Interpreting The Macroeconomic Time Series Facts: The Effects of Monetary Policy

Download as pdf or txt
Download as pdf or txt
You are on page 1of 20

INTERPRETING THE MACROECONOMIC TIME SERIES FACTS:

THE EFFECTS OF MONETARY POLICY


By Christopher A. Sims
August 1991

ABSTRACT

Existing theory and evidence on the effects of monetary


policy are reviewed. Substantial room for disagreement
among economists remains. New evidence, based on multiva-
riate time series studies of several countries, is present-
ed. While certain patterns in the data consistent with
effective monetary policy are strikingly similar across
countries, others, particularly the tendency of interest
rate increases to predict high inflation, are harder to
reconcile with effective monetary policy.

----------------------------------------------------------------------------------------------------------

This paper was prepared for presentation at the International Seminar on Macroecono-
mics in Madrid in July of 1991, and discussion by participants there contributed to
its current form.
INTERPRETING THE MACROECONOMIC TIME SERIES FACTS:
THE EFFECTS OF MONETARY POLICY
By Christopher A. Sims
August 1991

I. The Range of our Ignorance


Though many macroeconomists would profess little uncertainty about it, the profession
as a whole has no clear answer to the question of the size and nature of the effects
of monetary policy on aggregate activity. Economists closely connected with policy
tend to view the monetary authority as capable of controlling nominal short-term
interest rates and thereby of strongly influencing the level of aggregate activity.
While there is formal statistical evidence and theoretical support for such a view,
many of these economists, in the United States at least, regard such support as
scarcely needed, given what seems to them the clear pattern of tightened monetary
policy preceding practically every postwar U.S. recession. On the other hand many of
the most active academic macroeconomic researchers are pursuing the agenda of the
real business cycle (RBC) school, which has often worked with models that include
only real variables. Where RBC school models have included nominal variables, it has
usually been to show that some of their correlations with real variables can be
reproduced in models where nominal aggregate demand management and monetary policy
have no important role.

In what follows we first discuss the existing state of support in studies of data for
these conflicting views. We then present some new reduced-form evidence on the
dynamic interactions of real and monetary variables, showing that certain interesting
characteristics of reduced form impulse-responses are robust both across countries
and across definitions and lists of included variables. In many respects these facts
merely reinforce what is known from previous studies, but there are a few patterns
that seem not to have been noted before. Finally we discuss the degree to which the
conflicting views on the effects of monetary policy are consistent with this quanti-
tative evidence. Our conclusion is that support in the data for both the RBC view
and the view that monetary policy is clearly important is weaker than proponents of
those views might think. We propose that further progress will depend on work with
models that include characteristics and mechanisms not now common in the field. The
models should specify endogenous monetary policy, confronting the identification
problems this raises directly. The should include an investment function rather than

1
just a demand for capital (i.e. adjustment costs or a relative price of capital).
And the should include nominal variables along with real variables, so that the facts
of smooth price and wage reactions to disturbances are confronted.

II. Existing Theory and Evidence


The theory supporting the view that monetary policy can strongly influence aggregate
activity is well worked out, by the standards of the macroeconomics of the 1950’s and
60’s. In that period, the differences between Keynesians and monetarists seemed to
many to be differences over the relative sizes of certain elasticities in a system
whose formal structure was more or less agreed-upon. This framework survives in most
intermediate macroeconomics textbooks, where IS and LM curves are derived and shifted
about to give students a vocabulary with which to discuss macroeconomic policy. In
this framework, which is essentially static, policy can increase the money stock,
thereby shifting the LM curve outward under the assumed constancy of prices. Assum-
ing the IS curve is downward-sloping and holds still under this policy action,
interest rates will drop and output will increase. Of course this is only the
"demand side" of the policy effect, and in a later chapter of the typical textbook
qualifications to this picture are introduced -- Can prices be assumed to remain
constant? What mechanism induces the increased labor input required to produce the
higher output? It is recognized that the answers to these questions could overturn
the standard ISLM conclusions, but such issues are treated as peripheral complica-
tions, rather than part of the central framework.

By modern standards, this theory is full of gaps. Of course it is a disequilibrium


theory, but it provides no explicit discussion of which conditions for individual
optimization are being relaxed. It uses an important idea of disequilibrium dynamics
-- that trade occurs at out-of-equilibrium prices and this has real effects -- but
does not have a plausible complete dynamic theory surrounding it. It does assume
that labor markets don’t clear, meaning that workers and/or firms are not equating
marginal utilities and/or marginal products to the wage. This is a plausible postu-
late. But it pays no attention to optimization conditions that are more likely to be
satisfied -- for example the connection of nominal interest rates to real rates and
expected inflation and the connection of investment to expected future marginal
products. If such conditions are simply tacked on to the ISLM framework, to form a
Keynesian model with rational expectations in asset markets, the conclusions about

2
the effects of monetary policy from the usual ISLM framework are not necessarily
preserved. Keynesians may object that rational expectations is an unreasonable
assumption, but that expectations are volatile and sensitive to the current environ-
ment seems indisputable -- even seems a Keynesian notion. That expectations may not
be rational only reinforces the point that recognition of their volatility and
endogeneity makes the conclusions from Keynesian ISLM reasoning suspect.

I lay out these objections to Keynesian and monetarist theory here not because I
think they are fatal. The basic idea of this type of theory is the following se-
quence of events. Because of the sluggishness of nominal prices and/or wages,
increases in nominal money stock or reductions of nominal interest rates can make the
real value of government liabilities held by the public exceed their equilibrium
level. In this situation output will rise as the public tries to spend their excess
assets, and there will be some accompanying price rise. Building a model that
behaves like this, concentrating disequilibrium in markets for labor and/or current
output, assuming rational behavior for investors and bondholders, is not trivial, but
it is possible. Here we simply note my conviction that there is nothing naive or
self-contradictory in supposing that such a model can be constructed, despite the
old-fashioned style of the much of the Keynesian and monetarist theoretical
literature.

And these old-fashioned theories remain influential, despite their lack of recent
polishing by academic macrotheorists, because they are indeed strongly supported by
historical evidence. There is a familiar set of facts. Monetary aggregates tend to
move in the same direction as aggregate economic activity, as has been repeatedly
documented. Simple co-movements could in principle easily be accounted for as
passive response of money demand to changes in the level of activity not generated by
monetary policy. Friedman and Schwartz [1963] and others therefore paid special
attention to the timing of movements in monetary aggregates and aggregate activity,
documenting a tendency for money in some sense to lead income. Tobin [1970] showed
that both the tendency to co-movement and the timing of movements of money and income
could roughly match observations in a model where monetary policy played no role in
generating aggregate fluctuations. Furthermore it was widely recognized that timing,
as measured by the location of peaks in cross-correlation functions or in relative

3
leads and lags of turning points, can shift drastically as series are differenced or
differentiated. That much monetarist work related changes in money stock to levels
of income, automatically inducing a phase shift in money via the difference operator,
weakened the timing evidence.

Economists have therefore tried to dig deeper, trying to isolate periods when mone-
tary policy variables moved for reasons that cannot be connected to any previous
developments in the private sector. Friedman and Schwartz [1963] tried to do this in
some passages, and more recently Romer and Romer [1989] have done the same thing more
systematically. If we can examine the aftermath of such periods, it is thought, we
will see the effects of policy unclouded by the effects of other disturbances that
might also shift policy. Such studies are attempts to solve an identification
problem informally, using the same intuition that leads to study of responses to
reduced form innovations in multivariate models.

Responses to innovations generate timing patterns that are at least partly immune to
the effects of differencing. The response of income to an innovation in money is
invariant to differencing of the money data. When the responses show the extreme
asymmetry of zero response of money to income innovations, this condition -- Granger
causal priority of money -- is invariant to differencing of either income or money.
It was surprising (to me at least) that money was Granger causally prior to nominal
income in U.S. postwar data through the 60’s [1971], and I recently verified [1989]
that that result still holds in using data through the 80’s.

The historical "event studies" using monetary aggregates and the study of impulse
responses lead to the same conclusion: that when monetary aggregates increase
unexpectedly, nominal income subsequently rises. Variation in income predicted by
monetary innovations constitutes a large part of total variation in income. The
variation in monetary aggregates predicted by income innovations is small. This is
completely consistent with the view that random disturbances in monetary policy
generate a large fraction of the observed business cycle, and it is difficult or
impossible to explain with most RBC models in the literature.

4
In larger multivariate time series models, including a nominal interest rate, money
stock innovations become smaller, as interest rate innovations predict a considerable
fraction of movement in the money stock. Further, the remaining money stock innova-
tions have less predictive power for income. These facts do not fit the most extreme
monetarist view -- that monetary aggregates alone are a complete measure of the
stance of monetary policy -- but raise no difficulty for a broader view that monetary
policy disturbances are important in generating aggregate fluctuations. Indeed this
view is strengthened by the pattern of responses to interest rate innovations --
there is a long, slow decline in output and a long, slightly less slow decline in
money stock. The part of output predictable from interest rate innovations is
substantial. Indeed this quantitative statistical evidence is a kind of summary of
the more anecdotal support cited in the opening paragraph of this paper -- sudden
rises in interest rates have regularly been followed by recession and drops in rates
by recovery.

The results summarized above have been found both by me and by others, for various
time periods, time units, definitions of variables, and countries. However some
researchers have criticized them as unstable, noting that they change in some res-
pects under particular choices of time period, time unit or variable list. Recently
Richard Todd [1991] has undertaken a systematic exploration of the sensitivity of the
results, for systems including interest rates, to these factors. While he does find
some sensitivity, the qualitative pattern of the results as described above is quite
robust. His paper includes a survey of the empirical literature in this area.

While there is intuitive appeal to these event-studies and innovation-response


studies, they do rest on implicit identifying assumptions. It is assumed that an
innovation in a "monetary policy variable" -- either interest rates or money stock --
is generated by monetary policy autonomously. Yet it is possible that there are
within-period feedbacks, so that unpredictable movements in monetary policy variables
are generated in part by disturbances originating elsewhere in the economy. In an
earlier paper [1985], I found that looser identifying assumptions leave the basic
interpretive picture the same, however: monetary policy disturbances are estimated
to have substantial and plausible effects on real output.

5
This picture of strong empirical support was weakened, however, by my paper [1989].
There I showed that a particular RBC-style model could mimic some of the reduced form
impulse response behavior observed in the data. The model differs from other RBC-
style models in the literature in that it has the monetary authority using the
interest rate as its instrument, raising it in response to rapid growth in money. It
also differs in having a two-good technology, so investment goods can become expen-
sive and interest rates low even when rental rates on installed capital are high.
Without this element it is difficult or impossible for a rise in demand to simultane-
ously increase employment, create expectations of price rises, and reduce interest
rates. The model includes a small real role for transactions balances in the techno-
logy, so monetary policy does have real effects, but they are small and not of the
form implied by monetarist or ISLM theory. Nonetheless it produces a strong Granger
causal ordering from money to nominal income, just as in actual data. Furthermore,
when interest rate data from the model are included in the reduced form, the interest
rate innovations predict declines in money stock and output very much like those
observed in reduced forms fit to actual data. The model is not a complete success.
It does not divide the impulse responses of nominal income into real and price
components the way impulse responses fit to actual data do. In particular, the price
responses are stronger and the real responses weaker in the model data than in actual
data. Nonetheless the occurrence of this strong predictive value for monetary policy
variables in a model where monetary policy is unimportant must weaken the credibility
of evidence based on event studies and impulse responses. Putting it briefly, this
model shows that, because interest rates and money are closely linked to investment
portfolio decisions, they tend to react quickly to new information, as other asset
market variables do. Money and interest rates have strong predictive value for
aggregate activity for the same reason that stock prices do. One can imagine, in
other words, that the historical pattern of monetary tightness preceding recessions
is misleading. High interest rates might "produce" contractions in activity the way
the cock’s crow produces the sunrise.

It goes without saying that the theoretical underpinnings of RBC models are, at least
in comparison with ISLM models, modern, complete, and fully dynamic. RBC models
(with the exception of my own venture into this area [1989]) have, however, not
confronted the documented impulse-response facts about interactions of monetary and

6
real variables. Given the profession’s long experience with the Keynesian-monetarist
debate, in which the limitations of evidence on correlations, timing, and (combining
these) cross-correlation functions came to be widely appreciated, it is surprising
that these tools are again receiving so much attention. It is not surprising that
policy-oriented economists, well-taught by the earlier debates how easy it is to
match a given pattern of correlations or timing with models that have widely diffe-
rent interpretations, do not regard this kind of exercise with RBC models as convinc-
ing. My own RBC-style model [1989] does confront the impulse-response evidence and
indeed partly succeeds in matching it. But the failure of that model to match
observed price behavior is crucial. An ISLM or monetarist framework would predict
strong real effects of monetary policy because of slow price adjustment. An RBC-
style model where monetary policy has little real effect, but prices counterfactually
respond strongly and quickly to disturbances of all kinds, is not a direct competitor
with ISLM or monetarist interpretations that can rationalize behavior of all of R, M,
P, and Y at once.

III. New Evidence


Vector autoregressive (VAR) systems were estimated for data from France, Germany,
Japan, the U.K., and the U.S. All data were taken from the IFS data base, and the
series numbers and detailed descriptions are given in the data appendix. The data
are monthly and include for each country a short interest rate R, a monetary aggre-
gate M (some variety of "M1"), a consumer price index P, an industrial production
index Y, an index of the foreign exchange value of domestic currency XR, and a
commodity price index PC. The commodity price index is the same across countries, not
converted to domestic currency. Production Y is seasonally adjusted in all coun-
tries, while M is adjusted in some countries but not others. A complete set of
seasonal dummy variables was included in each equation for each country in any case.
Each variable entered each equation with 14 lags. They began between 1957:1 and
1964:1 and ended in 1990 or 1991 in all countries but Germany. In Germany the sample
was truncated at the end of 1989 to avoid distortion from the effects of unification.
Exact sample periods are given in the titles of the charts. All variables but R
entered as logarithms, while R entered as a per cent.

7
Charts 1-5 show the estimated impulse responses for the 6-variable VAR estimates for
the five countries.1 Responses are shown over an expanse of 48 months. The
responses are for orthogonalized innovations, with the ordering as shown in the
charts -- that is, contemporaneous R innovations impact all other variables, while
contemporaneous Y innovations impact no other variables. However the correlations
among innovations are so low that the orthogonalization has little effect, as can be
seen from the fact that in the charts initial heights of off-diagonal responses are
seldom far from zero.2 The response graphs in a given row all have the same scale,
with the maximum and minimum heights shown on any graph in the row noted at the left.
The filled-in areas in the graphs in a given row provide a visual measure of the
relative contributions of innovations listed in the columns to explaining variance in
the variable listed at the left of the row.

-----------------------------------------------------------------------------------------------------------------------------

1The nature of VAR’s, their impulse responses, and orthogonalizations of them, is


described, e.g., in my paper [1980].
2However, these monthly sample periods are long enough to make quite a few of the
small correlations significant according to the asymptotic distribution theory.
Table 1 displays these correlations. Note that two standard errors from zero is
approximately .10 for the correlations shown.

8
Table 1
Correlation Matrices for Innovations
6 variable systems

R XR PC M P Y
France
1.000
-0.127 1.000
-0.089 0.172 1.000
-0.076 0.038 -0.022 1.000
0.078 -0.021 0.005 0.071 1.000
-0.007 -0.021 0.033 -0.041 0.081 1.000
Germany
1.000
-0.062 1.000
-0.035 0.159 1.000
-0.067 -0.061 0.008 1.000
0.092 -0.080 0.040 -0.030 1.000
-0.070 0.026 0.095 0.065 -0.158 1.000
Japan
1.000
0.002 1.000
-0.071 0.036 1.000
0.037 -0.108 0.004 1.000
0.092 -0.085 0.069 -0.056 1.000
0.095 0.034 0.019 -0.260 -0.078 1.000
U.K.
1.000
-0.269 1.000
0.022 0.183 1.000
-0.059 0.071 0.003 1.000
-0.014 0.048 -0.018 0.027 1.000
0.054 0.032 0.079 -0.074 -0.073 1.000
U.S.
1.000
0.174 1.000
-0.164 -0.191 1.000
-0.010 -0.053 0.117 1.000
-0.027 -0.029 0.080 0.053 1.000
0.136 0.116 0.029 0.020 0.019 1.000

9
Table 2
Correlation Matrices for Innovations
4-variable systems

R M P Y
France
1.000
-0.081 1.000
0.100 0.075 1.000
0.027 -0.107 0.068 1.000
Germany
1.000
-0.069 1.000
0.075 -0.039 1.000
-0.044 0.032 -0.161 1.000
Japan
1.000
0.004 1.000
0.089 -0.024 1.000
0.099 -0.238 -0.072 1.000
U.K.
1.000
-0.088 1.000
0.025 0.024 1.000
0.082 -0.061 -0.056 1.000
U.S.
1.000
-0.002 1.000
-0.007 0.057 1.000
0.175 0.022 0.046 1.000

Note that the persistent negative response of money stock and output to positive
innovations in interest rates -- the fourth and sixth rows of the first column on
each chart -- is consistent across all five countries. The size of these responses
varies, but not their general form. (The response of M to R in the UK, which appears
small, is actually larger than in the other countries except Germany, because of
differences in scale of the graphs across countries.) Their form fits the moneta-
rist/ISLM explanation: interest rate surprises represent monetary policy shocks, and
monetary contraction generates declining M and Y. The innovation accounting is less
robust across countries than these forms of response. That is, looking across the
bottom row of each chart, the interest rate response is not a uniquely dominant off-
diagonal source of variation in output in any country, and it ranges from being the
most important one in Germany and the U.S. to being fourth-ranked in Japan. Of
course, if these interest rate disturbances were money supply disturbances, we would

10
expect them to have similar effects across countries, but not necessarily to be of
similar quantitative importance across countries.

The responses of prices to interest rate shocks show some consistency -- they are all
initially positive -- but the positive response is strong and persistent only in
France and Japan. In the other countries the responses are weaker and eventually
become negative. This pattern does not fit so easily into a monetarist/ISLM inter-
pretation. If monetary contraction reduces nominal aggregate demand, lowering output
through the interaction of deflationary pressure with price stickiness, how can it be
associated with previously unanticipated inflation rather than deflation? We will
come back to this point.

Money innovations show some consistency in their effects on prices -- the effects are
all positive. However the effects are absolutely small and relatively unimportant in
Germany and the United Kingdom -- the two countries where the money innovations
generate the most sustained increase in the money stock. The responses of output to
money innovations are either absolutely and relatively small, or predominantly
negative, in all countries except Germany. Money stock innovations therefore are not
a good candidate for interpretation as a monetarist/ISLM monetary policy shock.

To complete our discussion of the responses to interest rate and money shocks we need
to discuss the top three rows of the charts. The entries in the M column here are all
fairly small. The responses of commodity prices to interest rate innovations are all
sustained and negative, except in the case of Japan. This supports interpretation of
interest rate innovations as monetarist/ISLM monetary policy shocks -- contraction is
deflationary. The only caveat is that it is perhaps surprising that four of these
five countries’ monetary policies could all independently have such strong influences
on a single international commodity price index. It would be desirable to have a
model that allowed explicitly for international interactions of monetary policies.
The responses of currency value to interest rate innovations raise more difficulties.
A monetary policy contraction should raise the value of domestic currency, other
things being equal. This more or less what occurs in Japan, the U.K., and the U.S.
in response to interest rate innovations. But in France and Germany interest rate

11
innovations are followed by large and persistent declines in the value of domestic
currency.

IV. Contemplating Policy Endogeneity


The strong positive responses of prices and negative responses of currency value to
interest rate innovations in some countries raise serious difficulties for interpret-
ing these innovations as monetary policy shocks. The interpretation can possibly be
rescued, however, if we suppose that monetary policy authorities have information
about inflationary pressures better than what can be obtained from the variables in
these models. Policy authorities might know that inflationary pressure is about to
arrive and contract to dampen the effects of these pressures. Then prices would rise
after the monetary contraction (though by less than they would have without the
contraction) and output would fall because of the standard effects of nominal demand
contraction on real output. The falling currency value in the wake of the interest
rate rise is consistent with this story.

This interpretation can be given either an ISLM/monetarist or RBC twist. The RBC
theorist would claim that the same foreseen disturbance that generates the inflation-
ary pressure generates the output decline. Interest rate rises precede output
declines only because the monetary authorities tend to raise interest rates when they
accurately anticipate negative supply shocks. Monetary policy does not generate
subsequent output decline. The monetarist/ISLM theorist would claim instead that the
inflationary pressures generate lower output mainly because of the reaction to them
by the monetary authorities.

The first round of statistical work for this paper was carried out with four-variable
systems, excluding XR and PC. These same patterns and issues of interpretation arose
in that work. The introduction of the XR and PC variables was meant to shed light on
these identification questions. Charts 6-10 show impulse responses for four-variable
systems (solid lines) along with corresponding responses taken from the 6-variable
systems displayed in Charts 1-5. Note that the large, sustained, mostly positive
responses of price level to interest rate innovations that in the 6-variable system
appears only in France and Japan, in the four-variable systems appears also in the
U.K. and more weakly in the U.S. and Germany. In no case has a positive response of
prices to interest rate innovations become larger rather than smaller in moving from

12
the 6 to the 4 variable systems. Note also that the negative responses of output to
interest rate innovations are only slightly reduced in size in the 6-variable sys-
tems, more or less corresponding to the slight reductions in the size and persistence
of interest rate movements themselves in the wake of an interest rate innovation.

To make this interpretation more concrete, consider the case of the U.K., where the
effect of the additional variables on the price responses is probably strongest.
From Charts 4 and 9 it is clear that much of the price variance that was, in the 4-
variable system, being attributed to interest rate innovations, is now being attri-
buted to PC innovations. These innovations generate a large rise in PC, followed by
a smoother and somewhat smaller rise in UK prices. Interest rates rise promptly
after the PC increase, and money stock declines smoothly. Output does not decline.
This fits the picture of a monetary contraction reacting to, without eliminating,
anticipated inflationary pressure signaled by a jump in commodity prices.

Somewhat similar patterns in PC columns of Tables 1-5 occur in the other countries.
The XR columns are also similar across countries, and are consistent with a story in
which rises in currency value have deflationary impacts partially offset by the
loosened monetary policy they induce.

On the whole, comparison of the 6 and 4 variable systems suggests that exchange rates
and commodity prices do influence monetary policy, and that accounting for this can
be important for our interpretation of the data. The fact that responses of output
to interest rate innovations remain so stable as the additional variables are intro-
duced supports the monetarist/ISLM interpretation. But it remains true even in the
6-variable systems that the only country in which the price response to an interest
rate innovation is primarily negative is the United States, and even there the
negative response does not appear for about a year. This is an embarrassment for a
monetarist/ISLM interpretation.

From a RBC perspective, the results are mixed. The repeated finding in actual data
of declining output and money stock in response to interest rate innovations has not
been reproduced in any existing RBC model. My own RBC model [1989] did at least
examine the model’s implied responses to innovations, but did not reproduce this

13
observation. Most RBC-style modelers have limited their aims to matching qualitative
timing patterns or bivariate cross-correlations in the data, aspects of the data we
should know by now are much more easily reproduced than multivariate impulse res-
ponses. Until RBC-style modelers confront the multivariate time series data at this
level of detail, they will not be making a serious contribution to reducing our
ignorance in this area.

But in order to confront data at this level, some aspects of RBC-style modeling that
have become conventional will need to be relaxed (though it is perhaps not clear just
which ones.) The bottom rows of Charts 1-5 show a relatively even scattering of
predictive power for output across 6 mutually independent sources of disturbance.
Models built around a one-dimensional source of stochastic shocks (usually neutral
technology shocks in RBC models) will not match this pattern. To explain the predic-
tive value of nominal interest rates for output from an RBC perspective seems from
this paper’s results to require recognition that monetary policy-makers attempt to
anticipate inflationary pressures and pressures on exchange rate values -- and do a
better job of anticipating such pressures than can be done with a small aggregate
time series model. Such behavior will not be matched by a model that treats monetary
policy variables as determined by an exogenous sequence of random draws or by a model
that treats it as a simple function of the main aggregates in the economy. A serious
treatment of monetary policy behavior and our ignorance about it is required. That
is, we will need to see RBC models that recognize and deal with the identification
problem that is the source of the continuing disagreement about the effects of
monetary policy.

But though this paper documents the robustness of multivariate time series facts that
RBC modelers have not yet confronted, and though it shows that an informal moneta-
rist/ISLM-style interpretation of them can go some way toward making sense of them,
the paper also suggests that improved RBC-style modeling might eventually undercut
some or all of the apparent support in the data for monetarist/ISLM interpretations.
Much of what appears in small systems to be independent variation in monetary policy
instruments is found in large systems to represent responses to disturbances that can
be measured from knowledge of exchange rates and international commodity prices.
Cleaning out these endogenous components of interest rate disturbances does not

14
succeed in generating price responses to interest rate innovations that can comfort-
ably be interpreted in the monetarist/ISLM framework. It thus remains possible that
in a larger model, with a more detailed set of technology-related variables, we would
find interest rate innovation effects on real variables almost entirely dwindling
away.

V. A Research Agenda
It will be worthwhile to proceed further to study the statistical facts presented
here. We have displayed no error bands about the estimated impulse responses, for
which there is no excuse but time pressure. The models have been estimated indepen-
dently across countries, except for the appearance of the same variable (internation-
al commodity prices) in each country’s system. This is not only unsatisfying, it is
logically inconsistent -- we have 5 different equations for predicting the same
variable. Though the cross-country correlations of innovations (not reported in the
paper) are not large here, there can be little doubt that developments in different
economies are related, and leaving PC as the only route for such linkage does not
seem realistic. It is in principle possible that most of the apparent "robustness"
of these impulse responses across countries is simply a common component of sampling
error. We cannot assess this possibility without a model that takes explicit account
of international stochastic dependence.

The identification schemes used to interpret the data in this paper have relied
mainly on postulating that innovations in a particular variable - short interest
rates -- represent monetary policy disturbances. Because our interpretations have
been informal, we have been able to discuss qualitatively the possibility of contami-
nation of this interpretation by non-policy components of interest rate innovations,
but we have not used any model that takes formal account of this. Elsewhere [1985],
and in early stages of work on this paper, I have used formal "identified VAR"
frameworks that might be thought to be helpful here. It is worth noting that they
turn out not to be very helpful here. The problem is that in these monthly data,
correlations among innovations are so weak that the identified VAR framework gives
poor results. Without going into detail, we can characterize the problem as like
what would happen in a cross-section simultaneous equations model in which all
exogenous and endogenous variables turned out to have approximately zero sample
cross-correlations. No number of zero-restrictions on structural equations would in

15
that case give identification. Exogenous variables would always turn out to be
uncorrelated with the endogenous variables they might be used as instruments for.3

The plausibility of Wold-style identification, in which each structural equation is


identified with a unique dependent variable, is increased by the near-diagonality of
the innovation covariance matrix. If we believe that the U.S. Federal Funds Rate is
for practical purposes directly controlled by the Federal Reserve from month to
month, so that any change in monetary policy must impact it within the month, then
the lack of correlation of interest rate innovations with contemporaneous innovations
in other variables strengthens the case that they can be interpreted as policy
shocks. Nonetheless this case can never be ironclad. It is always possible, e.g.,
that interest rate and money innovations emerge as uncorrelated because of the
canceling effects of monetary policy shocks, which move interest rates and money in
opposite directions, and money demand shocks, which move them in the same direction.
One cannot study these possibilities without going beyond the framework of identifi-
cation based on the contemporaneous impacts alone.

As already outlined in the previous section, this paper’s results suggest an agenda
of improvements in RBC modeling. Such modeling could combine the economic theore-
tical rigor of RBC models with the unblinkered confrontation of the data of identi-
fied VAR’s. But ideally there should be a similar movement from the monetarist/ISLM
side of this discussion. It should be quite possible to formulate models with
complete dynamics and explicit treatment of disequilibrium elements -- workers who
search rather than freely adjust their level of work effort, firms that advertise and
adjust firm-specific wages rather than freely adjust their level of employment for
example. If such a model could generate the monetarist/ISLM policy implications,
while matching the facts of nominal as well as real aggregate fluctuations, it would

-----------------------------------------------------------------------------------------------------------------------------

3It is perhaps worth noting that in work with quarterly U.S. data in a 6-variable
system, introduction of an identification scheme like that in my earlier paper [1985]
produced shifts in the responses to interest rate innovations very much like those
displayed in Table 10. With the quarterly data the correlation matrix of innovations
is not so nearly diagonal, giving the identification scheme more to work with.
However the results with such schemes for other countries produced little new insight
with the quarterly data. With the monthly data, there was in every case a likelihood
maximum that very nearly treated each variable innovation as structural, but also
other local maxima with nearly the same likelihood that were quite different.

16
provide a more direct challenge to the RBC models than existing informal appeals to
time series facts.

REFERENCES
Friedman, Milton and Anna J. Schwartz [1963] A Monetary History of the United States
1867-1960, Princeton University Press.

Tobin, James [1970]. "Money and Income: Post Hoc Ergo Proper Hoc?" Quarterly
Journal of Economics.

Romer, Christine and Paul Romer [1989]. "Does Monetary Policy Matter?: A New Test
in the Spirit of Friedman and Schwartz," NBER Macroeconomics Review.

Sims, Christopher [1971]. "Money, Income and Causality," American Economic Review.

___ [1989]. "Models and Their Uses," American Journal of Agricultural Economics.

___ [1986]. "Are Forecasting Models Usable for Policy Analysis?" Quarterly Review
of the Federal Reserve Bank of Minneapolis.

___ [1980]. "Macroeconomics and Reality," Econometrica.

Todd, Richard M. [1991]. "Vector Autoregression Evidence on Monetarism: Another Look


at the Robustness Debate," Quarterly Review of the Federal Reserve Bank of
Minneapolis, Spring.

17
APPENDIX

Below is a description of the data series. In each case the series is named and the
identifying number of the series in the IMF publication International Financial
Statistics is given.

Commodity price index.


WORLD EXPORT COMMODITY PRICE INDEX. IFS 00176AXD

Exchange Rates.
The raw data were:
British Pounds per SDR. IFS 112..AA.
Yen per SDR. IFS 158..AA.
Deutsche Marks Per SDR. IFS 134..AA.
French Francs PER SDR. IFS 132..AA.
US Dollars per SDR. IFS 111..SA.
The exchange rates actually used were, for each country except the US, the ratio
of the SDR rate above to the US SDR rate. For the US the exchange rate was the
ratio of the US SDR rate to the average of the other four countries’ SDR rates.

Industrial Production.
US Industrial Production, SA. IFS 11166..C
UK Industrial Production, SA. IFS 11266..C
JAPAN Industrial Production, SA. IFS 15866..C
GERMANY Industrial Production, SA. IFS 13466..C
FRANCE Industrial Production, SA. IFS 13266..C

Price Index.
US CPI. IFS 11164...
UK CPI. IFS 11264...
JAPAN CPI. IFS 15864...
GERMANY CPI. IFS 13464...
FRANCE CPI. IFS 13264...

18
M1.
US M1. IFS 11134...
UK M1. IFS 11234..B
Japan M1. IFS 15834...
Germany M1. IFS 13434...
France M1. Deterministic ssnl. adj. of (IFS 13234...) over 64:1-77:12, spliced
to the official SA series (IFS 13234..c), for 77:12-90:2

Interest Rate
US Federal Funds Rate. IFS 11160B..
UK Treasury Bill Rate. IFS 11260C..
Japan Call Rate. IFS 15860B..
Germany Call Rate. IFS 13460B..
France Call Rate. IFS 13260B.. 86:3-5 interpolated from (IFS 13260BS.)

19

You might also like