate equations 8 and 9
with no zero restrictions on either a 2 or y2, it
is necessary to proceed in some other way.
Gordon and Leeper (1994) and Gali (1992)
provide examples of how this might be done.
For example, Gordon and Leeper estimate
the money-supply disturbance from a struc
tural model of seven variables, z = Im, r, u, y,
p, r10, cp]', in which the money demand and
supply block has the form
45
5 Thornton actually uses differences
rather than levels of the variables in
his regression and adds in a lagged
dependent variable. The latter is
not a good idea. Equation 12 sug
gests that one should be oble to
omit variables other than the
money shocks without producing
biases, although any omitted vari
ables will cause serial correlation
that would be important for stan
dard error computations. If one
had a model in which the true rela
tion is of the form ff= y m ,+ e |
and m,= pmH + e ” then r,=
pr,., + e f + e ; - p e ;.,,an d a
regression off, on and lagged
values e ” may well indicate that
only the first response is non-zero.
One actually sees this effect if
logged is added to the regres
sions for Table 3.
MAY/JUNE
1995
Ta b le 4
Estimates of G ordon and Leeper Demand and Supply M odel
FIML (1982:12-1992:04)
FF
FIML (1982:12-1993:12)
IV (1982:12-1993:12)
Demand
Supply
Demand
Supply
Demand
Supply
(TR)
IFF)
(TR)
(FF)
(TR)
(FF)
-.0 2 8
(.014)
-.0 0 9 9
-.0 2 6
(.013)
(.010)
22.578
(9.987)
30.28
(10.95)
TR
23.529
(13.629)
P
.958
(.752)
1.007
(.603)
.796
(.486)
Y
.495
(.315)
.650
(.311)
.353
(.237)
CP
varfe,)
(. 10 1 )
.341
(.099)
(. 010 )
1.893
(1.326)
1.879
(1.410)
1.890
(1.418)
.456
RIO
6.61e-5
5.04e-2
6.73e-5
4.98e-2
.342
-.3 8 5
corrdje^,)
over-id test
p-value
5.31 e-2
4.62e-5
.09
.04
.41
.06
Gordon and Leeper estimate the model using a six-lag VAR and data from 1982:12 to 1992:04 (T = 107). We use data from 1982:12 to 1993:12
(T = 127). Asymptotic standard errors are reported in parentheses below the point estimates. Note that the reported estimates ore for the contem
poraneous coefficients of the supply and demand functions. The dynamics are left unrestricted, and are partialled out by fitting the VAR.
10 Adding the IV supply-equation resid
ual as an additional instrument for
FF in the demand equation yields
almost exactly the FIML estimate of
the demand-equation parameters.
11 To estimate the system, we follow
LS and rewrite it to involve real
money in place of pt, as that
enables us to impose zero restric
tions upon the log distributions on
ra, in each equation. At the end,
we convert back to the system
involving p„ m„ r,and yt.
Comparing the IV and FIM L results
reported in Table 4, we see there is a
close correspondence between the IV and
FIM L estimates of the supply equation.
In contrast, the IV demand elasticity esti
mate is much larger than the corresponding
FIM L estimate ( —0.01 vs. —0.026) and
is no longer significantly negative.10 A
negative correlation between the structural
errors would be expected to produce a
negative bias in the FIML estimator of
the demand elasticity, and this leads to a
smaller magnitude for the liquidity effect
for a given supply elasticity estimate. The
inconsistency will be proportional to the
actual correlation between
and e\ when
E (X 'e ™) = 0. Against this, the supply elas
ticity estimate itself may be biased due to
weak instruments. The net outcome of
these two effects is indeterminate but does
cast some doubt on whether the liquidity
effect uncovered by Gordon and Leeper
is a real one.
Another approach to estimating equations
8 and 9 that eschews recursive assumptions
is to impose some long-run restrictions upon
the impact of monetary shocks. Lastrapes
and Selgin (1 9 9 4 ) and Gali (1 9 9 2 ) impose a
variety of these. Lastrapes and Selgin begin
by postulating that a unit shock in the money
supply causes prices to rise by a unit in the
long run, that is, real money balances do not
change, while there is a zero long-run impact
on output and interest rates. As explained
in the preceding section, when discussing
the SYS3 procedure, such restrictions free
up instruments that can be used to estimate
the elements of B0. Taking the system to
be estimated as (where all lagged values
are suppressed)
FEDERAL RESERVE B A N K OF ST. L OU I S
46
REVIEW
MAY/JUNE
(27)
the three long-run restrictions on the impact
of money supply shocks on prices, output
and interest rates hold, as well as analogous
ones involving money demand and aggre
gate demand shocks.
Ap, = -b°n Ay, - b°3Am, - b°,Ar, + e u
Ay, = ~ b 2l AP, ~ b 23A m t - K Art + e 2,
Am, = -b°31APt - b 32Ayt - b°4Ar, + e 3t
Ar, = - b 41Ap, - b ° 2Ay, - b ° 3Am, + e 4(
imposition of the long-run restrictions on
each of the equations for pt, y, and r, enables
the estimation of three of the b| ."
Before further analysis, one has to consider
why the system above is measured in differ
ences, whereas most of the systems described
previously are in levels. Lastrapes and Selgin
(1 9 9 4) argue that the variables m„ p„ r, and
y t are integrated but not cointegrated. If
equation 27 was written in levels, the error
terms must be Integrated of order one 1(1);
otherwise, the equations would represent
cointegrating relations among the variables.
Hence, it is appropriate to transform all the
variables by differencing. Suppose, instead,
that one proceeded to impose the long-run
restrictions upon the levels model. To make
the analysis simple, focus on the equation for
output and assume that the only right-hand
side variables are mt and mM. Then, as pre
viously explained in the second section, one
would be using mM as an instrument for Am,
when the equation is re-parameterized to have
Am, and m,_, as the two regressors (mM is elim
inated because its coefficient is the long-run
response of zero, leaving the only regressor as
Am,). This estimator is b23=b23—(T '£m MAm,)1
( T lZm,_le2i)- If m, is 1(1), both the numerator
and denominator are asymptotically random
variables, and the instrumental variables esti
mator converges asymptotically to a random
variable, failing to even be consistent. The use
of differenced variables obviates this problem
as the new re-parameterized equation features
A2m, as regressor and Am,.! as instrument, and
T 'XAmMA2mt will converge to a constant.
Now, let us consider the various estimates
that might be made of the initial impulse
response of r, to shocks in m,. To estimate this,
we need to be able to form B^1. Accordingly,
six restrictions need to be placed upon the
system to identify the elements in B0. It is
useful to draw these from one of the following
five alternatives:
2. The three long-run money supply shock re
strictions hold, along with b°l2 = b°4 = b% = 0.
3. Long-run restrictions on the effect of money
supply shocks on prices and output hold
(but not on interest rates), along with
1,0
_ uo
d 12 _
_ 1)0
_ LO
n
d 14 ~ D 2 4 ~ ° 3 4 — u -
4. Only the long-run restriction on the effect
of money supply shocks on prices holds,
along with b°2 = b°4 = b% = b 3i = b23 = 0.
5. There are no long-run restrictions, and B0
is lower triangular, that is, the system is
recursive.
The first of these is what Lastrapes and
Selgin actually use. Most of their paper specif
ically mentions only three long-run restric
tions, but this fails to identify the magnitude
of the responses, and the quantitative results
they present require the extra long-run restric
tions. As an experiment, we consider other
ways of estimating B0 that impose only the
long-run restrictions emphasized by Lastrapes
and Selgin, allied with various short-run
assumptions. In particular, we build up to a
recursive system [p, y , m, r) by progressively
removing the long-run assumptions. Given
these choices, and with m being base money
and r the three-month T-bill rate, the impact
multipliers are, respectively, —63, —20, —8, 5
and 7, showing that the long-run restrictions
do indeed help to identify a liquidity effect.
The magnitude of the effect is large if six
long-run restrictions are imposed, but if only
the three restrictions Lastrapes and Selgin
discuss are adopted, the magnitude is much
the same as found with simple recursive
systems featuring NBR and FE
Clearly, there are a number of econometric
estimation issues raised by the work with
non-recursive models such as those of Gordon
and Leeper, Lastrapes and Selgin, and Gali,
and some of these are explored in detail in
Pagan and Robertson (1995). For instance, it
is shown there that the instruments implicitly
used by all three studies are very weak, and
this leads to biases in the estimated impulse
1. The matrix of long-run impulse responses,
C ( l), is lower triangular. This implies that
1995
47
REVIEW
MAY/JUNE
199
structural models in the various studies,
most of the empirical models are estimated
using different sample periods. There are a
number of ways of examining the robustness
of results from changing the sample period,
some of which are considered here. First,
the estimates could be sensitive to estimation
over a sub-period. Examining the impulse
responses for the CP and ER models when
estimated only with observations from the
period 1982:1 2 -1 9 9 3 :1 2 , we find that each
model produces small negative initial effects
on interest rates and that the largest negative
effects, after three of four periods, are around
one-third of what was in evidence over the
period 1974:01-1993:12. Compare Figures 4a
and 5a. Moreover, while the price responses
are similar for both models (see Figure 5b), the
income responses are perverse (see Figure 5c).
To understand why the conclusions
drawn from the models fitted over the
1982 :1 2 -1 9 9 3 :1 2 sub-sample are so different,
we might start by examining the underlying
structural relations. As mentioned earlier, in
recursive systems like the ER and CP models,
the initial effect of money shocks on interest
rates requires that one only examine the
interest elasticity of money demand drawing
our attention to the estimated money demand
curves in each period. The im plicit contem
poraneous components of the demand equa
tions corresponding to those in equations 24
and 25 for the 1 9 8 2 :1 2 -1 9 9 3 :1 2 period are
F ig u re 5 a -c
NBR + FF, 1 9 7 4 : 0 1 - 1 9 9 3 : 1 2
Basis points
Periods
N B R -^P , 1 9 8 2 : 1 2 - 1 9 9 3 : 1 2
Log
Periods
N B R -^ Y , 1 9 8 2 : 1 2 - 1 9 9 3 : 1 2
Log
CP: F F = —.14N B R + 12.42Y
+ 18.28P4-2.97C P
ER: FF = .2 6 N B R + 12.12Y + 11.14P
+ .19R F + 3.05C P .
Over the longer period, the interest rate
coefficient was strongly negative so that the
estimated liquidity effect was genuine. In this
shorter sample, the situation is not as clear.
A comparison of the two sets of estimates
points to instability in the money-demand
equation. On the basis of this evidence, one
would have to be skeptical about the presence
of a liquidity effect, although an alternative
interpretation might be that the observations
from the 1982-93 decade are ju st uninforma
tive about the size of the interest rate coeffi
cient, and that a longer series of data has
Periods
response functions, raising the possibility
that the observed magnitudes for the various
responses are partly an artifact of the estima
tion procedures adopted.
Different Data Samples
Compounding the difficulties arising
from the use of different sets of variables and
48
REVIEW
MAT/JUNE
managed to produce more precise estimates
of that parameter.12
To shed further light on this issue, we
estimated the money-demand equation from
the CP model using varying-coefficient tech
niques. Figure 6a presents the recursive esti
mate of the NBR coefficient in the FF equa
tion. W hat is striking in this graph is that
the magnitude of the liquidity effect increased
very sharply after the change of operating
procedures of the Fed in October 1979. In
light of the standard errors, the evidence for
a liquidity effect in pre-1979 data does not
seem very convincing, and there is a sugges
tion that the 1982-93 decade may be closer
to the pre-1979 period in what it says about
liquidity effects. To assess this latter propo
sition, we re-estimated the NBR coefficient,
but now with a moving sample window of
120 months so that the last point estimate
uses data from 1983:12-1993:12. Figure 6b
presents this information. It is very clear
from this graph that 1979-82 is a watershed
period when it comes to empirical work on
the liquidity effect. If it is omitted from the
data, it would be very hard to believe that
the initial impact on interest rates of money
supply movements is not close to zero.13
Given the sensitivity of results to the
sample period, it is desirable to investigate
the uncertainty about the estimates in more
detail. Here we encounter some difficulties.
The presence of (near) unit roots in the data
means that standard asymptotic formulae for
standard errors, based on the assumption that
the random variables are stationary, will be
incorrect and parametric simulation methods
seem to be the best approach to producing
standard errors. Even then, there are problems
in implementing the simulations. One of these
arises from the fact that, over any period
incorporating 1979-82, there is extensive
ARCH in the VAR equations for interest rates
and money. The dependence introduced by
the ARCH errors means that one cannot
assume that the shocks are i.i.d, and, there
fore, simple bootstrapping methods are not
strictly appropriate in this context.14 We have
ignored the effects of ARCH and have deter
mined percentile-based, 90 percent confidence
intervals for the CP model by re-estimating
the impulse responses from 1,000 samples of
1995
F ig u re 6 a , b
Recursive Estimates of
Contem poraneous FF
Response to NBR Shock
+ / - 2SE (CP M o d e l)
Basis points
12 The ordering is therefore {? K MO,
H3}, which reverses Pand I'from
the CP model. This has little effect.
For example, for the recursive
model, if one orders /first, the
response at impact is nine rather
than seven. There seems no good
reason to choose one ordering over
the other.
10 -Y e a r Rolling Estimates
of Contem poraneous FF
Response to NBR Shock
+ / - 2SE (CP M o d e l)
Basis points
artificial data bootstrapped from the estimated
CP m odel.15 Figures 7a-c present the com
puted confidence intervals for the income
responses over the three sample periods of
1959:0 1 -1 9 9 3 :1 2 , 1974 :0 1 -1 9 9 3 :1 2 and
198 2 :1 2 -1 9 9 3 :1 2 , respectively. We find that
the incom e responses could easily be zero
for the first few periods, and are then only
positive in subsequent periods for models
fit using the longer samples. The corre
sponding results for prices are presented in
Figures 7d-f, and these show that negative
price responses are easily realized from a
model that has positive point estimates
for price responses. Finally, as Figures
7g-i show, one gets a well-defined liq
uidity effect over the first two periods
49
13 This is consistent with Cochrane
(1989) ond Gordon and Leeper
(1992), who find a strong liquidity
effect using single-equation, distributed-lag techniques on data for the
period 1979 to 1982, whereas
similar analyses using data prior to
1979 were unable to find evidence
for the liquidity effect.
14 There are mony other problems
that arise in computing confidence
intervals which are not adequately
dealt with in the literature. First,
some studies use a Monte Cado
integration procedure in RATS,
which assumes that VAR parameter
estimators ore normally distributed,
and this will be incorrect in the
presence of unit roots. Second,
because the information presented
is the whole impulse response func
tion, the standard errors computed
for any given response (say the
k'th step) do not capture the range
of uncertainty about the whole
function. Finally, the impulse
responses ore functions of the VAR
parameters. If there are more of
the former than the latter, estima
tors of the former must have a sin
gular distribution. Since one some
times sees hundreds of impulses
displayed on o page, it is very like
ly that the distributions are singular.
15 Similar results to those reported
here are obtained when the error is
simulated from a NID(0, %) distri
bution instead.
REVIEW
MAY/JUNK
1995
F ig u r e 7 a - i
N B R -^ Y and 9 0 % Cl. CP M odel
(a) 1959:01-1993:12
(b) 1974:01-1993:12
(0 1982:12-1993:12
log
Log
0.6
Log
0.25
0.20
0.15
0.10
0.05
0
-0 .05
- 0.10
-0 .15
p i i i m i l i i i i i i i i i |i i i i i i ................... i .. i ■i |
1
10
20
30
40
1
10
20
30
0.20
-0 .25
40
1
10
Periods
Periods
20
30
40
Periods
N B R -^P and 9 0 % Cl. CP M odel
(d) 1959:01-1993:12
(e) 1974:01-1993:12
Log
Log
(f) 1982:12-1993:12
Log
0.35
0.25
0.30
0.20
0.25
0.15
0.20
0.10
0.15
0.05
0.10
0
0.05
0
-0 .05
-0 .05
-
0.10
1
Periods
10
Periods
20
30
Periods
N B R -^FF and 9 0 % Cl. CP M odel
(g) 1959:01-1993:12
(h) 1974:01-1993:12
(i) 1982:01-1993:12
Basis points
Basis points
Basis points
30
40
20
30
15
20
10
10
5
0
0
-10
-5
-20
-1 0
-3 0
-1 5
-20
-4 0
10
20
Periods
30
20
10
Periods
20
30
Periods
but not over the last. Notice also that,
particularly for prices and output, the
confidence intervals are asymmetric. This
asymmetry may be due to the non-stationarity
in the data. Some previous studies have
assumed that the estimated coefficients
can be drawn from a normal distribution,
whereas it is known theoretically that
they should be sampled from a skewed
distribution if there are unit roots in
the data. Sampling from a normal density
will induce the confidence intervals
to look symmetric. Lastrapes and Selgin
(1994) are an exception, and they find
asymmetry in their bootstrapped confidence
intervals. In their case, however, we suspect
the asymmetries are the result of biases in
the point estimates arising from the use of
first-differenced variables as instruments (see
Pagan and Robertson, 1995, for details).
THE EFFECTS O N TH E TER M
STRUCTURE
Relatively little attention has been paid
to the impact of monetary policy upon the
complete term structure of interest rates,
despite the fact that the results will be
50
REVIEW
MAY/JUNE
important to an understanding of the trans
mission mechanism. There is a voluminous
literature on the term structure in both
finance and economics which concentrates
upon the slope of the term structure and the
number of factors influencing it. Rarely are
the factors decomposed into those that are
monetary and those that are not. Cook and
Hahn (1989) study the immediate changes
seen in longer-term rates in response to an
announced change in the federal funds rate,
concluding that this effect becomes small for
longer maturities. However, this does not
address the question of the influence of a
monetary policy change, since the federal
funds rate is influenced by many factors, and
we might expect them to have different influ
ence at different points in the term structure.
One way to proceed would be to utilize
the expectations theory of the term structure,
which links long-term rates to the average of
expected short-term rates.16
1995
F ig u re 8 a -c
N B R -^ R IO , R 3, FF. CP M o d e l,
1959:01-1993:12
Basis points
Periods
N B R -^ R IO , R 3 , FF. CP M o d e l,
1974:01-1993:12
Basis points
r,L= n“‘Et[ X rw ] i= 0
Using the expression for r, in equation 12
and taking derivatives with respect to
dr,I /de™
= n“‘ V c,M
I
l’
Periods
i= 0
N B R -^ R IO , R 3 , FF. CP M od e l,
1982:12-1993:12
we can obtain the long-run responses by
summing the short-term ones. For one-unit
shocks to £tmin the CP model over the full
sample period, these are —12.4(n = 1 ) ,
— 19.3(n = 4) and 3.18(n = 120), which are
of the same order of magnitude as for the
federal funds rate but of opposite sign at
longer maturities.
An alternative method, which does not
depend upon the expectations theory holding,
is to simply add longer-term rates to the VAR
and to directly compute impulse responses
for various interest rates. These are presented
in Figure 8 for FF, R3 and RIO using an aug
mented CP model ordered as (Y, P, CP, NBR,
FF, R3, RIO, TR}, and estimated over the three
sample periods used in the paper. For the two
longer periods, the outcomes resemble those
noted by Cook and Hahn (1 9 8 9 ), but the
period 1982:12-1993:12 shows the greatest
effect of monetary variations to be on the
long-term rate.
Basis points
Periods
C O N C LU S IO N
That the Fed can influence the federal
funds rate on a daily basis is scarcely debat
able. W hat is puzzling has been the failure of
these actions to show up in data. Perhaps this
simply reflects the fact that most empirical
51
16 In fad, this is o linearization of the
precise formulo, and higher-order
terms in the Taylor series expansion
show that the long-term rote will
depend upon higher-order moments
of the conditional density.
REVIEW
MAY/JUNK
there are a number of caveats. Foremost
among these are: The models do not seem
to be very robust to data coming from the
1980s; The implied structural models can
sometimes be implausible; The estimation
procedures often rely on weak information
and, for recursive models, the long-run
multipliers can be contrary to a priori
beliefs. How much damage these features
do to the new view is an unsolved puzzle.
If one encounters odd results, it is hard to
know what their cause is without some
underlying economic model. It may be
that one can produce the observed responses
within a plausible econom ic model as a con
sequence of choosing a particular calibration
of it. Research in the past five years has to
be credited with directing attention to the
fact that analyses of the transmission mecha
nism require a systems perspective, but it is
not clear that the recursive systems chosen
for the investigation are as useful as they
might be. Once unexpected results are found,
the lack of a structure makes it very hard to
account for them. In our view, the natural
progression has to be toward non-recursive
models with less profligate dynamics. The
attempt to say nothing about dynamics
has inevitably lead to a focus upon a set of
variables that may be too narrow to capture
the main interactions in an economy.
F ig u r e 9
M on th ly Change in FF and FF*, CP M odel
Percent
4
2
0
-2
-
-4
-6
-8
Apr-60 Apr-64 Apr-68 Apr-72 Apr-76 Apr-80 Apr-84 Apr-88 Apr-92
17 Actually, we do not believe tbat
these innovations represent policy.
As we emphasized at the beginning
of the article, ideally one wants to
measure the effects of policy-relat
ed shifts in the money supply
curve, ond the dynamic effects of
such chonges could be identified
with the impulse responses with
respect to supply innovations.
However, this does not justify treat
ing innovations as policy. It is our
belief that the innovations can be
best thought of os a mixture of poli
cy ond "noise," the latter stem
ming from the fact that we are
working with a model. Setting the
innovations to zero therefore over
corrects for policy chonges,
although in this instance the
implied policy component would
still be small.
199
work does not use daily data, or it might
be a consequence of reactions within the
economy offsetting the initial im pact over
a longer time period. It is therefore reassuring
that recent work seems to have isolated
a liquidity effect with monthly data. How
large is the effect? If one takes nonborrowed
reserves as the relevant money variable, the
immediate response of the federal funds
rate in the CP model might be taken to be
around —13 basis points as a consequence
of a 1 percent point rise in the level of NBR.
How large this is obviously depends on the
feasible range of variation in NBR. Histori
cally, the average absolute change in NBR
innovations (1 9 5 9 :0 1 -1 9 9 3 :1 2 ) is for a
0.9 percent rise, but it is only around 0.7 per
cent during the 1990s. Consequently, the
measured effect does seem to be small.
Even if we cumulate the multipliers until
they turn positive, it would be rare for the
sum to be smaller than—60 basis points, so
that m ost of the factors historically driving
the federal funds rate do not seem to be due
to the Fed once one looks at it from a monthly
viewpoint. Figure 9 illustrates this, plotting
AFF and AF F *, where F F * is the (one-step)
predicted value of FF using the CP model
after setting the NBR innovation to zero, that
is, assuming there is no policy action. Most
of the variation in interest rates seems to be
explained by factors other than those directly
attributed by the model to monetary policy.17
Even if one accepts the “new” view
regarding the presence of a liquidity effect,
REFERENCES
Bernonke, Ben S., and Alan S. Blinder. "The Federal Funds Rate and the
Channels of Monetary Transmission," The American Economic Review
(September 1992), pp. 901-21.
Boschen, John F., and Leonard 0. Mills. "The Narrative Approach to
Evaluating Monetary Policy: Consistency of Interpretation and the
Relation of Monetary Activity," working paper (1993), Federal
Reserve Bonk of Richmond.
Cochrane, John H. "The Return of the Liquidity Effect: A Study of
the Short-Run Relation Between Money Growth and Interest Rates,"
Journal of Business and Economic Statistics, (January 1989),
pp. 75-83.
Cooley, Thomas F., and Stephen F. LeRoy. "Atheoretical
Macroeconomics: A Critique," Journal of Monetary Economics
(November 1985), pp. 283-308.
Cook, Timothy, ond Thomas Hahn. "The Effect of Changes in the
Federal Funds Rate Target on Market Interest Rates in the 1970s,"
Journal of Monetary Economics (November 1989), pp. 331-51.
NK OF ST. L OU I S
52
REVIEW
MAY/JUNE
Christiano, Lawrence J., and Martin Eichenbaum. "Identification and
the Liquidity Effect of a Monetary Policy Shock," in Alex Cuikerman,
Zvi Hercowitz and Leonard Leiderman, eds., Business Cycles, Growth
and Political Economy. MIT Press, 1992, pp. 335-70.
199
Nelson, Charles R., and Richard Startz. "The Distribution of the
Instrumental Variables Estimator and its T-ratio When the Instrument is
a Poor One," Journal of Business (Port 2, January 1990), pp. 125-40.
Pagan, Adrian R. "Econometric Issues in the Analysis of Regressions with
Generated Regressors," International Economic Review (February 1984),
pp. 221-47.
_ _ _ _ _ _ , _ _ _ _ _ _ and Charles Evans. "The Effects of Monetary
Policy Shocks: Some Evidence from the Flow of Funds," National
Bureau of Economic Research Working Paper No. 4699 (April 1994).
_ _ _ _ _ _ and John C. Robertson. "Structural Models of the Liquidity
Effect," The Australian National University Working Paper No. 283
(1994).
Eichenbaum, Martin. "Comments" on "Interpreting the Macroeconomic
Time Series Facts: The Effects of Monetary Policy" by C. Sims,
European Economic Review (June 1992), pp. 1001-11.
_ _ _ _ _ _ and Y. Jung. "Understanding the Failure of Some
Instrumental Variable Estimators," working paper (1993), The
Australian National University.
_ _ _ _ _ _ and Charles Evans. "Some Empirical Evidence on the Effects
of Monetary Policy Shocks on Exchange Rates," National Bureau of
Economic Research Working Paper No. 4271 (February 1993).
Reichenstein, William. "The Impact of Money on Short-Term Interest
Rates," Economic Inquiry (January 1987), pp. 67-82.
Goli, Jordi. "How Well Does the IS-LM Model Fit Postwar U.S. Data?"
Quarterly Journal of Economics (May 1992), pp. 709-38.
Romer, Christina D., and David H. Romer. "Does Monetary Policy Matter?
A New Test in the Spirit of Friedman and Schwartz," in Oliver Jean
Blanchard and Stanley Fischer, eds., National Bureau of Economic
Research Macroeconomics Annual. MIT Press, 1989, pp. 121-70.
Gilles, Christian, John Coleman and Pamela Labadie. "Identifying
Monetary Policy with a Model of the Federal Reserve," Board of
Governors of the Federal Reserve System finance and Economics
Discussion Series No. 93-24 (July 1993).
Sarte, P.G. "On the Identification of Structural VAR's," working paper
(1994), University of Rochester.
Gordon, David B., and Eric M. Leeper. "The Dynamic Impacts of
Monetary Policy: An Exercise in Tentative Identification," Federal
Reserve Bank of Atlanta Working Paper No. 93-5 (April 1993).
Shapiro Matthew D., and M.W. Watson. "Sources of Business Cycle
Fluctuations," in Stanley Fischer, ed., National Bureau of Economic
Research Macroeconomics Annual. MIT Press, 1988, pp. 111-48.
Kocherlakoto, Narayana R. "On Tests of Representative Consumer Asset
Pricing Models," Journal of Monetary Economics (October 1990),
pp. 285-304.
Sims, Christopher A. "Interpreting the Macroeconomic Time Series
Facts: The Effects of Monetary Policy," European Economic Review
(June 1992), pp. 975-1000.
Koopmans, Tjalling C., H. Rubin and R.B. Leipnik. "Measuring the
Equation Systems of Dynamic Economics," in Tjalling C. Koopmans, ed.,
Statistical Inference in Dynamic Economic Models, Cowles Commission
Monograph 10. John Wiley and Sons, 1950.
_ _ _ _ _ _ . "Macroeconomics and Reality," Econometrica
(January 1980), pp. 1-48.
Lastrapes, W.D., and G. Selgin. "The Liquidity Effect: Identifying ShortRun Interest Rate Dynamics Using Long-Run Restrictions," working
paper (1994), University of Georgia, Department of Economics.
Staiger, D., and J.H. Stock. "Instrumental Variables Regression with
Weak Instruments," working paper (1993), Harvard University,
Kennedy School of Government.
Leeper, Eric M., and David B. Gordon. "In Search of the Liquidity
Effect," Journal of Monetary Economics (June 1992), pp. 341-69.
Strongin, Steven. "The Identification of Monetary Policy Disturbances:
Explaining the Liquidity Puzzle," Federal Reserve Bank of Chicago
Working Paper No. 92-27 (November 1992).
McKibbin, Warwick J., and Jeffrey D. Sachs. Global Linkages:
Macroeconomic Interdependence and Cooperation in the World
Economy. Brookings Institution, 1991.
Thornton, Daniel L. "The Effect of Monetary Policy on Short-Term
Interest Rates," this Review (May/June 1988), pp. 53-72.
Mishkin, Frederic S. The Economics of Money, Banking, and Financial
Markets. Harper Collins, 1992.
_ _ _ _ _ _ . "Monetary Policy and Short-Term Interest Rates: An
Efficient Markets-Rational Expectations Approach," Journal of Finance
(March 1982), pp. 63-72.
_ _ _ _ _ _ . "Monetary Policy and Long-Term Interest Rates: An
Efficient Markets Approach," Journal of Monetary Economics
(January 1981), pp. 29-55.
S3
REVIEW
MAY/JUNI
199
A p p e n d ix
D A TA A N D D A TA SOURCES
Other Series:
Except for the commodity price series
the data are sourced from CITIBASE. The
corresponding CITIBASE mnemonics are
reported in parentheses. The data are m onth
ly from 59:01 to 93:12. All series except
interest rates and the exchange rate are sea
sonally adjusted.
Y
(IP) = log of industrial production
index.
P
(PUNEW ) = log of consumer price
index, urban.
CP (76AXD) = log of industrial country
commodity price index. From the
IMF International Financial
Statistics data tape.
U (LHUR) = unemployment rate, all
workers 16 and over.
ER (EXRUS) = log o f weighted-average
exchange rate.
Money (seasonally adjusted):
M2 (FM 2) = log of M2.
M l (FM 1) = log of M l.
MO (FMBASE) = log of money base
(Federal Reserve Bank of St Louis
definition).
TR (FMRQA+F6CMRE) = log of total
reserves.
NBR(FMRNBC) = log of non-borrowed
reserves plus extended credit.
BR (FMRRA - FMRNBC) = log of
borrowed reserves excluding
extended credit.
NBRX = ratio of non-borrowed to total
reserves (proportion).
Interest Rates (percent, not seasonally
adjusted):
RIO (FY G T10) = 10-year Treasury
Note yield.
R3 (FYGM 3) = three-month Treasury
bill yield (secondary market).
FF (FY FF) = federal funds rate.
RD (FYGD) = discount rate.
RF (FW AFIT) = weighted-average
foreign interest rate.
54
Lowrence J. Christiano is a professor of economics at Northwestern University. The author is grateful to Manuel Balmaseda for excellent
research assistance, and to V.V. Chari and Martin Eichenbaum for discussions.
Com m entary
Law rence J . C h ristia n o
he Pagan and Robertson article pre
sents a useful review of the evidence
on the empirical status of the liquidity
effect proposition— that an exogenous
increase in the money supply drives the
rate of interest down. It discusses how the
consensus in the empirical literature has
shifted from an initial one of skepticism to
what Pagan and Robertson call “the new
view”: that the liquidity effect proposition
has substantial empirical support. In my
comment, I offer an alternative perspective
on the evolution of the empirical literature,
one which focuses on the dynamic correla
tions between three monetary aggregates
and the federal funds rate.
A valuable contribution of the Pagan and
Robertson article is to document evidence to
suggest that the liquidity effect may have
gotten smaller in the years since 1982. This
is an important observation which deserves
more attention to determine exactly what it
means. It may simply be a statistical artifact,
reflecting the relatively small amount of
information in the post-1982 data. Assessing
this is complicated by the fact, documented
further below, that most of the evidence of a
change reflects sub-sample variation in the
estimated variance-covariance matrix of vector
autoregression (VAR) residuals. As Pagan
and Robertson note, these residuals appear
to be characterized by autoregressive condi
tional heteroskedasticity (ARCH) effects and,
under these circumstances, it may be difficult
to identify a true change in an unconditional
variance-covariance matrix. But, assuming
that the change in the variance-covariance
matrix of VAR residuals is in fact real, then
this raises further interesting questions of
interpretation: Has the liquidity effect in fact
gotten smaller, or is the evidence of a reduction
an artifact of an error in the specification of
T
monetary policy? The calculations that pro
duce evidence of a change in the liquidity effect
assume there has been no change in monetary
policy. Most commentators on Fed policy think
that there was a shift in policy in late 1982.
W H A T IS TH E " L IQ U ID IT Y
EFFEC T" A N D W H Y CARE
A B O U T IT?
I begin my discussion by defining what
I mean by a liquidity effect, which I take to be
a property of an econom ic model. An eco
nomic model possesses a liquidity effect if it
has the following characteristic: An exogenous,
persistent, upward shock in the growth rate of
the monetary base, engineered by the central
bank and not associated with any current or
prospective adjustment in distortionary taxes,
drives the nominal rate of interest down for a
significant period of time.
This definition of the liquidity effect can
be distinguished from the traditional, partialequilibrium liquidity effect in the literature.
That refers to the fall in the interest rate that
is required by a downward-sloped money
demand schedule when the money supply
increases and there is no change in the price
level and level of income. Many existing
general-equilibrium models that do not
possess a liquidity effect in the sense that I
define it do display a partial-equilibrium
liquidity effect.
The basic question addressed in the Pagan
and Robertson article, and in the empirical
liquidity effect literature, is: W hat do the data
say about the relative plausibility of the fol
lowing two types of models: models with a
liquidity effect and models with the implication
that an exogenous increase in the monetary
base drives the nominal rate of interest up?
The reason why this question is
interesting is that the answer one selects
has important implications for the construc
tion of quantitative m acroeconomic models
with money. This is discussed further in
Christiano (1 9 9 1 ) and Christiano and
Eichenbaum (1995).
REVIEW
MAY/JUNE
(Examples include King, 1983; Melvin, 1983;
and Mishkin, 1983). This had an impact on
the development o f monetary business cycle
models. For example, Barro (1987, p. 521)
and Robert King (1 9 9 1 ) cite these findings as
evidence in support of the first wave of mone
tized real business cycle models. These models
have the implication that an exogenous
increase in money growth, if persistent, leads
to a rise in the nominal rate of interest. Now
the consensus has returned to the traditional
position in favor of liquidity effects. This in
turn has sparked efforts to identify frictions
which allow monetary models to display a
liquidity effect.
A case can be made that this evolution
in thinking reflects early analysts’ tendency
to focus exclusively on broader monetary
aggregates and their tendency to ignore the
sources of endogeneity in money. To gain
insight into the role played by these consid
erations, consider the results reported in
Figures 1-3, taken from Christiano and
Eichenbaum (1 9 9 2 ). They display the crosscorrelation between different monetary aggre
gates and the federal funds rate (black lin e),
together with plus-and-minus one standarddeviation confidence bands (blue line). The
monetary aggregates examined include non
borrowed reserves (NBR), the monetary base
(MO) and M l. Both the interest rate and the
monetary aggregates have been logged and
Hodrick-Prescott filtered prior to the compu
tations.1 The data display three key features:
(1) The broad monetary aggregates covary
positively with current and future values of
the interest rate; (2) negatively with past values
of the interest rate; and (3) NBR covaries
negatively with current and future values of
the interest rate.
In view of the first feature, it is perhaps
not surprising that analysts who assumed the
endogenous component of money is small
and focused on broader monetary aggregates,
arrived at the view that the evidence does not
support an important liquidity effect. Early
research which recognized the potential role
of endogeneity took the view that the Fed
conducts monetary policy by targeting the
nominal interest rate. (See, for example,
Bernanke and Blinder, 1992; and Sims, 1986.)
Under this view, exogenous innovations in
F ig u re 1
Correlation Between the Fed Funds
Rate(f) and N BR (f-Ir)
Correlation
k (quarters)
Notes: Doto are quarterly and cover the period 1959:1-1991:4. Money data have been logged,
and both series have been HP-filtered prior to doing the computations.
F ig u re 2
Correlation Between the Fed Funds
Rate(f) and the M o n e ta ry Base(f-Ir)
Correlation
k (quarters)
Notes: See notes to Figure 1.
1 The nonborrowed reserves data
were obtained from Steve Strongin.
The other data were token from
CITIBASE. The federal funds rate,
monetary base and Ml have
mnemonics FFYF, FMBASE and
FM1, respectively. The results
reported in Figures 1-3 ore robust
to alternative detrending procedures
and sample periods. See Christiono
and Eichenbaum (1992) for details.
1995
E V O L U T IO N O F V IE W S O N
TH E EM PIR ICAL STATUS O F
TH E L IQ U ID IT Y EFFECT
Historically, economists have taken the
plausibility of the liquidity effect for granted.
This is reflected in standard intermediate
macroeconomics textbooks, which feature
models exhibiting liquidity effects. However,
when researchers initially attempted to quan
tify the liquidity effect using data, they came
away quite skeptical as to its plausibility.
NK OF ST. L OUIS
56
REVI EW
MAY/JUNE
base growth engineered by the central bank
are associated with innovations in the interest
rate. Feature two of the data helps explain
why these analysts favor the liquidity effect
view that an upward revision in the Fed’s
interest rate target is implemented by engi
neering a reduction in the money supply.
Finally, beginning with Thornton (1988),
researchers have recently begun working
with NBR. In light of feature three, it is
perhaps not surprising that they have tended
to conclude that the evidence favors the liq
uidity effect view.
W hile the correlations I ju st described
go a long way toward explaining why different
researchers reached different conclusions about
the empirical status of liquidity effects, they do
not tell the whole story. That is because the
liquidity effect pertains to the sign of the cor
relation between the components of interest
rates and money that reflect exogenous dis
turbances to monetary policy. Raw correla
tions, by contrast, reflect the jo in t movements
of interest rates and money arising due to
the effects of all shocks, not ju st exogenous
monetary policy shocks. To see why this dis
tinction probably matters, consider the cor
relation between logged and detrended gross
domestic product and NBR in Figure 4 .2 The
fact that the contemporaneous correlation is
significantly negative may reflect a policy of
“leaning against the wind” at the Fed. If so,
then the raw correlation between interest rates
and NBR reflects in part the response of both
variables to whatever shocks are driving GDP.
Such shocks could in principle produce a
positive or negative correlation between money
and interest rates, independent of whether the
liquidity effect is operative.
Coleman, Gilles and Labadie (1995), CGL,
present a couple of hypothetical examples that
illustrate very nicely how this could happen.
The examples underscore the importance of
isolating the exogenous monetary policy
component of a monetary indicator variable.
They are also useful for illustrating the kind
of steps researchers take in practice to build
confidence that the shocks they have isolated
are indeed monetary policy shocks and not
something else. In one of CGI’s examples, the
economy is driven by a single shock, one that
is non-monetary in origin. CGL assume that
199
F ig u r e 3
Correlation Between the Fed Funds
Rate(f) and M l ( f - l t )
Correlation
k (quarters)
Notes: See notes to Figure 1.
F ig u re 4
Correlation Between G D P (f)
and N B R (f-k )
Correlation
k (quarters)
Notes: Data ore quarterly and cover the period 1959:1-1991:4. Both variables were logged
and HP-filtered prior to the compuutations.
the shock drives up the equilibrium nominal
rate of interest, and that this produces an
accommodation at the Federal Reserve’s dis
count window. The Federal Open Market
Committee (FOM C) is assumed to partially
offset the impact of this on total bank reserves
by undertaking contractionary open market
operations which have the effect of reducing
nonborrowed reserves. In an economy like
this, there would be a negative correlation
between the rate of interest and NBR, even
NK OF S T . L O U I S
57
J The gross domestic product data
are taken from CITIBASE.
BMW
MAY/JUNK
reserves produces an immediate rise in the
interest rate. Suppose the rise in the interest
rate results in an accommodation at the dis
count window, so that to insulate total reserves
from this, the Fed must reduce nonborrowed
reserves. In a world like this, one would
expect a negative correlation between non
borrowed reserves and the interest rate, even
though there is no liquidity effect.
It is in an effort to avoid the sort of
pitfalls illustrated by the CGL examples that
the recent literature has taken great pains to
isolate the exogenous component of monetary
policy in monetary indicator variables. The
assumptions made to do this are called iden
tifying assumptions, and they typically involve
incorporating more variables into the analysis.
Additional steps are taken to further reduce
the likelihood of the kind of problems empha
sized in the CGL examples. One strategy for
doing so is pursued in Christiano, Eichenbaum
and Evans (1994), CEE. To build confidence
that their shocks correctly isolate the exoge
nous shock to policy, CEE analyze the impact
of their monetary policy shock measures on
many macroeconomic variables. Based on
their findings, they conclude that their mon
etary policy shock measures probably do not
suffer significantly from the sort of distortions
illustrated in the two CGL examples. For
example, it seems unlikely that the CEE policy
shock really measures the private economy
shock in CGLs first example. That’s because
CEE find that a negative shock to nonbor
rowed reserves leads to a rise in unemployment
and inventories, and a fall in output, employ
ment, profits, and the broad monetary aggre
gates. It seems hard to imagine a reasonable
model in which a non-monetary shock would
have these effects. Finally, CGLs second
example seems implausible in light of the CEE
finding that a negative shock to NBR leads to
a fall in the broader monetary aggregates.
In sum, the basic outlines of the story
describing the evolution of thinking about
liquidity effects can be understood with ref
erence to simple correlations between various
monetary aggregates and the interest rate. The
full story is more complicated and involves a
broader set of variables. These are used first
to isolate a measure of the exogenous com
ponent of monetary policy, and then to “test”
F ig u r e 5
Interest Rate Response to
O rth o g o n a lize d NBR Shock
Basis points
Months
Notes: Impulse response based on 14-lag, six-variable VAR estimated using monthly data,
1959:1-1991:10.
F ig u re 6
Interest Rate Response to
O rth o g o n a lize d NBR Shock
Basis points
Months
Notes: Impulse response function based on six-lag, six-variable VAR estimated for period
1982:12-1991:10. Confidence interval is the one implied by the 14-lag, six-variable VAR
fit to whole sample.
1995
though there are no monetary policy
shocks at all.
CGLs second example illustrates how
an economy with monetary policy shocks,
but only an anticipated inflation effect and
no liquidity effect, could also generate a
negative correlation between nonborrowed
reserves and the interest rate. Suppose the
Fed signals policy shifts in advance of actually
implementing them, and that a signal of an
imminent increase in the growth of total
58
REVIEW
MAY/JUNE
hypothesis is rejected.5 This test suggests that
the reduction in the liquidity effect in the
1980s is more than what one would expect
given that the 1980s constitute a relatively
small sample of data.
The primary reason for the shift in the
impulse response function appears to lie in a
shift in the variance-covariance matrix of the
VAR disturbances. One way to see this is to
note that the biggest change in going from the
full sample to the short sample is in the esti
mated impact effect of an orthogonalized NBR
shock. That object is a direct function of the
variance-covariance matrix of the fitted dis
turbances. (In particular, it is the 5,4 element
in the lower triangular Choleski decomposition
of the variance-covariance matrix.)
Another way to see this is to consider
Figure 7. That reproduces the impulse
response functions reported in Figures 5
and 6 for convenience. In addition, Figure 7
reports an impulse response function obtained
by combining the lagged coefficients from the
14-lag VAR fit to the period 1959:01-1991:10,
with the variance-covariance of the sub-set of
its fitted disturbances covering the period
1982:12-1991:10.6
Note that the resulting impulse response
function resembles the one fit to the post-1982
data in that it implies a small liquidity effect.
Thus, in essence the statistical test reported
in the previous paragraph (and, presumably,
in Pagan and Robertson too) is a rejection of
the null hypothesis of constancy of a particular
function of the VAR disturbance variancecovariance matrix.
The fact that the smaller liquidity
effect in the 1980s reflects instability in the
estimated variance-covariance of fitted dis
turbances raises two questions. First, Pagan
and Robertson have emphasized that there is
“extensive ARCH in the VAR equations for
interest rates and money.” But the procedure
I used (following Pagan and Robertson) to
deduce that there is statistically significant
instability in the impulse response functions
assumes the disturbances are iid. Under these
circumstances, one presumes that extensive
ARCH in the disturbances would greatly
increase the probability of false rejections in
tests of the null hypothesis of no change in a
variance-covariance matrix. This is because
that the resulting measure does not confound
shocks that are non-monetary in origin. This
part of the story involves many assumptions.
Significantly, researchers using a wide variety
of plausible assumptions have reached the
conclusion that the data support the liquidity
effect view.
A V A N IS H IN G L IQ U ID IT Y
EFFECT?
Pagan and Robertson report calculations
that suggest the liquidity effect may be smaller
in the 1980s than before. To see this, first
consider Figure 5, which displays the response
of the interest rate to an orthogonalized inno
vation in nonborrowed reserves. The response
is based on what Pagan and Robertson call the
CP model of Christiano, Eichenbaum and
Evans (1994). The underlying six-variable,
14-lag VAR was estimated using the period
1959:01 to 1991:10. The blue lines are
90 percent confidence intervals computed by
the bootstrap method outlined in the Pagan
and Robertson article.3 Note the statistically
significant negative initial response of the
interest rate. A 1 percent rise in nonborrowed
reserves drives the funds rate down about
15 basis points (annual rate) in the current
month, and 25 basis point in the next month.
The Pagan and Robertson observation can be
seen by comparing Figure 5 with Figure 6,
which displays the interest rate response based
on a six-lag, six-variable CP model estimated
over the sample 1982:12 to 1991:01.4 This
impulse response function has the implication
that a 1 percent rise in nonborrowed reserves
leads to a contemporaneous rise of 1 basis
point in the funds rate, followed by relatively
small reductions of 3, 8 and 13 basis points
in the first, second and third months, respec
tively, after a shock. After that, the point
estimates in Figures 5 and 6 are quite similar.
The other curves in Figure 6 enable one to
test the null hypothesis that the data from the
later sample are consistent with the model fit
to the whole sample. They define a 90 percent
bootstrap confidence interval, constructed
using the 14-lag VAR model and its fitted
residuals estimated for the entire sample.
Note that the first two impulses lie outside
this confidence interval, so that the null
199
59
3 That is, I used random samples of
the fitted VAR disturbances, together
with the estimated VAR and actual
U.S. data for the required 14 initial
conditions to generate 1,000 artificial
data sets of 388 observations each,
for all six variables in the VAR. In
each artificial data set, I fit a 14-log,
six-variable VAR and computed an
impulse-fesponse function using the
procedure undedying the computa
tions for the point estimates in
Figure 5. Let a,(k) denote the r
month's response of the interest
rate to o policy shock, /= 1... 36,
on the It1* artificial data set, k = 1,
...,1000. Then, for each /,
was ordered from largest to smallest.
The 50th ond 950th elements are
reported as the top ond bottom
curves in Figure 5.
4 The logs lengths of the two models
correspond to the choices made by
Pagan ond Robertson. A six-lag
model for the long sample period
does not work well. The Ljung-Box
(^statistic at lag 24 computed on
the residuals for the interest rote
equation has a value of 43, with a
significance level of 1 percent. This
convinced me that six lags is too
short for this sample. The Q-statistic computed for the six-lag VAR fit
to the post-1982 sample did not
show any evidence of serial correla
tion in the residuals.
5 The bootstrap confidence intervals
were computed os follows. Using
U.S. data for the required 14 initial
conditions, the empirically estimated
14-lag VAR was used to simulate
1,000 artificial data sets of 388
observations each. The residuals
for each data set were obtained by
random sampling from the fitted
residuals. The last 107 observations
in each sample were used os the
estimation period for fitting a six-lag
VAR ond computing on impulse
response function like the middle
(continued on following p a g e)
mm
MAY/JUNK
it is not a statistical artifact, it would be inter
esting to investigate exactly what it means.
Does it reflect specification error due to a
change in policy regime? Does it reflect that
the liquidity effect actually was smaller in
the 1980s, perhaps because agents became
more sensitive to news about inflation?
To assess the results in this article, it is
important to recall what is at stake here. Views
about the presence or absence of a liquidity
effect in the data determine what kind of
monetary models macroeconomists use to
conduct policy analysis. In early monetized
real business cycle models, the interest rate
money dynamics were dominated by strong
anticipated inflation effects. The PaganRobertson article presents no evidence to
support the notion that there is a strong rise
in interest rates in response to an expansionary
monetary policy shock, as these models
require. Instead, all the point estimates indi
cate a fall in the interest rate in the wake o f a
positive monetary policy shock. In particular,
the Pagan and Robertson article provides
no evidence that macroeconomists should
abandon models exhibiting liquidity effects
and go back to simple monetized real business
cycle models.
F ig u re 7
Interest Rate Response to
O rth o g o n alize d NBR Shock
15
10
5
0
-5
-10
-15
-20
-25
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39
Months
* obtained bycombining lagged VARcoefficients fromVARfit to whole sample with
innovationvariance-covariance matrix frompost-1982 period.
(footnote 5 cont.)
line in Figure 6. The confidence
intervals were computed using the
1,000 impulse response functions
with the same method as the one
underlying Figure 5.
6 That is, let Y,=4 (0 I'm + (/t>
where t/f are the fitted VAR distur
bances and A(L) denotes the fitted
14-lag matrix polynomial of VAR
coefficients. Let /denote the varioncecovariance matrix of u,covering
the period 1982:12-1991:10 only.
Let CC'= /be the lower triangular
Choleski decomposition of V. Then,
the numbers in Figure 7 are the
coefficients in the 5,4 element of
the matrix polynomial [l->l(L)]''C.
7 Another interesting question is
whether such an analysis could be
reconciled with dynamic macroeco
nomic theory.
1995
under ARCH, a sample variance-covariance
matrix can display substantial time variation,
even though the underlying unconditional
variance-covariance matrix is constant. Thus,
it remains an open question whether the
smaller estimated liquidity effect in the 1980s
is simply a statistical artifact.
Second, the apparent instability in the
variance-covariance matrix of VAR distur
bances suggests it might be fruitful to explore
the possibility of policy shifts using the
“identified VAR” identification strategy pur
sued by Bernanke (1986) and Sims (1986).
There are two reasons for this: (1) It is widely
thought that policy did change across the
1979-82 and 1982-present periods; and (2)
the Bernanke-Sims style approach would
predict a change in the variance-covariance
matrix of residuals under a change in policy.
W hether it predicts precisely the instability
observed is an open question.7
REFERENCES
Barro, Robert J. Macroeconomics, 2nd edition. John Wiley & Sons, 1987.
Bernanke, Ben S. "Alternative Explanations of the Money Income
Correlation," in Karl Brunner and Allan Meltzer, eds., Carnegie Rochester
Conference Series on Public Policy (autumn 1986), pp. 49-100.
Bernanke, Ben S., and Alan S. Blinder. "The Federal Funds Rate and the
Channels of Monetary Transmission," The American Economic Review
(September 1992), pp. 901-21.
Christiano, Lawrence J. "Modeling the Liquidity Effect of a Monetary
Shock," Federal Reserve Bank of Minneapolis Quarterly Review
(winter, 1991), pp 3-34.
_ _ _ _ _ _ and Martin Eichenbaum. "Identification and the Liquidity
Effect of a Monetary Policy Shock," in Alex Cukierman, Zvi Hercowifr
and Leonard Liederman, eds., Business Cycles, Growth and Political
Economy. MIT Press, 1992, pp. 335-70.
C O N C LU S IO N
To summarize, the authors draw attention
to a reduction in the estimated size of the liq
uidity effect in the 1980s. This certainly
deserves attention. However, the right statis
tical techniques have not yet been applied to
determine whether the apparent change is
statistically significant, or ju st an artifact of
the small number of observations. Assuming
_ _ _ _ _ _ and_ _ _ _ _ _ _ . "Liquidity Effects, Monetary Policy and
the Business Cycle," National Bureau of Economic Research Working
Paper No. 4129 (August 1994).
_ _ _ _ _ _ , _ _ _ _ _ _ and Charles Evans. "The Effects of Monetary
Policy Shocks: Evidence From the Flow of Funds," Review of
Economics and Statistics (forthcoming).
i NK O F S T . L O U I S
60
Gali, Jordi. "How Well Does the IS-LM Model Fit Post Wat U.S. Data,"
Quarterly Journal of Economics (Moy 1992), pp. 709-38.
King, Robert. "Money and Business Cycles," working paper (1991),
University of Rochester.
King, Stephen R. "Real Interest Rates and the Interaction of Money,
Output, and Prices," working paper (1983).
Melvin, Michael. "The Vanishing Liquidity Effect of Money on Interest:
Analysis and Implications for Policy," Economic Inquiry (April 1983),
pp. 188-202.
Mishkin, Frederic S. A Rational Expectations Approach to Macroeconomics:
Testing Policy Ineffectiveness and Efficient-Markets Models. University
of Chicago Press, 1983.
Thornton, Daniel L "The Effect of Monetary Policy on Short-Term
Interest Rates," this M e w (May/June 1988), pp. 53-72.
REVIEW
MAY/JUNE
1995
R. Glenn Hubbard is Russell L. Carson professor of economics and finance at Columbia University. The author is grateful to Allen Berger, Phillip
Cagan, Richard Cantor, Mark Gertler, Simon Gilchrist, Anil Kashyap, Don Morgan, Glenn Rudebusch, Bruce Smith and participants at the confer
ence for helpful comments and suggestions. The author also acknowledges financial support from the Center for the Study of the Economy
and the State of the University of Chicago, and the Federal Reserve Bank of New York.
I
Is There a
"Credit Channel"
for M o n e ta ry
Policy?
sidestep the credit view language per se, and
instead focus on isolating particular frictions
in financial arrangements and on developing
testable implications of those frictions. To
anticipate that analysis a bit, I argue that
realistic models of “financial constraints” on
firms’ decisions imply potentially significant
effects of monetary policy beyond those
working through conventional interest rate
channels. Pinpointing the effects of a narrow
“bank lending” channel of monetary policy
is more difficult, though some recent models
and empirical work are potentially promising
in that regard.
I
begin by reviewing the assumptions
and implications of the money view of the
monetary transmission mechanism and by
describing the assumptions and implications
of models of financial constraints on borrow
ers and models of bank-dependent borrow
ers. The balance of the article discusses the
transition from alternative theoretical models
of the transmission mechanism to empirical
research, and examines implications for
monetary policy.
R. G le n n H u b b a rd
nderstanding the channels through which
monetary policy affects econom ic vari
ables has long been a key research topic
in macroeconomics and a central element of
econom ic policy analysis. At an operational
level, a “tightening” of monetary policy by the
Federal Reserve implies a sale of bonds by the
Fed and an accompanying reduction of bank
reserves. One question for debate in academic
and public policy circles in recent years is
whether this exchange between the central
bank and the banking system has consequences
in addition to those for open market interest
rates. At the risk of oversimplifying the debate,
the question is often asked as whether the
traditional interest rate or “money view”
channel presented in most textbooks is aug
mented by a “credit view” channel.1
There has been a great deal of interest in
this question in the past several years, moti
vated both by developments in economic
models (in the marriage of models of infor
mational imperfections in corporate finance
with traditional macroeconomic models) and
recent events (for example, the so-called credit
crunch during the 1990-91 recession).2 As I
elaborate below, however, it is not always
straightforward to define a meaningful credit
view alternative to the conventional interest
rate transmission mechanism. Similar diffi
culties arise in structuring empirical tests of
credit view models.
This paper describes and analyzes a broad,
though still well-specified, version of a credit
view alternative to the conventional monetary
transmission mechanism. In so doing, I
U
H O W R EA S O N A B LE IS THE
M O N E Y V IE W ?
Before discussing predictions for the
effects of alternative approaches on monetary
policy, it is useful to review assumptions
about intermediaries and borrowers in the
traditional interest rate view of the monetary
transmission mechanism. In this view, finan
cial intermediaries (banks) offer no special
services on the asset side of their balance sheet.
On the liability side of their balance sheet,
banks perform a special role: The banking
system creates money by issuing demand
deposits. Underlying assumptions about
borrowers is the idea that capital structures
do not influence real decisions of borrowers
and lenders, the result of Modigliani and
Miller (1 9 5 8 ). Applying the intuition of the
Modigliani and Miller theorem to banks, Fama
(1 9 8 0 ) reasoned that shifts in the public’s
portfolio preferences among bank deposits,
63
1 For descriptions of the debote, see
Bernonke ond Blinder (1988) ond
Bernanke (1993).
2 For on onolysis of the "credit
crunch" episode, see Kliesen ond
Tatom (1992) ond the studies in
the Federal Reserve Bank of New
York (1994). The paper by Cantor
and Rodrigues in the New York Fed
studies considers the possibility of o
credit crunch for nonbank interme
diaries.
REVIEW
MAY/JUNE
3 Foma's insight amplifies the earlier
contribution of Brainard and Tobin
(1963) that monetary policy can
be analyzed through its effects on
investor portfolios.
4 More generally, in o model with
many assets, this description would
assign to the money view of the
transmission mechanism effects on
spending arising from any changes
in the relative prices of assets.
While this simple two-asset-model
description of the money view is
highly stylized, it is consistent with
a number of alternative models
beyond the textbook IS-LM model
(see, for example, Hubbard, 1994),
including dynamic-equilibrium cashin-advance models (for example,
Rotemberg, 1984; and Christiano
and Eichenboum, 1992).
5 For on empirical description of this
transmission mechanism in the con
text of the Federal Reserve's forecast
ing model, see Mauskopf (1990).
6 See, for example, "limited partici
pation" models as in Lucos (1990)
and Christiano and Eichenbaum
(1992).
7 See, for example, analyses of
inventory investment in Kashyop,
Stein and Wilcox (1993) and
Gertier and Gilchrist (1993). See
also the review of empirical studies
of business fixed investment in
Chirinko (1993) and Cummins,
Hassett and Hubbard (1994).
8 This current fashion actually has a
long pedigree in macroeconomics,
with important contributions by Fisher
(1933), Guriey and Shaw (1955,
1960), Minsky (1964,1975) and
Wojnilower (1980). Some econo
metric forecasting models have also
focused on financial factors in prop
agation mechanisms (see, for
example, the description for the DRI
model in Eckstein and Sinai, 1986).
Cagan (1972) provides on empirical
analysis of money and bank lending
views. An early contributor to the
contemporary credit view literature
is Bernanke (1983).
in short-term rates is not obvious a priori
based on conventional models of the term
structure. Empirical studies, however, have
documented a significant, positive relationship
between changes in the (nominal) federal
funds rate and the 10-year Treasury bond rate
(see, for example, Cohen and Wenninger, 1993;
and Estrella and Hardouvelis, 1990). Finally,
although many components of aggregate
demand are arguably interest-sensitive (such
as consumer durables, housing, business fixed
investment, and inventory investment), output
responses to monetary innovations are large
relative to the generally small estimated effects
of user costs of capital on investment.7
I
shall characterize the money view as
focusing on aggregate, as opposed to distribu
tional, consequences of policy actions. In
this view, higher default-risk-free rates of
interest following a monetary contraction
depress desired investment by firms and
households. W hile desired investment falls,
the reduction in business and household
capital falls on the least productive projects.
Such a view offers no analysis of distribu
tional, or cross-sectional, responses to policy
actions, nor of aggregate implications of this
heterogeneity. I review these points not to
suggest that standard interest rate approaches
to the monetary transmission mechanism are
incorrect, but to suggest strongly that one
ought to expect that they are incomplete.
bonds or stocks should have no effect on real
outcomes; that is, the financial system is
merely a veil.3
To keep the story simple, suppose that
there are two assets— money and bonds.4
In a monetary contraction, the central bank
reduces reserves, limiting the banking system’s
ability to sell deposits. Depositors (house
holds) must then hold more bonds and less
money in their portfolios. If prices do not
instantaneously adjust to changes in the money
supply, the fall in household money holdings
represents a decline in real money balances. To
restore equilibrium, the real interest rate on
bonds increases, raising the user cost of capital
for a range of planned investment activities,
and interest-sensitive spending falls.5
W hile the money view is widely accepted
as the benchmark or “textbook” model for
analyzing effects of monetary policy on eco
nomic activity, it relies on four key assump
tions: (1) The central bank must control the
supply of “outside money,” for which there are
imperfect substitutes; (2) the central bank can
affect real as well as nominal short-term interest
rates (that is, prices do not adjust instanta
neously); (3) policy-induced changes in real
short-term interest rates affect longer-term
interest rates that influence household and
business spending decisions; and (4) plausible
changes in interest-sensitive spending in
response to a monetary policy innovation
match reasonably well with observed output
responses to such innovations.
In this stylized view, monetary policy is
represented by a change in the nominal supply
of outside money. O f course, the quantity
of m uch of the monetary base is likely to be
endogenous.6 Nonetheless, legal restrictions
(for example, reserve requirements) may
compel agents to use the outside asset for
some transactions. In practice, the central
bank’s influence over nominal short-term
interest rates (for example, the federal funds
rate in the United States) is uncontroversial.
There is also evidence that the real federal
funds rate responds to a shift in policy (see,
for example, Bernanke and Blinder, 1992).
Turning to the other assumptions,
that long-term rates used in many saving
and investment decisions should increase or
decrease predictably in response to a change
1995
H O W R EA S O N A B LE IS THE
CREDIT V IE W ?
The search for a transmission mechanism
broader than that ju st described reflects two
concerns, one “macro” and one “micro.” The
macro concern, mentioned earlier, is that
cyclical movements in aggregate demand—
particularly business fixed investment and
inventory investment— appear too large to
be explained by monetary policy actions that
have not generally led to large changes in
real interest rates. This has pushed some
macroeconomists to identify financial factors
in propagating relatively small shocks, fac
tors that correspond to accelerator models
that explain investment data relatively well.8
Indeed, I use the term “financial accelerator”
(put forth by Bernanke, Gertier and Gilchrist,
64
REVIEW
MAY/JUNK
forthcoming) to refer to the magnification of
initial shocks by financial market conditions.
The micro concern relates to the emer
gence of a growing literature studying infor
mational imperfections in insurance and credit
markets. In this line of inquiry problems of
asymmetric information between borrowers
and lenders lead to a gap between the cost
o f external finance and internal finance.
The notion of costly external finance stands
in contrast to the more complete-markets
approach underlying the conventional interest
rate channels, which does not consider links
between real and financial decisions.’
Although a review of this literature is
beyond the scope of this article, I want to
mention three common empirical implica
tions that have emerged from models of the
financial accelerator.10 The first, which I ju st
noted, is that uncollateralized external finance
is more expensive than internal finance.
Second, the spread between the cost of external
and internal finance varies inversely with the
borrower’s net worth— internal funds and
collateralizable resources— relative to the
amount of funds required. Third, an adverse
shock to a borrower’s net worth increases the
cost of external finance and decreases the
ability of the borrower to implement invest
ment, employment and production plans.
This channel provides the financial accelerator,
magnifying an initial shock to net worth.
(See, for example: Fazzari, Hubbard and
Petersen, 1988; Gertler and Hubbard, 1988;
Cantor, 1990; Hoshi, Kashyap and Scharfstein,
1991; Calomiris and Hubbard, forthcoming;
Hubbard and Kashyap, 1992; Oliner and
Rudebusch, 1992; Fazzari and Petersen, 1993;
Hubbard, Kashyap and W hited, forthcoming;
Bond and Meghir, 1994; Cummins, Hassett
and Hubbard, 1994; Carpenter, Fazzari and
Petersen, 1994; and Sharpe, 1994.)" Links
between internal net worth and broadly defined
investment (holding investment opportunities
constant) have been corroborated in a number
of empirical studies.12
Let me now extend this argument to
include a channel for monetary policy.13 In
the money view, policy actions affect the
overall level of real interest rates and inter
est-sensitive spending. The crux o f models
of information-related financial frictions is
1995
a gap between the cost of external and internal
finance for many borrowers. In this context,
the credit view offers channels through which
monetary policy (open market operations or
regulatory actions) can affect this gap. That
is, the credit view encompasses distributional
consequences of policy actions, because the
costs of finance respond differently for different
types of borrowers. Two such channels have
been discussed in earlier work: (1) financial
constraints on borrowers and (2) the exis
tence of bank-dependent borrowers.
Financial Constraints On Borrowers
Any story describing a credit channel for
monetary policy must have as its foundation
the idea that some borrowers face high costs
of external finance. In addition, models of
a financial accelerator argue that the spread
between the cost of external and internal funds
varies inversely with the borrowers’ net worth.
It is this role of net worth which offers a
channel through which policy-induced
changes in interest rates affect borrowers’
net worth (see, for example, Gertler and
Hubbard, 1988). Intuitively, increases in the
real interest rate in response to a monetary
contraction increase borrowers’ debt-service
burdens and reduce the present value of
collateralizable net worth, thereby increasing
the marginal cost o f external finance and
reducing firms’ ability to carry out desired
investment and employment programs. This
approach offers a credit channel even if open
market operations have no direct quantity
effect on banks’ ability to lend. Moreover,
this approach implies that spending by lownet-worth firms is likely to fall significantly
following a monetary contraction (to the
extent that the contraction reduces borrowers’
net worth).
The Existence of Bank-Dependent
Borrowers
The second channel stresses that some
borrowers depend upon banks for external
funds, and that policy actions can have a direct
impact on the supply of loans. W hen banks
are subject to reserve requirements on liabilities,
a monetary contraction drains reserves,
65
9 Potential effects of adverse selec
tion problems on market allocation
have been addressed in important
papers by Akerlof (1970) and
Rothschild and Stiglitz (1976), and
have been applied to loan markets
by Jaffee and Russell (1976) and
Stiglitz ond Weiss (1981), and to
equity markets by Myers and
Majluf (1984). Research on principal-agent problems in finance hos
followed the contribution of Jensen
and Meckling (1976). Gertler
(1988), Bernanke (1993) and
King and Levine (1993) provide
reviews of related models of infor
mational imperfections in capital
markets.
10 See also the review in Bernanke,
Gertler and Gilchrist (forthcoming).
These implications are consistent
with a wide class of models, includ
ing those of Townsend (1979),
Blinder and Stiglitz (1983),
Farmer (1985), Williamson
(1987), Bernonke and Gertler
(1989,1990), Calomiris and
Hubbard (1990), Sharpe (1990),
Hart and Moore (1991), Kiyotaki
ond Moore (1993), Gertler
(1992), Greenwald and Stiglitz
(1988,1993) and Lamont (1993).
" For households, Mishkin (1977,
1978) and Zeldes (1989) provide
evidence of effects of household
balance sheet conditions on con
sumer expenditures.
12 The appendix presents a simple
model that illustrates these predic
tions.
13 For broader descriptions of credit
view arguments, see Bernanke
(1993), Friedman and Kuttner
(1993), Gertler (1993), Gertler
and Gilchrist (1993) ond Kashyap
and Stein (1994). An early expo
sition of a role for credit availability
appears in Roosa (1951).
REVIEW
MAY/JUNK
14 Models of equilibrium credit
rationing under adverse selection
(for example, Stiglitz and Weiss,
1981) offer another mechanism
through which an increase in the
level of default-risk-free real interest
rates reduces loan supply. Credit
rationing is not required for the
bank-dependent-borrower channel
to be operative. Instead, what is
required is that loons to these bor
rowers are an imperfect substitute
for other assets and that the bor
rowers lack alternative sources of
finance.
15 Calomiris and Kahn (1991) offer a
model of demandable debt to
finance bank lending.
16 A substantial body of empirical evi
dence supports the idea that banks
offer special services in the lending
process. For example, James
(1987) and Lummer and
McConnell (1989) find that the
announcement of a bank loan, all
else equal, raises the share price of
the borrowing firm, likely reflecting
the information content of the
bonk's assessment. In a similar
spirit, Fama (1985) and James
(1987) find that bonks' borrowers,
rather than banks' depositors, bear
the incidence of reserve require
ments (indicating that borrowers
must not have easy access to other
sources of funds). Petersen and
Rajon (1994) show that small
businesses fend to rely on local
banks for external funds.
17 See, for example, the discussion in
Petersen ond Rajan (1994).
18 Owens and Schreft (1992) discuss
the identification of "credit crunch
es." See also the description in
Hubbard (1994).
possibly decreasing banks’ ability to lend. As
a result, credit allocated to bank-dependent
borrowers may fall, causing these borrowers
to curtail their spending. In the IS-LM
framework of Bem anke and Blinder (1988),
both the IS and LM curves shift to the left in
response to a monetary contraction. Alterna
tively, an adverse shock to banks’ capital could
decrease both banks’ lending and the spending
by bank-dependent borrowers. Such bank
lending channels magnify the decline in out
put as a result of the monetary contraction,
and the effect of the contraction on the real
interest rate is muted. This basic story raises
three questions, relating to: (1) why certain
borrowers may be bank-dependent (that is,
unable to access open market credit or borrow
from nonbank financial intermediaries or
other sources), (2) whether exogenous changes
in banks’ ability to lend can be identified, and
(3) (for the analysis of open market operations)
whether banks have access to sources of funds
not subject to reserve requirements.
The first question is addressed, though
not necessarily resolved, by the theoretical
literature on the development of financial
intermediaries14. In much of this research
(see especially Diamond, 1984; and Boyd
and Prescott, 1986), intermediaries offer
low-cost means of monitoring some classes
of borrowers. Because of informational fric
tions, non-monitored finance entails dead
weight spending resources on monitoring. A
free-rider problem emerges, however, in pub
lic markets with a large number of creditors.
The problem is mitigated by having a finan
cial intermediary hold the loans and act as a
delegated monitor. Potential agency problems
at the intermediary level are reduced by having
the intermediary hold a diversified loan port
folio financed principally by publicly issued
debt.15 This line of research argues rigorously
that borrowers for whom monitoring costs
are significant will be dependent upon inter
mediaries for external finance,16 and that costs
of switching lenders will be high.17 It does not,
however, necessarily argue for bank dependence
(for example, finance companies are interme
diaries financed by non-deposit debt).
Second, even if one accepts the premise
that some borrowers are bank-dependent in
the sense described earlier, one must identify
1995
exogenous changes in banks’ ability to lend.
Four such changes have been examined in
previous research. The first focuses on the
role played by banking panics, in which
depositors’ flight to quality— converting bank
deposits to currency or government debt—
reduces banks’ ability to lend (for empirical
evidence, see Bernanke, 1983, and Bernanke
and Jam es, 1991, for the 1930s and Calomiris
and Hubbard, 1989, for the National Banking
period).
A second argument emphasizes regulatory
actions, such as that under binding Regulation
Q ceilings in the United States (see, for
example, Schreft, 1990; Kashyap and Stein,
1994; and Romer and Romer, 1993) and reg
ulation of capital adequacy (see, for example,
Bernanke and Lown, 1992; and Peek and
Rosengren, 1 9 9 2 ) . Empirical evidence for
this channel is quite strong. Third, Bizer
(1 9 9 3 ) suggests that increased regulatory
scrutiny decreased banks’ willingness to
lend in the early 1990s, all else equal.
The fourth argument stresses exogenous
changes in bank reserves as a result of shifts
in monetary policy. In principle, such a shift
in monetary policy could be identified with a
discrete change in the federal funds rate in the
aftermath of a dynamic open market operation
or with a change in reserve requirements.
Because the effects on reserves of changes in
reserve requirements are generally offset by
open market operations, bank-lending-channel stories are generally cast in terms of open
market operations.
An illustration of the gap between models
and practice surfaces in addressing the third
question of the ease with which banks can raise
funds from non-deposit sources (for example,
CDs), when the Fed decreases reserves. Romer
and Romer (1 990) have pointed out, for
example, that if banks see deposits and CDs
as perfect substitutes, the link between open
market operations and the supply of credit to
bank-dependent borrowers is broken. Banks
are unlikely, however, to face a perfecdy elastic
supply schedule for CDs at the prevailing CD
interest rate. Since large-denomination CDs
are not insured at the margin by federal deposit
insurance, prospective lenders must ascertain
the quality of the issuing bank’s portfolio.
Given banks’ private information about at
66
REVIEW
MAY/JUNE
banks or other financial intermediaries) facing
incomplete financial markets. Examining
links between the volume of credit and eco
nomic activity in aggregate data (with an eye
toward studying the role played by bankdependent borrowers) requires great care.
Simply finding that credit measures lead output
in aggregate time-series data is also consistent
with a class of models in which credit is passive,
responding to finance expected future output
(as in King and Plosser, 1984). Consider the
case of a monetary contraction, for example.
The effect of the contraction on interest rates
could depress desired consumption and
investment spending, reducing the demand
for loans.
In a clever paper that has stimulated a
number of empirical studies, Kashyap, Stein
and W ilcox (1 9 9 3 )— henceforth, KSW—
examine relative fluctuations in the volume
of bank loans and a close open market sub
stitute, issuance o f commercial paper. In the
KSW experiment, upward or downward
shifts in both bank lending and commercial
paper issuance likely reflect changes in the
dem and for credit. However, a fall in bank
lending while commercial paper issuance is
rising might suggest that bank loan supply is
contracting. To consider this potential co
movement, KSW focus on changes over time
in the mix between bank loans and commer
cial paper (defined as bank loans divided by
the sum of bank loans and commercial paper).
They find that, in response to increases in
the federal funds rate (or, less continuously,
at the times of the contractionary policy
shifts identified by Romer and Romer, 1989),
the volume of commercial paper issues rises,
while bank loans gradually decline. They
also find that policy-induced changes in the
mix have independent predictive power for
inventory and fixed investment, holding con
stant other determinants.19
The aggregate story told by KSW masks
significant firm-level heterogeneity, however.
The burden of a decline in bank loans fol
lowing a monetary contraction is borne by
smaller firms (see Gertler and Gilchrist, 1994).“
Moreover, the evidence in Oliner and
Rudebusch (1 993) indicates that once trade
credit is incorporated in the definition of
small firms’ debt and once firm size is held
least a portion of their loan portfolio, adverse
selection problems will increase the marginal
cost o f external finance as more funds are
raised (see, for example, Myers and Majluf,
1984; and Lucas and McDonald, 1991). In
addition, as long as some banks face constraints
on issuing CDs and those banks lead to
bank-dependent borrowers, a bank lending
channel will be operative.
W hile the foregoing discussion centers
on open market operations, regulatory actions
by the central bank— credit controls, for
example— represent another way in which
monetary policy can have real effects through
influencing the spending decisions of bankdependent borrowers. Here the effects are
likely to be more pronounced than for the
case of open market operations, since the
question of the cost of non-deposit sources
of funds is no longer central, and the effec
tiveness of such regulatory actions depends
only on the existence of bank-dependent
borrowers.
G O IN G FROM MODELS TO
EMPIRICAL RESEARCH
Both the financial-constraints-on-borrowers and bank-lending-channel mechanisms
imply significant cross-sectional differences
in firms’ shadow cost o f finance and in the
response of that cost to policy-induced changes
in interest rates. Accordingly, empirical
researchers have attempted to test these
cross-sectional implications. As I examine
this literature, I explore how Modigliani-Miller
violations for nonfinancial borrowers, financial
intermediaries or both offer channels for
monetary policy beyond effects on interest
rates. The appendix frames this discussion
using a simple model; an intuitive presenta
tion follows.
EMPIRICAL RESEARCH ON
THE CREDIT VIEW
Studies Using Aggregate Data
The microeconomic underpinnings of
both financial accelerator models and the
credit view of monetary policy hinge on
certain groups of borrowers (perhaps including
1995
67
19 Oliner and Rudebusch (1 9 9 3 ) and
Friedman and Kuttner (1 9 9 3 ) have
disputed the KSW interpretation of
the mix as measuring a substitution
between bank loans and commer
cial paper. They orgue that, during
a recession, shifts in the mix are
explained by an increase in com
mercial paper issuance rather than
by o decrease in bank loans.
20 Morgan (1 9 9 3 ) finds a similar
result in an analysis of loon com
mitments. After an episode of
monetary contraction, firms without
loan commitments receive a small
er share of bank loons.
REVIEW
MAY/IUNE
constant, monetary policy changes do not
alter the mix.
It also does not appear that bank-dependent borrowers switch to the commercial
paper market following a monetary contrac
tion. Instead, the increase in commercial
paper issuance reflects borrowing by large
firms with easy access to the commercial
paper market, possibly to smooth fluctua
tions in their flow of funds when earnings
decline (Friedman and Kuttner, 1993) or to
finance loans to smaller firms (Calomiris,
Himmelberg and Wachtel, forthcoming).
two-year period following a monetary tight
ening, results consistent with the financial
accelerator approach. They also demonstrate
that the effects of shifts in monetary policy
on the small-firm variables are sharper in
periods when the small-firm sector as a whole
is growing more slowly, also consistent with
the financial accelerator approach. Finally,
they show that the ratio of cash flow to interest
expense (a measure of debt-service capacity)
is associated positively with inventory accu
mulation for small, but not for large, manu
facturing firms.
The Gertler and Gilchrist results, which
are very much in the spirit of the earlier
cross-sectional tests of financial accelerator
models, have been borne out for studies of
fixed investment by Oliner and Rudebusch
(1 9 9 4 ) and for inventory investment by
Kashyap, Lamont and Stein (1994).21 In addi
tion, Ramey (1 993) shows that, for forecast
ing purposes, the ratio of the sales growth of
small firms to that for large firms offers sig
nificant information about future GDR
Finally, using the firm-level data
underlying the aggregates summarized in
the Quarterly Financial Reports, Bernanke,
Gertler and Gilchrist (forthcoming) analyze
the differences in sales and inventories between
large and small manufacturing firms by
two-digit industry. They find that fluctua
tions in the large firm-small firm differences
are roughly the same size as fluctuations in
the corresponding aggregate fluctuations for
the manufacturing sector. Because small firms’
sales (as they define small firms) comprise
about one-third of the sales of the manufac
turing sector, roughly one-third of cyclical
fluctuations in manufacturing sales can be
explained by large firm-small firm differences.
Studies Focusing on Cross-Sectional
Implications
21 Toword this end, more direct com
parisons of borrowing by bankdependent and nonbank-dependent
borrowers have been offered.
Using firm-level data, Kashyap,
Lomont and Stein (1 9 9 4 )—
henceforth KLS— follow the
Fazzari, Hubbard ond Petersen
(1 9 8 8 ) approach of classifying
groups of firms os a priori financeconstrained (in this case, bankdependent) or not. In particular,
they study inventory investment by
publicly traded firms with ond with
out bond ratings, as a proxy for
bank dependence. Focusing on the
1982 recession (as an indirect
means of identifying a period fol
lowing a tight money episode),
they find that inventory investment
by non-roted firms wos influenced,
all else equal, by the firms' own
cash holdings, an effect not present
for the inventory investment by
rated firms. In subsequent boom
years (which KLS identify with an
easy money episode), they find lit
tle effect of cosh holdings on inven
tory investment for either non-roted
or rated companies. These patterns
lead KLS to conclude that a bank
lending channel was operative in
response to the monetary contrac
tion. However, the KLS results are
consistent with a more general
model in which low-net-worth firms
foce more costly external finance in
downturns.
More convincing empirical tests focus
on the cross-sectional implication of the
underlying theories— namely that creditmarket imperfections affect investment,
employment or production decisions of some
borrowers more than others. At one level,
existing cross-sectional empirical studies have
been successful: There is a substantial body
of empirical evidence documenting that proxies
for borrowers’ net worth affect investment
more for low-net-worth borrowers than for
high-net-worth borrowers (holding constant
investment opportunities). This suggests
that, to the extent that monetary policy can
affect borrowers’ net worth, pure interest
rate effects of open market operations will
be magnified.
The second body of empirical analysis of
information-related imperfections focuses on
the effects of monetary policy on borrowers’
balance sheets. Gertler and Hubbard (1 9 8 8 )
conclude that, all else equal, internal funds
have a greater effect on investment by nondividend-paying firms during recessions.
The evidence of Gertler and Gilchrist (1994)
is particularly compelling here. Analyzing
the behavior of manufacturing firms summa
rized in the Quarterly Financial Reports data,
Gertler and Gilchrist consider differences in
small and large firms’ responses to tight
money (as measured by federal funds rate
innovations or the dates identified by Romer
and Romer, 1989). In particular, small firms’
sales, inventories and short-term debt
decline relative to those for large firms over a
1995
Assessing the Bank Lending
Channel
W hile the principal empirical predictions
of the financial accelerator approach have been
corroborated in micro-data studies and lownet-worth firms appear to respond differen
tially to monetary contractions, the question
of the role of banks remains. I consider this
question below in three steps.
First, is there evidence of significant
68
REVIEW
MAT/JUNE
emerges when loans are disaggregated to
include ju st commercial and industrial loans.
One possible explanation for the Kashyap
and Stein pattern is that a monetary contrac
tion weakens the balance sheet positions of
small firms relative to large firms. If small
firms tend to be the customers of small banks
and large firms tend to be the customers of
large banks, a fall in loan demand (by small
borrowers) for small banks could be consis
tent with the differential lending responses
noted by Kashyap and Stein. To examine
this possibility, Kashyap and Stein analyze
whether small banks increase their holdings
of securities relative to large banks during a
monetary contraction. They actually find
that small banks’ securities holdings are less
sensitive to monetary policy than large banks’
securities holdings, though the difference in
the responses is not statistically significant.
The use of bank size as a measure to gen
erate cross-sectional differences does not corre
spond precisely to the underlying theoretical
models, which stress the importance of net
worth. In this context, bank capital may be
a better proxy. Peek and Rosengren (forth
coming) analyze the lending behavior of New
England banks over the 1990-91 recession.
Their results indicate that the loans of well
capitalized banks fell by less than the loans
of poorly capitalized banks.22 Hence, as with
the Kashyap and Stein findings, their evidence
suggests there are effects of informational
imperfections in financial markets on the
balance sheets of intermediaries as well
as borrowers.
departures from Modigliani and Miller’s
results for certain groups of banks in the
sense that have been identified for firms?
Second, is there evidence that small- or lownet-worth firms are more likely to be the
loan customers of such banks? Finally, do
low-net-worth firms have limited opportuni
ties to substitute credit from unconstrained
financial institutions when cut off by con
strained financial institutions?
Applying the Modigliani and Miller
Theorem for Banks
Kashyap and Stein (1994) apply the
intuition of the models o f effects of internal
net worth on investment decisions by nonfinancial firms to study financing and lending
decisions by banks. This is an important
line of inquiry in the bank lending channel
research agenda, because it addresses the ease
with which banks can alter their financing
m ix in response to a change in bank reserves
and the effect of changes in the financing
m ix on the volume o f bank lending. Ju st as
earlier studies focused on cross-sectional dif
ferences in financing and real decisions of
nonfinancial firms of different size, Kashyap
and Stein analyze cross-sectional differences
in financing and lending decisions of banks
of different size. To do this, they use data
drawn from the quarterly “Call Reports”
collected by the Federal Reserve.
Kashyap and Stein construct asset size
groupings for large banks (those in the 99th
percentile) and small banks (defined as those
at or below the 75th, 90th, 95th or 98th
percentiles). They first show that contrac
tionary monetary policy (measured by an
increase in the federal funds rate) leads to
a similar reduction in the growth rate of nom
inal core deposits for all bank size classes.
They find significant heterogeneity across
bank size classes, however, in the response
of the volume of lending to a change in mon
etary policy. In particular, a monetary con
traction leads to an increase in lending in
the short run by very large banks. This is in
contrast to a decline in lending in the short
run by smaller banks. These do not simply
reflect differences in the type of loans made
by large and small banks. A similar pattern
1995
Matching Borrowers and Lenders
The last two questions relate to the
matching of borrowers and lenders. The
former asks whether the firms identified by
empirical researchers as finance-constrained
are the loan customers of the constrained
(small) banks such as those identified by
Kashyap and Stein. This line o f inquiry
requires an examination of data on individ
ual loan transactions, with information on
characteristics of the borrower, lender and
lending terms.23 One could establish whether
constrained firms are the customers of con
strained banks and whether such firms
69
21 Using data on commercial bonks
nationwide over the 1979-92 period,
Berger ond Udell (1 9 9 4 ) found lit
tle evidence that the introduction of
risk-based capital requirements per se
affected credit allocation. Hancock,
Loing ond Wilcox (1 9 94 ) also use
quarterly dato on individual bank's
portfolios to estimate the respon
siveness of portfolio composition to
changes in capital requirements.
They find that "capital shortfall"
institutions reduced their C&l loans
response by larger total amounts,
oil else equal, than "capital surplus"
institutions.
13 Anil Kashyap, Darius Palia and I ore
currently engaged in such an analysis.
REVIEW
MAY/JUKI
24 Another possibility is that the
weakened balance sheet positions
of mony borrowers precipitates
a "flight to quality'by lenders
generally, increasing the demand
for commercial paper issues of
large firms.
25 The dates of monetary policy
contractions suggested by Romer
and Romer (1 9 89 ) have generated
significant controversy. Shapiro
(1 9 9 4 ) argues, for example,
that empirical evidence favors the
hypothesis that several Romer dates
ore predictable using measures
of unemployment and inflation
as determinants of actions by the
Federal Open Market Committee;
see also the discussion in Cecchetti
(1 9 9 5 ). Hoover and Perez (1994)
offer a number of criticisms of the
Romers' approach.
24 Such relationships are typically
estimated as:
Y(t) = a + b Y (t- i)
- c H (t-i) - d F lt -i) ,
where Pis the percentage change
in real GDP relative to potential
GDP, H is the percentage change in
the high-employment federal bud
get surplus, F is the change in the
federal funds rate, f is the current
time period, and / denotes lags.
See, for example, Hirtle ond Kelleher
(1 9 90 ), Perry ond Schultze (1992)
ond Cohen and Wenninger (1 9 9 3 ).
17 Cover (1 9 9 2 ) finds still stronger
evidence of asymmetric effects
when monetary aggregates ore
used as the policy indicator instead
of the federal funds rate.
ponents of outside money over which the
central bank can exercise exogenous control.
First identifying exogenous changes in mone
tary policy is difficult.25 Recent research by
Bernanke and Blinder (1 9 9 2 ) and Christiano,
Eichenbaum and Evans (forthcoming) offers
promising strategies for studying the effects
of monetary policy shocks.
In addition, recent analyses of policyreduced-form models document a significant,
negative relationship in quarterly data between
the percentage change in real GDP relative to
potential GDP and the change in the federal
funds rate.26 Such studies must first confront
the possibility that the measured interest
sensitivity of output reflects links between
interest rate and net worth changes for certain
groups of borrowers/spenders. A second
issue, noted by Morgan (1 993) and Cohen
and Wenninger (1 9 9 3 ), is that quarterly
residuals from estimated policy-reduced-form
equations display large negative errors during
recessions, suggesting the possibility of an
asymmetric response of econom ic activity to
increases or decreases in the federal funds
rate.27 Finally, more theoretical and empirical
research is needed to examine links between
changes in short-term real interest rates (which
are significantly influenced by policy actions)
and changes in long-term real interest rates
(which affect firms’ cost of capital).
switch from constrained banks to uncon
strained ones during episodes of monetary
contractions. Theories emphasizing the impor
tance of ongoing borrower-lender relationships
imply that such switches are costly and unlikely.
If true, part of the monetary transmission
mechanism takes place through reductions
in loan supply by constrained banks.
The latter of the two questions suggests
the need to study a broader class of lenders
than banks. If borrowers from constrained
banks can switch at low cost to nonbank
lenders following a monetary contraction,
the narrow bank credit channel of monetary
policy is frustrated. In this vein, Calomiris,
Himmelberg and Wachtel (forthcoming) ana
lyze firm-level data on commercial paper
issuance and argue that large, high-quality
commercial paper-issuing firms increase
paper borrowings during downturns to
finance loans to smaller firms.” They note
that accounts receivable rise for paper-issu
ing firms, supporting the notion that these
firms may serve as trade credit intermedi
aries for smaller firms in some periods.
From the standpoint of the bank lending
channel, it is important to establish what
happens to the costs and terms imposed by
these intermediaries. If, on the one hand,
such terms are no more costly than bank
intermediary finance, then the switch of bor
rowers from being bank customers to being
trade credit customers entails very limited
macroeconomic effects. On the other hand,
if large, paper-issuing firms accept their
intermediary role reluctantly, very costly
trade credit may exacerbate a downturn
by raising the cost of funds for constrained
firms. More empirical investigation of
trade credit terms is needed to resolve
this question.
CONCLUSION AND IMPLI
CATIONS FOR M ONETARY
POLICY
This survey argues that the terms money
view and credit view are not always welldefined in theoretical and empirical debates
over the transmission mechanism of m one
tary policy. Recent models of information
and incentive problems in financial markets
suggest the usefulness of decomposing the
transmission mechanism into two parts: one
related to effects of policy-induced changes
on the overall level of real costs of funds; and
one related to magnification (or financial
accelerator effects) stemming from impacts
of policy actions on the financial positions of
borrowers and/or intermediaries.
Two observations emerge clearly from
the literature. First, the spending decisions
Empirical Research on Conventional
Interest Rate Channels
More empirical research is also needed
to assess the validity of the basic money
view. A central problem is that, while most
empirical studies focus on monetary aggre
gates such as M2, the theoretical description
offered in the first section suggests an emphasis
on outside money and, importantly, on com
1993
70
REVIEW
MAY/JUNI
Berger, Allen N., and Gregory F. Udell. "Did Risk-Based Capital Allocate
Bank Credit and Cause a 'Credit Crunch' in the U.S.?" (part two)
Journal of Money, Credit and Banking (August 1994), pp. 585-628.
of a significant group of borrowers are influ
enced by their balance sheet condition in the
ways described by financial accelerator models.
Second, even in the presence of more sophis
ticated financial arrangements, there are still
information costs of screening, evaluation
and monitoring in the credit process, impart
ing a special role for intermediaries (be they
banks or other lenders) with cost advantages
in performing these tasks.21
The first observation suggests that
financial factors are likely to continue to play
a role in business fluctuations. The second
suggests that regulatory policies affecting
information-specializing intermediaries are
likely to affect the cost of credit for at least
some borrowers. In part because of interest
in alternative views of the monetary trans
mission mechanism and in part because of
concern over the effects of institutional change
in the financial system, academics and poli
cymakers are analyzing whether the scope
for monetary policy to affect real outcomes
is becoming narrower. Both observations
noted above are consistent with a heightened
role for monetary policy in affecting real
decisions of firms with weak balance sheet
positions. Developing ways to incorporate
borrower heterogeneity in both economic
models of money and credit and in forecasting
is an important, practical task for economic
modelers and policymakers.
W hether the simplest bank lending
channel— that a fall in banks’ reserves fol
lowing contractionary open market operations
decreases both banks’ ability to lend and bor
rowers’ ability to spend— is operative is not
clear, however. More micro-evidence at the
level of individual borrower-lender transac
tions is needed to resolve this question. At
the same time, proponents of the simplest
characterization of an interest rate channel
must address both the cross-sectional hetero
geneity in firms’ response to monetary policy
and the extent to which observed interest rate
effects on output reflect differentially large
effects of policy on certain classes of borrowers.
Bernanke, Ben S. "Credit in the Macroeconomy," Federal Reserve Bank
of New York Quarterly Review (spring 1993), pp. 50-70.
_ _ _ _ _ _ . "Nonmonetary Effects of the Financial Crisis in the
Propagation of the Great Depression," The American Economic Review
(June 1983), pp. 257-76.
_ _ _ _ _ _ and Alan S. Blinder. "The Federal Funds Rote and the
Channels of Monetary Transmission," The American Economic Review
(September 1992), pp. 901-21.
_ _ _ _ _ _ and_ _ _ _ _ _ _ . "Credit, Money, and Aggregate
Demand," Jhe American Economic Review (May 1988), pp. 435-39.
_ _ _ _ _ _ and Mark Gertier. "Agency Cost, Net Worth, and Business
Fluctuations," The American Economic Review (March 1989), pp. 14-31.
_ _ _ _ _ _ and_ _ _ _ _ _ _ . "Financial Fragility and Economic
Performance," Quarterly Journal of Economics (February 1990),
pp. 87-114.
_ _ _ _ _ _ , _ _ _ _ _ _ and Simon Gilchrist. "The Financial
Accelerator and the Flight to Quality," Review of Economics and
Statistics (forthcoming).
_ _ _ _ _ _ and Harold Jomes. "The Gold Standard, Deflation, and
Financial Crisis in the Great Depression: An International Comparison,"
in R. Glenn Hubbard, ed., Financial Markets and Financial Crises.
University of Chicago Press, 1991.
_ _ _ _ _ _ and Cara S. Lown. "The Credit Crunch." Brookings Papers
on Economic Activity (1992:2), pp. 205-39.
Bizer, David S. "Regulatory Discretion and the Credit Crunch," working
paper (April 1993), U.S. Securities and Exchange Commission.
Blinder, Alan S., ond Joseph E. Stiglitz. "Money, Credit Constraints, and
Economic Activity," The American Economic Review (May 1983), pp.
297-302.
Bond, Stephen, and Costas Meghir. "Dynamic Investment Models ond
the Firm's Financial Policy," Review of Economic Studies (April 1994),
pp. 197-222.
Boyd, John H., and Mork Gertier, "Are Banks Dead? Or, Are the Reports
Greatly Exoggerated?" working paper (May 1994), Federal Reserve
Bank of Minneapolis.
_ _ _ _ _ _ and Edward Prescott. "Financial Intermediary-Coalitions,"
Journal of Economic Theory (April 1986), pp. 211-32.
_ _ _ _ _ _ and Bruce D. Smith. "Capital Market Imperfections in a
Monetary Growth Model," Working Paper No. 533 (August 1994),
Federal Reserve Bank of Minneapolis.
Brainard, William C., and James Tobin. "Financial Intermediaries and
the Effectiveness of Monetary Controls," The American Economic
Review (May 1963), pp. 383-400.
REFERENCES
Akerlof, George. "The Market for Lemons: Quality Uncertainty and the
Market Mechanism," Quarterly Journal of Economics (August 1970),
pp. 488-500.
199
Cogan, Phillip. The Channels of Monetary Effects on Interest Rotes.
Columbia University Press, 1972.
71
28 For recent analyses of the future of
banking, see Gorton ond Pennacchi
(1 9 9 3 ), Edwards (1 9 93 ) and
Boyd ond Gertier (1 9 9 4 ).
Thornton (1 9 94 ) discusses likely
effects of recent financial innova
tions on the bonk lending channel.
REVIEW
MAY/JUNE
Calomiris, Charles W., Charles P. Himmelberg and Paul Wachtel.
"Commercial Paper and Corporate Finance: A Microeconomic
Perspective," Cornegie-Rochester Conference Series on Public Policy
(forthcoming).
1995
Eckstein, Otto, and Allen Sinai. "The Mechanisms of the Business Cycle
in the Postwar Era," in Robert J. Gordon, ed., The American Business
Cycle: Continuity and Change. University of Chicago Press, 1986.
Edwards, Franklin R. "Financial Markets in Transition— Or the Decline
of Commercial Banking," in Changing Capital Markets: Implications for
Monetary Policy. Federal Reserve Bank of Kansas City, 1993.
_ _ _ _ _ _ and R. Glenn Hubbard. "Firm Heterogeneity, Internal
Finance, and 'Credit Rationing,'" Economic Journal (March 1990),
pp. 90-104.
Estrella, Arturo, and Gikas Hordouvelis. "Possible Roles of the Yield
Curve in Monetary Policy," in Intermediate Targets and Indicators of
Monetary Policy: A Critical Survey, 1990. Federal Reserve Bank of
New York, 1990, pp. 339-62.
_ _ _ _ _ _ and_ _ _ _ _ _ _ . "Price Flexibility, Credit Availability, and
Economic Fluctuations: Evidence from the United States, 1894-1909,"
Quarterly Journal of Economics (August 1989), pp. 429-52.
Fama, Eugene F. "Banking in the Theory of Finance," Journal of
Monetary Economics (January 1980), pp. 39-57.
_ _ _ _ _ _ and_ _ _ _ _ _ _ . "Tax Policy, Internal Finonce, and
Investment: Evidence from the Undistributed Profits Tax of 1936-37,"
Journal of Business (forthcoming).
Fomo, Eugene F. "What's Different About Banks?" Journal of Monetary
Economics (June 1985), pp. 29-39.
_ _ _ _ _ _ and Charles Kahn. "The Role of Demandable Debt in
Structuring Optimal Banking Arrangements," The American Economic
Review (June 1991), pp. 497-513.
Farmer, Roger E.A. "Implicit Contracts with Asymmetric Information ond
Bankruptcy: The Effect of Interest Rotes on Layoffs," Review of
Economic Studies (July 1985), pp. 427-42.
Cantor, Richard. "Effects of Leverage on Corporate Investment and
Hiring Decisions." Federal Reserve Bank of New York Quarterly
Review (summer 1990), pp. 31-41.
Fazzari, Steven M., and Bruce C. Petersen. "Working Capital and Fixed
Investment: New Evidence on Financing Constraints." RAND Journal
of Economics (outumn 1993), pp. 328-42.
_ _ _ _ _ _ and Anthony P. Rodrigues. "Nonbank Lenders and
the Credit Slowdown," in Studies on Couses and Consequences
of the 1984-1992 Credit Slowdown. Federal Reserve Bank of
New York, 1994.
Fazzari, Steven M., R. Glenn Hubbard, and Bruce C. Petersen.
"Financing Constraints and Corporate Investment." Brookings Papers
on Economic Activity (1988:1), pp. 141-95.
Carpenter, Robert, Steven M. Fazzari and Bruce C. Petersen. "Inventory
(Dis)investment, Internal Finance Fluctuations, and the Business
Cycle," Brookings Papers on Economic Activity (1994:2).
Federal Reserve Bonk of New York. Studies on Causes and
Consequences of the 1989-1992 Credit Slowdown, February 1994.
Fisher, Irving. "The Debt-Deflation Theory of Great Depressions."
Econometrico 1 (October 1933), pp. 337-57.
Cecchetti, Stephen G. "Distinguishing Theories of the Monetary
Transmission Mechanism," this Review (May/June 1995),
pp. 83-100.
Friedman, Benjamin M., and Kenneth N. Kuttner. "Economic Activity
and the Short-Term Credit Markets: An Analysis of Prices and Quantities,"
Brookings Papers on Economic Activity (1993:2), pp. 193-266.
Chirinko, Robert S. "Business Fixed Investment Spending: A Critical Survey,"
Journal of Economic literature (December 1993), pp. 1875-911.
Christiano, Lawrence, and Mortin S. Eichenbaum. "Liquidity Effects ond
the Monetary Transmission Mechanism," The American Economic
Review (May 1992), pp. 346-53.
Gerrier, Mark. "Comment" on Christina D. Romer and Dovid H. Romer,
"Credit Channel or Credit Actions?: An Interpretation of the Postwar
Transmission Mechanism," in Changing Capital Markets: Implications
for Monetary Policy. Federal Reserve Bank of Kansas City, 1993.
_ _ _ _ _ _ , _ _ _ _ _ _ , and Chorles Evans. "The Effects of Monetary
Policy Shocks: Evidence from the Flow of Funds," Review of
Economics and Statistics (forthcoming).
_ _ _ _ _ _ . "Financial Capacity and Output Fluctuations in on
Economy with Multi-period Financial Relationships," Review of
Economic Studies (1992), pp. 455-72.
Cohen, Gerald, and John Wenninger. "The Relationship Between the
Federal Funds Rate and Economic Activity," working paper (December
1993), Federal Reserve Bank of New York.
_ _ _ _ _ _ . "Financial Structure and Aggregate Economic Activity: An
Overview," Journal of Money, Credit and Banking (August 1988, part 2),
pp. 559-88.
Cover, John Perry. "Asymmetric Effects of Positive and Negative Money
Supply Shocks," Quarterly Journal of Economics (November 1992),
pp. 1261-82.
_ _ _ _ _ _ and Simon Gilchrist. "Monetary Policy, Business Cycles, and
the Behavior of Small Manufacturing Firms," Quarterly Journal of
Economics (May 1994), pp. 309-40.
Cummins, Joson G., Kevin A. Hassett and R. Glenn Hubbard. "A
Reconsideration of Investment Behavior Using Tax Reforms as Natural
Experiments," Brookings Papers on Economic Activity (1994:2),
pp. 1-74.
_ _ _ _ _ _ and_ _ _ _ _ _ _ . "The Role of Credit Market Imperfections
in the Monetary Transmission Mechanism: Arguments and Evidence,"
Scandinavian Journal of Economics (No. 1, 1993), pp. 43-64.
Gertler, Mark, and R. Glenn Hubbard. "Corporate Financial Policy,
Taxation, and Macroeconomic Risk." RAND Journal of Economics
(summer 1993), pp. 286-303.
Diamond, Douglas W. "Financial Intermediation and Delegated
Monitoring," Review of Economic Studies (July 1984), pp. 393-414.
72
HfVIEN
MAY/JUNK
_ _ _ _ _ _ and_ _ _ _ _ _ _ . "Financial Factors in Business
Fluctuations," Financial Market Volatility. Federal Reserve Bank of
Kansas City, 1988.
199
James, Christopher. "Some Evidence on the Uniqueness of Bank Loans,"
Journal of Financial Economics (1987), pp. 217-36.
Jensen, Michael, and William Meckling. "Theory of the Firm:
Management Behavior, Agency Costs, and Ownership Structure,"
Journal of Financial Economics (October 1976), pp. 305-60.
Gorton, Gary, and George Pennacchi. "Money Market Funds and Finance
Companies: Are They Banks of the Future?" in Michael Klausner and
Lawrence J. White, eds., Structural Change in Banking. Irwin Publishing,
1993, pp. 173-214.
Kashyap, Anil K., and Jeremy C. Stein. "Monetary Policy and Bonk
Lending," in N. Gregory Mankiw, ed., Monetary Policy. University of
Chicago Press for the National Bureau of Economic Research, 1994a,
pp. 221-62.
Greenwald, Bruce C., and Joseph E. Stiglitz. "Financial Market
Imperfections and Business Cycles," Quarterly Journal of Economics
(February 1993), pp. 77-114.
_ _ _ _ _ _ and_ _ _ _ _ _ _ . "The Impact of Monetary Policy on Bank
Balance Sheets," Carnegie-Rochester Conference Series on Public
Policy (forthcoming).
_ _ _ _ _ _ and_ _ _ _ _ _ _ . "Information, Finance Constraints, and
Economic Activity," in Meir Kohn and S.C. Isiang, eds., Finance
Constraints, Expectations, and Economic Activity. Oxford University
Press, 1988.
_ _ _ _ _ _ , _ _ _ _ _ _ and David Wilcox. "Monetary Policy and
Credit Conditions: Evidence from the Composition of External
Finance," The American Economic Review (March 1993), pp. 78-98.
Gurley, John, and Edward Shaw. Money in a Theory of Finance.
Brookings Institution, 1960.
_ _ _ _ _ _ and_ _ _ _ _ _ _ . "Financial Aspects of Economic
Development," The American Economic Review (September 1955),
pp. 515-38.
_ _ _ _ _ _ , Owen A. Lamont and Jeremy Stein. "Credit Conditions and
the Cyclical Behavior of Inventories: A Case Study of the 1981-82
Recession," Quarterly Journal of Economics (August 1994), pp. 56592.
Hancock, Diana, Andrew J. Laing and James A. Wilcox. "Bank Capital
Shocks: Dynamic Effects on Securities, Loans, and Capital," working
paper (August 1994), Board of Governors of the Federal Reserve
System.
King, Robert G., and Charles Plosser. "Money, Credit and Prices in a
Real Business Cycle," The American Economic Review (June 1984),
pp. 363-80.
_ _ _ _ _ _ and Ross Levine. "Finance and Growth: Schumpeter Might
Be Right," Quarterly Journal of Economics (August 1993), pp. 717-38.
Hardouvelis, Gikas, and Thierry Wizman. "The Relative Cost of Capital
for Marginal Firms Over the Business Cycle," Federal Reserve Bank of
New York Quarterly Review (autumn 1992), pp. 59-68.
Kiyotaki, Nobuhiro, and John Moore. "Credit Cycles." mimeo (March
1993), London School of Economics.
Hart, Oliver, and John Moore. "A Theory of Debt Based on the
Inalienability of Human Capital," Discussion Paper No. 129 (1991),
London School of Economics, Financial Markets Group.
Kliesen, Kevin L., and John A. Tatom. "The Recent Credit Crunch: The
Neglected Dimensions," this Review (September/October 1992),
pp. 18-36.
Hirtle, Beverly, and Jeanette Kelleher. "Financial Market Evolution and
the Interest Sensitivity of Output." Federal Reserve Bank of New York
Quarterly Review (summer 1990), pp. 56-70.
Lamont, Owen S. "Corporate Debt Overhang and Macroeconomic
Vulnerability," working paper (1993), Massachusetts Institute of
Technology.
Hoover, Kevin D., and Stephen J. Perez. "Post Hoc Ergo Propter Once More:
An Evaluation of 'Does Monetary Policy Matter?' in the Spirit of James
Tobin," Journal of Monetary Economics (August 1994), pp. 47-73.
Lucas, Robert E., Jr. "Liquidity and Interest Rates," Journal of Economic
t o y (1990), pp. 237-64.
Hoshi, Takeo, Anil K. Kashyap and David Scharfstein. "Corporate Structure,
Liquidity and Investment: Evidence from Japanese Panel Data,"
Quarterly Journal of Economics (February 1991), pp. 33-60.
Lucas, Deborah J., and Robert L. McDonald. "Bonk Financing and
Investment Decisions with Asymmetric Information About Loan
Quality," RAND Journal of Economics (winter 1991), pp. 86-105.
Hubbard, R. Glenn. Money, the Financial System and the Economy.
Addison-Wesley, 1994.
Lummer, Scott, and John McConnell. "Further Evidence on the Bank
Lending Process and Capital Market Response to Bank Loan
Agreements," Journal of Financial Economics (1989), pp. 99-122.
_ _ _ _ _ _ and Anil K. Kashyap. "Internal Net Worth and the
Investment Process: An Application to U.S. Agriculture," Journal of
Political Economy (June 1992), pp. 506-34.
Mauskopf, Eileen. "The Transmission Channels of Monetary Policy:
How They Have Changed," Federal Reserve Bulletin (December
1990), pp. 985-1008.
_ _ _ _ _ _ , _ _ _ _ _ _ and Toni Whited. "Internal Finance and Firm
Investment," Journal of Money, Credit and Bonking (forthcoming).
Minsky, Hyman P John Maynard Keynes. Columbia University Press, 1975.
Jaffee, Dwight, and Thomas Russell. "Imperfect Information,
Uncertainty and Credit Rationing," Quarterly Journal of Economics
(November 1976), pp. 651-66.
_ _ _ _ _ _ . "Financial Crisis, Financial Systems, and the Performance
of the Economy," in Commission on Money and Credit, Private Capital
Markets. Prentice-Hall, 1964.
73
REVIEW
MAY/JUNE
Mishkin, Frederic S. "The Household Balance Sheet and the Great
Depression," Journal of Economic History (1978), pp. 918-37.
_ _ _ _ _ _ and_ _ _ _ _ _ _ . "Does Monetary Policy Matter?: A New
Test in the Spirit of Friedman and Schwartz," in Olivier J. Blanchard
and Stanley Fischer, eds., National Bureau of Economic Research
Macroeconomics Annual. MIT Press, 1989, pp. 121-70.
Mishkin, Frederic S. "What Depressed the Consumer?: The Household
Balance Sheet and the 1973-75 Recession," Brookings Papers on
Economic Activity (1977:1), pp. 123-64.
Roosa, Robert V. "Interest Rates ond the Central Bank," in Money,
Trade, and Economic Growth: Essays in Honor of John H. Williams.
Macmillan, 1951.
Modigliani, Franco, and Merton Miller. "The Cost of Capital, Corporation
Finance and the Theory of Investment," The American Economic
Review (June 1958), pp. 261-97.
Rotemberg, Julio J. "A Monetary Equilibrium Model with Transactions
Costs," Journal of Political Economy (February 1984), pp. 40-58.
Morgan, Donald P. "The Lending View of Monetary Policy and Bank
Loan Commitments." working paper (July 1993), Federal Reserve
Bank of Kansas City.
Rothschild, Michael, and Joseph E. Stiglitz. "Equilibrium in Competitive
Insurance Markets: An Essay on the Economics of Imperfect
Information," Quarterly Journal of Economics (November 1976),
pp. 630-49.
_ _ _ _ _ _ . "Asymmetric Effects of Monetary Policy." Federal Reserve
Bank of Kansas City Economic Review (second quarter 1993), pp. 21-33.
Schreft, Stocey. "Credit Controls: 1980," Federal Reserve Bank of
Richmond Economic Review (November-December 1990), pp. 25-55.
Myers, Stewart, and Nicholas Majluf. "Corporate Financing and
Investment Decisions When Firms Have Information that Investors Do
Not Have," Journal of Financial Economics (1984), pp. 187-221.
Shapiro, Matthew D. "Federal Reserve Policy: Cause and Effect," in N.
Gregory Mankiw, ed., Monetary Policy. University of Chicago Press,
1994, pp. 207-34.
Oliner, Stephen D., and Glenn D. Rudebusch. "Is There a Bank Credit
Channel for Monetary Policy?" Board of Governors of the Federal
Reserve System, Finance and Economics Division Discussion Paper No.
93-8 (February 1993).
Sharpe, Steven. "Asymmetric Information, Bank Lending, and Implicit
Contracts: A Stylized Model of Customer Relationships," Journal of
Finance (September 1990), pp. 1069-87.
_ _ _ _ _ _ and_ _ _ _ _ _ _ . "Sources of the Financing Hierarchy for
Business Investment," Review of Economics and Statistics (November
1992), pp. 643-54.
Stiglitz, Joseph E., and Andrew Weiss. "Credit Rationing in Markets with
Imperfect Information," The American Economic Review (June 1981),
pp. 393-410.
Owens, Raymond, ond Stacey Schreft. "Identifying Credit Crunches,"
Federal Reserve Bank of Richmond Working Paper No. 92-1
(March 1992).
Thornton, Daniel L. "Financial Innovation, Deregulation and the "Credit
View" of Monetary Policy," this Review (January/February 1994),
pp. 31-49.
Peek, Joe, and Eric Rosengren. "The Capital Crunch: Neither a Borrower
Nor a Lender Be," Journal of Money, Credit and Banking (forthcoming).
Townsend, Robert. "Optimal Contracts and Competitive Markets with
Costly State Verification," Journal of Economic Theory (October 1979),
pp. 265-93.
_ _ _ _ _ _ and_ _ _ _ _ _ _ . "The Capital Crunch in New England."
Federal Reserve Bonk of Boston New England Economic Review
(May-June 1992), pp. 21-31.
Williamson, Stephen. "Costly Monitoring, Optimal Contracts, and
Equilibrium Credit Rationing," Quarterly Journal of Economics
(February 1987), pp. 135-45.
Perry, George L., and Charles L. Schultze. "Was This Recession Different?
Are They All Different?" Brookings Papers on Economic Activity
(1993:1), pp. 145-211.
Wojnilower, Albert. "The Central Role of Credit Crunches in Recent
Financial History," Brookings Papers on Economic Activity (1980:2),
pp. 277-326.
Petersen, Mitchell A., and Raghuram G. Rajan. "The Benefits of FirmCreditor Relationships: Evidence from Small Business Data," Journal of
Finance (March 1994), pp. 3-38.
Zeldes, Stephen P. "Consumption and Liquidity Constraints: An
Empirical Investigation," Journal of Political Economy (April 1989),
pp. 305-46.
Ramey, Valerie. "How Important is the Credit Channel in the Transmission
of Monetary Policy?" Carnegie-Rochester Conference Series on Public
Policy (December 1993), pp. 1-45.
Romer, Christina D., and David H. Romer. "Credit Channel or Credit
Actions?: An Interpretation of the Postwar Transmission Mechanism,"
in Changing Capital Markets: Implications for Monetary Policy.
Federal Reserve Bank of Kansas City, 1993.
_ _ _ _ _ _ and_ _ _ _ _ _ _ . "New Evidence on the Monetary
Transmission Mechanism," Brookings Papers on Economic Activity
(1990:1), pp. 149-213.
1995
74
REVIEW
MAY/JUNK
1995
A p p e n d ix
THE FINANCIAL ACCELERATOR AN D THE CREDIT VIEW
cave (where/(0) = 0 ,/ '(0 ) =
and/'(z) —> 0
as z —> « ) ; Kl + Kh = 1; 0 < a < 1; V > 0; and
the random productivity realization is idio
syncratic.
The structure of the problem guarantees
that the firm will either use VK units of soft
capital or none. For simplicity, assume that
it is always efficient to employ soft capital.
(Formally, this requires one to assume that
(;r* + 7tba ) /( 1 + v) > a ) .
If there are no informational imperfec
tions, the firm’s investment decision is intu
itive. It chooses K to satisfy
There are three basic conclusions of
models of financial frictions relating to asym
metric information between borrowers and
lenders: (1) Uncollateralized external finance
is more costly than internal finance; (2) the
spread between the cost of external and
internal funds varies negatively with the level
of the borrower’s internal funds; and (3) a
reduction in internal funds reduces the bor
rower’s spending, holding constant underlying
investment opportunities. I illustrate these
conclusions (and link them to empirical tests
of credit view models) below in a simple
model of firm investment decisions adapted
from Gertler and Hubbard (1988).
Consider two periods— zero and one.
In the first, a risk-neutral borrower uses inputs
to produce output Y to sell in the second
period. These inputs are hard capital, K—
say, machinery— and soft capital, C— inputs
which improve the productivity of hard capital
(such as organizational or maintenance
expenditures). The production technology
is risky, with two possible productivity states,
“good” and “bad”; uncertainty is realized
after the investment decision is made.
To make the example as simple as possi
ble, suppose the firm can increase the chance
of a good output realization if it uses a suffi
cient quantity of soft capital, where sufficient
is defined by a level proportional to the
quantity of hard capital used. In particular,
let output Y satisfy:
(1A)
(3A)
and
Y = a f(K ), with probability 7tb,
if
C > VK,
and
Y = a f( K ) ,
if
C< VK,
where/(K) is twice continuously differen
tiable, strictly increasing, and strictly con
nba )f'(K )-(l + v )r = 0,
where r is the gross interest rate faced by the
firm. Equation 2A simply states that, at the
optimum, the expected marginal benefit
from an additional unit of hard capital (given
a complementary addition of V units of soft
capital) equals the marginal cost of investing.
The value of K that satisfies equation A2 —
call it K *— does not depend on any financial
variables; that is, the Modigliani and Miller
theorem applies.
The traditional interest rate channel
often identified with the money view mecha
nism is easy to illustrate in this example.
Suppose for simplicity that the interest rate
paid on deposits is zero, so that r represents
the gross required rate of return on lending.
To the extent that an open market sale raises
r, investment demand falls. This is the usual
textbook interest rate channel for monetary
policy.
Under asymmetric information, the
story is more complicated. Consider, for
example, a simple agency problem:
Expenditures on hard capital are observable
by outside lenders, while expenditures on
soft capital are not. In this case, the manager
may be tempted to divert soft capital funds
to personal gain. Such perquisite consump
tion can take a number of forms. For sim
plicity, assume that the manager can invest
the funds (say, in a Swiss bank account) to
yield a gross interest rate, r.
Y =/(K), with probability 7ig,
(2A)
(7Tg +
75
MAY/JUNI
Lenders understand this temptation,
and modify the financial contract to mitigate
incentives to cheat. As shown below, one
consequence of this modification is that desired
capital, K *, may exceed actual capital, K, and
this gap will depend inversely on the bor
rower’s net worth. Suppose the firm signs a
loan contract with a competitive financial
intermediary. The firm has some initial liq
uid asset position, W, and collateralizable
future profits, V, in period one, worth a pre
sent value of V/r. Hence, the firm’s initial net
worth is (W + V/r). To make the story inter
esting, assume that W < K * ; that is, the firm
would like to borrow. (For a richer descrip
tion of the role of internal net worth in the
contracting problem, see Gertier, 1992.)
The firm-intermediary loan contract
specifies the amount borrowed (in this case,
(1 + v)K — W )),a payment Pg to the inter
mediary in the event that the project yields
the “good” output level, and a payment P1’ in
the event of the “bad” output level. These
contractual features are chosen to maximize
the firm’s expected profits:
1995
to cheat is to increase the amount of Pb that
the firm must pay the intermediary in the
event of a bad outcome. The firm, however,
can only credibly promise to pay available
assets in the bad state. That is, a limited lia
bility constraint influences the contract:
(7A)
Ph < a f ( K ) + V.
From the intermediary’s perspective, the
loan contract must offer an expected return
equal to its opportunity cost of funds, which
equals the gross interest rate r times the
quantity borrowed:
To summarize, the contracting problem
involves the selection of K, Pg and Ph to max
imize equation 4A subject to equations 5A,
6A and 7A. One case is easy: As long as the
incentive constraint does not bind, actual
investment, K, simply adjusts to desired
investment K *. In addition, the pattern of
contract payments is indeterminate. (For
simplicity, I am abstracting from a richer
structure that would lead to both debt and
equity contracts and tax considerations; see,
for example, Gertier and Hubbard, 1993, for
such a treatment.)
W hen the incentive constraint in equa
tion 6A binds, financing and investment
decisions are no longer independent. First,
note that when the incentive constraint
binds, it is desirable to raise Pb to the m axi
mum extent possible; therefore, the limited
liability constraint in equation 7A also binds.
Using 5A and 7A, one can eliminate Pg and
Pb from equation 6A, and thereby obtain a
relation among K, the interest rate and inter
nal net worth:
(5A)
(8A)
(4A)
(tt* + n ba ) j { K ) - n gP g - n bPb.
n gPg + n bPb = r[(l + v )K —w ].
That is, for simplicity, assume that the inter
mediary simply channels funds from savers
to borrowers, and uses no resources.
Given the underlying incentive problem,
the contract must give the firm the incentive
to invest in soft capital as a complementary
input to hard capital. That is, the contract
must satisfy the “incentive constraint:”
(6A)
|n g + 7tba j f ( K )
~ [r(l + 2v)] K + r ( W + V/r) = 0.
As long as equation 8A holds, investment
K is an increasing function of the borrower’s
net worth (W + V/r), holding investment
opportunities constant:
(9A)
( n g+ 7 tba ) f ( K ) - ( 7 t gP g+7CbPb)
,
*
d (W + V/r)
(l + 2 v ) - [ n g + 7tba j f ( fcj /r
>0.
> ( c c f ( K ) - P h)J+rvK.
The explanation for this effect is that,
when the incentive constraint binds, an
increase in internal net worth increases the
amount of feasible investment.
The existence of the net worth channel
precludes neither the traditional interest rate
channel nor the bank lending channel. To
Equation 6A ju st states that the manager’s
expected gain from honest action exceeds
the gain from diverting the soft capital funds
to personal use.
One way in which the intermediary
could reduce the entrepreneur’s temptation
76
REVIEW
MAY/JUNI
see the former, note an increase in lenders’
opportunity cost of funds on account of a
monetary contraction reduces desired invest
ment K * (since K * is determined by { n i +
K ha ) f' ( K ) = (1 + v)r). To see the latter, note
that, to the extent that banks face a higher
marginal opportunity cost of funds because of
a less than perfectly elastic supply schedule
for managed liabilities (and borrowers lack
access to nonbank finance), the increase in r
lowers both desired and actual investment.
This simple framework is consistent
with the description of the financial accelerator
mechanism: The cost of uncollateralized
external finance exceeds that for internal
finance. This gap varies inversely with the
internal net worth of the borrower and a
decline in net worth reduces the borrowers’
spending, all else equal. The framework also
yields simple testable predictions related to
these money view and credit view arguments:
(1) W hen informational imperfections are
ignored, an increase in real interest rates
following a monetary contraction should
affect investment (broadly defined) simi
larly for borrowers of a given type (for
example, with similar technology and
risk characteristics).
(2) If informational imperfections are signif
icant only on the borrower side, all else
equal, spending by borrowers with lower
levels of internal net worth should fall
relative to spending by borrowers with
higher levels of internal net worth.
1995
(3) For bank-dependent borrowers, the avail
ability of monitored bank credit can be
thought of as a substitute for internal net
worth. Changes in the availability of bank
credit can influence the ability of bankdependent borrowers to finance spending.
(4) The model’s intuition can apply to banks
as well as nonfinancial borrowers. A
decline in banks’ net worth raises banks’
opportunity cost of external funds (say, in
the CD market). As a result, the cost of
funds to bank-dependent borrowers rises.
(5) If relationships between borrowers and
specific banks are important, shocks to
the balance sheet positions of individual
lenders affect credit availability (at any
given open market interest rate) to their
borrowers.
UlS
77
RtVIEW
MAY/JUNK
199
Bruce D. Smith is a professor of economics at Cornell University.
Com mentary
W hile the Hubbard paper cites any
number o f references on financial market
imperfections and their effects on firm invest
ment behavior, to my knowledge none of the
papers he cites presents a general-equilibrium
model of an economy with money, capital and
a credit market friction. So, at this point, I
have the following questions:
Bruce D. Smith
1 for example, some monetary
models give rise to Modigliani and
Miller's theorems for open market
operations. These theorems state
conditions under which open mar
ket activity has no implications for
interest rates, or anything else.
For a discussion of such theorems,
and their empirical relevance, see
Wallace (1 9 8 1 ), Chomley and
Polemorchakis (1 9 8 4 ), Sargent
(1 9 8 2 , 19 8 7), Sargent and Smith
(1 9 8 7 ) or Smith (1 9 9 4 ).
lenn Hubbard’s paper considers what is
perhaps the m ost basic question in
monetary economics: How does mone
tary policy “work”? It suggests that informa
tional frictions affecting capital markets create
additional mechanisms— beyond those of
conventional textbook models— through
which monetary policy operates. In particular,
Hubbard suggests that “realistic models of
financial constraints on firms’ decisions imply
potentially significant effects of monetary
policy beyond those working through con
ventional interest rate channels.” Now I
personally feel that there are a number of
serious issues about what these “conventional
interest rate channels” are,1but that takes us
beyond the scope of the present paper. So, for
the purpose of discussion, let’s imagine that
we accept that there are such channels,2 and
consider how the presence of financial con
straints impacts the scope for monetary policy
to have other effects.
As the previous quotation suggests, there
ought to exist models in which there are infor
mational (or other) frictions affecting firm
investment decisions, and in which there is
scope for monetary policy to operate. This
requires a model with— at a minimum— money,
capital and a credit market friction. Moreover,
I would argue that an interesting model for
analyzing the role of monetary (or other)
policies in an economy with a financial market
friction should be a general-equilibrium model,
since we would like to know the answers to
at least two questions: not only
2 The issue in the literature discussed
in footnote 1 is: Under what condi
tions would there be such chan
nels? That literature suggests that
these channels exist only if policy
is conducted in a way which is
intentionally redistributive.
(1) W hat can monetary policy do?
but also,
(2) W hat should monetary policy do? (That
is, what are the welfare implications of
alternative methods of conducting mon
etary policy?)
G
• W hat are the models of financial con
straints implying these magnified effects of
monetary policy (presuming, of course,
the need for general-equilibrium models)?
• W hat are the implications of these models
for the effects (and welfare consequences)
of various methods for conducting mone
tary policy?
The Hubbard paper comes in two parts:
Its appendix contains a suggestive model of
a single firm undertaking credit-financed
investment, subject to a moral hazard prob
lem, along with a proposed list of empirical
implications derived from the literature that
the model represents. The text of the paper
presents a discussion of the empirical litera
ture on how monetary policy does (or can)
affect the investment behavior of individual
firms. To a large extent, I very much like
both the model o f the paper and the discus
sion of the empirical evidence. I do think,
however, there is a serious question about
how these two parts of the paper fit together.
Let me therefore add to my list o f questions:
• If we do have general-equilibrium models
of capital accum ulation in the presence of
money and financial market frictions, what
do these models imply about the conse
quences of various monetary policy actions?
• W hat is (or could be) the empirical
evidence on these implications?
• How does the empirical evidence discussed
in the Hubbard paper bear on them?
Before proceeding to a discussion of
these issues, let me say that I intend to focus
my discussion most where the Hubbard dis
cussion focuses least— on the theoretical
aspects of monetary growth models with
78
REVIEW
MAY/JUNK
informational frictions. In large part, this is
because Glenn is a pioneer in, and a major
continuing contributor to, the empirical lit
erature on these topics, and his discussion of
this literature is thoughtful and easy to follow.
Thus, while admitting Glenn may have
absolute advantage along both dimensions,
considerations of comparative advantage
suggest that I should primarily concentrate
on theoretical issues.
The Hubbard paper identifies three
common implications of the models he has
in mind, and which he identifies with the
credit view:
of consumption foregone today becomes one
unit of capital after one period. And, again
as in traditional monetary growth models,
there is no role for banks or other financial
market institutions.
Boyd and Smith modify the Diamond
model to allow for two classes of agents in
each generation. One class of agent has access
to a stochastic linear technology for converting
current goods into future capital, the other
type does not. In all other respects, the two
types are identical.
The capital production technology
considered by Boyd and Smith is subject to
a standard costly state verification (CSV)
problem o f the type considered by Townsend
(1 9 7 9 ) and, more specifically, Gale and
Hellwig (1 9 8 5 ), W illiamson (1 9 8 6 , 1987)
and Bernanke and Gertler (1 9 8 9 ). As is con
ventional in such models, each operator of the
capital production technology must produce
at some fixed, indivisible scale. Thus, to
finance capital investments, young investors
must combine their own young period income,
along with funds obtained externally.
Under the assumption of risk-neutral
firms and fixed verification costs, this setup
yields an optimal capital structure and
financing arrangement for firms producing
capital goods. Such firms should be (com
pletely) debt-financed, and it is efficient for
them to borrow from financial intermedi
aries. Presumably this captures the notion
of “bank-dependent borrowers” discussed
by Hubbard. Moreover, this model would
produce, at the individual firm level, the
three key results of models that Hubbard
associates with the credit view.
In this model, the amount of internal
finance provided by investors is endogenous,
depending on the young period wage income
of borrowers. Internal finance is valuable
because it helps to mitigate the CSV problem.
In addition, as in Gale and Hellwig (1 9 8 5 )
and W illiamson (1 9 8 6 , 1987), the presence
of the CSV problem permits credit rationing
to be observed for exactly the reasons dis
cussed by Stiglitz and Weiss (1 9 8 1 ): Because
of the costs of verifying project returns when
borrowers default, raising the interest rate
charged on loans affects a lender’s.expected
return in a non-m onotonic fashion. Thus,
1. “Uncollateralized external finance is more
costly than internal finance.
2. “The spread between the cost of external
and internal funds varies negatively with
the level of the borrower’s internal funds.
3. “A reduction in internal funds reduces the
borrower’s spending, holding underlying
investment opportunities constant.”
W hile the discussion on these points is
somewhat vague, from my knowledge of the
literature I take these to be partial equilibrium
results that apply to a particular borrower,
holding aggregate conditions fixed. W hat
monetary growth models exist, then, that
would deliver these as implications at the
level of an individual firm?
To my knowledge, there is exactly one
such model— that of Boyd and Smith (1994).
Let me sketch the main features of this model
and then describe its implications for the kinds
o f issues that come up in the Hubbard dis
cussion.
The Boyd and Smith model uses as its
basic framework the neoclassical growth model
of Diamond (1965), which allows for outside
assets in a general-equilibrium model of cap
ital accumulation. The Diamond model is a
two-period, overlapping-generations model
in which all agents supply one unit of labor
inelastically when young, earning the pre
vailing real wage rate. These agents are retired
when old. They save for old-age retirement
by accumulating either capital or money (or,
more generally, outside assets).
Capital accumulation in the Diamond
model— as in most traditional monetary
growth models3— is a “black box”; one unit
199
79
3 See, for instance, the models of
Sidrauski (1 9 67 a ,b), Brock
(1 9 7 4 ) or Tirole (1 9 8 5 ).
MAY/JUNE
the interest rate charged on loans can be “bid
up” to a level that maximizes the expected
return to a lender; thereafter, increases in the
interest rate reduce a lender’s expected return
and are counterproductive. As a result, if the
demand for credit exceeds its supply and
interest rates are raised to their expected return
maximizing level, there is no action that an
unfunded (or rationed) borrower can take to
obtain a loan. This presumably maximizes
the scope for monetary factors to “matter,”
since availability of credit becomes an issue
of central concern.
Boyd and Smith consider the situation
where credit is rationed, and examine the
following policy regime. The monetary
authority fixes, once and for all, a rate of
money growth. In the Diamond model, the
fixed rate of money growth determines the
steady-state real rate of interest. This formu
lation gives the Hubbard analysis its best case
scenario, parenthetically, since it allows the
monetary authority— at least potentially— the
power to control real interest rates directly.4
The Boyd and Smith modification of the
Diamond model is, superficially, very minor.
But it has dramatic implications for the prop
erties of monetary equilibria in the Diamond
model. Most of these implications are, I think,
bad news from the standpoint of the kind of
analysis conducted in the Hubbard paper,
although there is one piece of good news. I will
now review some of the relevant implications.
The effects of a monetary policy action
depend— very strongly, as it turns out— on
which equilibrium path the economy is fol
lowing. Moreover, for some parameter con
figurations there exist equilibria which
approach no steady state; that is, lim it cycles
can be observed. Changes in monetary policy
can change the entire set of equilibria, creating
scope for equilibria that did not exist under
other configurations of policy.
These possibilities are of some interest
from a theoretical perspective. They imply
that the interaction of policy choices with
the operation of financial markets subject to
frictions creates a scope for the indeterminacy
of equilibrium and for “excessive fluctuations,”
a point emphasized by Simons (1948) and
Friedman (1 9 6 0 ). However, they also imply
that there is no unique answer to the ques
tion: How do credit channels affect the
consequences of monetary policy?
W hy do credit market frictions create
indeterminacies and render questions about
their effects on policy actions problematic?
The answer has to do with exactly the feature
most emphasized by Hubbard: the impor
tance of internal finance, and the fact that
the ability to provide internal finance is going
to be (at least partly) endogenous in a general
equilibrium model. In the Boyd-Smith model,
the monetary authority controls the real rate
of interest (at least in steady-state equilibria).
Borrowers are then forced to deliver this policydetermined real rate of return on funds they
obtain. In a steady-state equilibrium, there
are typically two ways to do this. One is to
have a low capital stock, a correspondingly
high marginal product o f capital, and low
incomes (low levels of internal finance). The
other is to have a high capital stock, a corre
spondingly low marginal product of capital,
and high incomes enabling borrowers to pro
vide a lot of internal finance. Since internal
finance mitigates the CSV problem, it offsets
the low marginal product of capital and per
mits borrowers to offer lenders the necessary
expected return.
The key element in this analysis, of course,
is the endogeneity of the amount of internal
finance. Once this is endogenous, models
representing what Hubbard calls the credit
view cannot generally be expected to deliver
THE BAD NEWS
4 Notice that this does not require the
existence of any nominal rigidities,
os Hubbard asserts.
Traditional monetary growth models
have the property that there is a unique
monetary steady-state equilibrium, which is
a saddle. Thus, one can unambiguously
identify the monetary equilibrium of such a
model, and can unambiguously discuss the
effects of monetary policy actions on the
equilibrium. The kind of model that Hubbard
apparently has in mind may, however, have
multiple equilibria, and multiple possible
effects of a monetary policy action.
The Boyd and Smith model has (typically)
two monetary steady state equilibria. It can
easily transpire that one is a sink and one is
a saddle, so both can be approached. Thus,
there is a continuum of monetary equilibria.
199
80
MWN
MAY/JUNE
equilibria, any discussion of its empirical
implications must confront the difficulties
associated with the empirical analysis of
models displaying multiple equilibria. This
is a difficult issue, and one that I am not cur
rently prepared to take on. However, Hubbard
argues that the money and the credit views
have the following implications:
unique equilibria, and questions about
“the effects” of monetary policy will not
be well-posed.
To underscore this point, a monetary
expansion in the Boyd and Smith model (a
higher rate of money growth) increases the
capital stock, output, and credit extension
in the low-capital-stock steady state. All of
these effects are reversed in the high-capitalstock steady state.
“W hen informational imperfections are
ignored, an increase in real interest rates
following a monetary contraction should
affect investment (broadly defined) simi
larly for borrowers of a given type (for
example, with similar technology and
risk characteristics).5
SOME G O O D NEWS AND
SOME BAD NEWS
The good news is that— in the low-capi
tal-stock steady state, where expansionary
monetary policy actions are actually expan
sionary— the Boyd and Smith model predicts
that capital market imperfections will enhance
the effects of a given change in monetary
policy. In particular, a given change in the
rate of money growth has a larger effect on
output in the presence of the credit market
friction than is the case under full informa
tion. In this sense, one prediction of the
credit view is borne out.
However, even this effect does not occur
for the reasons discussed by Hubbard.
According to his analysis,
“If informational imperfections are signif
icant only on the borrower side, all else
equal, spending by borrowers with lower
levels of internal net worth should fall
relative to spending by borrowers with
higher levels of net worth.6
“The model’s intuition can apply to
banks as well as non-financial borrow
ers. A decline in banks’ net worth raises
banks’ opportunity cost of external funds
(say in the CD market). As a result, the
cost of funds to bank-dependent bor
rowers rises.”
“...the crux of models of informationrelated financial frictions is a gap
between the cost of external and internal
finance for many borrowers. In this
context, the credit view offers channels
through which monetary policy can
affect this gap.”
I am not sure what the practical empirical
content of the first implication is likely to be,
since we do not typically observe the techno
logical characteristics or demand conditions
of individual firms directly. I am also unclear
as to why borrowers with similar net worth
cannot be affected differentially by monetary
policy under the credit view. (This is, in fact,
what happens in the Boyd and Smith model.)
And, indeed, it is easy to produce certain
kinds of counter-examples to the second
claim in models that seem perfectly consis
tent with the credit view.7 Finally, creditview models, like W illiamson’s (1 9 8 7 ), tell
us that the effects o f increases in the costs of
external funds can depend very heavily on
the nature of how interest rates are determined.
In particular, the “incidence” of higher costs
depends heavily on whether credit is rationed,
on the interest elasticity of the supply and
the demand for funds, and so on. It is there
fore not clear to me why it follows that an
In the Boyd and Smith model, monetary
policy can have heightened effects, but not
because it affects the differential between the
cost of internal and external funds in this way.
Indeed, it is possible to show that, in the steady
state equilibria they examine, monetary policy
cannot affect this differential (appropriately
defined). Nonetheless, in one of their steady
state equilibria, credit market frictions do
magnify the impact of monetary policy.
EMPIRICAL IMPLICATIONS
Since the credit view applied to monetary
models seems prone to delivering multiple
1995
81
s This implication applies, of course,
to the money view.
6 This implication applies under the
credit view.
7 For example, the Boyd and Smith
model can easily be modified to
allow for borrowers with different
levels of net wortfi. In that model,
increases in interest rates will affect
only the spending of marginal bor
rowers (who, in many data sets,
would then disappear from the
sample). Changes in interest rates
would not affect infra-marginal bor
rowers. This point is illustrated, for
example, in the model of Mo and
Smith (1 9 9 3 ).
R[VI[N
MAY/JUNK
_ _ _ _ _ _ . "The Ends of Four Big Inflations," in Robert E. Hall, ed.,
Inflation: Causes ond Effects. National Bureau of Economic Research,
University of Chicago Press, 1982.
increase in the costs of external funds for
banks must be borne by borrowers.
In short, it is not transparent that there
exist any sharp empirical hypotheses distin
guishing the money view from the credit view
at the firm level. Perhaps we are best advised
to take seriously the notion that the credit view
predicts the possibility of multiple equilibria,
with some equilibria displaying endogenously
enhanced volatility and to pursue the empirical
implications of that idea.
_ _ _ _ _ _ and Bruce D. Smith. "Irrelevance of Open Market
Operations in Some Economies with Government Currency Being
Dominated in Rate of Return," The American Economic Review
(March 1987), pp. 78-92.
Simons, Henry. Monetary Policy for a Free Society. University of
Chicago Press, 1948.
Smith, Bruce D. "Mischief ond Monetary History: Friedmon and
Schwartz Thirty Years Later," Journal of Monetary Economics
(August 1994), pp. 27-45.
REFERENCES
Stiglitz, Joseph, and Andrew Weiss. "Credit Rationing in Models with
Imperfect Information," The American Economic Review (June 1981),
pp. 393-410.
'Bemanke, Ben S., and Mark Gertler. "Agency Costs, Net Worth, ond
Business Fluctuations," The American Economic Review (March 1989),
pp. 14-31.
Tirole, Jean. "Asset Bubbles and Overlapping Generations,"
Econometrica (November 1985), pp. 1499-528.
Boyd, John H., and Bruce D. Smith. "Capital Market Imperfections in a
Monetary Growth Model," working paper (1994), Federal Reserve
Bank of Minneapolis.
Townsend, Robert M. "Optimal Contracts and Competitive Markets
with Costly State Verification," Journal of Economic Theory
(October 1979), pp. 265-93.
Chomley, Christophe, ond Heroklis Polemarchakis. "Assets, General
Equilibrium, ond the Neutrality of Money," Review of Economic
Studies (January 1984), pp. 129-38.
Wallace, Neil. "A Modigliani-Miller Theorem for Open Market Operations,"
The American Economic Review (June 1981), pp. 267-74.
Diamond, Peter A. "National Debt in a Neoclassical Growth Model,"
The American Economic Review (December 1965), pp. 1126-50.
Williamson, Stephen D. "Costly Monitoring, Loan Contracts, and
Equilibrium Credit Rationing," Quarterly Journal of Economics
(February 1987), pp. 135-45.
Friedman, Milton. A Program for Monetary Stability. Fordham
University Press, New York, 1960.
_ _ _ _ _ _ . "Costly Monitoring, Financial Intermediation, and
Equilibrium Credit Rationing," Journal of Monetary Economics
(September 1986), pp. 159-79.
Gale, Douglas, and Martin Hellwig. "Incentive-Compatible Debt
Contracts: The One-Period Problem," Review of Economic Studies
(October 1985), pp. 647-63.
Mo, Chien-hui, and Bruce D. Smith. "Credit Morket Imperfections and
Economic Development: Theory and Evidence, Journal of
Development Economics (forthcoming).
Sargent, Thomas J. Dynamic Macroeconomic Theory. Harvard University
Press, 1987.
199
82
REVIEW
MAY/JUNE
1995
Stephen G. Cecchetti is professor of economics at Ohio State University. The author thanks Margaret Mary McConnell for able research
assistance, and Allen Berger, Ben Bernanke, Anil Kashyap, Nelson Mark, Alan Viard and the participants at the conference for comments and
suggestions. The author also expresses gratitude to the National Science Foundation and the Federal Reserve Bank of Cleveland for financial
and research support.
Distinguishing
Theories of the
M onetary
Transmission
Mechanism
Stephen G . Cecchetti
raditional studies of monetary policy’s
impact on the real economy have
focused on its aggregate effects.
Beginning with Friedman and Schwartz
(1 9 6 3 ), modern empirical research in mone
tary economics emphasizes the ability of pol
icy to stabilize the macroeconomy. But casual
observation suggests that business cycles
have distributional implications as well. One
way of casting the debate over the relative
importance of different channels of monetary
policy transmission is to ask if these distrib
utional effects are sufficiently important to
warrant close scrutiny.
The point can be understood clearly by
analogy with business cycle research more
generally. If recessions were characterized by
a proportionate reduction of income across
the entire employed population— for example,
everyone worked 39 rather than 40 hours
per week for a few quarters— then economists
would pay substantially less attention to
cycles. It is the allocation of the burden or
benefit of fluctuations, with some individuals
facing much larger costs than others, that is
of concern. There are two ways for an econ
omist to address this problem. The first is to
attempt to stabilize the aggregate economy,
the traditional focus of policy-oriented
macroeconomics. The second is to ask why
the market does not provide some form
of insurance.
The recent debate over the nature of the
monetary transmission mechanism can be
T
thought of in similar terms. According to
the original textbook IS-LM view of money,
changes in policy are important only insofar
as they affect aggregate outcomes. Only the
fluctuation in total investment is important
since policies only affect the required rate of
return on new investment projects, and so it
is only the least profitable projects (economywide) that are no longer funded. But since
the most profitable projects continue to be
undertaken, there are no direct efficiency
losses associated with the distributional
aspects of the policy-induced interest
rate increase.
In contrast, the “lending” view focuses
on the distributional consequences of mone
tary policy actions. By emphasizing a com bi
nation of capital market imperfections and
portfolio balance effects based on imperfect
asset substitutability, this alternative theory
suggests the possibility that the policy’s inci
dence may differ substantially across agents
in the economy. Furthermore, the policy’s
impact has to do with characteristics of the
individuals that are unrelated to the inherent
creditworthiness of the investment projects.
An entrepreneur may be deemed unworthy
of credit simply because of a currently low
net worth, regardless of the social return to
the project being proposed. It is important
to understand whether the investment
declines created by monetary policy shifts
have these repercussions.1
In this essay, I examine how one might
determine whether the cross-sectional effects
of monetary policy are quantitatively impor
tant. My goal is to provide a critical evaluation
of the major contributions to the literature
thus far. The discussion proceeds in three
steps. I start in the first section with a
description of a general framework that
encompasses all views of the transmission
mechanism as special cases, thereby high
lighting the distinctions. In the second section,
I begin a review of the empirical evidence
with an assessment of how researchers typi
cally measure monetary policy shifts. The
following two sections examine the methods
1 The financial accelerator, in which
the impact on investment of small
interest changes is magnified by
balance sheet effects, is also on
important part of many discussions
of the lending view.
REVIEW
MAT/JUNE
used for differentiating between the theories.
Studies fall into two broad categories
depending on whether they use aggregate or
disaggregate data. The third section discusses
the aggregate data, while the fourth section
describes the use of disaggregate data. A
conclusion follows.
possible to compute the utility maximizing
portfolio weights. These will depend on the
mean and variance of the returns,^ and F,
the moments o f the consumption process,
call these fi c, and a vector of taste parameters
that I will label cf>and assume to be constants.
The utility maximizing asset demands can be
expressed as X * = w *,(z , jT, /jlc, (f>)W.2
This representation makes clear that
asset demands can change for two reasons.
Changes in either the returns process ( z , F )
or macroeconomic quantities (/a , W) will
affect the X*'s.3
At the most abstract level, financial
intermediaries exist to carry out two functions.
First, they execute instructions to change
portfolio weights. That is, following a change
in one or all of the stochastic processes driving
consumption, wealth or returns, the interm e
diary will adjust investors’ portfolios so that
they continue to maximize utility. In addition,
if one investor wishes to transfer some wealth
to another for some reason, the intermediary
will effect the transaction.
W hat is monetary policy in this stylized
setup? For policy to even exist, some gov
ernment authority, such as a central bank,
must be the monopoly supplier of a nominally
denominated asset that is imperfectly substi
tutable with all other assets. I will call this
asset “outside money.” In the current envi
ronment, it is the monetary base. There is a
substantial literature on how the demand for
outside money arises endogenously in the
context of the type of environment I have
ju st described.4 But in addition, as Fama
emphasizes, there may be legal requirements
that force agents to use this particular asset
for certain transactions. Reserve requirements
and the use of reserves for certain types of
bank clearings are examples.
W ithin this stylized setup, a policy action
is a change in the nominal supply o f outside
money. For such a change to have any
effects at all, (1) the central bank controls
the supply of an asset that is both in demand
and for which there is no perfect substitute,
and (2) prices must fail to adjust fully and
instantaneously. Otherwise, a change in the
nominal quantity of outside money cannot
have any impact on the real interest rate, and
will have no real effects. But, assuming that
M ONETARY POLICY:
THEORY
A General Framework
2 See Ingersoll (1 9 8 7 ) for a com
plete description of this problem.
3 Following the traditional financial
economics approach, I have avoided
discussing demand and supply
explicitly. Insteod, the asset
demands ore derived from the idea
of arbitrage relationships among all
of the assets.
4 These include the limited participa
tion models based on Lucas
(1 9 9 0 ). See the survey by Feurst
(1 9 9 3 ), as well as the summary in
Christiano and Eichenbaum
(1 9 9 2 ).
1995
One way of posing the fundamental
question associated with understanding the
monetary transmission mechanism is to ask
how seemingly trivial changes in the supply
of an outside asset can create large shifts in
the gross quantity of assets that are in zero
net supply. How is it that small movements
in the monetary base (or nonborrowed
reserves) translate into large changes in
demand deposits, loans, bonds and other
securities, thereby affecting aggregate invest
ment and output?
The various answers to this puzzle can
be understood within the framework origi
nally proposed by Brainard and Tobin (1963).
Their paradigm emphasizes the effects of
monetary policy on investor portfolios, and
is easy to present using the insights from
Fama’s (1980) seminal paper on the relation
ship between financial intermediation and
central banks.
Fama’s view of financial intermediaries
is the limit of the current type of financial
innovation, because it involves the virtual
elimination of banks as depository institu
tions. The setup focuses on an investor’s
portfolio problem in which an individual
must choose which assets to hold given the
level of real wealth. Labeling the portfolio
weight on asset i as wP and total wealth as
W, then the holding of asset, i— the asset
demand— is ju st X = wW.
In general, the investor is dividing
wealth among real assets— real estate, equity
and bonds— and outside money Each asset
has stochastic return, zp with expectation z,;
and the vector of asset returns, z, has a
covariance structure F. Given a utility func
tion, as well as a process for consumption, it is
84
REVIEW
MAT/JUNE
the policymaker can change the real return
on the asset that is monopolistically supplied,
investors’ portfolio weights must adjust in
response to a policy change.
The view of financial intermediaries that
is im plicit in this description serves to high
light the Brainard and Tobin (1963) insight
that monetary policy can be understood by
focusing solely on the endogenous response
of investor portfolios. Understanding the
transmission mechanism requires a charac
terization of how asset holdings change in
response to policy actions.
Second, even though there need b e no
banks as we know them, there will surely be
intermediaries that perform the service of
making small business loans. The agency
costs and monitoring problems associated
with this type of debt will still exist, and spe
cialists in evaluation will emerge. W hile they
will have such loans as assets, they most
likely will not have bank deposits as liabilities.
Such entities will be brokers, and the loans
will be bundled and securitized.
W ith this as background, it is now
possible to sketch the two major views of the
monetary transmission mechanism. There
are a number of excellent surveys o f these
theories, including Bemanke (1993a),
Gertler and Gilchrist (1 9 9 3 ), Kashyap
and Stein (1994a) and Hubbard (1995).
As a result, 1 will be relatively brief in
my descriptions.
that the shift in the w*'s for all of the assets
excluding outside money are equal.
An important implication of this tradi
tional model of the transmission mechanism
involves the incidence of the investment
decline. Since there are no externalities or
market imperfections, it is only the least
socially productive projects that go unfunded.
The capital stock is marginally lower. But,
given that a decline is going to occur, the
allocation of the decline across sectors is
socially efficient.
This theory actually points to a measure
of money that is rarely studied. Most empiri
cal investigations of monetary policy trans
mission focus on M2, but the logic of the
portfolio view suggests that the monetary
base is more appropriate. It is also worth
pointing out that investigators have found
it extremely difficult to measure economically
significant responses of either fixed or inven
tory investment to changes in interest rates
that are plausibly the result of policy shifts.
In fact, most of the evidence that is interpreted
as supporting the money view is actually evi
dence that fails to support the lending view.
THE LENDING VIEW :
BALANCE SHEET EFFECTS
The second theory of monetary trans
mission is the lending view.6 It has two parts,
one that does not require introduction of
assets such as bank loans, and one that does.
The first is sometimes referred to as the broad
lending channel, or financial accelerator, and
emphasizes the im pact of policy changes on
the balance sheets of borrowers. It bears
substantial similarity to the mechanism oper
ating in the money view, because it involves
the impact of changes in the real interest rate
on investment.
According to this view, there are credit
market imperfections that make the calcula
tion of the marginal efficiency of investment
schedule more complex. Due to information
asymmetries and moral hazard problems, as
well as bankruptcy laws, the state of a firm’s
balance sheet has implications for its ability
to obtain external finance. Policy-induced
increases in interest rates (which are both real
and nominal) can cause a deterioration in
THE M O N EY VIEW
The first theory, commonly labeled
the money view, is based on the notion that
reductions in the quantity of outside money
raise real rates of return.5 This, in turn,
reduces investment because fewer profitable
projects are available at higher required rates
of return— this is a movement along a fixed
marginal efficiency of investment schedule.
The less substitutable outside money is
for other assets, the larger the interest
rate changes.
There is no real need to discuss banks
in this context. In fact, there is no reason
to distinguish any of the “other” assets in
investors’ portfolios. In terms of the simple
portfolio model, the money view implies
1995
85
5 Terminology has the potential to
creote confusion here. I have cho
sen the traditional term for this
textbook IS-LM or "norrow" money
view. I do not meon to imply that
this is the 'monetarist' view of the
transmission mechanism.
6 1 follow Kashyap ond Stein's
(1 994o) terminology rather than
the more common credit view to
emphasize the importance of loons
in the transmission mechanism.
Bernanke and Gertler (1989,
1990) provide the original theoreti
cal underpinnings for this view.
INIEN
MAY/JUNE
7 Bernanke, GerHer and Gilchrist
(1 9 9 4 ) refer to this as a financial
accelerator since it causes small
changes in interest rates to have
potentially large effects on invest
ment and output.
* It may be particularly difficult to dis
tinguish these effects from those
that arise from varying cyclicality of
different firms' sales and profitability.
9 See James (1 9 8 7 ) for a discussion
of the uniqueness of bank loans.
10 With nominal rigidity, a decrease in
outside money reduces the price
level slowly, and so the reol return
to holding money increases. This
channel of transmission requires
that investors shift away from loans
in response.
11 Kashyap and Stein (1994b) point
out that lorge b a n b can issue CDs
in o way that insulates their bal
ance sheets from contraction in
deposits, but small bonks cannot.
So long as small b a n b are an
important source of funds for some
bank-dependent firms, there will
still be a bank lending channel. In
other words, for bonk lending to be
on important part of the transmis
sion mechanism, credit market
imperfections must be important
for banb.
the firm’s net worth, by both reducing expected
future sales and increasing the real value of
nominally denominated debt. With lower net
worth, the firm is less creditworthy because it
has an increased incentive to misrepresent
the riskiness of potential projects. As a result,
potential lenders will increase the risk pre
mium they require when making a loan. The
asymmetry of information makes internal
finance of new investment projects cheaper
than external finance.
The balance sheet effects imply that the
shape of the marginal efficiency of investment
curve is itself a function of the debt-equity
ratio in the economy and can be affected by
monetary policy.7 In terms of a simple text
book analysis, policy moves both the IS and
the LM curves. For a given change in the rate
of return on outside money (which may be
the riskless rate), a lender is less willing to
finance a given investment the more debt a
potential borrower has. This points to two
clear distinctions between the money and
the lending views— the latter stresses both
the distributional impact of monetary policy
and explains how seemingly small changes
in interest rates can have a large impact on
investment (the financial accelerator).
Returning to the portfolio choice model,
the presence of credit market imperfections
means that policy affects the covariance
structure of asset returns. As a result, the
w*’s will shift differentially in response to
monetary tightening as the perceived riski
ness of debt issued by firms with currently
high debt-equity ratios will increase relative
to that of others.8
the quantity of loans. It is not necessary to
have a specific institutional framework in
mind to understand this. Instead, it occurs
whenever loans and outside money are
complements in investor portfolios; that is,
whenever the portfolio weight on loans is a
negative function of the return on outside
money for given means and covariances of
other asset returns.10
The argument has two clear parts. First,
there are borrowers who cannot finance
new projects except through loans, and
second, policy changes have a direct effect
on loan supply. Consequently, the most
important impact of a policy innovation is
cross-sectional, as it affects the quantity of
loans to loan-dependent borrowers.
Most of the literature on the lending view
focuses on the implications of this mechanism
in a world in which banks are the only source
of loans and whose liabilities are largely
reservable deposits. In this case, a reduction
in the quantity of reserves forces a reduction
in the level of deposits, which must be
matched by a fall in loans. The resulting
change in the interest rate on outside money
will depend on access to close bank deposit
substitutes. But the contraction in bank bal
ance sheets reduces the level of loans. Lower
levels of bank loans will only have an impact
on the real economy insofar as there are
firms without an alternative source of
investment funds.
As a theoretical matter, it is not necessary
to focus narrowly on contemporary banks in
trying to understand the different possible
ways in which policy actions have real effects.
As I have emphasized, bank responses to
changes in the quantity of reserves are ju st
one mechanism that can lead to a comple
mentarity between outside money and loans.
As pointed out by Romer and Romer (1 9 9 0 ),
to the extent that there exist ready substitutes
in bank portfolios for reservable deposits such
as CDs, this specific channel could be weak
to nonexistent.11 But it remains a real possi
bility that the optimal response of investors
to a policy contraction would be to reduce
the quantity of loans in their portfolios.
The portfolio choice model also helps to
make clear that the manner in which policy
actions translate into loan changes need not
THE LENDING VIEW: LOANS
FROM INTERMEDIARIES
The second mechanism articulated by
proponents of the lending channel can be
described by dividing the “other” assets in
investors’ portfolios into at least three cate
gories: outside money, “loans” and all the
others. Next, assume that there are firms for
which loans are the only source of external
funds— some firms cannot issue securities.9
Depending on the solution to the portfolio
allocation problem, a policy action may
directly change both the interest rate and
1995
86
REVIEW
MAT/JUHI
be a result of loan rationing, although it
may.12 As Stiglitz and Weiss (1981) originally
pointed out, a form of rationing may arise in
equilibrium as a consequence of adverse
selection. But the presence of a lending
channel does not require that there be
borrowers willing to take on debt at the cur
rent price who are not given loans. It arises
when there are firms which do not have
equivalent alternative sources of investment
funds and loans are imperfect substitutes in
investors’ portfolios.
Obviously, the central bank can take
explicit actions directed at controlling the
quantity of loans. Again, lowering the level
o f loans will have a differential impact that
depends on access to financing substitutes.
But the mechanism by which explicit credit
controls influence the real economy is a
different question.13
This immediately suggests that looking
at aggregates for evidence of the right degree
of imperfect substitutability or timing of
changes may be very difficult. W hat seems
promising is to focus on the other distinction
between the two views— the lending view’s
assumption that some firms are dependent
on loans for financing.
In addition to differences stemming
from the relative importance o f shifts in loan
demand and loan supply, the lending view
also predicts cross-sectional differences arising
from balance sheet considerations. These are
also likely to be testable. In particular, it
may be possible to observe whether, given
the quality of potential investment projects,
firms with higher net worth are more likely
to obtain external funding. Again, the major
implications are cross-sectional.
EMPIRICAL EVIDENCE:
PRELIMINARIES
DISTINGUISHING THE
TW O VIEW S: GENERAL
CONSIDERATIONS
Before discussing any empirical exami
nation of the monetary transmission mecha
nism, two questions must be addressed.
First, do nominal shocks in fact have real
effects? Unless monetary policy influences
the real economy, it seems pointless to study
the way in which policy changes work.
Second, how can we measure monetary
policy? In order to calculate the impact of
monetary policy, we need a quantitative
measure that can reliably be associated with
policy changes.
Here I take up each of these issues. In
the following section, I will weigh the evi
dence on the real effects of money. This is
followed by a discussion of ways in which
recent studies have attempted to identify
monetary shocks.
Distinguishing between these two views
is difficult because contractionary monetary
policy actions have two consequences,
regardless of the relative importance of the
money and lending mechanisms. It both
lowers current real wealth and changes the
portfolio weights.14
Assuming that there are real effects,
contractionary actions will reduce future
output and lower current real wealth, reducing
the demand for all assets. In the context of
standard discussions of the transmission
mechanism, this is the reduction in investment
demand that arises from a cyclical downturn.6
The second effect of policy is to change
the mean and covariance of expected asset
returns. This changes the wj(’s. In the simplest
case in which there are two assets, outside
money and everything else, the increase in
the return on outside money will reduce the
demand for everything else. This is a reduc
tion in real investment.
The lending view implies that the
change in portfolio weights is more complex
and in an important way. There may be
some combination of balance sheet and loan
supply effects.
1995
THE REAL EFFECTS OF
NO M IN AL SHOCKS
Modern investigation of the impact of
money on real econom ic activity began with
Friedman and Schwartz (1 9 6 3 ). In many
ways, this is still the most powerful evidence
in support of the claim that monetary policy
plays an important role in aggregate fluctua
tions. Through an examination that spanned
87
12 Since there must be firms that are
loan-dependent, there is still some
form of rationing in the security
market.
,3 See Romer and Romer (1 9 93 ) for
a concise discussion of recent
episodes in which the Federal
Reserve has attempted to change
the composition of bank balance
sheets through means other than
standard policy actions.
14 The change in portfolio weights can
arise either from any combination
of a change in the return on the
outside asset, a change in the
covariance structure of returns, or a
shift in the consumption process.
15 In general equilibrium, there is an
offsetting effect that arises from
the increase in the interest rate. All
other things equal, this would
increase saving and therefore
investment. But we can be fairly
confident that so long as monetary
policy tightening con cause a reces
sion, the impact of the income and
wealth declines will be large
enough that investment will fall.
REVIEW
MAY/JUNI
16 The equivalent open economy
observation is that in small open
economies, exchange rates move in
response to changes in policy.
17 Boschen and Mills (1 9 9 2 ) describe
a related technique.
" See Hamilton (1 994) for a complete
description of the methodology.
” The exact measures and sample fol
low those of Kashyap and Stein
(1 9 94 a ), who kindly supplied the
data.
of money are a combination of endogenous
responses to real shocks (King and Plosser,
1984) and shifts in money demand
(Bernanke and Blinder, 1992).
There have been two reactions to the
fact that monetary aggregates provide little
insight into policy actions. Both begin by
looking at the functioning of the Federal
Reserve and examining how policy is actually
formulated. The first, due to Bernanke and
Blinder (1 9 9 2 ), note that the federal funds
rate is the actual policy instrument that is used
on a day-to-day basis. This suggests that
innovations to the federal funds rate are likely
to reflect, at least in part, policy disturbances.
The main justification for their conclusion
comes from examining the institutions of how
monetary policy is carried out.
Romer and Romer (1 9 8 9 ) suggest a
second method. By reading the minutes of
the Federal Open Market Committee (FOMC)
meetings, they have constructed a series of
dates on which they believe policy
became contractionary.17
numerous monetary regimes, they argue that
apparently exogenous monetary policy actions
preceded output movements.
Recent researchers use more sophisticated
statistical tools to study the correlations
between money and income. This “moneyincome causality” literature is largely inconclu
sive, because it fails to establish convincingly
either that money “caused” output or the
reverse. In the end, the tests simply establish
whether measures of money forecast output,
not whether there is causation. Given that
outside money— the monetary base— is less
than 10 percent of the size of M2, it is not
surprising that economists find the simul
taneity problems inherent in the question
too daunting and give up.
Two pieces of evidence seem reasonably
persuasive in making the case that money
matters. First, the Federal Reserve seems to
be able to change the federal funds rate vir
tually without warning. (I am not arguing
that this is necessarily a good idea, ju st that
it is possible.) In the very short run, these
nominal interest rate changes cannot be
associated with changes in inflationary
expectations, and so they must represent
real interest rate movements. Such real
interest rate changes almost surely have an
impact on real resource allocations.16
The second piece of evidence comes
from the examination of the neutrality of
money in Cecchetti (1986, 1987). In those
papers, I establish that output growth is
significantly correlated with money growth
at lags of up to 10 years! There are several
possible interpretations of these findings,
but they strongly suggest that monetary
shocks have something to do with aggregate
real fluctuations.
Innovations to the Federal Funds Rate
To understand the shortcomings of these
two approaches, I will describe how each is
used. In the first, researchers begin by specify
ing a vector autoregression. For the purposes
of the example, I will use the formulation in
Bernanke and Blinder’s (1992) Section IV
They employ a six-variable specification with
the total civilian unemployment rate, the log of
the CPI, the federal funds rate, and the log of
three bank balance sheet measures, all in real
terms: deposits, securities and loans. The
assumption is that the federal funds rate is a
“policy” variable, and so it is unaffected by all
other contemporaneous innovations.18
Following Bernanke and Blinder, I esti
mate the VAR with six lags using seasonally
adjusted monthly data.19 Figures 1 and 2
plot some interesting results from this VAR.
The first figure shows the estimated residuals
from the federal funds rate equation. The
solid vertical lines are National Bureau of
Econom ic Research (NBER) reference cycle
peaks and troughs, while the dashed vertical
lines are the Romer and Romer dates, intend
ed to indicate the onset of contractionary
M EASURING INNOVATIONS
TO M ONETARY POLICY
It stands to reason that before one can
study the monetary transmission mecha
nism, it is necessary to identify monetary
shocks. A number of authors have argued
convincingly that policy disturbances cannot
be gauged by examining movements in the
monetary aggregates. The reason is that the
variance in the innovations to broad measures
1995
88
REVIEW
MAT/JUNE
monetary policy episodes.
This series looks extremely noisy and
it is hard to see how it could represent policy
changes. The 1979-82 period is the only
one with large positive or negative values.
Although it is surely the case that there are
unanticipated policy changes both when the
Federal Reserve acts and when it does not,
one would expect small normal shocks with
occasional spikes. If decisions are really this
random, there is something fundamentally
wrong with the policymaking apparatus.
Furthermore, since the federal funds rate
itself is the equilibrium price in the reserves
market, given technicalities of the way that
monetary policy is actually carried out, the
market-determined level of the funds rate is
not a policy instrument.20
The second figure shows the response of
the log of the CPI to a positive one percentage
point innovation in the federal funds rate. To
understand how this is computed, begin by
writing the vector autoregression as
1995
F ig u re 1
Estim ated In n o va tio n s to the Fe d e ra l
Funds Rates
Percentage points
R
R
R R
R
F ig u re 2
Response of the Log CPI to a C han ge in
the F e d e ra l Funds Rate
A(L)y, = £t ,
where A(L) is a matrix of polynomials in the
lag operator L (L'yt = y ti) , y is the vector
of variables used in the estimation, and et is
mean zero independent (but potentially heteroskedastic) error. The first step is to esti
mate the reduced form version of equation 1
by assuming that no contemporaneous vari
ables appear on the right-hand side of any
equations (A(0) = I). This results in an esti
mate A it ) along with an estimated covariance
matrix for the coefficient estimates— call
this ft. The impulse response functions
are obtained by inverting the estimated lag
polynomial B(L) = A 1(L ).21
But the point estimate of the impulse
response function is not really enough to
allow us to reach solid conclusions. It is also
important to construct confidence intervals
for the estimates. There are two ways to do
this. The first involves the technique that
has been called Monte Carlo Integration.
This is a Bayesian procedure that involves
presuming that the distribution of the vector
of errors in equation 1— the et’s— is i.i.d.
normal.22 To avoid making such stringent
assumptions, I choose to estimate confidence
bands using an alternative technique grounded
Horizon in months
in classical statistics.
The delta method is the simple proce
dure that comes from noting that if the esti
mates of the coefficients in lag polynomials
are asymptotically normally distributed, then
any well-behaved function of these parameters
will also be asymptotically normally distrib
uted. Stacking all of the parameters in A(L)
and calling the result 6, then
V t (0 - ■0)~N ( 0 ,f t ) .
It follows that any function of these
parameters/(0)— for example, the impulse
response function— will be asymptotically
normally distributed,
89
20 This entire discussion ignores the
possibility that anticipated mone
tary policy matters— something
that researchers should consider
bringing into the discussion.
21A simple way to calculate the vec
tor moving-overage form of equa
tion 1 is to construct the compan
ion form of the VAR os described in
Sargent (1 9 8 7 ). This is also dis
cussed in Homilton (1 9 9 4 ).
22 See Doan (1 9 9 0 ).
tions, because they feel that expansions were
more ambiguous. Since most models predict
symmetric responses to positive and negative
monetary innovations, this strategy throws
out information.
But the main issue is the exogeneity
of the policy shifts. It is difficult to believe
that the actions of the FOMC, as reported in
the minutes of the meetings, are truly exoge
nous events. There have been two responses
to this. First, Hoover and Perez (1 991) pro
vide a lengthy discussion of why Romer and
Romer’s methods are not compelling in iden
tifying output fluctuations induced by exoge
nous monetary shocks.
Taking a slightly different approach,
Shapiro (1 9 9 4 ) examines whether the FOMC
is responding to changes in econom ic condi
tions, and so there is some reaction function
im plicit in policy. He estimates a probit
model for the Romer and Romer dates using
measures of inflation and unemployment,
both as deviations from a carefully constructed
target level, as determinants. Figure 3 repro
duces his estimates of the probability of a
date, with the vertical lines representing the
dates themselves. The unanticipated policy
action is 1 minus the estimated probability.
As is clear from the figure, several o f the
dates were largely anticipated, and there
were some periods when policy shifts were
thought to be likely, and then did not occur.
Overall, Shapiro’s results suggest that the
standard interpretation of the dummy vari
ables as exogenous is incorrect to varying
degrees over time.
There seems to be no way to measure
monetary policy actions that does not raise
serious objections. Given this, it might seem
difficult to see how to proceed with the study
of different theories of the transmission
mechanism. But the literature proceeds in
two directions. The first uses these measures
directly in an attempt to gauge the influence
of policy changes directly. The conclusions
of these studies must be viewed with some
degree of skepticism. The alternative
approach is to note that investment declines
account for the major share of output reduc
tions during recessions. If one is able to
show that the distribution of the contraction
in investment is correlated with variables
Jr(f[e]-f[e]yN(o,n}),
where
s dd'
23 A test for whether all of the reac
tions are jointly zero rejects ot the
5 percent level for the first seven
months, and then fails to reject.
The test for whether all the effects
ore zero simultaneously for all 24
months foils to reject with o p-volue
of 0 .77.
24 There is a further reason to view
this measure of policy with some
skepticism. Because of the lorge
number of parameters estimated,
the regressions are usually overfit
ted. As a result, they normally
have very poor out-of-sample fore
casting properties.
25 See Friedman (1 9 9 0 ) for a discus
sion of the strategy of using policy
makers' statements to gauge their
actions.
de
which can be estimated numerically.
The result plotted in Figure 2 was first
pointed out by Sims and is known as the
“price puzzle.” Paradoxically, the VAR esti
mates imply that monetary policy contrac
tions lead to price increases! As is clear from
the estimated standard-error bands, this price
rise is significantly positive for approximately
the first year. After two years, however, it is
not possible to reject the hypothesis that a
funds rate increase has no effect on the price
level.23
The standard conclusion is that the VAR
is misspecified in some way. One strong pos
sibility is that the funds rate is not exogenous
in the way that is required for this identifica
tion to be valid, and so these innovations do
not accurately reflect policy movements.24
My conclusions may be too harsh for the
following reason. As Ben Bernanke pointed
out in the conference, the estimated innova
tions are the sum of true policy innovation,
policy responses to omitted variables, and
more general specification errors in the VAR.
As a result, one would expect them to be
noisy. Furthermore, as pointed out by Adrian
Pagan, since one is primarily interested in the
impulse response functions— the impact of
unanticipated policy on output, prices and
the like— then it may be immaterial that the
estimated policy innovations are noisy even
if the true innovations are not.
The Romer and Romer Dates
The Romer and Romer dates have been
both widely used and extensively criticized.25
They suffer from both technical and substan
tive problems. First, they are discrete.
Presumably, policy changes have both an
intensity and a timing. Ignoring the size of
policy changes must have an impact on
results. Second, Romer and Romer choose to
focus their inquiry only on policy contrac
90
R[VI[N
MAY/JUNE
related to a firm’s balance sheet and its access
to bank loans, then this strongly suggests the
existence of a lending channel.
1995
F ig u re 3
Estim ated P ro b a b ility of a
R o m er a n d R o m er D ate
(Shapiro probit model using inflation and unemploymnt)
USING AGGREGATE DATA
0.6
Numerous studies have used aggregate
data in an attempt to distinguish the channels
of monetary transmission. This literature
can be divided into three categories: The
first looks at the relative forecasting ability
of different quantity aggregates; the second
studies differences in the timing of the
response of aggregate quantities to presumed
policy shocks; and the third examines the
behavior o f interest rates.
Before examining the work on quantities,
I will discuss the use of interest rate data.26
As is clear from the discussion in the first
section, the lending view does allow for
movements in market interest rates.
Furthermore, these movements are in the
same direction as those predicted by the
money view, and their magnitude depends
solely on the degree of substitutability
between outside money and various other
assets. Where the two views differ is in their
predictions for movements in the interest
rate on loans. But since there is currently no
secondary market for these securities, it is
impossible to determine the interest rate on
these loans.27 This implies that market interest
rates are of virtually no use in this exercise.
There is no sense in which the behavior of
interest rates could serve to distinguish
between the money or lending views.
I now turn to the work on quantities. In
the following section, I examine tests involving
the relative forecasting ability of measures of
money and credit. This is followed by a dis
cussion of papers that emphasize aggregate
timing relationships.
0 .5 ■
0.4 ■
0.3 ■
■
0.1
■
1953 56
59
62
65
. J
L
71
74
I
l
A number of papers have examined the
ability of different financial aggregates to
forecast output (or unemployment) fluctua
tions. Ramey (1993) is a recent example.
The main methodology here is to ask whether
measures of credit are informative about
future output movements, once money has
Aggregate Timing Relationships
The second use of aggregate data has
been to examine the response of various
91
U
u
77
been taken into account. The problem with
this is that credit is usually ju st a broader
measure of money. To put it slightly differently,
the balance sheet identity of the banking
system implies that bank assets equal bank
liabilities. As Bernanke (1993b) points out,
monetary aggregates are a measure of bank
liabilities, while credit aggregates are measures
of bank assets. Since these are calculated
slightly differently, they will not be identical.
But it is these technical measurement differ
ences that are likely to account for the differ
ences in forecasting ability, not anything
about the transmission mechanism.
More generally, the main finding is that
credit lags output. Unfortunately, this tells
us nothing about the transmission mechanism.
The aggregate data do show that aggregate
credit is countercyclical, but it is easy to find
explanations for this that are consistent with
the lending view. For example, Kiyotaki and
Moore (1 993) present a model in which indi
viduals must continue to service credit even
after income falls, and so credit falls after
incom e even though it is the fundamental
source of fluctuations. In the end, it is diffi
cult to see how aggregate timing relationships
can tell us anything at all about the way in
which monetary policy affects real activity.28
Relative Forecasting Ability
0.2
83
L
86
*
89 1992
26 Miron, Romer and Weil (1 994)
study preWorld War II interest
rates in an attempt to address
these questions.
27 In the presence of rationing, there
is the odded complication that one
would need observations on the
shadow price for a loan to a bor
rower who is deemed not to be
creditworthy given the economic
environment. Obviously, there is
no eosy way to infer such a price.
28 This point is also mode by
Bernanke, Gertler ond Gilchrist
(1 9 9 4 ).
impact of federal funds rate innovations on
various parts of bank balance sheets. These
results are based on the estimation of a large
number of parameters with a relatively small
amount of data— this VAR has 237 parameters
and 354 data points— and so the estimates
are fairly imprecise.30
But even if one were to find that the
impulse responses differed significantly, this
would only bear on the substitutability of the
assets, and not directly on the validity of the
lending view. Both the prices and quantities
of perfect substitutes must have the same
stochastic process, and so finding that this
particular partial correlation is different
would be evidence of imperfect substitutabil
ity. As Bernanke and Blinder (1 9 9 2 ) make
clear in discussing their findings, this is a
necessary but not a sufficient condition for
the lending view to hold. It is not possible,
using reduced-form estimates based on
aggregate data alone, to identify whether
bank balance sheet contractions are caused
by shifts in loan supply or loan demand.
W hat is needed is a variable that is known
to shift one curve but not the other.
Kashyap, Stein and W ilcox (1 993) also
provide evidence based on aggregate timing.
They compare the response of bank loans to
that o f commercial paper issuance following
policy innovations. They find that monetary
policy contractions seem to decrease the
mix of loans relative to commercial paper.
Borrowers that can move away from direct
bank finance following a tightening appear
to do so. Both Friedman and Kuttner (1 9 9 3 ),
and Oliner and Rudebusch (1 993) take issue
with these findings and show that changes in
the mix are due to increases in the amount of
commercial paper issuance during a recession,
but that the quantity of bank loans does not
change. In addition, Oliner and Rudebusch
show that once firm size is taken into account,
and trade credit is included in the debt of the
small firms, the mix of financing is left unaf
fected by policy changes.
It is worth making an additional point
about the commercial paper market. First,
Post (1 9 9 2 ) documents that all commercial
paper rated by a rating agency must have a
backup source of liquidity, which is generally
a bank line of credit or a standby letter of
F ig u re 4
B e rn a n k e a n d B lin d e r P lot of Responses
to F e d e ra l Funds Rate Shock
Horizon in months
29 Morgan (1 9 92 ) and Strongin
(1 9 9 1 ) ore also examples of this
type of work.
301 have tried two variants of the
Bernanke and Blinder VAR method
that might seem promising ways of
addressing the problem of measur
ing monetary policy changes. In
the first, I substituted the funds rote
target as reported in Sellon (1 9 94 )
for the actual funds rate. This has
very little impact on the results, as
the Fed comes extremely close to
hitting the targets. Second, I made
the alternative extreme assumption
that the funds rate itself is exoge
nous. This has very dramatic
effects on the results, as Bernanke
ond Blinder's conclusions ore com
pletely unsupported. If all move
ments in the funds rote ore
assumed to represent exogenous
policy actions, it would be extreme
ly difficult to claim that loans and
securities responded differently to
policy shifts.
financial quantities to policy innovations.
Returning to Bernanke and Blinder (1 9 9 2 ),
they study whether bank loans and securities
respond differently to federal funds rate
innovations.29 The standard methodology is
to calculate the impulse responses for the
two variables and note that they look different.
Figure 4 reports the common finding, calcu
lated using the six-variable Bernanke and
Blinder VAR estimated over the 1959-90
sample. In response to a positive 1 percentage
point innovation in the federal funds rate,
the unemployment rate rises by nearly 0.1
percentage point after one-and-a-half years,
while bank securities fall 0.07 percent and
loans decline 0.0 2 percent. Securities fall
both by a larger amount and more quickly
than loans.
But point estimates of these impulse
responses do not tell the entire story. In
Figure 5 , 1 plot the point estimate and two
standard error bands for the difference
between the impulse response for loans and
securities. This allows an explicit test of
whether these two assets are imperfectly sub
stitutable in response to the shock. The dif
ferences are individually greater than zero in
only a few months, and a jo in t test of the first
24 months of the impulse response, which is
asymptotically distributed as a Chi-squared,
has a p-value of 0.70.
My conclusion is that reduced-form vec
tor autoregressions are nearly incapable of
providing convincing evidence of a differential
92
REVIEW
MAY/JUNI
credit. This means that commercial paper is
an indirect liability of banks, albeit one that
is not on their balance sheet. Furthermore,
Calomiris, Himmelberg and W achtel (1994)
suggest that increases in commercial paper
issuance are accompanied by an increase in
trade credit. This means that a policy con
traction may simply cause a re-shuffling of
credit by forcing banks to move liabilities off
of their balance sheet such that large firms
issue commercial paper in order to provide
trade credit to small firms that would have
otherwise come from banks.
1995
F ig u r e 5
D ifference B e tw e e n Loan a n d Se curity
Response to F e d e ra l Funds Rate Shock
(p e rc e n ta g e ch an ge a t a n n u a l ra te w ith tw o s ta n d a rd -d e v ia tio n b a n d s)
USING CROSS-SECTIONAL
DATA
Horizon in months
There is a large empirical literature using
cross-sectional data that is relevant to under
standing the channels of monetary policy.
These studies fall into groups that separately
address the two parts of the lending view.
The first set of papers tries to gauge the
importance of capital market imperfections
on investment, and so is related to the balance
sheet effects described in the first section.
The second set, which is fairly small, examines
time-series variation in cross-sectional data
in an attempt to characterize the distributional
effects of monetary policy directly. I will
briefly describe each of these strategies.
pioneered the technique of dividing firm-level
data into groups using measures thought to
correspond to the project monitoring costs
created by information asymmetries, and
then seeing if the correlation between invest
ment and cash-flow measures varies across
the groups. The finding in a wide range of
studies is that investment is more sensitive
to cash-flow variables for firms who have
ready access to outside sources of funds.31
The main issue in interpreting these
results is whether the characteristics of the
firm used to split the sample are exogenous
to financing decisions. Measures of firm
size, dividend policies, bond ratings and the
like may be related to the quality of investment
projects a firm has available, and so lender
discrimination may not be a consequence of
asymmetric information.
There are several examples in which
researchers identify potentially constrained
firms based on institutional characteristics, and
so the endogeneity problems are mitigated.
I will m ention two. Hoshi, Kashyap and
Scharfstein (1 991) find that investment by
Japanese firms that were members of a
keiretsu, or industrial group, was not influ
enced by liquidity effects. Using data on
individual hospitals, Calem and Rizzo (1994)
find that investment depends more heavily on
cash-flow variables for small, single-unit hos
Measuring Capital Market
Imperfections
The literature on capital market imper
fections is an outgrowth of the vast work
done on the determinants of investment.
The general finding in this literature is that
internal finance is less costly than external
finance for firms that have poor access to
primary capital markets.
The empirical studies fall into two cate
gories. The first examines reduced-form
correlations, while the second looks directly at
the relationship between the cost and expected
return to a marginal investment project— they
estimate structural Euler equations.
pitals than for large, network-affiliated ones.
In the most convincing study of this
type, Calomiris and Hubbard (1 9 9 3 ) study
Reduced-Form Correlations
Fazzari, Hubbard and Petersen (1988)
NK OF ST. L OU IS
93
si Beman);e>Gertier and Gilchrist
(1994) survey the large number of
studies that use this approoch.
review
MAY/JUNE
the undistributed corporate profits tax in
1936 and 1937 to estimate the differences in
financing costs directly from firms’ responses
to the institution of a graduated surtax
intended to force an increase in the dividend
payout rate. Their results, holding investment
opportunities fixed, are that investment
spending is affected by the level of internal
funds only for those firms with low levels of
dividend payments and high marginal tax
rates. Furthermore, these tended to be
smaller and faster growing firms.
sequence of shifts in loan demand or loan
supply. My conclusion is that these studies
fail to establish the desired result in a con
vincing way. Instead, they provide further
evidence of capital market imperfections.
Three major studies use data on manu
facturing firms. In the first, Gertler and
Hubbard (1 988) find that the impact of cash
flow on investment is higher during recessions
for firms that retain a high percentage of
their earnings. The second, by Kashyap,
Lamont and Stein (1 9 9 2 ), shows that during
the 1981-82 recession, the inventories of
firms without ready access to external
finance fell by more when their initial level
of internal cash was lower. On the other
hand, the inventory investment behavior of
firms with ready access to primary capital
markets showed no evidence of liquidity
constraints. In the third, Gertler and
Gilchrist (1 994) use the Quarterly Financial
Report f o r M anufacturing Corporations (QFR)
to divide firms into asset-size categories and
find that small firms account for a dispropor
tionate share of the decline in manufacturing
following a monetary shock.32
Both Kashyap and Stein (1994b ) and
Peek and Rosengren (forthcoming) focus on
the behavior of lenders rather than borrow
ers. By examining the cyclical behavior of
banks, Kashyap and Stein hope to find evi
dence for the importance of loan supply
shifts. The strongest result in their paper is
that, following a monetary contraction, the
total quantity of loans held by small banks
falls while that of large banks does not. By
contrast, Peek and Rosengren study New
England banks during the 1990-91 recession
and find that poorly capitalized banks shrink
by more than equivalent institutions with
higher net worth. My interpretation is that
both of these show that the capital market
imperfections commonly found to apply to
manufacturing firms apply to banks as well.
There are two difficulties inherent in any
attempt to establish that the important trans
mission mechanism for monetary policy
shocks is through bank loan supply shifts.
First, as described at length in the second
section, there is the problem of empirically
identifying monetary policy. Beyond this,
there is the subtlety of distinguishing loan
Structural Estimation
The neoclassical theory of investment
allows one to derive the complex equilibrium
relationship among the capital stock, rates of
return, future marginal value products and
project costs that form the first-order condi
tions for a firm’s problem. W ith the appro
priate data, it is then possible to see whether
these Euler equations hold. Hubbard and
Kashyap (1992) is an interesting use of this
technique. Following the work of Zeldes
(1989) on consumption, they examine
whether the ability of agricultural firms
to meet this first-order condition depends
on the extent of their collaterizable net
worth. They find that during periods when
farmers have high net worth, and so have
better access to external financing, their
investment behavior is more likely to look as
if it is unconstrained.
These investment studies have been very
successful in establishing the existence of
capital market imperfections as well as their
likely source in information asymmetries
arising from monitoring problems. W hile
the work has little to say about monetary
policy directly, it does provide an excellent
characterization of the distributional effects
of changes in the health of firms’ balance
sheets regardless of the source.
Time-Series Evidence
32 Oliner and Rudebusch (1 9 94 ) and
Betnanke, Gertler and Gilchrist
(1 9 9 4 ) obtain similar results using
the QFR data os well.
The strategy in the second set of studies
is to use the cross-sectional dimension to
identify the transmission mechanism. The
goal is to determine whether the reduction in
loans during monetary contractions is a con
199
94
n
MAY/JUNE
over time. With the introduction of interstate
banking and the development of more
sophisticated instruments aimed at trading
pools of loans, it is only the balance sheet
effects that will remain. As a result, it is
important to know which is the more impor
tant channel of monetary policy transmission.
supply shifts from the balance sheet effects.
Is the observed reduction in loans a conse
quence of their complementarity with outside
money caused by the structure of the banking
system, or is it the result of changes in the
shape of the marginal efficiency of investment
schedule brought on by the balance sheet
effects? Kashyap, Lamont and Stein (1992)
suggest one possible way of distinguishing
these possibilities. If one can find a reces
sionary period that was not preceded by a
monetary contraction, and show that interest
rates rose but that bank dependence was
irrelevant to individual firms’ experiences,
this would mean that banks are responsible
for the distributional effects induced by
monetary shocks. Unfortunately, such
evidence is not readily available.
REFERENCES
Bernanke, Ben S. "How Important is the Credit Channel in the
Transmission of Monetary Policy? A Comment," Carnegie-Rocbester
Conference Series on Public Policy 39 (December 1993) pp. 47-52.
_ _ _ _ _ _ . "Credit in the Macroeconomy," Federal Reserve Bonk of
New York Quarterly Review(spring 1993) pp. 50-70.
_ _ _ _ _ _ and Mark Gertler. "Financial Fragility and Economic
Performance," QuarterlyJournal of Fconomics (February 1990), pp.
87-114.
_ _ _ _ _ _ and Mark Gertler. "Agency Cost, Net Worth and Business
Fluctuations," Tbe American Economic Review (March 1989),
pp. 14-31.
CONCLUSION
After a survey of the work that attempts
to distinguish theories of the monetary trans
mission mechanism, where do we stand?
My conclusion is that the myriad studies
have succeeded in establishing the empirical
importance of credit market imperfections.
This means that monetary policy shifts have
an important distributional aspect that
cannot be addressed within the traditional
money view. It is the smaller and faster
growing firms that bear a disproportionate
share of the burden imposed by a recession.
Since these are likely to be firms with highly
profitable investment opportunities, this has
important implications for social welfare.
Not only are recessions associated with
aggregate output and investment declines,
but the declines are inefficient.
Beyond this, there is the issue of distin
guishing the two parts of the lending view.
Do we care if we can distinguish changes in
investment opportunities resulting from
financial accelerator effects from bank loan
supply shifts per se? Does the conclusion
have implications for the actual conduct of
monetary policy? I believe the answer is yes.
If the complementarity of bank loans and
outside money arises largely as a result of the
financial regulatory environment, then, with
financial innovation and liberalization, these
effects are likely to become less important
1995
_ _ _ _ _ _ ond Alan S. Blinder. "The Federal Funds Rate and the
Channels of Monetary Transmission," Tbe American Economic Review
(September 1992), pp. 901-21.
_ _ _ _ _ _ Mark Gertler ond Simon Gilchrist. "The Financial Accelerator
and the Flight to Quality," National Bureau of Economic Research
Working Paper No. 4789 (July 1994).
Broinard, William C., and James Tobin. "Financial Intermediaries and
the Effectiveness of Monetary Controls," Tbe American Economic
Review (May 1963), pp. 383-400.
Boschen, John F., and Leonard 0. Mills. "The Effects of Countercyclical
Monetary Policy on Money and Interest Rates: An Evaluation of
Evidence from F0MC Documents," working poper (1992), College of
William and Mary.
Colem, Paul, ond John A. Rizzo. "Financing Constraints and Investment:
New Evidence from Hospital Industry Data," working paper (1994),
Yale University, School of Medicine, Deportment of Epidemiology and
Public Health.
Calomiris, Charles W., Charles P. Himmelberg and Paul Wochtel.
"Commercial Paper and Corporate Finance: A Microeconomic
Perspective," working paper (April 1994), Stern School of Business,
New York University.
_ _ _ _ _ _ and R. Glenn Hubbard. "Tax Policy, Internal Finance, and
Investment," Internal Finance and Investment: Evidence from the
Undistributed Corporate Profits Tax of 1936-37, National Bureau of
Economic Research Working Paper No. 4288 (March 1993).
Cecchetti, Stephen G. "Testing Short Run Neutrality: International
Evidence," Reviewof Economics and Statistics (February 1987)
pp. 135-40.
95
_ _ _ _ _ _ . "Testing Short Run Neutrality," Journal of Monetary
Economics (May 1986) pp. 409-23.
James, Christopher. "Some Evidence on the Uniqueness of Bank
Loons," Journal of Financial Economics (1987), pp. 217-36.
Kashyap, Anil K., and Jeremy C. Stein. "Monetary Policy and Bank
Lending," in N. Gregory Mankiw, ed., Monetary Policy. University of
Chicogo Press for the National Bureau of Economic Research, 1994,
pp. 221-62.
_ _ _ _ _ _ and_ _ _ _ _ _ _ . "The Impact of Monetary Policy on Bank
Balance Sheets," working paper (March 1994), University of
Chicago, Graduate School of Business.
_ _ _ _ _ _ , Jeremy C. Stein and David W. Wilcox. "Monetory Policy
and Credit Conditions: Evidence from the Composition of External
Finance," The American Economic Review (March 1993), pp. 78-98.
_ _ _ _ _ _ , Owen A. Lamont and Jeremy C. Stein. "Credit Conditions
and the Cyclical Behavior of Inventories: A Case Study of the 19811982 Recession," National Bureau of Economic Research Working
Paper No. 4211 (November 1992).
King, Robert G., and Charles I. Plosser. "Money, Credit and Prices in A
Real Business Cycle," The American Economic Review (June 1984),
pp. 363-80.
King, Stephen R. "Monetory Transmission: Through Bank Loans or
Bank Liabilities?" Journal of Money, Credit and Bonking (August
1986), pp. 290-303.
Kiyotaki, Nobuhiro, and John Moore. "Credit Cycles," working paper
(Morch 1993), London School of Economics.
Lucas, Robert E., Jr. "Liquidity and Interest Rates," Journal of Economic
Theory (1990), pp. 237-64.
Miron, Jeffrey, Christina D. Romer and David Weil. "Historical
Perspectives on the Monetary Transmission Mechanism," in N. Gregory
Mankiw, ed., Monetary Policy. University of Chicago Press for the
National Bureau of Economic Research, 1994, pp. 263-306.
Morgan, Donald P. "The Lending View of Monetary Policy and Bank
Loan Commitments," working paper (December 1992), Federal
Reserve Bonk of Konsos City.
Oliner, Stephen, and Glenn D. Rudebusch. "Is There a Bank Credit
Channel for Monetary Policy?" Board of Governors of the Federal
Reserve System, Finance and Economics Division Discussion Paper No.
93-8 (February 1993).
Peek, Joe, and Eric Rosengren. "The Capital Crunch: Neither a Borrower
Nor a Lender Be," Journal of Money, Credit and Banking
(forthcoming).
Post, Mitchell A. "The Evolution of the U.S. Commercial Paper Market
Since 1980," Federal Reserve Bulletin (December 1992),
pp. 879-91.
Ramey, Valerie. "How Important is the Credit Channel in the
Transmission of Monetory Policy?" Camegie-Rochester Conference
Series on Public Policy (December 1993), pp. 1-45.
Romer, Christina D., and David H. Romer. "Credit Channels or Credit
Actions? An Interpretation of the Postwar Transmission Mechanism,"
in Changing Capital Markets: Implications for Monetary Policy.
federal Reserve Bank of Kansas City, 1993.
Christiano, Lawrence, and Martin S. Eichenbaum. "Liquidity Effects and
the Monetary Transmission Mechanism," The American Economic
Review (May 1992) pp. 346-53.
Doan, Thomas A. User's Manual, RATS Version 3.10. VAR Econometrics,
1990.
Fama, Eugene E. "Bonking in the Theory of Finance," Journal of
Monetary Economics (January 1980) pp. 39-57.
Fazzari, Steven M., R. Glenn Hubbard ond Bruce C. Petersen.
"Financing Constraints and Corporate Investment," Brookings Papers
on Economic Activity (1988:1) pp. 141-95.
Feurst, Timothy. "Liquidity Effect Models and Their Implications for
Monetory Policy," working paper (April 1993), Bowling Green State
University, Department of Economics.
Friedman, Benjamin M. "Discussion of 'New Evidence on the Monetary
Transmission Mechanism," Brookings Papers on Economic Activity
(1990:1), pp. 204-9.
_ _ _ _ _ _ and Kenneth N. Kuttner. "Economic Activity ond the ShortTerm Credit Markets: An Analysis of Prices and Quantities," Brookings
Papers on Economic Activity (1993:2), pp. 193-266.
Friedman, Milton, ond Anna Schwartz. A Monetary History of the United
States, 1867-1960. Princeton University Press, 1963.
Gertler, Mork, and Simon Gilchrist. "Monetary Policy, Business Cycles,
and the Behavior of Small Manufacturing Firms," Quarterly Journal of
Economics (May 1994) pp. 309-40.
_ _ _ _ _ _ and_ _ _ _ _ _ _ . "The Role of Credit Market Imperfections
in the Monetary Transmission Mechanism: Arguments ond Evidence,"
Scandinavian Journal of Economics (1993) pp. 34-64.
_ _ _ _ _ _ and R. Glenn Hubbard. "Financial Factors in Business
Fluctuations," National Bureau of Economic Research Working Paper
No. 2758 (November 1988).
Hamilton, James D. Time Senes Analysis. Princeton University Press, 1994.
Hoover, Kevin D., and Stephen J. Perez. "Post Hoc Ergo Propter Hoc
Once More: An Evaluation of 'Does Monetary Policy Matter?' in the
Spirit of James Tobin," Journal of Monetary Economics (February
1992), pp. 3-24.
Hoshi, Takeo, Anil K. Kashyap and David Scharfstein. "Corporate
Structure, Liquidity and Investment: Evidence from Japanese Panel
Data," Quarterly Journal of Economics (February 1991), pp. 33-60.
Hubbard, R. Glenn. "Is There a 'Credit Channel' for Monetary Policy?"
this Review (May/June), pp. 63-82.
_ _ _ _ _ _ and Anil K. Kashyap. "Internal Net Worth ond the
Investment Process: An Application to U.S. Agriculture," Journal of
Political Economy (June 1992) pp. 506-34.
_ _ _ _ _ _ and_ _ _ _ _ _ _ . "New Evidence on the Monetary
Transmission Mechanism," Brookings Papers on Economic Activity
(1990:1), pp. 149-213.
Ingersoll, Jonathan E. Theory of Financial Decision Making. Rowman
and Littlefield, 1987.
FE D E R A L RESERVE B A N K OF ST . L OU I S
96
REVIEW
MAT/JUNE
_ _ _ _ _ _ and_ _ _ _ _ _ _ . "Does Monetary Policy Matter? A New
Test in the Spirit of Friedman and Schwartz," National Bureau of
Economic Research Macroeconomics Annual. MIT Press, 1989,
pp. 121-70.
Stiglitz, Joseph E., and Andrew Weiss. "Credit Rationing in Markets with
Imperfect Information," The American Economic Review (June 1981),
pp. 393-410.
Strongin, Steven. "The Identification of Monetary Policy Disturbances:
Explaining the Liquidity Puzzle," Federal Reserve Bank of Chicago
Working Paper No. 92-27 (November 1992).
Sargent, Thomas J. Macroeconomic Theory, Second Edition. Academic
Press, 1987.
Sellon, Gordon H.( Jr. "Measuring Monetary Policy," working paper
(August 1994), Federal Reserve Bank of Kansas City.
Zeldes, Stephen P. "Consumph'on and Liquidity Constraints: An
Empirical Investigation," Journal of Political Economy (April 1989),
pp. 305-46.
Shapiro, Matthew D. "Federal Reserve Policy: Cause and Effect," in N.
Gregory Mankiw, ed., Monetary Policy. University of Chicago Press for
the National Bureau of Economic Research, 1994, pp. 307-34.
Sims, Christopher. "Interpreting the Macroeconomic Time Series Facts:
The Effects of Monetary Policy," European Economic Review (May
1992), pp. 975-1000.
1995
97
REVIEW
MAY/JUNK
199
Mark Gertler is a professor of economics ot New York University.
Commentary
In the benchmark case of perfect capital mar
kets, Q( = 1 . As credit constraints tighten,
internal funds become more valuable and,
consequently, Qt rises. Notice that the spread
M ark Gertler
between the cost of external and internal
funds is approximately Q, - 1.
teve Cecchetti has written a very nice
Now suppose that X ( is a decision variable
survey. A central point of his article,
in an intertemporal choice problem. In the
with which I completely agree, is that in
case of a firm, for example, X ( could represent
assessing the empirical relevance of the credit either capital investment or inventory invest
channel literature, it is incorrect to focus on
ment. In the case of a household, it could
credit aggregates. Perhaps contrary to con
represent saving or the acquisition of a con
ventional wisdom, these theories do not
sumer durable. At the optimum, the agent
imply that credit aggregates should forecast
adjusts Xt to the point where the marginal
output better than standard indicators of
cost equals the discounted marginal benefit:
monetary policy, nor do they imply that
credit aggregates should lead output over the
(1) MC, (X t) = [1/Qt (1 + p()]E {M B ltl(X()}.
business cycle.1 Because I agree with virtually
everything Steve said, I would like to take
Since Qt influences the effective discount
the opportunity to clarify what I believe are
rate, it ultimately influences the choice of Xt.
the central elements of the literature.
Up to this point, we have taken Qt as
In my view, a credit channel for monetary
given. In the theoretical literature on the
financial propagation mechanism, Q( is
policy is a special case of a financial propaga
tion mechanism. To understand the former,
derived endogenously, and in equilibrium it
therefore, it is first useful to define the latter.
is determined jointly with X (. Roughly
Suppose that 1 + pt is the gross borrowing
speaking, the theory suggests that Qt should
rate that an economic agent would face at
depend on financial variables such as the
agent’s net worth (that is, collateral and
time t if capital markets were perfect. The
internal funds) and the availability of inter
net rate pt equals the sum of the safe rate
plus a premium that depends on the system
mediary credit. That is, we can write:
atic risk of the agent’s investment.
Suppose now that we allow for the pos
(2) Q, = Q(NWt, IC(),
sibility that the agent may have imperfect
access to the capital market. In this
where NW is the agent’s net worth and 1C is
instance, due to information and enforce
an indicator of the availability of intermediary
ment problems, the price of external funds
credit.
will exceed the (risk-corrected) opportunity
Now to the point: A financial propaga
cost of internal funds.2 The effective bor
tion mechanism amplifies the impact of a
rowing rate that the agent faces may be
primitive disturbance on X ( via an impact on
expressed as Q(( l + p(), where Qt > 1. This
Qt. One example is Bernanke’s (1 983) theory
rate is the explicit cost of funds if there are
of the Great Depression. The bank runs
no non-price terms to the loan. Otherwise,
associated with the initial downturn reduced
it is the implicit cost, after taking into
the availability of intermediary credit, forcing
account the effect of the non-price terms.
up Qt, further depressing spending and out
The multiplier Qt is interpre table as the
put. Another example is the financial accel
shadow cost of a unit of internal funds. It is
erator theory of investment described in
the maximum the agent would be willing to
Bernanke and Gertler (1 9 8 9 ). In that frame
pay for an additional unit of internal funds.
work, endogenous procyclical movements
S
1 In addition to Cecchetti (1 9 9 5 ),
see the discussion in Bernanke,
GerHer and Gilchrist (forthcoming),
and Kashyap and Stein (1 9 9 4 ).
2 For more detail, see Bernanke,
GerHer ond Gilchrist (forthcoming)
98
REVIEW
MAY/JUNE
in the strength of firm balance sheets
induce countercyclical movements in Qt.
The resulting countercyclical movements
in the spread between the cost of external
and internal funds serve to amplify
investment fluctuations.
A credit channel for monetary policy is a
financial propagation mechanism, in which
the primitive impulse is monetary policy. A
credit channel thus amplifies the impact of a
shift in interest rates induced by monetary
policy by causing an associated movement in
the spread between the cost of external and
internal funds; that is, by altering how
smoothly funds flow between lenders and
borrowers. In my view, the variety of types
of mechanisms in the literature that have
received the label “credit channel” fit this
broad definition.
Let me illustrate the impact of a credit
channel with the following simple neoclassical
investment problem. Let Y be the firm’s out
put, K its capital stock, I the rate of invest
ment, 6 a technology parameter and 8 the
rate of depreciation. Then suppose
are fixed at unity. A shift in monetary policy
alters the short-term interest rate, directly
influencing the discount rate 1/(1 +pt) and,
in turn, the firm’s investment decision. W ith
a credit channel present, however, Qt and
possibly also Q( +| change in a way that mag
nifies the overall impact.
The key point I wish to emphasize is
that relatively small changes in Q( may have
a large effect (that is, the propagation mecha
nism may be strong). A numerical example is
helpful. Consider the impact of 1 percentage
point change in Q( on the firm’s investment
decision. This corresponds to a 100 basis
point rise in the cost of external funds relative
to internal funds. If the rise in Qt is persistent
(so that Qt+1 adjusts to keep Q( unchanged),
then investment drops 12.5 percent, assum
ing conventional values for the exogenous
parameters.3 That is a large effect.
If the rise in Q( is purely transitory, so
that Q[+l is unchanged, then investment drops
a whopping 250 percent. In this latter case,
there is a strong intertemporal substitution
effect. Because the shadow price of internal
funds is high today relative to tomorrow, the
firm defers investment. More informally, if
credit constraints are tight today but expected
to be lax tomorrow, then the firm has a strong
incentive to defer investment. To be sure, in
constructing this example, I have abstracted
from factors such as physical adjustment costs
that will dampen the impact. Nonetheless, it
is interesting that movements in the spread
can have such a large impact in a fairly
conventional framework.
Finally, let me fill in the details of how a
credit channel may produce a shift in the
spread between external and internal funds
that complements the movement in interest
rates associated with monetary policy.4
There are two distinct but complementary
ways in which the credit channel may work.
The first is via the impact on borrower balance
sheets. A rise in short-term interest rates
induced by monetary tightening may weaken
borrower balance sheets in several different
ways. The rise in interest expenses on
short-term debt directly reduces the supply
of internal funds by reducing net cash flows
(after interest payments). In addition, the
where by definition
K , + 1 = I , + ( l - S ) K t.
At the optimum, the firm chooses invest
ment to equate the marginal cost of a unit of
capital, normalized at unity, with the dis
counted expected marginal gain:
1 = [l /Qt( l + p t)] ■E {aG K “- 1 + Qt+1( l - 5 )}.
3 In porticulor, I take a = 0 .33, 8
= 0.1, p = 0.04 and the steadystate growth rate of the economy
equal to 0.02.
* See the discussion in Gertier and
Gilchrist (1 9 9 3 ), and tfie Cecchetti
and Hubbard papers in this issue of
the Review (May/June 1995).
1995
The marginal benefit of a unit of capital in
the next period is the sum of the marginal
product of the next period and the value of
the undepreciated capital. The latter is valued
at the shadow price of internal funds of the
next period, Qt+1. This reflects the fact that
credit market imperfections make a unit of
capital worth more if it is already inside the
firm than if the firm had to borrow to buy it.
In the conventional description of the
transmission mechanism, both Qt and Q(+1
99
REVIEW
MAT/JUNE
Gertler, Mark, and Simon Gilchrist, "The Role of Credit Market
Imperfections in the Monetary Transmission Mechanism: Arguments
and Evidence," Scandonovian Journal of Economics (No. 1, 1993),
pp. 43-64.
associated decline in asset prices may reduce
the value of the borrower’s collateral. Either
of these effects works to raise the shadow
price of internal funds and, in this way, mag
nifies the impact of monetary policy on the
borrower’s effective discount rate. The mag
nified impact on the discount rate translates
into a magnified effect on spending.
The second and perhaps more controver
sial way in which the credit channel may
work is via the impact of a shift in bank
reserves on the loan supply schedule that
commercial banks may offer. Here, tightening
of monetary policy forces banks to contract
reservable deposits, reducing banks’ source
of funds available to service loan demand.
This forces the bank loan rate up relative to
the open market rate, raising the cost of capital
for bank-dependent borrowers. In terms of
the language here, the rise in the bank loan
rate due to the contraction in deposits raises
the shadow price of internal funds for
bank-dependent borrowers. As pointed out
by Romer and Romer (1 9 9 3 ), Gertler and
Gilchrist (1993) and Thornton (1 9 9 4 ), how
ever, a critical premise is that banks cannot
perfectly decouple deposits from loans by
elastically issuing managed liabilities at the
margin. Although this premise seems to be a
reasonable description of financial markets
prior to the financial deregulation that began
in 1980, it is less clear that it is applicable in
the contemporary financial climate. The key
issue concerns the liquidity of the market for
large CDs. None of the existing studies have
really addressed this difficult question.
Hubbard, R. Glenn, "Is There A "Credit Channel" for Monetary Policy,"
this Review (May/June 1995), pp. 63-82.
Kashyap, Anil, and Jeremy Stein. "Monetary Policy and Bonk Lending,"
in N. Gregory Mankiw, ed., Monetary Policy. University of Chicago
Press for the National Bureau of Economic Research, 1994,
pp. 221-62.
Romer, Christina, and David Romer. "Credit Channels or Credit Actions?:
An Interpretation of the Postwar Transmission Mechanism," in
Changing Capitol Markets.: Implications for Monetary Policy. Federal
Reserve Bank of Kansas City, 1993.
Thornton, Daniel L. "Financial Innovation, Deregulation and the "Credit
View" of Monetary Policy," this Review (January/February 1994),
pp. 31-49.
REFERENCES
Bernanke, Ben S. "Nonmonetary Effects of the Financial Crisis in the
Propagation of the Great Depression," The American Economic Review
(June 1983), pp. 257-76.
_ _ _ _ _ _ and Mark Gertler. "Agency Costs, Net Worth and Business
Fluctuations," The American Economic Review (March 1989), pp. 14-31.
_ _ _ _ _ _ , _ _ _ _ _ _ and Simon Gilchrist. "The Financial
Accelerator and the Flight to Quality," National Bureau of Economic
Research Working Paper No. 4789 (July 1994).
1995
100
Rmfw
MAY/JUNE
1995
Allan H. Meltzer is a professor of political economy and public policy at Carnegie Mellon University. An earlier version of this article was pre
sented at the 1994 Konstanz Seminar in Insel Reichenau, Germany, and some material is included from earlier work with Karl Brunner and
Alex Cukierman. The author thanks Dale Henderson and Bennett McCallum for comments on earlier drafts.
Information,
Sticky Prices and
Macroeconomic
Foundations
believe its success will be limited. Success
is relative, of course. Real business cycle
theory has developed an explicit analysis of
the transmission of productivity and terms of
trade shocks. These shocks, though widely
recognized earlier, had not been made the
subject of an explicit model. Neo-Keynesians
and others have produced some suggestions
about pricing. Overlapping-generations
models of money, intertemporal substitution
theories of unemployment, and productivity
shock theories of the business cycle have not
proved fruitful. These models have not pro
duced either an accepted foundation for
macro theory or a verified theory of aggregate
output, prices and interest rates. Perhaps
they will in the future, but the results to date
are not promising.
There are two main ways to tie micro
and macro theories. The first is aggregation.
Aggregation is an important and much
neglected issue, but it is not my main theme.
Most current or recent research dispenses
with the aggregation problem by assuming a
representative individual. Here the old and
new macroeconomics are equally deficient
and open to charges of “arm waving” and “ad
hocery.” The representative individual is a
useful working assumption for some purpos
es, but the representative individual discards
an important difference between markets
and individuals.
The second problem is the specific
micro-foundation used for macro theory.
Most of my discussion is directed there. I do
not attempt to survey a large literature on
micro theory, uncertainty and industrial orga
nization. The implications of some of this
literature for price stickiness is ably summa
rized in Gordon (1 9 9 0 ). This paper is a
personal statement, reflecting jo in t work
as noted earlier.
The problem with existing m icro-foun
dations that I emphasize is the neglect of
costs of information. In Walrasian micro
foundations, all trades take place at market
clearing prices, and there is no cost of
acquiring information. An auctioneer calls
Allan H. M eltzer
or decades, macroeconomists have lis
tened to criticism from their professional
colleagues about the absence o f micro
foundations from most of what they say and
do. A typical comment is: “I don’t understand
much about macroeconomics; how is it relat
ed to econom ics?” A possibly less-frivolous
comment is that nothing can be said about
macroeconomic policy until economists
develop a macroeconomic theory from a
microeconomic foundation. Anything less is
branded “ad hoc” and dismissed.
Much of the work in macroeconomics of
the past two decades has responded to these
criticisms by attempting to build macroeco
nomics on an explicit micro-foundation.
That is a worthwhile goal but it opens the
question: W hich foundation should that be?
The current generation o f academic
researchers are as divided about the appro
priate analytic model as their predecessors.
Analytic paradigms now include multiple
equilibrium, real business cycle, neoKeynesian, monetary-rational expectations,
and eclectic models. Most have a micro
foundation, but it is not the same foundation.
1 do not question the presumption that a
micro-foundation is useful. At issue is what
the foundation should be. I question the
relevance for monetary and macroeconomics
of micro-foundations which feature a repre
sentative agent who trades on a complete set
of Arrow-Debreu markets. Despite some
limited successes, it is time to question
whether this now widely accepted approach
is likely to be fruitful and to suggest why I
F
101
adjust quickly. Large industrial firms—
General Motors, Mitsubishi— know that they
are a relatively large part o f the economy, so
they should adjust to aggregate demand
more promptly than commodity producers.
Typically, they do not.
Missing from Gordon’s analysis is the
difference in information and in costs of
acquiring information. Commodities are
traded on open markets, so prices promptly
reflect changes in the factors affecting
demand and supply. Prices of autos, steel,
heavy industrial products and consumer
durables are not set in organized markets.
Information on which to base prices is
more uncertain.
As Keynes recognized, a principal miss
ing element is uncertainty about the future.
Uncertainty and its twin, costs of informa
tion, make it rational for some firms to
adjust prices slowly Prices may be “sticky,”
as econom ists have observed for about as
long as there has been a discipline.
Some economists will scoff that sticky
prices are irrational or equivalent to leaving
five, 50 or 500 dollar bills in the street.
This is an error arising from dependence on
Walrasian foundations and neglect of infor
mation costs. Yet, there is nothing novel
about invoking costs of acquiring information
to explain sluggish price adjustment or
sticky prices. The positive slope of Lucas’
(1 9 7 2 ) aggregate supply curve arises from
confusion between relative and absolute
price changes; prices in that model do not
immediately adjust to new information. The
cost of learning whether a change in demand
is an aggregate or relative change is infinite
for one period and zero thereafter. In the
neo-Keynesian models discussed in Ball and
Mankiw (1 9 9 4 ), costs of price adjustment—
so-called menu costs— and “real rigidities”
are assumed to be present. The authors
accept that one of the principal costs of price
adjustment is the cost of acquiring informa
tion relevant for a decision about how much
to change price.
Given this widespread acceptance of
information or transaction costs, what
remains to be done? Lucas’ (1 9 7 2 ) model
implies more rapid adjustment of prices and
output to new information than we observe
out the prices at zero cost and does not close
transactions until all transactors are at an
equilibrium. There is a numeraire, but there
is no way for money to disturb the real value
of production or purchases. Non-neutrality
cannot arise. Attempts to graft a monetary
disturbance onto these micro-foundations
seem misdirected.
The hypothesis that all observed prices
are market clearing prices does not imply
that all individuals, behaving rationally,
know those prices and fully adjust to them.
Information is costly to acquire; time and
resources must be used to collect, process
and interpret— the latter especially— new
data. Even in markets dominated by price
takers, there are differences between the
information processed and known to arbi
tragers and specialists and what is known to
non-specialists. The representative individ
ual paradigm ignores this distinction.
One way to proceed is by aggregating
heterogenous individuals who face different
costs of acquiring information, and much new
work has taken this path. As Gordon (1990)
emphasizes, at any time there are many differ
ent layers of pricing and output included
in aggregate prices and output— suppliers,
suppliers of suppliers, foreign producers, and
so on. Instead of trying to aggregate over
these many, diverse and changing levels of
decisions, it may prove more useful to treat
them as part of a stochastic process.
One of the problems with the hypothesis
that “sticky” prices reflect decisions at differ
ent levels of aggregation comes out clearly
in Gordon’s paper. He argues that sluggish
price adjustment reflects marginal cost pric
ing. “ [Agents] care about the relation of
their own price to their own costs, not to
aggregate nominal demand. Unless a single
agent believes that the actions of all other
agents will make its marginal costs mimic
the behavior of nominal demand with m ini
mal lags, the aggregate price level cannot
mimic nominal demand, and Keynesian
output fluctuations result.”
This argument has a perverse implication.
Farmers or farms that operate in commodity
markets with many competitors should be
slowest to adjust to aggregate demand
shocks. Yet commodity prices typically
F E D E R A L RESERVE B A N K OF ST . L OU I S
102
REVIEW
MAT/JUNE
change. Delays in recognizing permanent
changes in money or its growth rate would
affect only real balances. At best, we would
be forced to fall back on the real balance
effect in consumption to explain short-term
real effects of changes in money. A rational
reason for setting some, but not all, nominal
prices permits a more direct effect through
inventory adjustment. Firms that hold
inventories can both anticipate future price
changes and buffer current transitory price
changes by varying inventories. Because
some prices are determined in auction mar
kets, price setting introduces different speeds
of adjustment and relative price changes.
My extensive work with Brunner emphasized
the relative prices of assets and output.
This introduces the difference between the
replacement cost, or the cost of current
production, and the market price of existing
assets. Asset prices adjust more rapidly than
output prices particularly for assets that
trade in organized markets. Hence, informa
tion costs (and transaction costs) are implicit
in that framework as an explanation of
changes in relative prices. My emphasis
here is on information as in Brunner,
Cukierman and Meltzer (1 9 8 3 ). Relative
price changes in response to nominal shocks
are part of that story, but they remain in
the background.
in practice. Could it be rational to adjust
more slowly than in one week or month?
Is the compelling fact that data on the price
level are released monthly? Or is this infor
mation subject to error so that it takes longer
to disseminate, interpret and act on these
data? Neo-Keynesian models build on the
model of an imperfectly competitive firm,
with strong implications for profits and
excess capacity that do not find empirical
support at the micro level. If monopolistic
elements and menu costs are the source of
sticky prices, there are testable implications
for the profits of different types of firms; there
is a significant problem of reconciling contin
uous excess capacity with rational behavior.
Further, as Gordon (1990) has noted, costs
of adjusting output are neglected in these
models. Yet these costs may be larger for
many firms than costs of price adjustment.
My principal criticism of both neo
classical and neo-Keynesian approaches is
that they seek to explain why firms delay
using available information. By putting the
issue in that framework, they neglect the
uncertainty that surrounds much of the
aggregate and disaggregated data. This paper
uses several strands of earlier work to explain
rational price setting and gradual adjustment
as a response to uncertainty about what cur
rent information implies. Information is
costly to acquire and to interpret as in
Brunner and Meltzer (1971) or Alchian
(1 9 7 7 ). As in Bomhoff (1 9 8 3 ), a principal
difficulty in interpreting information is
uncertainty about how long changes will
persist. This is the central idea developed
in Brunner, Cukierman and Meltzer (1983),
but we took the idea from Muth’s (1961)
seminal paper on rational expectations. In
Meltzer (1 9 8 2 ), I used these ideas to discuss
price setting.
There are three separable aspects of pric
ing to consider. First, (some) prices are set
by firms. Second, firms choose nominal
values; they do not index or set a relative
price. Third, prices that are set change less
frequently than prices in auction markets.
Each aspect is important for macroeco
nomics. If some prices were not fixed in
nominal value for at least one period, relative
prices would be invariant to a monetary
1995
H O W STICKY ARE PRICES?
Postwar recessions typically last about
nine months on average. Costs of informa
tion or other explanations of sluggish adjust
ment must be able to explain this timing,
and estimates of price stickiness should be
consistent with data on cyclical fluctuations.
I digress therefore to consider a recent
attempt to measure stickiness. Blinder
(1 9 9 1 ) reports preliminary findings from a
survey of pricing decisions by managers of a
random sample of corporations with annual
sales of $10 million located in the northeast
ern United States. Blinder’s survey produced
two very different sets of estimates.
First, Blinder reports the answer to a
question about how frequently a firm
changes its price. He finds that the median
firm changes price about once a year. Nearly
103
RfVKW
MAY/JUNK
1 Beaulieu and Mattey (1 9 94 )
estimate the effects on price
dispersion of factors such as
advertising, transport costs, and
concentration using cross-section
data for specific commodities.
They find effects of some of these
elements on price dispersion.
early results confirm that delivery lags
and service are moderately important for
65 percent of the firms in his sample; he
reports delivery lags and service are the most
commonly cited reason for price stickiness.1
Delivery lags and service are one way of
adjusting the theoretical term “price” while
leaving the quoted price unchanged.
Assume that firms initially respond to
changes in cost and demand by adjusting
deliveries, advertising allowances, discounts,
and so on while leaving quoted prices
unchanged for several months or longer. If
managers are uncertain about the duration of
changes in demand or cost, they can change
other components of the price vector to test
the market’s response. By changing delivery
terms, or offering or removing discounts,
firms can change their revenues or the
buyers cost without changing the quoted
price. This pricing model can be used to
rationalize the familiar Keynesian supply
curve— a reverse L— when quoted prices are
distinguished from other terms in the price
vector. Equally, the model can explain the
difference in response to the questions in
Blinder’s survey.
Figure 1 shows the initial response to an
increase in demand. On the left, the quoted
price (p) does not respond to a perceived
change in demand when output is below
capacity; the firm (or industry) increases
output (q) with little or no change in quoted
price. On the right, the price vector (p),
includes other dimensions of the firm’s price;
supply is drawn as a positively sloped linear
function. An increase in perceived demand
from dg to d t induces the firm to reduce
advertising allowances, remove discounts,
or change some other component of the
price vector, so (p) increases.
If the firm’s initial response reflects
uncertainty about the persistence of
increased demand, the response changes as
information accrues. An increase in the
quoted price may substitute for or supple
ment other components of the price vector.
As perception of the magnitude of the per
manent increase becomes clearer, the firm
may recognize that the new demand curve is
at, or to the right of, d r Or the firm may
raise p, with p unchanged.
40 percent of the firms were at the median.
More than one-sixth of the firms changed
prices less frequently than once a year, so 55
percent of the prices change no more than
once a year.
Second, Blinder also reports the respon
dents’ estimate of the mean lag of actual
prices behind increases and decreases in
demand and cost. There is considerable
uniformity in these responses. The mean lag
for increases or decreases in demand and
cost is three to four months.
The second set of data suggest consider
able uniformity of response to the four shocks.
There is little evidence of the asymmetric
adjustment associated with Keynesian down
ward price inflexibility. The data also sug
gest an inconsistency that Blinder does not
mention. How do we reconcile a mean delay
of three to four months in response to changes
in demand and cost with the report that 76
percent of the sample changes price no more
than twice a year and, as noted, 55 percent
changes price no more than once a year?
Are demand and cost changes infrequent?
It is difficult to reconcile the assumption that
costs change infrequently with evidence
showing that commodity prices and other
open market prices change daily and typically
fall in recessions and rise in expansions.
Blinder did not ask whether firms adjust
their prices fully in response to changes in
demand and cost. They may adjust partially,
as implied by rational behavior under uncer
tainty about the persistence or permanence
of announced changes, or they may antici
pate future changes and adjust prices more
than current changes in cost or demand.
One way to reconcile the different responses
is to assume that respondents answer the
two questions in different ways. Suppose
they treat price as a scalar when asked about
the frequency of price changes but include in
price adjustment more than ju st price setting.
On this interpretation, quoted prices are one
component of a vector of terms and conditions
relevant to sellers and buyers. The theoretical
term price used in economics typically sub
sumes delivery time, discounts, advertising
allowances, volume rebates, payment terms
and other conditions used to adjust the buyer’s
cost and the seller’s net receipts. Blinder’s
1995
104
MAT/JUNE
We need not explore the many possibili
ties. The main point is that uncertainty about
the degree of persistence and the use of a
price vector permit this hypothesis to account
for very different responses within a rational
expectations framework. In particular, the
firm’s revenues may respond instantly to
increases in demand, but quoted prices may
adjust with a long lag, as Blinder found.
The reasoning is symmetric. Perceived reduc
tions in demand may induce firms to offer
discounts and allowances with p unchanged.
As information accrues and perceptions
change, the actual price, p, is reduced.
The Keynesian supply function does not
work for changes in cost if the demand curve
is not kinked. Changes in cost shift the
supply curve, so prices change instantly,
whether the supply has a reverse L-shape or
is monotonically increasing. Since Blinder’s
survey finds that price responds about as
promptly to changes in cost as to changes in
demand, this evidence rejects the Keynesian
supply curve .2
The distinction between price as a vector
and a scalar does not reconcile the differences
in timing reported in the survey. Although I
believe that prices, terms and conditions
change at different rates when there is uncer
tainty about the persistence of the market
conditions inducing firms to change prices,
those differences are neglected hereafter.
Price stickiness will mean that it takes about
three or four months on average for prices to
respond to demand and cost changes. The
three-to-four month delay that Blinder reports
includes at least one quarterly reporting period
at which managements announce earnings
and sales and, most importantly, observe
reported earnings, inventories and sales of
competing firms .3 Knowledge of competitors’
results helps the firm to supplement trade
gossip and other informal data sources to
decide whether a persistent change in market
demand (or industry costs) has occurred.
F ig u r e 1
section discusses one reason that is rather
general and can be incorporated readily
into macroeconomics. Price setting is
considered as a response to uncertainty or
costs of acquiring information about the
market clearing price by at least one (large)
group of market participants .4 At times,
prices convey information known to the
seller but not to the buyer. At times,
neither the buyer nor the seller is certain
about the market clearing price. Among
the possible reasons on the cost and demand
sides, this section emphasizes rational
reasons why buyers and sellers do not
instantly know the permanence of changes
in demand. They develop contracts and
market arrangements to deal with
this uncertainty.
This approach differs from the literature
on so-called menu costs. The menu cost lit
erature emphasizes costs of changing prices .5
These costs are recognized as relatively small
(Ball and Mankiw, 1994). Moreover, the menu
cost model does not explain why sellers face
different costs and adjust at different speeds.
I do not challenge the existence of menu costs,
but the emphasis here is on uncertainty and
information costs— the cost of learning about
current and prospective market conditions.
Information costs differ by firm and industry.
They depend on the way in which markets
are organized and the types of contracts
that emerge to reduce costs of uncertainty,
information and moral hazard.
INFORM ATION AND PRICE
SETTING
Anyone familiar with literature on the
behavior of firms knows that there are many
rational reasons for firms to set prices. This
1995
105
2 Ball and Mankiw (1 9 94 ) use
the ability to generate a reverse
L-shape (Keynesian) supply curve
as evidence for a model of menu
costs and imperfect competition.
3 This should not suggest that the
timing is constant. As in Muth
(1 9 6 1 ), the timing of the response
depends on the relative variances
of permanent and transitory compo
nents. See below.
4 Much of this section is taken from
Meltzer (1 9 8 2 ), ond especially
Brunner and Meltzer (1 9 9 3 ).
Additional examples of rational
price setting when information is
costly to acquire con be found in
these references. A recent paper
by Heaton and Lucas (forthcoming)
analyzes the effects of transaction
costs, such os bid-osk spreads,
investor heterogeneity and persis
tence of shocks. They find that
many puzzles in asset pricing can
be reduced or removed by using
models with transaction costs and
persistence. Balvers and Cosimano
(1 9 9 0 ) model price adjustment
when firms must learn about the
persistence of changes in demand.
s See McCollum (1 9 89 ) for a critical
discussion of menu costs as a basis
for rational price-level slickiness.
RfVKW
MAY/JUNE
Market Organization
large number of market participants also
imposes severe constraints. For many reasons,
including the law of large numbers, people
choose different times for their market
activities. The commitment to assemble at
pre-specified times for all marketing purposes
is costly, if not impossible. The use of an
agent introduces costs of monitoring
and supervision.
In practice, markets operate in many
different ways. One alternative commonly
found in oriental countries is called a bazaar.
Prices are not posted, and there is no auctioneer
mechanism. Buyers and sellers negotiate
(haggle) until a transaction is completed or the
negotiation terminates. Considering a bazaar
brings out the importance of information.
The bazaar requires an investment of
time by transactors. W here the auction
market requires simultaneous arrival of all
participants or their agents, the bazaar
depends on a trickle of arrivals. It cannot
cope with large groups arriving simultane
ously because each negotiation is separate.
It is not surprising that the bazaar is found
in comparatively simple, low-income
economies. In these circumstances, the
allocation of time to the bargaining process
may be partly a consumption good. W ith
rising opportunity costs of time, the disad
vantages of the bargaining process exceed
the benefits.
The working of a bazaar restricts its
application. There are severe limits on the
number of transactions per period. Sellers
cannot serve several customers simultane
ously. Delegating bargaining to employees
poses both an incentive and a moral hazard
problem. Buyers incur costs to learn about
reservation prices, since information is
revealed only by commitments to transact
firm offers to buy or sell. An accepted offer
to purchase may be above the seller’s reserva
tion price; a refusal to sell may be based on
an incorrect inference that the buyer is willing
to pay more. Learning the reservation price
and negotiating the market price often requires
a series of offers. There is no certainty that
the reservation price is revealed. Subsequent
transactors on one side of the market do not
know the history of past transactions. They
must invest their own time.
Frank Knight was an early expositor of
the economics of information. Knight (1933),
argued that decisions about how much
and when to produce require a pooling of
information about individual decisions to
purchase. W hen aggregated, the individual
decisions constitute a demand curve.
Firms reduce uncertainty for consumers
by pooling information. In principle, indi
viduals can contract in advance for the goods
and services that they want. Organizing
retail firms that pool information is an efficient
alternative to futures contracts if individuals
are less certain about the magnitude and tim
ing of their purchases than firms are about
market demand. Knight appeals to the law
of large numbers to explain firms’ advantage.
He compares pooling by firms to insurance
and concludes that the two differ in an
important way; a firm’s pricing and output
decisions are non-insurable because they
require more subjective judgment, and errors
are less likely to cancel across firms.
In Knight’s view, firms produce for inven
tory using pooled information about expected
demand. This arrangement shifts uncertainty
from individuals to firms and, by pooling,
reduces the social cost of bearing uncertainty.
Knight’s argument is one of several that
links information costs and uncertainty to
price setting. A non-uniform distribution of
information is critical for these arguments.
In an auction market, all market participants
must have information about the qualities
of the goods traded and their prices. In the
standard Walrasian model, this is accom
plished by: ( 1 ) assuming the presence of
an auctioneer who calls out the prices;
(2) allowing recontracting; and (3) letting all
trades be made simultaneously. These assump
tions are necessary for equilibrium. They leave
no role for monetary disturbances.
The necessary conditions are frequently
violated in practice. Some people have a
comparative advantage in acquiring informa
tion. Some receive information about market
conditions as a by-product o f other activity.
For example, in securities markets, there are
brokers, dealers and market makers who
acquire specialized information in the course
of trading. Assembling all or a sufficiently
199
N K OF ST . L O U I S
106
REVIEW
MAY/JUNE
A seller who posts prices reduces the
buyers’ costs o f acquiring information about
prices. Buyers’ costs of comparing prices by
different sellers are reduced. The social
advantage of price setting is greatest where
one party to the transaction has more infor
mation about market conditions. Some
examples illustrate this argument.
Information about prices conveys more
than ju st purchase cost. Posted prices can
also reduce costs of acquiring information
about quality. For example, a restaurant
owner must decide on the market he wants
to serve. This decision influences the kind
and quality of food served, the services offered,
and the prices charged. By posting prices, the
owner informs the buyer about his choices.
Although the buyer must sample to judge
quality, the correlation between quality and
price helps the buyer to decide whether to
sample. A policy of frequently changing
prices reduces information and places the
restaurant at a competitive disadvantage.
The organization of the diamond market
provides additional evidence on the role of
information in market organization. The
wholesale diamond market is an auction
market dominated by buyers and sellers who
are specialists. Traders rely on their own
skill in judging quality, knowledge of prices
and other attributes. The retail market is
very different. The sellers are mainly special
ists; the buyers typically have much less
information than the sellers. By posting
prices, sellers exploit the correlation between
quality and price to inform buyers. Buyers
find it less costly to invest in information
about the seller than to invest in information
about the quality of diamonds, so sellers use
resources to build reputation. If costs of
acquiring information about the quality of
diamonds were to fall to a minimal value,
these arrangements would change. Diamonds
might be sold in supermarkets or in retail
auction markets.
Price setting is valued by transactors even
in some auction markets. In well-organized
auction markets, we find people willing to
pay for the right to purchase or sell at fixed
nominal prices. The contracts expressing
these rights, known as “put” and “call”
options, give the owner the right to buy or
199
sell at a fixed price within a fixed time period.
The prices of the puts and calls are determined
in auction markets. The price of these options
is the cost that people pay for the right to trade
in the future at prices fixed today Similarly,
in commodity markets, hedgers pay to
change uncertain future prices into known
values. They pay a fee for the right to buy or
sell at fixed nominal prices. In such markets,
information about current and currently
anticipated future prices is available. The fee
permits transactors to avoid uncertain future
price changes.
Costs of information and transactions
are not uniform across goods, so no single
form of market organization dominates all
others. The specific reasons transactors are
willing to pay for puts and calls differ from
the reason for price setting in the diamond
market, ju st as the diamond market differs
from the restaurant. Each is related, however,
to costs of information. The organization
of the diamond market reduces the costs of
bearing uncertainty about quality. The market
for puts and calls permits asset owners or
speculators to limit risk of wealth changes.
And there are other market arrangements
where price setting is useful. Catalogues
must post prices. Contracts fix prices for a
year or more on housing rentals, magazine
subscriptions, and automobile leases.
These many different examples suggest
that there are advantages in different types
of contracts.
The examples suggest a way to model
price setting formally. The seller has infor
mation that is costly for the buyer to acquire.
The seller internalizes the cost of acquiring
the information; it is part of his specialized
knowledge and he revises his information in
the process o f buying inputs. By posting a
price, he exploits the correlation between price
and quality. In goods markets, the seller may
offer a particular type of put— an option for
the buyer to return the merchandise if the
quality is not as represented or, perhaps, if the
buyer finds the same merchandise at a lower
price. The buyer pays for the good and for
the put, but the purchase cost is lower than
under alternative forms of organization. In
service markets, the buyer may purchase
increased certainty that he will not have to
107
REVIEW
MAT/JUNE
relocate or search for a particular input at an
inconvenient time.
and by workers if negative. That we do not
observe contracts of this kind suggests that
assessments differ even after the event and
that reaching a common assessment o f the
past is costly. O f course, setting nominal
wages is not costless either. Bargaining or
negotiating to correct for unforeseen events
can be costly, privately and socially, if there
are strikes or layoffs .6
Differences in information can explain
price setting, but they do not fully explain
why firms and employees often set nominal
wages or prices. There must be some addi
tional cost of setting relative prices or benefit
from setting nominal wages (or prices). One
explanation is that the parties do not agree
on the interpretation of real wages and, par
ticularly at low rates of inflation, have diffi
culty agreeing on an appropriate index.
There are two different meanings of real
wages. One meaning expresses real wages in
terms of the product of the firm at which the
worker is employed. The second refers to the
basket of goods and services that the worker
can purchase. The problems of setting con
tract terms differ in the two cases.
Contracts that set wages in relation to
productivity require a satisfactory solution
to the measurement of productivity. Where
precise measurement is difficult, as in service
industries or managerial tasks, real wage
contracts are difficult to write. Even
when productivity is measured reliably as
in piece-work systems, the measure does
not translate directly into a real wage
rate. Valuation is required; often some
(more or less) arbitrary system must be used
to impute the price of the final product to
the various inputs. Imputation and valuation
bring two additional problems. One is moral
hazard; the employer has some incentive to
adopt a cost accounting system that benefits
him. The other is the difference between
relative and absolute prices. The employer is
mainly concerned with the relation of wages
to product prices. The employee is concerned
also, and perhaps most, about the relation of
wages to the price of consumables— the sec
ond meaning of real wages. A frequent
compromise in periods of inflation is partial
indexation to the price of consumables.
Real wage setting with full indexation
Labor Markets
6 Large Japanese firms that offer
lifetime employment partially index
by tying bonuses to some measure
of the firm's profitability. With
lifetime employment, it may be
rational to assume that errors will
(approximately) cancel over time.
Lifetime employment contracts
impose costs on firms (owners) if
output falls below some expected
value and impose costs on workers
if output is above the expected
value. The latter can he compen
sated more readily by adjusting
bonuses, but negative bonuses are
not observed. An unexpectedly
long period of slow growth elimi
nated some of these contracts.
The conditions leading to price setting
also apply to the labor market. The terms
negotiated and the time horizon built into an
agreement depend on the assessments of the
parties. There are no organized futures mar
kets for labor. Both parties use all available
information to form their uncertain assess
ments. An assessment of the market is a
(subjective) probability distribution. The
more diffuse the distribution, the shorter the
time covered by the arrangement.
Realizations often deviate from the
expectations implied by the subjective prob
ability distributions. Both parties have to
infer from realizations whether the unexpected
changes are transitory or permanent.
A transitory change does not change
the expected value, so the unanticipated
gains and losses do not change the informa
tion on which the agreement was based.
Either party may believe that a costless
revision of the bargain to adjust to a transitory
change would be beneficial, but attempts at
revision for each such change raise the cost
of transacting and eliminate the benefits of a
longer-term agreement.
A more permanent change in conditions
poses a different problem. The initial assess
ment of at least one party must be revised in
the light of the new information. If the stakes
are sufficiently high, the permanent change
may justify the cost of renegotiation.
Negotiations proceed more smoothly
when both parties share the reassessment of
market conditions. Differences in assessment
provide evidence on the extent of uncertainty.
Strikes and lockouts increase with differences
in assessment, for example, at the start of a
period of inflation or disinflation. If both
parties could agree on the actual shocks that
occurred in the past and on how long they
will persist, they could contract in advance
to compensate for unanticipated changes
after they occur. Permanent changes in
nominal values would not be allowed to affect
real wages. Permanent changes in produc
tivity would be paid to workers if positive
1 »«S
108
REVIEW
MAY/JUNK
is rare. This suggests that at low levels of
inflation, buyers and sellers prefer ex post
adjustment through negotiation to reliance
on an imprecise or imperfect index. As infla
tion increases, costs of non-indexation rise
relative to costs of indexation. More parties
choose a mixed strategy of partial indexation.
Experience in Israel and Brazil suggests that
at relatively high rates of inflation, indexation
is nearly, but never fully complete. There is
always some lag in adjustment. Data for
countries with high inflation also suggest
that workers willingly forego indexation if
offered relatively stable prices.
The choice of an index is a problem for
both parties. Some of the problem would
be removed if shocks could be identified
unambiguously, if one-time price changes
(transitory changes in the rate of inflation)
could be separated from permanent changes
in inflation, or if all shocks were of one
kind— for example, permanent, nominal
or real aggregative, or real allocative. The
absence of reliable information prevents settle
ment on an optimal indexation formula.
Nominal wage contracting is also not
ideal. Different types of contracts are used to
adjust nominal wages for inflation. In periods
of low or moderate inflation, we observe
contracts that differ in duration, in the extent
of formal indexation, and in the use of clauses
permitting reopening of the wage agreement
during the life of the contract. We observe
also that the types of contracts change with
the rate of inflation and that employers can
be induced to compensate for (som e) past
price changes when (non-indexed) contracts
are renewed. At high rates of inflation, firms
and other market participants monitor the
rate of inflation. Costs that were previously
marginal costs of information became fixed
or quasi-fixed costs of information. Nominal
prices adjust more frequently.
Retail store leases differ from wage
contracts. Leases are often indexed to the
volume of sales. Sales are more easily
monitored and therefore less subject to moral
hazard. Valuation is based on receipts, so
measurement is not as much of a problem as
for profits or productivity. Both parties have
an interest in maintaining the property.
Bond contracts provide another example
1995
of the problem of choosing an index. Private
parties do not issue price-level linked bonds.
Under the gold standard, however, firms
offered to pay in gold. Buyers and sellers
could agree on this index of long-term value.
Once this common measure, related to the
value of money, became less relevant, indexed
bonds were rare. Inability to agree on an
index left no agreement. In Britain, Israel,
Brazil and a few other countries, the govern
ment resolved this problem by issuing an
indexed bond.
Comparison of the choices made in
markets for labor, rental property and bonds
suggests that agreement on an index is most
difficult when prices can change because of
real and nominal shocks, and changes can be
permanent or transitory. This should not
suggest that non-indexation is optimal.
Contracting parties find many different solu
tions but, as experience in labor markets
shows, full indexation is rare.
A Stylized Model
In several papers, I have used a model of
permanent and transitory changes based on
Bomhoff (1 983) to study the frequency of
shocks and their interaction. A main conclu
sion of this work is that there is no reason to
expect constancy or even repetition in the
frequency distribution of shocks. The public
cannot use data from the past to anticipate
the future relative frequency of permanent
and transitory shocks or real and nominal
shocks. These frequencies change with public
policy decisions, policy rules at home and
abroad, weather, inventions or innovations,
changes in market structure, and other factors
affecting an economy’s structural relations.
The model has three types of shocks:
transitory shocks to the level; permanent
shocks to the level (which are transitory
shocks to the growth rate); and permanent
shocks to the growth rate. Let x t be the current
value of real output and pt the current value
of the price level. Both prices and output can
be affected by each of the three shocks so that
x t~ x t + e,
x ,=
+ xt + y t
*. = *,-i + P.
109
REVIEW
MAT/JUNE
the price level reflects current information
about money Money is neutral. Incomplete
information of some sort is a necessary
condition for significant monetary effects,
but it is not sufficient.
Lucas’ (1 9 7 2 ) hypothesis about incom
plete information restricts uncertainty to
misperception of current shocks; people do
not know whether a shock to demand is spe
cific to their product or is a general increase
in demand. It is now generally recognized
that Lucas’ hypothesis produces a response
of real output to a nominal shock but does
not generate as m uch persistence as is found
in cyclical fluctuations of output and prices.
Persistence must be introduced. The problem
is to introduce persistence without introduc
ing an implausible reason for neglecting
information in current output, prices, interest
rates and other variables.
In Brunner, Cukierman and Meltzer
(1 9 8 3 ), a representative producer sets price
and output at the start of each period using
all available information at the time. This
includes, in particular, the variables in equa
tion 1 — inventory levels and changes, interest
rates and current policy. W hen making price
and output decisions, producers are uncertain
about the permanence o f observed shocks.
As in the earlier discussion, they do not
respond to changes that are perceived to
be transitory.
Let x * and m* denote the perceived per
manent components of x and m. Knowing
these values, producers set output and the
price level; y * and p * denote the producers’
decisions. They are determined from a subset
of the system of simultaneous equations:
and
P t = p t + Ut
p,= P '.l + Pt + Vt
Pt = Pt-, + •
There are stochastic elements in the growth
paths of output and inflation in addition to
transitory and permanent changes in the levels
of output and prices. Much confusion in the
discussion of inflation has been caused by
the use of “inflation” to refer both to level
changes such as oil shocks (distributed over
time) and persistent changes in the maintained
rate of change.
Suppose we now introduce a common
type of Phillips relation between p and x in
the neighborhood of p t = 1 .0 .
xf * . = a (P, - Pm) •
The way in which prices will change over time
depends on the permanence of the shocks.
It takes time for agents to learn whether
the shocks change ut, v( or z, and, from the
simultaneity of x and p, shocks to £(, y t and
pt. The path by which prices adjust— or the
degree to which they are sticky— depends on
the nature of the shocks. It is entirely ratio
nal in this framework for prices to be sticky
and for the speed of adjustment to differ
from one episode to another.
The model in Brunner, Cukierman and
Meltzer (1983) conveys some of the ideas
ju st discussed. The main idea of the model
can be written in a general way. Consider
the following system of simultaneous equa
tions for a macroeconomic system. In the
underlying micro model, firms set price at
the start of the period and hold inventories.
Shocks are revealed after production decisions
are made.
( 1)
(2) f l y * , P *. A h*, h(1, i *, r * , x * , m(*] = 0.
The actual values are x and m, but x - x * and
m - m * are ignored in adjusting prices and
output. Transitory changes are not innocuous.
W ith y * and p * adjusted to x * and m*, nom i
nal and real interest rates and the change in
inventories adjust to the perceived transitory
changes, shown as equation 3:
f [ y t, p t,A h t, h t_l t it, r t, x t, m t] = 0 ,
where y = output, p = price level, h = stock
of inventories, i = nominal interest rate,
r = real interest rate, x = exogenous real
shock, and m = nominal money stock. Under
full information about the structure of the
shocks and in the absence of transaction costs,
1995
(3)
f l y * , p * , Ah,, ht l , i„ rt, xt, m(] = 0.
Output, inventories and other real values
110
Iffl
MAY/JUNE
respond to both nominal and real shocks in
this model by amounts that depend on the
size of the misperception. As new information
arrives, producers revise their beliefs about
the permanence of shocks; p *, y * and all other
variables adjust to the changed perception.
An econom etrician examining the data
generated by this model would at times find
serially correlated changes in output and
serially correlated errors. Serial correlation
arises following a large permanent real or
nominal shock if the shock is believed for a
time to be transitory. As time passes, and
errors are repeated, perceptions adjust. Even
if the unconditional error in the population
is serially uncorrelated, misperception of the
permanence of shocks can lead producers to
make errors that, ex post, are serially correlat
ed. A model of this kind may explain why
some researchers have found ex post real
effects of anticipated changes in money. See,
inter alios, Mishkin (1983).
Although prices do not fully adjust to
shocks when they occur, decisions are entirely
rational. Producers use all available informa
tion, but they misinterpret the nature of the
shock. Once they perceive that the shock is
permanent, prices and output fully reflect
the information.
The length of the recognition lag depends
on the relative variance of permanent and
transitory shocks. The larger the variance of
permanent shocks, relative to transitory
shocks, the shorter the recognition lag. If all
shocks were permanent, prices and output
would adjust to p* andy* values as soon as the
shocks occurred. The lag in our model would
be one period, as in the Lucas model. If all
shocks are transitory, y and p change, but y *
and p * never adjust.
In Keynesian models, inflexible prices
(or wages) and gradual adjustment are taken as
evidence of disequilibrium. The information
structure of the model here implies that this
inference is invalid. Buyers and sellers use
all available information and adjust to a
market equilibrium.
There are three types of equilibrium.
At any moment, there is a permanent stock
equilibrium characterized by the state variables
x * and m*. This equilibrium occurs when
Ah = 0. The values of all variables are adjusted
199
to the perceived permanent shocks and
the condition of unchanging inventories.
Each firm uses resources at the profit m axi
mizing rate. No firm seeks to expand or
contract output or change its price and
inventory position.
A permanent equilibrium is less encom
passing. The state variables in this case are
x *, m* and ht ,. All other variables adjust to
these conditions. If x * and m* remain
unchanged, the permanent equilibrium
converges over time to the permanent
stock equilibrium.
A transitory equilibrium imposes an
adjustment of the system to given values,
m * m * , x ^ x * , y * , p * and ht r Inventory
adjustment and interest rate changes produce
the transitory short-run equilibrium.
In Brunner, Cukierman and Meltzer
(1 9 8 3 ) we show how real variables respond
to monetary shocks in a model of this kind.
All expectations are rational. No information
is wasted once it is correctly perceived.
Misperceptions occur, so the system adjusts
sluggishly to information that, ex post, turns
out to be permanent. The discussion here,
following the original, uses inventories as a
representative real variable but the response
of real output is similar.
Figure 2 shows the adjustment path.
Asterisks denotes permanent values, and a
superscript a denotes actual values.
“Up to period t, the econom y is at an
equilibrium: H( = H'\ During period t,
there is an unanticipated increase
in money growth. Interest rates
fall; with prices fixed for the period,
aggregate demand increases, and
inventories are reduced. Inability to
identify permanent shocks means that
the perceived value of m changes as
information about the rate of monetary
expansion becomes available. Forecast
errors remain on one side of zero for
several periods. Forecast errors rein
force the cyclical deviation of inventories.
A large permanent increase in money
growth, that is not immediately recog
nized as permanent, increases m(,
and adds to the cyclical deviation
of inventories.
i l l
REVIEW
M
component, adjustment to permanent
changes is relatively slow. Inventories
can fall below their expected value for
several periods and, thus, move away
from Hp.
F ig u r e 2
Inventory
“Additional information about the per
manence of the shock that first occurred
in period £ is revealed each period, so the
path of adjustment toward H p is not
smooth. As time passes, however, the
addition to information is small. After t+3
in fig. 2, inventories adjust toward Hp
unless another shock— another unanticipated increase in money growth— lowers
inventories and starts a new process of
learning and adjusting.
An asterisk in the table indicates the expected path of inventories.
The outer envelope shows the actual path.
“The reduction in inventories sets off
a process of adjustment of output and
inventories. The path along which
inventories are expected to adjust at
the onset of period £+1 is shown by the
positively sloped line from H “+1 to
the permanent level of inventories, Hp.
Along this path, a typical firm plans to
produce output in excess of expected
sales and build inventories. The expected
value of inventories by the end of period
£+1 is shown as Ht+r If the monetary
shock is correctly perceived as transitory
and there are no further shocks, firms
adjust along the planned path and
achieve the values of inventories, H!+2,
H1+3, in successive periods until the per
manent value, H p, is restored.
“Actual inventories follow the outer
envelope in fig. 2 ; expected inventories,
H, follow the adjustment paths that start
at the actual values for each period. The
figure shows principal features of our
model of inventory behavior, augmented
by the effect of permanent-transitory
confusion. Deviations from H p are on
one side of Hp for several periods
because of the slow adjustment of inven
tories. This feature occurs even if all
shocks are white noise. In addition,
information about the permanence of
shocks becomes available gradually.
People use all information and their
beliefs about permanent values to
determine the adjustment path, but
they make unavoidable errors because
they learn about the permanence of
shocks gradually.”
“Suppose, however, that the increase
in money growth persists. In period £+2,
interest rates are again pushed below the
value expected for the period, and actual
inventories, Hta+2, are, again, below the
value expected, Ht+2, as shown in fig. 2.
At the beginning of period £+2, firms
and households expect the economy to
adjust along the new path, from H''+2
to H p. The new path reflects all the
available information about the shocks,
including beliefs about the permanence
of the change in money growth and
knowledge of the structural parame
ters...If the variance of the transitory
component of money growth is large
relative to the variance of the permanent
SUM M ARY
My emphasis in this section is on micro
foundations that lead to price setting and
to gradual adjustment. I do not claim to
have uncovered a unique structure that
produces sticky prices under rational behavior.
There are many reasons and many valid
hypotheses that make both price setting and
gradual price adjustment compatible with
rational behavior.
Price setting is sufficient for real effects
of monetary shocks. I have discussed several
112
REVIEW
MAY/JUNE
reasons for price setting— menu costs,
Knight’s uncertainty argument for the
existence of firms, producers’ desire to
signal quality, purchasers’ gains from
lower cost of information and bargaining,
and other differences in costs of acquir
ing information.
In principle, sellers or buyers could
set relative, not absolute, prices. To do
so efficiently, buyers and sellers have to
agree on an index. Price index numbers
are subject to real and nominal shocks that
are sometimes permanent, sometimes
transitory, and sometimes alter the rate of
price change persistently. If these shocks
could be correctly identified ex post, and
if their expected duration were known, the
parties might agree to adjust prices after
shocks occur. Far more often, ex post
adjustment is done by negotiation, or
the parties agree to partial indexation and
negotiate about the remainder. These
arrangements are consistent with the presence
of relatively large costs of acquiring informa
tion and agreeing on what has occurred,
whether the change is permanent or transitory
and how long and how large future price
changes will be. Among these costs are the
costs associated with moral hazard if one
party controls relevant data.
Economic contractions typically last
nine months to one year. Considerable
evidence suggests that price changes may
lag as much as two years behind monetary
shocks. To yield the patterns of price and
output change observed in actual economies,
price setting must be joined to a rational
reason for persistence.
Permanent-transitory confusion— uncer
tainty about the duration or persistence
of shocks— provides one such condition.
Under this hypothesis, lags can be long or
short and ex post errors can be serially
correlated, if a large permanent shock is
perceived as transitory. If the variance of t
he permanent shocks is high relative to the
variance of transitory shocks, the lag is
relatively short, and there is no reason for
significant serial correlation to be observed
ex post. In this case, price setters believe
that most shocks are permanent, so they
adjust promptly.
1995
T a b le 1
Revisions to Reported Inflation
(percentage points)
Horizon
(quarters following)
1
2
3
4
Average
Revision
Range
0.2
0.3
0.3
0.3
-1.0 to 1.5
-1.2 to 1.7
-0.6 to 1.7
-0.6 to 1.7
SOME EVIDENCE
We cannot directly observe how people
decide on the degree of persistence in the
rate of inflation (or other variables). However,
we can measure some of the errors that
contribute to permanent-transitory confusion
and use econom etric methods to estimate the
variance of permanent and transitory errors
in the price level (or other variables). This
section considers these sources of evidence
on the relative size of permanent and
transitory changes.
Bullard (1 994) reported the size of revi
sions to quarterly reports of the rate of change
of the GNP deflator for the years 1986 through
1992. He found that the mean revision for
the 28 quarters was positive in this period;
early reports understated the rate of inflation
and later revisions added additional amounts.
Bullard also reports the range of revisions for
each of the four quarters following the period
considered. The reported ranges exclude the
most extreme 5 percent of the revisions in
each tail. Table 1 reproduces Bullard’s results.
The average rate of inflation for the period
is approximately 3 percent. The range of
revisions (excluding extreme values) is 2.3 to
2.9 percentage points. The revisions are from
70 percent to 130 percent of the (3 percent)
average reported change. The relatively large
size of the revisions suggests that it is rational
for the public to act as if the initial announce
ment is a very noisy indicator of the true rate
of inflation.
In several papers summarized in
Brunner and Meltzer (1 9 9 3 ), we report
forecast errors for the rate of change of output
or inflation using different models, methods
of forecast and countries. The forecasting
113
REVIEW
MAY/JUNE
methods include state-of-the-art econom etric
modeling, time-series analysis, judgm ent and
combinations o f these methods. A
rule-of-thumb summary is that mean
absolute errors for output growth in the
major industrial countries is 50 percent or
more of the average rate of change one
quarter or one year ahead. Inflation is less
variable than output over short periods, and
its forecast error is a smaller fraction of the
average rate of change. Still, errors in both
growth and inflation forecasts are large.
The data on forecast errors make a
persuasive case that forecasters frequently
misperceive future values. Data on revisions
suggest that current reports are subject
to large errors. Errors may be unbiased,
but that is a small consolation for those
who adjust to events that are found later
not to have occurred or those who do not
adjust to changes that did occur. The
permanent-transitory distinction implies that
partial adjustment is optimal. Ex post, it will
seem that adjustment has been sluggish.
Indeed, as noted, errors may appear to be
serially correlated following large changes.
A filter can be used to separate perma
nent and transitory components .7 Let the
error term in an aggregate such as the price
level or output have a permanent and transi
tory component,
(4 )
the quicker the economy adjusts to permanent
changes. A relatively large transitory variance
slows the response.
Meltzer (1 9 8 6 ) estimated the variances
of the permanent and transitory components
of inflation and growth for Canada, Germany,
the United Kingdom, and the United States
under fixed and fluctuating exchange rates.
The fixed exchange rate period runs from the
first quarter of 1960 through the third quarter
of 1971. Fluctuating rates begin in the fourth
quarter of 1971 and end in the fourth quarter
of 1984.8
Table 2 shows the ratios computed from
these data using the adjustment equations in
Muth (1 9 6 0 ). The estimated variance ratios
for inflation and growth changed under fluc
tuating rates, but the direction of change is
not uniform across countries. The length of
the adjustment lag required to distinguish
permanent from transitory shocks changed
much less. For inflation, seven of the eight
values of the adjustment lag (A) lie between
0.41 and 0.56. These values imply that, con
sistent with Blinder’s survey data, about half
of the shock to inflation is seen in the current
quarter. Between 82 percent and 96 percent
of the adjustment of permanent inflation
occurs within four quarters of the initial
shock to the inflation rate. This adjustment
is faster than the two-year average lag com
monly suggested. For real growth, seven of
the eight values of A lie between 0.45 and
0.67; within four quarters, 91 percent to 99
percent of the adjustment occurs. The speed
of output adjustment is broadly consistent
with the length of post-war recessions if
these recessions are interpreted as the cumu
lative adjustment set off by a monetary or
real shock. Inflation adjusts more slowly
than growth as many studies have shown.
Clarida and Gali (1 9 9 4 ) studied
the response of real exchange rates to nominal
shocks in four countries. For Canada and the
United Kingdom, they were unable to find
any structural effects— real or nominal— on
real exchange rates.
For Germany and Japan, the evidence
suggests that nominal shocks explain
45 percent and 34 percent, respectively, of
the four-quarter-ahead forecast error variances
of the log level of bilateral real dollar exchange
£ ( = e pt + e \ ,
where p and r denote the two components.
The transitory component is white noise.
The permanent component has the property
of zero mean and constant variance, a 2.
(5)
7 See Bomhoff (1 9 83 ) fora discus
sion of the Bayesian mulfrstote
Kalman filter.
' X is the adjustment log given by
ond o is the ratio of the variance of
the permanent component to the
variance of the transitory component.
A ep<~ N (0,
)
At any time t, the expected change in the
permanent component is zero.
People observe prices and output. As in
the model of the previous section, they cannot
separate the levels or changes of the permanent
or transitory components. The rate at which
they adjust depends on the relative variances
of the transitory and permanent components
of inflation or output growth. The larger the
variance of the permanent component relative
to the variance of the transitory component,
199
114
REVIEW
MAY/JUNE
rates. Clarida and Gali note that their
estimates are consistent with the evidence
from vector autoregressions reported in
Eichenbaum and Evans (1992).
Clarida and Gali (1994) use a trivariate
vector autoregression to estimate the transi
tory component of real exchange rates .9
Their model includes shocks to aggregate
demand and supply. The authors report the
ratio of the variance of transitory shocks to
the variance of actual shocks. Using their
data, we can compute the ratio of the variance
o f permanent shocks to the variance of tran
sitory shocks. The ratio covers a wide range
in these countries— from 0.42 in Germany to
3 .76 in Canada. These findings suggest that
most shocks to the Canadian real exchange
rate have been permanent while m ost shocks
to the German real exchange rate were tran
sitory. Hence, the Canadian real exchange
rate should adjust more rapidly to shocks
than the German real rate.
Meltzer (1993) used the permanenttransitory distinction to model the U.S.
multilateral real exchange rate. For both levels
and first differences under fixed and fluctuat
ing rates from 1960 to 1991, the data suggest
that there is a large permanent component in
the change of the real exchange rate and a
significant transitory component. Further,
the data suggest that the multilateral real
exchange rate responds to changes in the
nominal stock of money. The effect eventually
vanishes, but monetary changes have real
effects until prices adjust. These findings are
consistent with short-run non-neutrality and
long-run neutrality of money if permanent
changes in money were perceived as transitory
at the time they occurred or conversely.
The studies of prices, output and
exchange rates support the principal
arguments of the article. They are only a
small part of the evidence supporting ex
post, short-run non-neutrality. They are of
interest because they attribute slow adjust
ment of nominal values and real effects of
nominal shocks to the difficulty of discerning
the persistence of shocks. W ith a non-trivial
cost of acquiring information, price setting
and permanent-transitory confusion imply
that nominal changes have real effects that
persist for a time.
1995
T a b le 2
Relative Variances of the Perm anent and
Tra n sito ry Components
1960-71
Inflation
Xp
Real growth
Xy
Canada
Germany
U.K.
U.S.
0.29
0.41
0.62
0.57
1.56
0.69
0.65
0.54
0.37
0.45
0.36
0.45
0.71
0.56
1.11
0.64
Canada
Germany
U.K.
U.S.
0.56
0.52
1.35
0.67
0.30
0.42
1.00
0.62
0.31
0.42
0.04
0.18
0.71
0.56
0.86
0.59
1971-84
Inflation
Xp
Real growth
Xy
Other recent studies find evidence
of costs of acquiring information.
Investors frequently pay a premium to buy
country-specific mutual funds. The premium
implies that they could buy the individual
securities at lower cost. If they are uncertain
about which securities to buy and when to
buy or sell, it may be rational to pay for the
services of traders who specialize in the
particular market.
Smith (1 991) uses costs of acquiring
information as one reason for the absence
of optimal portfolio diversification of world
market securities. The degree of diversifica
tion depends on costs of acquiring information.
People know much more about values and
earnings in their own market than in foreign
markets. Prices in foreign markets may reflect
full information, but some investors either
do not have this information or cannot assess
whether changes are permanent. Hence,
they do not respond promptly to information
about each of these markets. They do not
hold the “true” equilibrium portfolio they
would hold if information costs were zero.
A principal cost in this case, as in others, is
the interpretation of available information.
Permanent-transitory confusion is one part
of the interpretation problem.
In Fuhrer and Moore (1993), the inflation
115
9 The trivariate system has four
lagged values of the change in the
log real exchange rate, the change
in the log ratio of United States home country real GDP, and the
difference between U.S. and for
eign inflation. Data are quarterly
from mid-1970 to the fourth
quarter of 1992.
REVIEW
MAY/JUNE
if the micro-foundations are appropriate for
the task. Standard micro theory, such as
Arrow-Debreu, imposes complete markets
and market clearing in each market. There
is no role for monetary disturbances. This is
not the appropriate micro-foundation for
macroeconomics. No amount of squeezing,
cutting and pasting will make it so.
Rational behavior and rational expecta
tions are entirely consistent with costs of
acquiring information and the inability to
fully identify permanent and transitory
shocks either when they occur or for several
quarters after. Indeed, Muth’s (1 9 6 1 ) initial
formulation of rational expectations is based
on the latter distinction.
One alternative explanation of sticky
prices in recent literature relies on menu
costs and imperfect competition. This expla
nation is a foundation for the so-called
L-shaped supply curve familiar from Keynesian
theories. I show that the evidence in
Blinder’s (1 9 9 1 ) survey rejects the L-shaped
supply curve. Further, the implications of
monopolistic competition, such as widespread
excess capacity, do not explain gradual price
adjustment in most service industries.
Gordon (1 9 9 0 ) proposes a disaggregated
system to take account of the information at
many levels of the economy. He argues that
prices respond to marginal cost but that mar
ginal cost for any firm depends on the pricing
strategy of its suppliers. Hence, such infor
mation enters firms’ decisions about price
and output adjustment. Information from
macrodata is much less relevant.
This argument captures some of the
dynamics of pricing, but it poses unresolved
challenges for aggregation over different
industry structures. Moreover, Gordon’s
framework implies that commodity producers
should adjust slowly to aggregate shocks and
that large firms in durable goods industries
should adjust promptly. The stylized facts
suggest that the opposite is true. One reason
is that organized commodity markets increase
the information available to commodity pro
ducers. There are no comparable markets for
consumers’ and producers’ durable goods out
put. Differences in information and the costs
of acquiring information are consistent with
the stylized facts on speed of adjustment.
rate is sticky. Firms adjust relative prices to
the average o f other sectors’ expected
relative prices over the life of existing
contracts. Firms also adjust for the
current and expected level of output. The
autocorrelation functions based on their
model have very similar shapes to the auto
correlations generated by an unconstrained
vector autoregression. In particular, they
show considerable persistence in inflation
and output movements and sustained
effects o f inflation on real output. In short,
Fuhrer and Moore provide evidence that is
consistent with a model in which there are
costs of learning about the permanence or
persistence of changes, and in which
people adopt strategies that leave room
for misperception and real effects of
nominal changes.
Earlier work by Boschen and
Grossman (1 9 8 2 ), Gordon (1982 ) and
Mishkin (1983) also provide evidence
that supports sticky prices. Indeed, the
evidence of gradual adjustment of prices
and of short-term real effects of monetary
change is common. These studies lack
micro-foundations. Price-setting in part of
the economy and permanent-transitory
confusion, as in Brunner, Cukierman and
Meltzer (1 9 8 3 ), reconciles this evidence
with rational behavior.
CONCLUSION
9 The trivoriote system has four
logged values of the change in the
log real exchange rate, the change
in the log ratio of United States ■
home country real GDP, ond the
difference between U.S. ond for
eign inflation. Dato are quarterly
from mid-1970 to the fourth
quarter of 1992.
The examples in the preceding section
are a small part of the recently accumulated
evidence showing that there is m uch more
than casual observation to support the main
propositions in this paper: Nominal prices
adjust with a lag. The lag is sufficiently
long that real variables respond to nominal
changes. Costs of acquiring information
about the persistence of observed changes—
permanent-transitory confusion— is a main
component of the cost and a main reason
for the lag. Even where prices reflect infor
mation fully, all individuals or firms may
not have adjusted fully to available but
costly information.
The oft-repeated comment that
macroeconomics should be built on
micro-foundations is correct if and only
1995
116
REVIEW
MAY/JUNE
Ball, Laurence, ond N. Gregory Mankiw. "A Sticky Price Manifesto,"
Carnegie Rochester Conference Series on Public Policy (December
1994), pp. 127-52.
The micro-foundations suggested in this
article use costs of acquiring information to
explain three common observations. First,
many prices are set. Second, price setters
choose nominal values. Third, the daily, week
ly, monthly or quarterly variances o f “set
prices” are small fractions of the variance of
prices in auction markets; set prices are stickier.
Households and firms do not operate in
a world of full information. Incomplete infor
mation and costs of acquiring information are
central problems of a monetary economy.
Information and transaction costs explain why
people hold and use money as a medium of
exchange (Brunner and Meltzer, 1971). There
is considerable evidence that these costs are
not trivial. The article cites revisions to report
ed data, forecast errors, incomplete informa
tion about costs, profits and strategies of
competing firms. These examples do not
exhaust the costs that firms and individuals
face.
Some of these costs can be reduced by
institutional and contractual arrangements.
The arrangements that people choose may
be optimal when contracts are written but,
with changes in the environment, there are
unforeseen gains and losses. If the gains and
losses are transitory, their expected value is
zero. It may not be worthwhile to change the
contract or the method of contracting. Once
the change is considered persistent, gains and
losses are expected to cumulate. Adjustment
or re-negotiation becomes more appealing to
at least one party.
Information about permanent and tran
sitory changes in profits, prices, wages and
other variables is costly to acquire. The
distribution of shocks between real and
nominal, permanent and transitory may differ
from one sample period to the next. People
learn to monitor events or changes that are
costly to ignore. But learning requires a
continuous process of monitoring both what
has happened and what should be observed.
This is a basic problem for firms and house
holds. As such, it is a more appropriate
micro-foundation for macroeconomics.
Balvers, R. J., and T. F. Cosimano. "Actively Learning About Demand
and the Dynamics of Price Adjustment," Economic Journal (September
1990) pp. 882-98.
Baumol, William J. "The Transactions Demand for Cash: An Inventory
Theoretic Approach," Quarterly Journal of Economics (November
1952), pp. 545-56.
Beoulieu, Joe., and Joe Mattey. "The Effects of General Inflation and
Idiosyncratic Cost Shocks on Within-Commodity Price Dispersion:
Evidence from Microdata," finance and Economic Discussion Series
No. 94-12 (May 1994), Board of Governors of the Federal Reserve
System.
Blinder, Alan S. “Why Are Prices Sticky? Preliminory Results from an
Interview Study," Tbe American Economic Review (May 1991),
pp. 89-96.
Bomhoff, Eduard J. Monetary Uncertainty. North Holland, 1983.
Boschen, John F., and Herschel I. Grossman. "Tests of Equilibrium
Macroeconomics Using Contemporaneous Monetary Data," Journal of
Monetary Economics (November 1982), pp. 309-33.
Brunner, Karl, and Allan H. Meltzer. Money and the Economy: Issues in
Monetary Analysis. Cambridge University Press, for the Raffoele
Mattioli Lectures, 1993.
_ _ _ _ _ _ and_ _ _ _ _ _ _ . "The Uses of Money: Money in the
Theory of an Exchange Economy," Tbe American Economic Review
(December 1971), pp. 784-805.
_ _ _ _ _ _ and_ _ _ _ _ _ _ . "Economies of Scale in Cash Balances
Reconsidered," Quarterly Journal of Economics (August 1967),
pp. 422-36.
_ _ _ _ _ _ , Alex Cukierman and Allan H. Meltzer. "Money ond
Economic Activity: Inventories and Business Cycles," Journal of
Monetary Economics (May 1983), pp. 281-319.
Bullard, James B. "How Reliable Are Inflation Reports?" Federal
Reserve Bonk of St. Louis Monetary Trends (February 1994), p. 1.
Clarida, Richard, and Jordi Gali. "Sources of Real Exchange Rate
Fluctuations: How Important Are Nominal Shocks?" Carnegie
Rochester Conference Series on Public Policy (December 1994),
pp. 1-56.
Eichenbaum, Martin, and Chorles Evons. "Some Empirical Evidence
on the Effects of Monetary Policy Shocks on Exchange Rotes,"
Federal Reserve Bonk of Chicago Working Paper No. 92-32
(December 1992).
Fuhrer, Jeffrey, and George Moore. "Inflation Persistence," Finance and
Economic Discussion Series No. 93-17 (May 1993), Board of
Governors of the Federal Reserve System.
Gordon, Robert J. "What Is New-Keynesian Economics?" Journal of
Economic literature (September 1990), pp. 1115-71.
REFERENCES
_ _ _ _ _ _ . "Price Inertia and Policy Ineffectiveness in the United
States, 1890-1980," Journal of Political Economy (December 1982),
pp. 1087-1117.
Alchian, Armen A. "Why Money?" Journal of Money, Credit ond
Banking (February 1977, part 2), pp. 133-40.
199
117
REVIEW
MAY/JUNI
Heaton, John, and Debra Lucas. "The Importance of Investor
Heterogeneity and Financial Market Imperfections for the Behavior
of Asset Prices," Carnegie Rochester Conference Series on Public
Policy (forthcoming).
1995
Muth, John F. "Rational Expectations and the Theory of Price
Movements," Econometrica (July 1961), pp. 315-35.
_ _ _ _ _ _ . "Optimal Properties of Exponentially Weighted Forecasts,"
Journal of the American Statistical Association (June 1960),
pp. 299-306.
Knight, Frank H. Risk, Uncertainty and Profit (No. 16 in reprints of
Scarce Classics in Economics). London School of Economics, 1933.
Smith, Clifford W., Jr. "Globalization of Financial Markets," Carnegie
Rochester Conference Series on Public Policy (spring 19 91),
pp. 77-96.
Lucas, Robert E., Jr. "Expectations and the Neutrality of Money,"
Journal of Economic Theory (April 1972), pp. 103-24.
McCollum, Bennett T. "New Classical Macroeconomics: A Sympathetic
Account," Scandinavian Journal of Economics (June 1989),
pp. 223-52.
Meltzer, Allan H. "Real Exchange Rates: Some Evidence from the
Postwar Years," this Review (March/April 1993), pp. 103-17.
_ _ _ _ _ _ . "Size, Persistence and the Interrelation of Nominal and
Real Shocks: Some Evidence from Four Countries," Journal of
Monetary Economics (January 1986), pp. 161-94.
_ _ _ _ _ _ . "Rational Expectations, Risk, Uncertainty and Market
Responses," in Paul Wachtel, ed., Crises in the Economic ond financial
Structure. Lexington Books, 1982, pp. 3-22.
Mishkin, Fredrick S. A Rahonal Expectations Approach to
Macroeconometrics: Testing Policy Ineffectiveness and Efficient
Markets Models. University of Chicago Press, 1983.
Ul S
118
REVIEW
MAY/JUNI
1995
Randall Wright is a professor of economics at the University of Pennsylvania and serves as a consultant to the Federal Reserve Bank of
Minneapolis. The author thanks Dan Thornton for useful comments on an earlier draft, and Lee Ohanian for several useful discussions.
Com m entary
assertion itself. To focus the discussion, I
propose to debate the following position:
We have m ade little progress in m acroeconom ics
since the Keynesian-m onetarist debates, and
existing m odels built on micro-foundations are
neither fruitful nor promising. I perceive this
position to be a fair reflection of the view
expressed in the paper. But even if this
is not exactly what Professor Meltzer had
in mind, I believe that it is an interesting
issue to debate. I hope the reader will
forgive me if it appears I am debating a straw
man, and indulge me the opportunity to
present some of my own views on the state
of macroeconomics.
As I see it, econom ists have made
remarkable progress in understanding things
that bewildered us ju st two or three decades
ago. I will describe this progress in four of
the most important areas of macroeconomics:
business cycles; the labor market; monetary
econom ics; and growth. I will also discuss
some more general methodological issues
toward the end. This is not meant to say
that I am totally unsympathetic to the views
of Professor Meltzer, merely that I think
he overstates the case when he asserts that
existing macroeconomic models are neither
fruitful nor promising.
Randall W right
llan Meltzer raises a variety o f issues,
and reviews and extends some research
he and his collaborators have been pur
suing over the years. Some o f the more or
less technical points he presents, both with
regard to the theory and the evidence, will
undoubtedly be of interest to many macro
economists. He does a good jo b of presenting
these technical points, and so my plan is not
to discuss them in any detail here. Instead, I
want to address some more general method
ological issues. That is, I plan to comment
mainly on some remarks Meltzer makes on
the state o f macroeconomics.
To provide some motivation for the
discussion, I would like to begin with a few
quotations from the Introduction to his
paper. Meltzer says that “For decades,
macroeconomists have listened to criticism
from their professional colleagues about the
absence of micro-foundations for most of
what they say and do...It is time to question
whether this now widely accepted approach
is likely to be fruitful.”
He appears from his remarks to be of the
opinion that the answer is no. W hile con
ceding that we may have learned one or two
things over the years, the suggestion is that
much of modem macroeconomics is at a dead
end. For example, “Overlapping generations
models of money, intertemporal substitution
theories of unemployment and productivity
shock theories of the business cycle have not
proved fruitful. ..[and] the results to date are
not prom ising” [emphasis added]. He further
suggests that the current state of affairs com
pares to the Keynesian-monetarist debates of
a generation or so ago.
Presumably, he puts forward this assertion
so that the reader will be more sympathetic
to the alternative approach provided in his
paper. But what I want to do is question the
BUSINESS CYCLES
Two decades ago, few would have believed
the following assertion: A frictionless,
competitive, non-monetary model built
around the one-sector growth model,
abstracting from heterogeneity, distortionary
taxation, and many other features of reality,
can generate time series that look like
those in the data when hit by impulses that
seem like a reasonable representation of
stochastic technological progress. Since
the work of Kydland and Prescott (1 9 8 2 ),
Hansen (1 9 8 5 ) and others, we know that the
assertion is true. But I am sure that even
Kydland and Prescott would not have
expected it ex ante.
Consider the original version of what I will
119
REVIEW
MAT/JUNI
are estimating dynamic GE models using
traditional econom etric methods (see,
for example, McGrattan, 1994).
call their dynamic general-equilibrium (GE)
model. (This is a more accurate label than
the more common real business cycle, or
RBC, model, given that many people work
with monetary versions of the m odel).
It included many complications, such as
time-to-build, non-time-separable utility,
and signal extraction problems, which, while
still included in some applications, are not
parts of the current standard benchmark
model. W hy were those complications there?
They thought a simple model wouldn’t stand
a chance. On one interpretation, the entire
exercise was to see ju st how bad things were,
so we would have some idea where to go
next (for example, in terms of adding other
impulses and propagation m echanism s).
To everyone’s surprise, however, even
very simple dynamic GE models do quite
well at replicating key aspects of the macro
time series. Output is more volatile than
consumption, not as volatile as investment,
and about as volatile as the labor input; and,
the coherence of all these series is high.
Furthermore, the model is consistent with
these features of the data at a quantitative
level, not ju st a qualitative level.
Traditional macroeconomists, especially
Keynesians, reacted to these findings with
much suspicion, and virtually every aspect
o f the analysis was called into question. In
retrospect, many controversial issues turned
out to be red herrings, including the following:
1 «»S
(4) The HP filter is simply a convenient
tool, and obsessing over its merits or
demerits is like debating whether the
mean, median or mode is the “correct”
measure of central tendency.
The consensus today is that the dynamic
GE models are useful tools for studying busi
ness cycles. O f course, this does not mean
that business cycle research is a solved prob
lem. There are many unanswered or partially
answered questions, such as the correlation
between employment hours and productivity,
the equity premium, and the relations between
real and nominal variables. M uch work has
been done to address these questions with
some, but not total, success. There are still
interesting puzzles out there— but this is
why working in the area is exciting. The
point is that we now have a standard m odel of
the business cycle, a base case from which to
generalize when the situation warrants it.
The dynamic GE approach is a tool
for macroeconomics the way that the
supply-and-demand approach is a tool for
microeconomics. One should not ask: “Is
the model true?” but only: “Is it useful?”
Have we made progress understanding busi
ness cycles? Yes. Are these models based on
m icroeconom ic foundations? Yes. They are
based on the standard econom ic principles of
constrained optimization (which, in a dynamic
context, obviously concerns intertemporal
substitution) and a coherent concept of
equilibrium. Can the base model accommo
date frictions, money, heterogeneity, private
information and so on? Yes. Do we need to
throw out dynamic GE theory in favor of
new micro-foundations or a retrograde macro
approach? No, no more than we need to throw
out supply-and-demand curves.
O f course, a base model is always sim
plistic. In the case of supply and demand,
for example, suppose we want to know what
will happen to the price of orange ju ice after
a frost in Florida or a Vitamin C craze. Is it
OK to abstract from private information,
strategic issues, reputation and so on, and
proceed by shifting the supply or demand
(1) Abstracting from heterogeneity (that
is, focusing on a representative agent) is
an assumption that, depending on the
questions, is sometimes appropriate and
sometimes inappropriate, but is never
good or bad as a matter of principle.
(2) Abstracting from market failures,
frictions and money is a proper first
step— no one should advocate complica
tion for its own sake— and the fact that
simple models can be solved efficiently
by exploiting the welfare theorems does
not mean that users of these models
believe the real world is “first best” nor
that policy is unworthy of discussion.
(3) Calibration is a way of taking models
to the data that avoids many complica
tions; although these days many of us
120
RlYIEN
MAY/JUNE
multiple equilibrium considerations, and so
on, each may have some elements of truth to
them. Moreover, these models are not mutu
ally inconsistent, but are complimentary
special cases of a general framework (see
Mortensen, 1989). We should not look for a
simple single answer.
Are frictions in the labor market impor
tant, as Meltzer suggests? Yes. Can private
information be a relevant consideration, as
Meltzer suggests? Yes. Does this mean a
move away from micro-foundations, or a
move to new micro-foundations, is the answer?
No. Many researchers have been working on
incorporating frictions and informational
considerations into the standard paradigm
for years. It has been successful. Given this
success, I do not see any reason to argue to
return to a reduced-form Phillips Curve
approach. My preferred alternative is to
learn search theory and forge ahead.
curve? I don’t know the answer definitively,
but I think, provisionally, yes. Similarly, if
we want to ask something basic about busi
ness cycles, it seems reasonable to use the
basic dynamic GE framework as the bench
mark. To readers interested in studying this
in more detail, I recommend the book
Frontiers o f Business Cycle Research, edited by
Thomas Cooley.
THE LABOR MARKET
It is commonly believed that unemploy
ment is a major economic and social problem.
We have not come up with a definitive model
that explains unemployment or gives us a
panacea to cure unemployment. That is, we
have not solved the problem. But we know
something about it and can study it scientifi
cally. We know that incentives matter,
whereby I mean things like unemployment
insurance, dismissal restrictions, tax policy
and so on. These things can be built into
econom ic models built on standard micro
foundations (constrained optimization and
coherent equilibrium concepts), and analyzed
both qualitatively and quantitatively.
Some of the issues are more or less
static: for example, the incentive effects of
unemployment insurance on layoffs and hours
per worker (see, for example, Burdett and
Wright, 1989). Others are intrinsically
dynamic. A major recent success concerns
the application of search models of labor
market dynamics to worker and jo b flow data.
Combining the jo b creation-job destruction
data analysis of Davis and Haltiwanger (1990)
with the theoretical framework laid out, for
example, in Pissarides (1990) has proved
fruitful. These authors have used dynamic
GE models based on search theory to
account for the main empirical features of
the labor market, like the jo b creation and
jo b destruction data (see, for example,
Mortensen, 1994). These models can be
used to study policy interventions qualita
tively and quantitatively.
O f course, as I stated earlier, there is
more than one model of unemployment.
This is as it should be. There is more than
one type of unemployment. Efficiency wage
considerations, insider-outsider considerations,
1995
M ONETARY ECONOMICS
Not so long ago, there did not exist in
the literature a serious formal model of a
dynamic monetary economy. The overlapping-generations (OLG) model, invented by
Samuelson (1958) and developed by many
people (see, for example, Wallace, 1980), has
remedied this deficiency That model has
been and continues to be an extremely
useful framework within which to illustrate
theoretical properties of monetary economies,
to interpret episodes in econom ic history, to
shed light on policy debates, and to discover
new things about econom ics generally.
Concerning the latter, it is worth remarking
that many technical discoveries, such as the
possible inefficiency of competitive equilibrium,
or the potential for endogenous limit cycles
and sunspot equilibria, revolved closely
around the analysis of OLG models (see, for
example, Azariadis, 1993). These discoveries
seem important for macroeconomics.
W hen Meltzer criticizes the OLG model,
perhaps what he has in mind is that there are
certain phenomena for which it is ill-designed
to explain. One could belabor the obvious
and argue that money in the OLG model is
only a store of value and not a medium of
exchange. But a model need not capture
121
HEVIIN
MAY/JUNE
every feature or nuance of money in order to
teach us something about monetary theory
or policy. More to the point, we now have
theoretical models in which money clearly
and indisputably is a medium o f exchange.
Some of these models are built around search
frictions that capture Jevons’ famous “double
coincidence of wants” problem with direct
barter; see, for example, Kiyotaki and Wright
(1989) or Trejos and Wright (1995 ). Others
are built around private information prob
lems; see, for example, W illiamson and
Wright (1994).
Allan Meltzer, along with Karl Brunner,
is on record as saying that private information
is the driving force behind the use of money
in modem economies. He reiterates this
position in the current paper. Some of us
who work in monetary theory have taken
his position to heart and have attempted to
formalize these ideas. We do not think of
ourselves as abandoning micro-foundations;
the models are built on search or private
information frictions incorporated into
microeconomic models with optimizing
agents and coherent equilibrium concepts.
I agree with Professor Meltzer when he
argues that we need to develop theories that
incorporate not only money, but also other
aspects of the real world, like brokers, dealers,
market makers, intermediaries and so on.
Given this, it seems that search-based models
of the sort analyzed by Rubinstein and
Wolinsky (1 9 8 7 ), for example, are promising.
Like much of the search-based monetary
theory, these models are primitive, but they
do address many of the issues that Meltzer
correctly identifies as important.
Due to their rudimentary nature, the
models to which I am referring are not yet
very good at providing policy guidance. They
do not answer, “W hat should we do at the
discount window next week?” I for one do
not think that this is the m ost interesting
question in monetary economics. Even if
one is interested mainly in policy, there is
potential value in building qualitative models
that help edify us and our students regarding
more basic issues. At the same time, monetary
dynamic GE models currently exist that,
although not as well-grounded in terms of
first principles of m icroeconomics, can be
1995
brought to bear on more mundane policy
affairs. We have seen some of them discussed
at this conference.
Is it a problem that pure and applied
monetary economics have not converged?
In any science, it should not be too surpris
ing that progress in pure and applied theory
proceeds in counterpoint and not in unison.
That is why monetary econom ics today is
vibrant and flourishing.
GROW TH
It was only about a decade ago that
macroeconomists were relatively uninterested
in growth theory and in policy directed toward
econom ic growth. One reason may be that
our attention was directed toward other
issues— business cycles, unemployment and
money. Another reason is that we were look
ing at the wrong models. Although models
with perpetual growth have been around for
decades, the standard Solow model in the
textbooks is in the unfortunate position of
not explaining growth, except as a transitory
phenomenon on the way to a steady state
or as the outcome of exogenous technical
progress. We owe something to Romer (1986)
and Lucas (1988) for redirecting our attention.
There is now a plethora of endogenous
growth theories— arguably, too many.
However, these models all have common
threads that I hope will allow us to distill
common essence. We know that growth is
important as a matter of welfare. As com
pared to eliminating cyclical fluctuations in
GNP, getting the growth rate up a few per
centage points is an order of magnitude more
important. Can we achieve higher growth by
better policy? Can we understand why dif
ferent economies grow at different rates?
The jury is still out, but these are obviously
interesting questions. The way to answer
them is with standard econom ic theory.
W hen I say “standard econom ic theory,”
I do not mean that we should stick to the
status quo when confronting issues for
which the textbook model is inappropriate.
But going from Solow’s leading example to a
model with non-decreasing returns is hardly
a scientific revolution. Is there a role for
private information or other frictions in
122
REVIEW
MAY/JUNI
modern growth theory? Potentially. The
interaction of growth with financial develop
ment and intermediation may be interesting
and important. Some work has been done and
more is in progress. But 1 did not see anything
in Meltzer’s approach that makes me want to
stray from mainstream growth theory.
approach. We have been since then mostly
engaged in what Kuhn (1 9 6 2 ) calls “normal
science.” To be sure, there are disagreements,
but there is a core of good people working
on interesting and important questions and
making progress.
The bottom line is that macroeconomics
has made impressive advances on a large
number of fronts. Today, we have models built
on first principles— that is, on constrained
optimization and a consistent concept of
equilibrium— of business cycles, unemploy
ment, money and growth. It is still true that
good economists often have difficulty with
questions like, “W hat should we do at the
discount window next week?” This may
suggest the questions are ill-posed (although
I do understand and have sympathy for the
many professional economists who cannot
ignore such questions because they get paid
to come up with answers).
Perhaps I am too sanguine. How about
our failures? One thing that Meltzer empha
sizes that we are not so good at explaining
is sticky prices. This may be because we
sometimes take the pricing aspect of the
Arrow-Debreu paradigm too seriously. We
know that there are many ways to decentralize
a given allocation. Contracts, core-like coali
tions, reputation and several other institutions
are also possibilities, as Meltzer mentions. It
may be that agents get the allocation right
without using prices in the way that our
textbooks assume. That is, in principle, prices
may “look” sticky but this need not have
implications for welfare or policy.
W hen do sticky prices matter? Sticky
prices can be studied in dynamic GE models,
as Ohanian and Stockman (1 9 9 4 ), Cho and
Cooley (1994) and others have shown. These
authors do not explain w hy prices are sticky;
rather, they investigate the implications of
varying degrees of exogenous stickiness.
Should we try to explain stickiness endoge
nously? Maybe, but I was not convinced by
Meltzer’s current article.
I would like to conclude by saying that I
have always learned from Professor Meltzer,
especially on questions in monetary econom
ics. It is worthwhile trying to take seriously
his notions of information theory as a foun
dation for monetary theory. In other areas, he
GENERAL METHODOLOGICAL
ISSUES
There have been many technical devel
opments that have paved the way for these
successes in macroeconomics. One obvious
innovation involves computational ability.
Graduate students now have machines on
their desks that allow them to solve and sim
ulate dynamic GE models as homework in a
good first year macro-course. If one really
thinks that heterogeneity, incomplete markets,
income distribution or related issues are
important, we now have the technology
and the power to solve models with these
complications; see Rios-Rull (1995).
There have been developments outside
the domain of hardware. The publication of
Stokey and others (1989) illustrates how we
now all have access to a set of tools that few
macroeconomists were comfortable with not
so long ago. The analysis of multiple equi
libria, including dynamic multiplicity,
endogenous limit cycles and phenomena like
sunspot equilibria have given us a new set of
ways to think about the world. Game theory
has provided us with new ways of posing
and solving strategic questions, including
bilateral bargaining problems that are central
to some of the phenomena that Meltzer
emphasizes (see the references in the section
of monetary econom ics). Analysis of data,
like the jo b creation and destruction data, or
the cross-country growth data, have given us
new things to think about and new ways of
confronting our models with reality.
Lucas (1980) emphasized the interplay
between technical developments, on the one
hand, and deviations between theory and
facts, on the other hand, as what leads to
progress and change. He argues that this
interplay was behind the emergence in the
1970s of “rational expectations” macroeco
nomics and the downfall of the IS-LM
199
123
REVIEW
MAY/JUNE
1995
Pissarides, Christopher. Equilibrium Unemployment Theory. Blackwell,
1990.
is also posing interesting questions. Standard
macroeconomic GE models provide a venue
for their analysis.
Rios-Rull, Jose-Victor. "Models with Heterogeneous Agents," in Thomas
F. Cooley (ed.) Frontiers of Business Cycle Research. Princeton
University Press, 1995, pp. 98-125.
REFERENCES
Romer, Paul M. "Increasing Returns and Long-Run Growth," Journal of
Political Economy (October 1986), pp. 1002-37.
Azariadis, Costas. Intertemporal Macroeconomics. Blackwell, 1993.
Burdett, Kenneth, and Randall Wright. "Unemployment Insurance and
Short-Time Compensation: The Effects on Layoffs, Hours per Worker,
ond Wages," Journal ofPolilical Economy (December 1989), pp.
1479-96.
Rubinstein, Ariel, and Asher Wolinsky. "Middlemen," Quarterly Journal
of Economics (August 1987), pp. 581-93.
Samuelson, Paul. "An Exact Consumption-Loan Model of Interest With
or Without the Social Contrivance of Money," Journal of Political
Economy (December 1958), pp. 467-82.
Brunner, Karl, ond Allan H. Meltzer. "The Uses of Money: Money in the
Theory of an Exchange Economy," The American Economic Review
(December 1971), pp. 784-805.
Stokey, Nancy L., Robert E. Lucas, Jr., and Edward C. Prescott. Recursive
Methods in Economic Dynamics. Harvard University Press, 1989.
Cho, Jang-Ok, and Thomas F. Cooley. "The Business Cycle with Nominal
Contracts," working paper (1994).
Trejos, Alberto, and Randall Wright. "On the State of Macroeconomics,"
Journal of Political Economy Wo. 1, 1995), pp. 118-41.
Cooley, Thomas F. Frontiers of Business Cycle Research. Princeton
University Press, 1995.
Wallace, Neil. "The Overlapping Generations Model of Fiat Money," in
John H. Kareken and Neil Wallace, eds. Models of Monetary
Economies. Federal Reserve Bonk of Minneapolis, 1980.
Davis, Steven J., and John Holtiwanger. "Gross Job Creation and
Destruction: Microeconomic Evidence and Macroeconomic
Implications," Notional Bureau of Economic Research Macroeconomics
Annual (1990).
Williamson, Steve, and Randall Wright. "Barter and Monetary Exchange
Under Private Information," The American Economic Review (March
1994), pp. 104-23.
Hansen, Gary D. "Indivisible Labor and the Business Cycle," Journal of
Monetary Economics (November 1985), pp. 309-27.
Lucos, Robert E., Jr. "On the Mechanics of Economic Development,"
Journal of Monetary Economics (July 1988), pp. 3-42.
_ _ _ _ _ _ . "Methods and Problems in Business Cycle Theory,"
Journal of Money, Credit and Banking (November 1980,
part 2), pp. 696-715.
Kiyotoki, Nobuhiro, and Randall Wright. "On Money os a Medium of
Exchange," Journal of Political Economy (August 1989), pp. 927-54.
Kuhn, Thomas S. The Structure of Scientific Revolutions. University of
Chicago Press, 1962.
Kydland, Finn E., ond Edward C. Prescott. "Time to Build and Aggregate
Fluctuations," Econometrica (November 1982), pp. 1345-70.
McGratton, Ellen R. "The Macroeconomic Effects of Distortionory
Taxation," Journal of Monetary Economics (June 1994),
pp. 573-601.
Mortensen, Dale T. "The Cyclical Behavior of Job and Worker Flows,"
Journal of Economic Dynamics and Control (November 1994),
pp.l 121-42.
_ _ _ _ _ _ . "The Persistence and Indeterminacy of Unemployment in
Search Equilibrium," Scandinavian Journal of Economics (June 1989),
pp. 347-70.
Ohanian, Lee E., and Alan C. Stockman. "Short Run Independence of
Monetary Policy under Pegged Exchange Rates," working paper
(1994).
UIS
124
REVIEW
MAY/JUNK
Reply to
W right's
Com m entary
tive or useful mediums of exchange, no dif
ferences in costs of acquiring information,
and no distinction between money, bonds
and capital that would enlarge the role of rel
ative price changes in the transmission of
monetary and real shocks. I am critical of
models with one open market interest rate
and no sticky prices, and models in which
money is introduced as a socially costly way
to overcome frictions in an otherwise frictionless Walrasian model. I am skeptical of
some of the conclusions drawn from such
models.
W right gives considerable space to
methodological issues. Although the models
we use influence the way we look at the
world, it is a mistake to confuse the model
with the world. Science, well done, does not
equate the model to the world; it recognizes
that all useful models generate refutable
propositions. The weak testing procedure
called calibration that is now fashionable is a
distant substitute for serious, careful assess
ment of competing hypotheses.
Much recent work, including real business
cycle models or overlapping generation models
of money, have implications that are readily
refuted. For example, it is well established
that all correlations between money and
income are not the result of “reverse causa
tion” and that all unemployment is not the
result of intertemporal substitution.
I have proposed an alternative model in
which uncertainty and costs of transactions
and information have a large role. Monetary
arrangements (and other institutions) reduce
these costs, but some costs are unavoidable.
Wright’s comment reports on some of
the research that is now under way or that
has been published in the recent past. As
always, some of this work will prove fruitful,
some not. There is a high cost of informa
tion and great uncertainty about which will
be successful.
It is my view that progress on the impor
tant issue about how a monetary economy
adjusts to changes or shocks will require
more attention to uncertainty about the
Allan H. M eltzer
andall Wright has several complimentary
things to say about some of my earlier
work. I thank him for those remarks.
Most of his comment, however, fails to discuss
my current paper, and the main comments he
makes about the paper are untrue. Specifically,
my paper does not attack theory or oppose
the development of micro-foundations for
macroeconomics. It is not about economic
policy. I am at a loss to understand how a
reader could come away with either an idea
or with a belief that I am critical of recent
work on econom ic development and growth.
The paper proposes specific hypotheses
for analyzing the role of money and uncer
tainty. The foundation is a micro-theory in
which there is production for inventory.
Uncertainty about the duration of observed
changes gives rise to costs of information.
The reason is that permanent and transitory
changes cannot be distinguished for some
time after a change occurs. In this model,
money is privately and socially valuable
because it reduces costs of transacting and
bearing uncertainty.
I use this framework to discuss three
problems related to the questions that the
organizers asked me to address: ( 1 ) why
some prices are set; ( 2 ) why some firms set
nominal prices; and (3) why some prices are
sticky-have substantially less variance weekly
or monthly than prices in auction markets.
I give explicit references to papers in which
the framework is more fully developed
or in which it has been used fruitfully in
empirical studies.
I contrast this framework with others in
which there are no sticky prices, no produc
R
199
125
MAT/JUNI
permanence of shocks and to costs of
information and transactions. That view
may prove right or wrong, but it is neither
atheoretical nor anti-theory.
I respect W right’s past work and looked
forward to his comments. I regret that he
avoided discussion of the issues raised in my
article and my proposals for dealing with
them. Such a discussion is overdue.
126
1995
REVIEW
MAY/JUNI
199
Ben S. Bernanke is a professor of economics at Princeton University's Woodrow Wilson School.
A Conference
Panel Discussion:
W hat Do We
Know About
How M onetary
Policy Affects
the Economy?
measure closely related to the ratio o f non
borrowed reserves to total reserves). Second,
estimate a standard VAR system including
the relevant endogenous variables and the
policy variable, with the policy variable
ordered last. This structure imposes the
assumptions that the policymaker (potentially)
responds to contemporaneous information,
but that shocks to policy feed back to the
economy with at least a one-period delay.
Finally, calculate the implied impulse
response functions for the endogenous
variables in the system; these provide esti
mates of the dynamic response of the economy
to an unanticipated policy change.
There are now a number o f studies that
show that this method can give robust and
plausible measures of the behavior of interest
rates, output and many other variables to a
monetary policy shock, despite the minimalist
identifying assumptions. Several caveats
should be offered, however:
(1) The method depends on the choice
of policy measure being a valid one. No simple
or mechanical criterion, such as forecasting
power, can determine the optimal policy
measure. For the case of monetary policy,
the choice of policy measure depends on the
way the Fed chooses to implement its policies,
for example, by an interest rate targeting rule
or by targeting a component of bank reserves.
As is well-known, the Fed’s operating proce
dures have changed over time and, hence, no
single policy measure may be best for an
extended sample period. In ongoing research,
Ilian Mihov and I have estimated models of
the Fed’s operating procedure for different
sub-periods. We find that the funds rate is
an excellent indicator of the stance of mone
tary policy for the 1965-79 period but, more
recendy, the best indicator is one that combines
information from both the funds rate and
measures of reserves.
(2) As Sims (1 9 9 2 ) was the first to note,
the VAR approach to identifying the results
of policy shocks will give invalid results if
the policy innovation is dominated by the
policymaker’s response to information not
Ben S. Bernanke
his conference addressed two broad
issues. First, can Fed policies affect real
and nominal interest rates; and, if so, by
what mechanisms? Second, by what chan
nels do Fed actions affect real economic
activity (if they do)?
On the issue of whether the Fed can
affect interest rates: We have always been
pretty sure that it could, but it’s nice that we
now have formal econom etric methods that
can both verify the existence of a “liquidity
effect” and perhaps also obtain quantitative
measures of the linkage between interest rate
changes and changes in output, prices and
other key macro variables. Since I have the
opportunity, let me put in a few good words
for one of these methods, the semi-structural
VAR approach employed by Bernanke
and Blinder (1992), Strongin (1992) and
Christiano, Eichenbaum and Evans (1994),
and discussed further here by Larry Christiano
in his comment on Adrian Pagan’s paper.
This method, as described in more detail in
the above-mentioned sources, involves three
basic steps. First, based on institutional
analysis (for example, of Fed operating pro
cedures), identify a variable or combination
of variables that measure the stance of policy
(for example, Bernanke and Blinder opt for
the federal funds rate; Strongin uses a
T
127
REVIEW
M A T / J UN I
captured in the VAR. This problem is the
source of the infamous “price puzzle,” the
finding in some cases that a tightening of
monetary policy is followed by a rise in the
price level. Sims showed that this problem
can be eliminated by including a variable in
the VAR that proxies for the Fed’s information
about future inflation (for example, a com
modity price index or the exchange rate).
Christiano, Eichenbaum and Evans (1994)
find that including a commodity price index
and measuring the general price level by an
index that treats housing costs correctly (for
example, the GDP or Personal Consumption
Expenditure, PCE, deflator) largely eliminates
the price puzzle. My own experimentation
with these systems suggests that the
Christiano, Eichenbaum and Evans result
is quite robust.
(3)
Finally, although the identification
method works by tracing out the effects of
unanticipated policy shocks, this approach
takes no stand on whether it is only unantic
ipated monetary policy that “matters.” It
may well be the case that forecastable changes
in policy have a stabilizing effect on the
economy; measuring this effect, however,
requires the imposition of more economic
structure in the analysis. Because the
semi-structural VAR method does not
account for the possibly stabilizing effects of
predictable policy changes, this approach
cannot tell us whether policy has, on net,
been stabilizing or destabilizing during the
sample period. Thus, mechanical variance
decompositions that attribute a given per
centage of the variance of output or prices to
monetary policy can be misleading. At best,
variance decomposition exercises may suggest
the amount by which more predictable policies
could have reduced the variance of output
and prices in a given sample period.
Given the empirical support for the
existence of a liquidity effect, the next task is
to find theoretical models that rationalize
this effect. Alan Stockman and Lee Ohanian’s
paper in this conference does a nice jo b of
surveying the leading approaches. 1 was
particularly interested in their model which
assumes the existence of both flexible-price
and sticky-price sectors; it seems both realis
tic and a promising source of empirical
128
1995
applications. A small suggestion: Stockman
and Ohanian find in some of their simulations
that the effect of a monetary shock on interest
rates is ambiguous because of countervailing
liquidity and Fisher effects. This ambiguity
may be the result of the assumption of
one-period price stickiness. I suspect that
allowing multi-period, overlapping price
contracts (thus adding more inertia to
inflation) would generate a finding that
monetary expansion unambiguously lowers
the nominal interest rate in their model.
Stockman and Ohanian also discuss
limited-participation models as an alternative
theory of the liquidity effect. I find much
interest in this approach also. In particular,
it is quite realistic to assume that, in the
short run, Federal Reserve purchases and
sales of securities are absorbed by a relatively
small number of Treasury dealers and other
financial market participants. My main objec
tion to existing limited-participation models
is that they combine the limited-participation
assumption with the “wrong” friction, that
is, most of these models are closed by a
structure that imposes a cash-in-advance
constraint on consumers and firms. Not
only is the cash-in-advance constraint not
particularly plausible economically, but models
that assume this constraint have great
difficulty generating persistent effects of
monetary policy changes.
I think a more promising approach
would be to combine the limited-participation
assumption with the assumption of sticky
prices. Allan Meltzer’s paper gives a spirited
defense of the price-stickiness assumption
based on the notions of pervasive economic
uncertainty and the difficulty in distinguishing
between permanent and transitory shocks.
More formally, recent work by Lucas and
Woodford (1 9 9 4 ) shows how price stickiness
and monetary non-neutrality can be an equi
librium outcome in a non-Walrasian
setting with sequential service of customers.
Allegorically, one may illustrate the
Lucas and Woodford model by thinking of
the owner of a general store in a gold-mining
town, who must set prices without knowing
how much gold will be discovered in the
surrounding hills that day. Although the
general-store owner is free to raise prices
m
MAY/JUNI
during the day if business is brisk, his inability
to re-contract with earlier customers guaran
tees that unexpectedly high gold discoveries
(positive monetary shocks) will be reflected
in higher economic activity.
The second broad issue considered at
this conference concerns the channels by
w hich monetary policy has its effects on the
economy. A common comparison is between
the “money view” and the “credit view” of
monetary transmission. Unfortunately, this
terminology has created a great deal of con
fusion (in particular, what some have called
the money view does little justice to the
views of people like Milton Friedman,
Karl Brunner and Allan Meltzer), and it
should be abandoned. A better distinction is
between the view represented by the standard
textbook IS-LM model and what might be
termed the capital-market-imperfections
approach. The capital-market-imperfections
approach is based on the premise that the
same informational and agency problems
that explain many aspects of financial struc
ture (for example, the existence of financial
intermediaries) also play a role in monetary
transmission. A notable difference between
the two approaches is that the IS-LM model
assumes the existence of only two assets
(money and “bonds”), while models based
on capital market imperfections generally
require a richer menu of assets.
As ably discussed in the papers by
Glenn Hubbard and Steve Cecchetti, the
capital market-imperfections approach
suggests two new channels of influence for
monetary policy, above and beyond the
standard IS-LM-type effects. The first of
these may be referred to as the balance sheet
or net-worth channel: Here, the idea is that
increases in interest rates weaken the financial
conditions of consumers and firms, making
it more difficult or costly for them to obtain
credit. More formally, reductions in borrower
net worth associated with a rise in interest
rates increase the agency and information
costs o f making loans; see Bernanke,
Gertier and Gilchrist (1994) for more discus
sion. For example, increased interest rates
worsen the cash flows of indebted firms (if
their debt is short-term or floating-rate) and
reduce the capital values of assets (such as
199 5
land) that are commonly used as collateral
for loans. Reduced access to credit may
lower both aggregate demand (because of
declines in purchases of capital goods, con
sumer durables, and so on) and aggregate
supply (because of reductions in working
capital). As was discussed at this confer
ence, there is a good deal of evidence for the
balance sheet channel. In particular, it seems
clear that monetary policy differentially
affects agents who are more subject to
agency and informational problems in credit
markets, such as small firms and potential
homebuyers.
The second channel suggested by
the capital-market-imperfections approach
may be referred to as the bank lending
channel (Bernanke and Blinder, 1988).
Briefly, put, the premise here is that a
reduction in bank reserves by the Fed also
reduces bank deposits and, hence, banks’
loanable funds. To the extent that bank
loans are imperfect substitutes for other
forms of short-term credit (which seems
incontrovertible), a reduced supply of
bank loans will lower econom ic activity
by bank-dependent borrowers.
Critics have noted that institutional
changes and financial innovation have likely
weakened the bank lending channel, if it
ever existed (Romer and Romer, 1990;
Thornton, 1994). Their strongest point is
that, under current arrangements, banks
need not rely on core deposits for funds.
Large banks, at least, are able to raise funds
by issuing certificates of deposits (CD ),
against which no reserve requirements are
imposed. The response to this point is
that the bank lending channel survives
(at least in theory) as long as the demand by
investors for bank CDs is not infinitely elas
tic: If demand is not perfectly elastic, that is,
larger issuances of CDs require banks to pay
higher rates, then the level of core deposits
will be relevant to banks’ willingness to
supply loans. Indeed, the spread between
CD rates and Treasury bill rates does
increase, sometimes spectacularly, during
periods of tight money.
There have been a number of interesting
attempts to test for a bank lending channel,
as Hubbard and Cecchetti describe. W hile
129
REVIEW
MAY/JUNE
_ _ _ _ _ _ , Mark Gertler and Simon Gilchrist. "The Financial
Accelerator and the Flight to Quality," Notional Bureau of Economic
Research Working Paper No. 4789 (July 1994).
the evidence does not contradict the existence
of this channel, a generic difficulty is that
most tests of the bank lending channel do
not cleanly distinguish it from the balance
sheet channel. For example, Kashyap and
Stein (1994) find that small banks reduce
lending following a monetary tightening
more than large banks do. Since small banks
have less access to the CD market, this finding
is consistent with the view that a drain of
reserves forces a reduced supply of loans by
small banks. Unfortunately, since a larger share
of small bank loans goes to small borrowers,
this result might also be explained by the
differential effect of monetary tightening on
small firms’ balance sheets, which dispropor
tionately reduce the effective credit demand
by those firms. Matched bank-borrower data
will probably be needed to resolve this issue.
Despite the difficulties, there are several
reasons to continue to do empirical work on
the links between credit market imperfections
and monetary policy. First, as was discussed
at this conference, there are serious quantita
tive problems with the IS-LM approach and
other leading models of the transmission
process; channels based on credit market
imperfections may be necessary to explain
the apparent strength and persistence of
monetary policy effects on the economy.
Second, making monetary policy in an
environment of ongoing institutional change
and financial innovation requires a sophisti
cated appreciation of how those changes
affect the potency of policy and the interpre
tation of policy indicators. Models based on
credit market imperfections, because they
analyze monetary transmission using the
same information-based theories that underlie
our understanding of financial structure and
function, are best placed to help us attain
that appreciation.
Christiano, Lawrence J., Martin Eichenbaum and Charles Evans. "The
Effects of Monetary Policy Shocks: Evidence from the Flow of Funds,"
working poper (March 1994), Northwestern University.
Kashyap, Anil K., and Jeremy C. Stein. "The Impact of Monetary Policy
on Bank Balance Sheets," working paper (March 1994), University of
Chicago.
Lucas, Robert E., Jr., ond Michael Woodford. "Real Effects of Monetary
Shocks in an Economy with Sequential Purchases," working paper
(April 1994), University of Chicago.
Romer, Christina D., and David H. Romer. "New Evidence on the
Monetary Transmission Mechanism," Brookings Papers on Economic
Activity (1990:1), pp. 149-213.
Sims, Christopher A. "Interpreting the Macroeconomic Time Series
Facts: The Effects of Monetary Policy," European Economic Review
(June 1992), pp. 975-1011.
Strongin, Steven. "The Identification of Monetary Policy Disturbances:
Explaining the Liquidity Puzzle," working paper (November 1992),
Federal Reserve Bank of Chicogo.
Thornton, Doniel L. "Financial Innovation, Deregulation and the 'Credit
View' of Monetary Policy," this Review (January/February 1994),
pp. 31-49.
REFERENCES
Bernanke, Ben S., and Alan S. Blinder. "The Federal Funds Rate and the
Channels of Monetary Transmission," The American Economic Review
(September 1992), pp. 901-21.
_ _ _ _ _ _ and_ _ _ _ _ _ . "Credit, Money, and Aggregate Demand,"
The American Economic Review (May 1988), pp. 435-9.
199
130
REVIEW
MAY/JUNE
1995
Thomas F. Cooley is Fred H. Gowen professor of economics at the University of Rochester.
Thomas F. Cooley
the environments that Glenn Hubbard
discussed, why is that important? One obvi
ous answer is that there could be important
growth or welfare consequences of these
policy shifts, even though they may have
little consequence for output at the business
cycle frequency.
The traditional view is that monetary
policy does have important effects on real
econom ic activity at the business cycle
frequency. Certainly the recent actions of
the Federal Reserve suggest that the current
interest rate smoothing policy is predicated
on the belief that the Fed can moderate the
growth of real output. The theoretical
evidence for this is somewhat weak and the
empirical evidence is extremely fragile.
There is also theoretical evidence that
monetary policy and the nature of financial
institutions are important for economic
growth but, again, the empirical evidence is
thin. But these are the reasons why monetary
econom ics is so appealing: We believe
monetary policy is important but the evidence
is elusive. For that reason it is important
that we consider evidence from a variety
of sources.
ow do changes in monetary policy get
transmitted to the real economy? The
papers presented at this conference have
been sharply focused on this question and
on three candidate answers. New research
was presented on the liquidity effects chan
nel. There was abundant discussion of the
credit channel and several summaries of
research on the sticky-price channel. The
only transmission mechanism not discussed
in these papers is the m ost venerable one:
rigid wages.
The discussion has been focused with
almost surgical precision on the circuitry of
monetary policy— how actions of the Federal
Reserve affect the behavior of banks, firms
and consumers. Taken almost for granted in
this discussion— I assume— is the view that
shifts in monetary policy have important
consequences for the real economy. Steve
Cecchetti summarized some o f the recent
empirical research on the output effects of
monetary policy. But, in general, the papers
do not address very explicitly the sense in
which monetary policy is important. Are
there important growth effects associated
with monetary policy? Are there important
distributional consequences o f monetary
policy? Are there significant output effects
at business cycle frequencies? These are
quite distinct questions and all of them are
important. Unfortunately, the papers
presented here are unnecessarily vague
about these bottom line issues.
If there is a liquidity effect in the sense
that monetary expansions cause nominal
interest rates to rise, but output is left
unchanged over a horizon of two quarters or
more as in the model economies studied by
Ohanian and Stockman— aside from the
descriptive value of understanding these
liquidity effects, why should we care?
Correspondingly, if some investment projects
are not undertaken as a consequence of a
shift in monetary policy, as occurs in some of
H
M ON EY AND
THE BUSINESS CYCLE
First, I want to discuss very briefly the
empirical evidence on the role of money in
business cycles and the efficacy of monetary
policy. The empirical evidence based on
VARs or structural VARs is well-known and
known to be very sensitive to the set of con
ditioning variables, the sample period used,
and the identification restrictions imposed.
Pure reduced-form estimates which treat
money as exogenous are meaningless.
Structural VARs based on ju st identifying
restrictions seem to be consistent with the
proposition that money is neutral in the long
run, but has a short-run effect on output.
This evidence too is fragile (Cooley, 1994).
More recently, economists have shifted to
studying specific monetary episodes rather
131
REVIEW
MAY/JUNE
than time-series models to identify more
clearly when monetary policy shifts are taking
place. Friedman and Schwartz (1 9 6 3 ) have
become the darlings of the new Keynesians
because of their documentation of specific
historical episodes when deliberate monetary
actions were followed by declines in real
econom ic activity. Romer and Romer (1989)
follow a similar methodology to identify
monetary disturbances in the post-war
United States. They associate these episodes
with recessions. But this evidence, like the
Friedman and Schwartz evidence, is far from
clear. Steve Cecchetti discussed some of the
objections to their analysis in his article. There
are other objections as well, some of them
touched on by Kevin Hoover. Many of the
episodes identified by the Romers are also
associated with changes in reserve require
ments, tax reforms, and other things that are
at least arguably regarded as real shocks.
Moreover, Hoover and Perez (1994a, b) have
shown that the evidence of the Romers does
not sustain the causal interpretation given to
it and that the methodology cannot distinguish
monetary shocks and oil shocks as a cause
of recessions.
W henever one raises these qualms about
the evidence on monetary policy, advocates
of the monetary view resort to their ultimate
weapon— the Volcker recession.
199
shocks explain the 1990-91 recession?”
To address this question, I use a model
similar to the basic real business cycle (RBC)
model but modified to take account of some
important features of the post-war U.S.
economy. The most important modification
is that there are three sectors producing
consumption goods, consumer durables and
producer durables. The technologies for
producing these goods include land explicitly
as a factor of production:
( 1)
C ^ z X l h i L 1-6'^
V
— 7
A dt ~ ^dt
7 If**'
^ t ^ 2 t n 2t
kt ~ ^ k t
3t
*
where K, h and L denote the stock of
capital, the hours and the stock of land
employed in each sector, respectively. The
variables Z dt and Z k are the investment-good,
sector technology shocks relative to the
consumption-goods sector technology shock.
Their inverses give the relative prices of
consumer durables and capital relative to
consumption. Specifying technology shocks
in this way makes it possible to capture the
fact that the relative prices of consumer and
producer durables have declined over the
post-war period. The processes for the Z’s are:
Z t = A z , , Z j, =
A Real View of the Volcker
Recession
z d t, Z ht = X k z kt
I°g2,+1 = ( l - p ) l o g z + plogZ( + £1+1
l ° g z dt+l = Pi log£di + £di+l
The Volcker recession seems to be
regarded as the incontrovertible evidence
that monetary policy— in this case, the
Volcker disinflation— can cause a decline in
real econom ic activity. The case seems pretty
strong. Paul Volcker announced his intention
to squeeze inflationary expectations out of
the economy and the FOMC acted to tighten
monetary policy in a decisive way. This
episode is a serious challenge for those who
view real shocks as the most powerful
driving forces of business cycles.
Could technology shocks also explain
the Volcker recession? To answer this
question, I conducted an exercise similar to
that reported by Hansen and Prescott (1 9 9 3 ),
who asked the question, “Can technology
3t n 3t
2t
logZfe+i = Pk l ° g z fet + £ bt+i.
The economy is populated by a continuum
of identical households of measure N that
grows at the rate 17-I. Households have
utility given by
u (C ,, D ,, h , ) = a log C ( + ( l - a ) lo g D ( - Ah„
where D represents the service flow provided
by the stock of durables and the linear term
in hours results from assuming that labor is
indivisible, as in Rogerson (1 988) and
Hansen (1 9 8 5 ). The rest of the details of the
model economy are exactly as in Hansen and
132
MAT/JUNE
Prescott (1993) so I won’t repeat them here.
The important features of the calibration
follow the procedures outlined in Cooley
and Prescott (1 9 9 5 ), except that for this
exercise we choose the parameters so that
the steady state for the model matches the
data for the first quarter of 1987. We then
construct the sequence of technology
shocks, Zt. These shocks are then fed
into the model to generate a sequence for
consumption, investment, productivity,
hours and output for the actual economy.
The results of this exercise are shown in the
next two figures.
Figure 1 shows the path of real GNP as
predicted by the basic real business cycle
model and as it is in the data. The vertical
line is approximately the trough o f the
Volcker recession. Figure 2 shows the path
of hours worked as predicted by the model
and as in the data. Hours in the model are
much less smooth than in the data because
the indivisible labor assumption causes them
to respond sharply to the technology shocks.
The behavior of the other variables is much
the same; the model tracks actual values
quite closely.
As the figures show, the basic real busi
ness cycle model can account quite well for
the Volcker recession without recourse to
a monetary mechanism. W hat are we to
conclude from this? One might assert that
this exercise reveals that the RBC modeling
strategy is completely vacuous: The identifi
cation of technology shocks is so imprecise
that monetary shocks— along with any other
econom ic variables legitimately affecting
output— are included in the estimated
technology shock series. There are several
reasons why I think such a conclusion would
be wrong. First, we know that there were
important real shocks occurring over this
period. There were oil price increases in
1979 and 1981 and changes in reserve
requirements in 1979 and 1980. There
were also some credit controls imposed in
1979-80. These are all the kinds of events
that would legitimately show up as technology
shocks because they change the productivity
of existing inputs. Second, the tax treatment
of capital changed fairly dramatically during
this period. The Econom ic Recovery Act
199 5
F ig u re 1
Gross N a tio n a l P ro duct: M o d e l a n d D a ta
197 9 :1
= 100
F ig u re 2
H ours W o rk e d : M o d e l a n d D ata
1 9 7 9 :1
= 100
of 1981 introduced major changes in the
econom ic life and cost recovery rules for
capital assets. The Act of 1982 reversed, at
least partially, many of those changes,
effectively increasing again the tax on capital
income. These were real shocks to the
economy that had a big affect on the invest
ment decisions of firms and, again, would
legitimately show up as technology shocks
in a highly aggregated model.
A better conclusion to draw from these
results is that models like this don’t go far
enough— they rely on a formulation of the
technology shock that is too abstract. Any
variable that helps to track output can get
rolled into the technology shock. To better
exploit their potential as analytical tools for
F ED ER A L RESERVE B A N K OF ST. L O U I S
133
REVIEW
MAY/JUNI
Signal Extraction
understanding the role of money in the
macroeconomy, we need to do two things.
First, if we want to understand a broader set
of observations than those captured by the
basic neoclassical growth model, then we
have to add more theory— theory that admits
the possibility that monetary shocks get
transmitted. Second, if we want to understand
the role of “shocks” in these models, we
need a more explicit account of what these
shocks are. Obviously, the nature of technol
ogy shocks is such that a lot of things can get
rolled into them. As noted above, one obvious
example is oil price shocks. A recent paper
by Finn (forthcoming) does an impressive
jo b of documenting how explicitly accounting
for oil price shocks and capacity utilization
improves the ability of models to match
features of the data and account for the
behavior of Solow residuals. W hat about
changes in reserve requirements, borrowing
constraints, the tax treatment of depreciation?
These also may be reflected in technology
shocks and the only way to try to sort out
their quantitative importance is to try to
construct economic environments that
explicitly account for them.
One of the advantages of using artificial
economies to study the role of monetary
shocks is that the questions addressed can be
made fairly precise. Thus, if we are interest
ed in studying the precise channels by which
changes in monetary policy affect the real
economy, then the challenge is to construct
plausible models that address this question.
For a long period, the main mechanism
that macroeconomic theory focused on as
the transmission mechanism for monetary
policy was signal extraction problems of the
sort made famous by Lucas. Kydland (1 989)
was the first to study signal extraction
problems in the context of an equilibrium
business cycle.1 Signal extraction problems
caused by monetary policy are proxied by
confronting agents with a signal extraction
problem. In these models, agents only
observe a noisy version of the shocks to
technology. This is intended to reflect the
signal extraction problem caused by imper
fectly observed monetary policy. Cooley and
Hansen (1 9 9 5 ) studied a similar model. The
conclusion of this work is that signal extraction
problems provide very little propagation of
monetary shocks. In fact, the addition of
“monetary noise” can actually reduce the
size of fluctuations in the economy.
Wage Rigidities
Cho (1 9 9 3 ) and Cho and Cooley
(forthcoming) study a standard real business
cycle model in which money is introduced
by a cash-in-advance constraint, and workers
and firms agree to some contracting rule
which specifies the nominal wage in advance.
Workers cede to firms the right to determine
the level of employment. In this setting,
monetary shocks do get propagated and the
most interesting finding is that it doesn’t take
a lot of rigidity for these shocks to have sub
stantial output effects. The major problem
with this account of how monetary shocks
have real effects is that the cross-correlations
in the data generated by these models are
inconsistent with the properties of U.S. data.
This suggests that money is not a primary
cause of output fluctuations.
CHANNELS OF MONETARY
POLICY
1 Kydland ond Prescott (1 9 8 2 ) built
this feature into their models but
didn't do much with it because it
didn't seem very important. Their
paper was actually written much
earlier in the 1980s. It pioneered
the analysis of monetary transmis
sion in real business cycle models.
It is easy to think of many objections
to the economic environment I used previ
ously to simulate the Volcker recession.
Nevertheless, the study of similar economic
environments that include money has yielded
some useful insights about the channels of
monetary policy. The evidence is far from
conclusive, but it probably compares favorably
with empirical evidence based on aggregate
data. Accordingly, it seems worthwhile to
review briefly some of the evidence from
artificial economies that have tried to incor
porate monetary transmission mechanisms.
1995
Sticky Prices
Cho (1 9 9 3 ), Cho and Cooley (forthcom
ing) and King (1 9 9 1 ) followed a similar
approach in studying sticky prices as a
propagation channel for monetary shocks.
Workers and firms agree in advance to fix
prices, firms agree to supply all that is demand-
134
m
m
MAY/JUNE
ed at that price. Money does have big output
effects in such economies as long as the
equilibrium quantities are determined by
the demand curve rather than the supply
curve. Only a very small amount of rigidity
is necessary for monetary shocks to have a
big output effect. Again, however, the
cross-correlations don’t match those observed
in U.S. data and this casts doubt on monetary
shocks as the mechanism that produces real
effects o f monetary shocks.
Ohanian, Stockman and Kilian (1994)
extend these sticky-price models in a useful
way. They consider a two-sector version in
which consumer goods prices are sticky for
one period but investment goods prices are
perfectly flexible. In this setting, they find
that monetary shocks have no big output
effects. This seems to cast further doubt on
this propagation mechanism. Monetary
shocks are more powerful in the multipleequilibrium setting of Beaudry and Devereux
(1 9 9 4 ). In their model, final goods are
produced under monopolistic competition
between firms using a technology that
exhibits increasing returns and requires mul
tiple intermediate goods as inputs. There is
a rudimentary intermediation sector in which
the Fed can manipulate total reserves. One
of the model’s equilibria in which prices are
fixed one period in advance, seems to match
the dynamic responses in the data very well.
However, the equilibrium selection story
seems very weak and the model requires an
implausibly high degree of increasing returns.
Pagan’s article and the ensuing discussion.
W hat seems clear is that the empirical
evidence for a liquidity effect is pretty strong
but the magnitude of it is probably very
small. More importantly, the empirical
evidence has been focused almost exclusive
ly on the liquidity effect itself without much
discussion of the corresponding output
effects. The quantitative evidence from
studying artificial economies in which this
channel is present shows that monetary
shocks will have small but significant effects
on real output.
Endogenous Monetary Policy
One of the biggest problems that plagues
empirical researchers is the issue o f defining
exactly what monetary policy shifts are and
the extent to which changes in monetary
policy can be treated as exogenous. Artificial
economies also help to understand this issue.
Coleman (1 994) studies an artificial economy
in which monetary policy is endogenous.
The monetary authority chooses a supply of
currency to meet inflation and nominal
interest rate targets. Banks provide checkable
deposits in the amounts desired by house
holds, given the supply of currency. In this
environment, the Federal Reserve can raise
interest rates in response to changes in output.
Coleman then estimates the parameters of
his model economy to determine how the
Fed responds. The estimated parameter values
imply that a substantial portion of the condi
tional and unconditional variance of nominal
interest rates is endogenous, but not all of it
is. Coleman analyzes the implications of his
estimated parameters for the cross-correlations
between money and output. He finds that
endogenous money creation causes money
growth to be more strongly correlated with
current and past output than with future
output growth. In U.S. data, it is more
strongly correlated with future output growth.
This at least suggests that endogenous money
creation cannot by itself explain the observed
correlations between money and output.
Limited-Participation Models
These models have been developed and
exploited by Lucas (1 9 9 0 ), Fuerst (1992)
and Christiano and Eichenbaum (1992).
Christiano and Eichenbaum have done the
most in exploiting the quantitative implications
of these models for output. In these models,
the financial arrangements break the temporal
link between the consumption decisions of
households and monetary injections. This
generates a transient liquidity effect in which
both real and nominal interest rates change.
The controversies regarding the existence
and size of this liquidity effect have been
pretty thoroughly represented in Adrian
1995
The Credit Channel
The research on the credit channel for
monetary policy was well-summarized in the
135
hivhw
MAY/JUNE
papers by Hubbard and Cecchetti. In a typical
credit channel model, informational asym
metries between borrowers and lenders in
credit markets imply that loan contracts are
constrained by moral hazard or adverse
selection. Entrepreneurs who must borrow
in credit markets to finance new investments
pay a higher price for borrowed funds or
may be denied credit altogether. These
asymmetries vary in degree over the business
cycle: In periods of expansion, the information
asymmetries are mitigated. A key feature of
the credit view is that most of the empirical
implications are cross-sectional and the map
ping between time-series and cross-sectional
evidence is not obvious or direct. There
have been only a few attempts to incorporate
this view into artificial economies.
Fisher (1994) constructs a model econo
my in which there is costly state verification
by lenders. He finds that this does lead to
asymmetric response of firms to monetary
shocks, but he also finds that the quantitative
impact of monetary shocks on output are
quite small. Fuerst (1994) studied the prop
erty of a model economy which incorporates
some elements of the credit channel view.
He finds that adding these features to the
basic real business cycle model adds little or
nothing to the basic, real business cycle
propagation mechanism.
I don’t think this is the final story
because there are aspects of the credit channel
view that Fuerst’s model may miss. As the
papers and discussions made clear, the
cross-sectional implications of credit issues
are manifest in the wealth of different agents
in the economy. It is difficult to capture this
heterogeneity in wealth and make it fit in
the context of a representative-agent type of
business cycle model. Introducing hetero
geneity in a serious way and keeping capital
accumulation in the model is at the frontier
of what we can analyze. The curse of
dimensionality restricts our ability to
analyze a heterogeneous-agent economy
with capital accumulation.
Furthermore, even if the output effects
at business cycle frequencies are small, previ
ous experience with heterogeneous-agent
models of money suggests that heterogeneity
and asymmetric information problems may
199
have very important welfare consequences.
Finally, a lot of the theoretical work referred
to stresses the im portant long-term growth
effects of financial intermediation and
borrowing constraints. In that respect, I
think a lot of the discussion has been
focused far too narrowly on the output
effects at the business cycle frequency
and not enough on these welfare and
growth implications.
REFERENCES
Beaudry, Paul, and Michael B. Devereux. "Monopolistic Competition,
Price Setting and the Effects of Real and Monetary Shocks," working
paper (1994), Queens University.
Cho, Jang-Ok. "Money and the Business Cycle with One-Period Nominal
Contracts," Canadian Journal of Economics (August 1993),
pp. 638-59.
_ _ _ _ _ _ and Thomas F. Cooley, "The Business Cycle With Nominal
Contracts," Economic Theory (forthcoming).
Coleman, Wilbur John. "Money and Output: A Test of Reverse
Causation," working paper (1994), Duke University.
Cooley, Thomas F. "Data Analysis Without Much Theory: A Critique of
Structural VARs," working paper (1994), University of Rochester.
_ _ _ _ _ _ and Gary D. Hansen. "Money and The Business Cycle" in
Thomas F. Cooley, ed., Frontiers of Business Cycle Research. Princeton
University Press, 1995, pp. 175-216.
_ _ _ _ _ _ and_ _ _ _ _ _ _ . "Tax Distortions in a Neoclassical
Monetary Economy," Journal of Economic Theory (December 1992),
pp. 290-316.
_ _ _ _ _ _ and Edward C. Prescott. "Economic Growth and Business
Cycles", in Thomas F. Cooley, ed., Fmiiers of Business Cycle
Research. Princeton University Press, 1995, pp. 1-38.
Christiano, Lawrence J., and Martin Eichenbaum. "Liquidity Effects,
Monetary Policy and the Business Cycle," National Bureau of
Economic Research Working Paper No. 4129 (August 1994).
Finn, Mary. "Variance Properties of Solow's Productivity Residual and
Their Cyclical Implications," Journal of Economic Dynamics and
Control (forthcoming).
Fisher, Jonas D.M. "Credit Market Imperfections and the Heterogeneous
Response of Firms to Monetary Shocks," working paper (1994),
University of Western Ontario.
Friedman, Milton, and Anna J. Schwartz. A Monetary History of The
United States, 1867-1960. Princeton University Press, 1963.
Fuerst, Timothy S. "Monetary and Financial Interactions in the Business
Cycle," working paper (1994), Federal Reserve Bank of Cleveland.
_ _ _ _ _ _ . "Liquidity, Loanable Funds and Real Activity," Journal of
Monetary Economics (February 1992), pp. 3-24.
Fullerton, Don, and Roger H. Gordon. "A Reexamination of Tax
Distortions in General Equilibrium Models" in Martin Feldsfein, ed.,
Behavioral Simulation Methods in Tax Policy Analysis. University of
Chicago Press, 1983, pp. 369-426.
136
_ _ _ _ _ _ ond Mario Karayonnis. "United States," in Dole W.
Jorgenson and Ralph Landou, eds., Tax M o m and the Cost of
Capitaf. An International Comparison. Brookings Institution, 1993.
Gomme, Paul. "Money ond Growth Revisited: Measuring the Costs of
Inflation in an Endogenous Growth Model," Journal of Monetary
Economics (August 1993), pp. 51-77.
Hansen, Gory D. "Indivisible Labor and The Business Cycle," Journal of
Monetary Economics (November 1985), pp. 309-25.
_ _ _ _ _ _ and Edward C. Prescott. "Did Technology Shocks Cause the
1990-1991 Recession?" The American Economic Review (May
1993), pp. 280-6.
Hoover, Kevin D., ond Stephen J. Perez. "Post Hoc Ergo Propter Once
More: An Evaluation of 'Does Monetary Policy Matter?' in the Spirit
of James Tobin," Journal of Monetary Economics (August 1994a),
pp. 47-73.
_ _ _ _ _ _ and_ _ _ _ _ _ _ . "Money May Matter But How Could You
Know?" Journal of Monetary Economics (August 1994b), pp. 89-99.
King, Robert E. "Money ond Business Cycles," working paper (1991),
University of Rochester.
Kydland, Finn E. "The Role of Money in a Competitive Model of
Business Fluctuations," working paper (1989), The Institute for
Empirical Macroeconomics.
_ _ _ _ _ _ and Edward C. Prescott. "Time To Build and Aggregate
Fluctuations," Econometrica (November 1982), pp. 1345-70.
Lucas, Robert E., Jr. "Lotteries ond Interest Rates," Journal of Economic
Theory (April 1990), pp. 237-64.
Ohanian, Lee E., Alan Stockman ond Lutz Kilion. "The Effects of Reol
and Monetory Shocks in Business Cycle Model With Some Sticky
Prices," working paper (1994).
Rogerson, Richard. "Indivisible Labor, Lotteries and Equilibrium," Journal
of Monetary Economics (January 1988), pp. 3-16.
Romer, Christina D., and David H. Romer. "Does Monetary Policy
Matter? A New Test in the Spirit of Friedmon and Schwartz,"
Notional Bureou of Economic Research Macroeconomics Annual.
MIT Press, pp. 121-70.
REVI EW
MAY/JUNE
1995
Manfred J.M. Neumann is a professor of economics at the University of Bonn.
Manfred J.M . Neumann
Keynesian transmission channel of a
single interest rate led neo-Keynesians (for
example, Tobin, 1961) as well as monetarists
(for example, Brunner, 1961) to adopt a
broader view.
The theory of relative prices provides
an encompassing view of the transmission
mechanism. It assumes that all assets, finan
cial and real, are imperfect substitutes. This
implies that a change in the stock supply of
base money or government debt affects all
relative prices and sets in motion a process
of portfolio adjustment that extends to the
full array of financial and real assets. The
speeds o f adjustment may differ between
markets due to differential adjustment costs.
As Brunner (1970, 1971) pointed out, the
degrees of imperfect substitutability are
shaped by differences in the levels of transac
tions costs and marginal information cost.
These costs are generally low for money and
securities, but much higher for loans and
non-securitized real assets. Brunner conjec
tured that the relative magnitude of these
costs changes with the level o f interest rates.
This led him to assume that securities are
close substitutes to money when interest
rates are low, but closer substitutes to real
capital than to money when interest rates
are high.
Although macroeconomic analysis can
only deal with a few, highly aggregated asset
markets, there is no compelling reason for
ignoring intermediate assets by restricting
the analysis to the components of private net
wealth. In fact, both Tobin (1 961) and
Brunner (1 9 6 1 ) already considered private
debt, the difference being that private debt
in Tobin’s pure-asset model has no particular
role to play, while the bank credit market in
Brunner’s analysis is a cornerstone of mone
tary transmission to aggregate demand.
Against the background of the general
transmission theory of relative prices, any
specific view or model of the transmission
mechanism rests on simplifying assumptions
that permit aggregating assets into a small
number of representative assets. Different
he recent undertaking by the proponents
of the credit view of broadening the sim
ple Keynesian transmission mechanism
is to be welcomed. Adding the credit chan
nel to the traditional money channel permits
studying the effects of monetary policy on
the process of intermediation and provides a
richer description of the transmission of pol
icy actions to the real economy. Studying the
interdependence of various types of credit
markets appears to rank highly on the
research agenda of this new literature.
Although this is of interest in itself, from a
macroeconomic point of view, it remains an
unsettled issue how far to go in disaggrega
tion, hence, differentiation of financial
assets.
I divide my contribution to this panel
into two parts. First, I will examine the
aggregative structure of the new credit view
and compare it with other theories of trans
mission, notably the monetarist analysis.
The conclusion will be that the latter theory
is the more comprehensive one and permits
studying the issues that are on the research
agenda of the credit view. Thereafter, I will
discuss whether this new view of the trans
mission mechanism has any novel implica
tions for monetary policy. I believe that this
is not the case.
T
THE CREDIT VIEW
IN COMPARISON TO
OTHER THEORIES OF
TRANSMISSION
To understand the contribution of this
recent literature to our knowledge about the
channels of monetary transmission, I believe
that it is useful to put the recent credit vs.
money debate into the broader perspective of
the transmission theory of relative prices.
This theory dates from the early 1960s, when
a growing dissatisfaction with the narrow
138
REVIEW
MAY/JUNE
aggregative structures yield different visions
of the way in which the economy works
(Leijonhufvud, 1968). I will compare three
specific views of the transmission process:
the traditional money view; the new credit
view; and the monetarist view.
The money view was introduced by
Keynes in his G eneral Theory. This view
aggregates all assets into two categories:
money and non-money. The non-money
asset represents all other financial assets as
well as existing capital goods. The distinction
between financial and physical non-money
assets is eliminated by the straightforward
assumption of perfect substitutability. For
Keynes the non-money asset was long-term
in nature, while Keynesians became used
to equating the non-money asset with a
short-term bond within the IS-LM framework.
The Keynesian IS-LM model provides
the most restrictive analysis of monetary pol
icy transmission. Due to the assumption
that non-money assets are perfect substi
tutes, monetary policy is transmitted to
aggregate demand through a single interest
rate, the bond rate, and the efficacy of policy
actions depends solely on the interest elastic
ity of money demand. The classroom inter
pretation of the result is: A reduction in the
money stock raises the “cost of borrowing,”
which reduces investment demand by elimi
nating marginal projects. However, taken lit
erally, the model does not contain a banking
sector— hence, there are no bank loans and
the money variable neither represents M l
nor M2, but ju st currency. The narrowness
of the setup is rightly criticized by the
proponents of the new credit view as it was
before by monetarists during the debate of
the late 1960s on whether money matters.
The credit view adds the credit
channel to the Keynesian money channel
by introducing bank loans as a third
(intermediate) asset. In Table 1 , 1 take the
model by Bernanke and Blinder (1988) as
representative of this view and compare it
with the monetarist view as presented by
the Brunner and Meltzer (1972, 1976)
model. To be sure, the monetarist view of
transmission is not to be equated with the
money view, contrary to Gertier and
Gilchrist (1 9 9 3 ), because the monetarist
1995
T a b le 1
A lte rn a tive V iew s of Transmission
Peer Group
Variable
IS-LM
Credit
Monetarist
Markets
Money
Government bonds
X
X
X
X
X
X
Bank loans
X
X
Real assets
Prites
Bond rate (/s)
X
X
X
Bank loan rate (/t)
X
Asset price ( f 4)
X
Real wealth (w)
Financial
X
X
Real assets
X
Aggregate demand
d(is)
d(iB, i L)
~
d(iCf PA'
-
^
+ +
Note: The credit view is based on Bernanke-Blinder (1988) and the monetarist view
based on Brunner-Meltzer (19 7 2 ,1 9 76 )
model also contains the bank credit market.
The credit view concentrates on the
substitution relations between money, bonds
and bank loans. Accordingly, the real loan
rate, iL, supplements the bond rate, iB, as a
determinant o f aggregate demand. W ith the
additional credit channel, the transmission
of monetary policy no longer depends on the
interest responsiveness of money demand
alone. This is an improvement over the
money view. But note that the credit view is
silent on the role of existing real capital.
Apparently, the implicit assumption is that
the relevant transactions costs are infinite.
The monetarist analysis, in contrast,
lumps together government bonds and bank
loans and extends the range of substitution
to the existing stock of real assets (equity,
real estate, and so on). The asset price level,
PA, enters the aggregate demand function
directly, reflecting the substitution between
existing and new capital goods, and indirectly
as a determinant of real wealth, w.
W hat are the implications of the credit
and the monetarist models regarding the
139
REVIEW
MAT/JUNE
transmission of shocks to aggregate demand?
In both models, monetary policy shocks and
money demand shocks affect the money
stock, the stock of bank loans and aggregate
demand in a comparable fashion regarding
the signs of first derivatives. However, the
early monetarist model of Brunner and
Meltzer implies much stronger effects than
the new credit model of Bernanke and
Blinder, because the former permits substitu
tion over the full array of financial and real
assets. The asset price level is a particularly
important transmission variable. Leaving it
out of the picture is leaving out Hamlet.
Changes in this price affect investment
demand by changing the relative price of
new capital goods (Tobin’s q), and they affect
the net worth of firms and households—
hence, creditworthiness and investment
demand as well as consumption demand.
Qualitative differences between the two
views arise when we study shocks to loan
demand (see Table 2). The credit view
attaches importance to such shocks,
although the origins of such shocks need
clarification. Let us assume these shocks
reflect productivity shocks. Both models
imply that an exogenous shock to the
demand for bank loans raises the loan
rate and the stock of loans. However, the
credit view predicts a contraction of the
money stock, while the monetarist view
predicts a rise. More importantly, since the
credit view assumes that the impact of the
loan rate on aggregate demand dominates
the impact of the bond rate, this view
predicts a fall in real income. The mone
tarist view, in contrast, derives the opposite
conclusion. A loan-demand shock effects an
increase in real income, because it induces a
rise in the asset price level, which dominates
the contractionary effect on aggregate
demand of the simultaneous increase in the
loan rate. In the following section, we will
check whether this conflicting result has any
policy implication.
Before I turn to this, however, let me
briefly point out two aspects of the credit
versus money debate which I find puzzling.
To begin with, I do not see why there is
a need to search for evidence in support
of the existence of the credit channel
199 5
(Bernanke, 1993). Since neither the
existence of the credit market nor the
existence of differences between financial
assets regarding transactions and information
costs can be disputed, so cannot the
existence of the credit channel. Once this is
acknowledged, the effort put in testing for
existence or relative importance of this
channel is surprising.
Next, the evidence collected by the
credit literature (for example, Bernanke
and Blinder, 1992) on timing relationships
between changes in monetary policy, banks’
securities holdings and bank loans confirms
the important role of differential information
cost, and it may be noted that the principal
pattern of adjustment— first securities, then
loans— was predicted by Brunner (1 970) as
an implication of his theory of the relative
price process. Banks hold stocks of informa
tion about customers and, hence, are reluctant
to respond to monetary tightening by imme
diately cutting tailored loans instead of selling
standardized securities first. Moreover, when
finally forced to adjusting the loan portfolio,
they will prefer to lend less to borrowers
whose activities are less well-known or are
less diversified and, hence, more risky.
However, in contrast to the credit view,
the encompassing transmission mechanism
of relative prices implies that the observed
temporal pattern of adjustment is not exclu
sively determined by the banks’ behavior.
Instead, it is the result of the interaction of
loan supply and loan demand. Any monetary
policy change affects the asset price level,
which is a determinant of loan demand (as
well as of aggregate demand). A negative
policy shock, for example, reduces the asset
price level which, in turn, induces a rise in
loan demand. Given that monetary policy
shifts both curves, loan supply and loan
demand, I do not see what we can learn from
the attempt at identifying whether bank
balance sheet contractions are due to shifts
in supply or in demand, not to mention
the identification problem raised by
Cecchetti (1 9 9 5 ).
Summing up, I conclude that the
monetarist view provides a more comprehen
sive theory of transmission than the new
credit view. Moreover, I believe that this new
140
REVIEW
MAT/JUNE
literature would gain from accepting the
monetarist framework and from employing
the monetarist credit market theory of the
money supply (Brunner and Meltzer, 1966)
as a point of departure for the analysis
of the issues that are on the new view’s
research agenda.
1995
T a b le 2
Effects of a Lo an-D em and Shock
Bank loan rate
Bond rate
Credit View
Monetarist View
+
+
-
Asset price level
+
POLICY IMPLICATIONS
Stock of bonk loons
+
+
Does the credit view have any novel
implications for monetary policy making?
I believe not, at least not if one compares the
credit view to the monetarist analysis instead
of the standard textbook model.
Suppose, first, that the monetary author
ities follow the traditional monetarist advice
of concentrating on the objective of providing
stable money rather than trying to dampen
business fluctuations. The most extreme
proposal is to provide a permanent rate of
inflation of zero or some low level. In this
case, under either view of the transmission
mechanism, it is sufficient to estimate the
long-run money demand function and use it
for determining the target rate of money
growth. Though the market for bank loans
is an important channel of transmission,
this has no bearing on the question of which
particular monetary aggregate to chose for
targeting. Also, implementation procedures
are unaffected.
However, let us consider the issue of
dampening the impact of money demand
shocks and of loan demand shocks on
aggregate demand. Regarding the negative
impact of money demand shocks, all views
of the transmission mechanism imply that
stabilizing the money supply path makes
things worse. Bernanke and Blinder (1988)
find that stabilizing the path of bank loans
provides a superior alternative. However,
since the analysis by Poole (1 9 7 0 ), we
know that the ideal policy for this case is
stabilizing the interest rate. Above, we saw
that the competing views of the transmission
mechanism deliver contradictory predictions
regarding the impact of stochastic loan
demand shocks on aggregate demand.
Nevertheless, both views imply that stabiliz
ing the money supply path would be an
appropriate policy response. Ironically, the
Money stock
-
+
Aggregate demand
-
+
Note: The credit view is based on Bernanke-Blinder (1988) and the monetarist view
is bosed on Brunner-Meltzer (19 7 2 ,1 9 76 )
monetarist view permits stabilizing the
aggregate loan portfolio o f banks as an
alternative while the credit view does not.
This is not to say that I recommend
targeting a loan aggregate instead of the
money stock. To make this change, one
would need to know much more, notably
the source of shocks to the demand for bank
loans. Are they produced by productivity
shocks or do they reflect shifts from credit
markets outside the banking system into the
market for bank loans? In the latter case, it
would require integrating the outside credit
markets into the analysis to know what
would be the net effect on money stocks,
bank loan aggregates, interest rates and
aggregate demand. Apart from this, and
more generally, my reading of the empirical
literature is that the attempts at detecting
loan demand functions that are more stable
than money demand functions have
been unsuccessful.
As a final remark, the credit view
collects evidence on the unfavorable
cross-sectional results of monetary
tightening. Not unexpectedly, the smaller
and financially weaker firms are hit the
hardest. Due to the global nature of
monetary policy, the authorities can do
nothing to avoid this except, of course,
that the results provide backing for the
monetarist advice to be steady and to avoid,
in particular, unnecessarily large swings
in the creation of reserves or the
monetary base.
NK OF ST. L O U I S
141
REVIEW
MAY/JUNE
199
Gertler, Mark, and Simon Gilchrist. "The Role of Credit Market
Imperfections in the Monetary Transmission Mechanism: Arguments
and Evidence," Scandinavian Journal of Economics (No. 1, 1993)
pp. 43-64.
REFERENCES
Bernanke, Ben S. "Credit in the Macroeconomy," Federal Reserve
Bank of New York Quarterly Review (spring 1993), pp. 50-70.
_ _ _ _ _ _ and Alan S. Blinder "The Federal Funds Rate and the
Channels of Monetary Transmission," The American Economic Review
(September 1992), pp. 901-21.
Leijonhufvud, Axel. On Keynesian Economics and the Economics of
Keynes: A Study in Monetary Theory. Oxford University Press, 1968.
_ _ _ _ _ _ and_ _ _ _ _ _ _ . "Credit, Money, and Aggregate
Demand," The American Economic Review (May 1988), pp. 435-39.
Poole, William. "Optimal Choice of Monetary Policy Instruments in a
Simple Stochastic Macro Model," Quarterly Journal of Economics
(May 1970), pp. 197-216.
Brunner, Kart. "A Survey of Selected Issues in Monetary Theory,"
Schweizerische Zeitschrift fur Volkswirtschaft uni Statistik
(March 1971), pp. 1-146.
Tobin, James. "Money, Capital, and Other Stores of Value," The
American Economic Review (May 1961), pp. 26-37.
_ _ _ _ _ _ . "Eine Neuformulierung der Quantitatstheorie des Geldes Die Theorie der relativen Preise, des Geldes des Outputs und der
Beschiiftigung," Kredit und Kapital I (1970), pp. 1-30.
_ _ _ _ _ _ . "Some Major Problems in Monetary Theory," The
American Economic Review (May 1961), pp. 47-56.
_ _ _ _ _ _ and Allan H. Meltzer. "An Aggregative Theory for a Closed
Economy," in Jerome L. Stein, ed., Monetarism. North-Holland
Publishing Co., 1976, pp. 69-103.
_ _ _ _ _ _ and_ _ _ _ _ _ _ . "A Monetarist Framework for
Aggregative Analysis," in Proceedings of the First Konstanzer Seminar
on Monetary Theory and Monetary Policy (supplement to Kredit und
Kapital1). Duncker & Humblot, 1972, pp. 31-88.
_ _ _ _ _ _ and_ _ _ _ _ _ _ . "A Credit Market Theory of the Money
Supply and an Explanation of Two Puzzles in U.S. Monetary Policy," in
Tullio Bagiotti, ed., Essays in Honour of Marco Eanno, vol. 2. Padova,
1966, pp. 151-76.
Cecchetti, Stephen G. "Distinguishing Theories of the Monetary
Transmission Mechanism," this Review (Moy/June 1995),
pp. 83-100.
142
A ll nonproprietary and nonconfidential data and programs for the articles published
in Review are available to our readers and can be obtained from three sources:
l.F R E D (F ed eral R e serv e Economic
2 . R esearch D ep artm ent, Federal
D ata), an e lectron ic bu lletin board
serv ice. You can access FRED by
dialing (3 1 4 ) 6 2 1 -1 8 2 4 through a
modem-equipped personal computer.
For a free brochure on FRED, please
Reserve Bank o f St. Louis, Post Office
Box 4 42, St. Louis, MO 63166 -0 4 4 2 .
You can request data and programs on
either disk or hard copy. Please include
the author, title, issue date and page
call (3 1 4 ) 444-8 8 0 9 .
numbers with your request.
3 . In ter-u n iv ersity Consortium for
P olitical and Social R esearch (ICPSR)
Member institutions can request
data through the CDNet Order facility.
Nonmembers should write to: ICPSR,
Institute for Social Research, PO. Box
1248 Ann Arbor, M ichigan 4 8 1 0 6 , or
call (3 1 3 ) 763-5010.
Federal Reserve Bank of St. Louis
Post Office Box 442
St. Louis, Missouri 63166-0442
recycled paper
@FedFRASER