Jean Boivin and Marc Giannoni

Assessing Changes in the

Monetary Transmission
Mechanism: A VAR Approach
1. Introduction then there is a good chance that once confronted again with
large successive shocks, the economy will again become more

S everal authors have documented a reduced variability of

output and inflation in the United States since the beginning
of the 1980s.1 In fact, a comparison of the 1980:1-2001:2 period
volatile. Alternatively, if most of the reduced volatility is due to
a change in the propagation of the disturbances, then it is
plausible to expect the greater stability to last. In the latter case,
with the two preceding decades shows that the standard it is also plausible to think that the monetary transmission
deviation of quarterly output growth has fallen 30 percent, mechanism has changed.
while the standard deviation of inflation has decreased more Several factors may have rendered the economy more
than 40 percent. These changes in the time series properties of immune to shocks. On the one hand, firms and consumers may
output and inflation raise a number of important questions for have changed their behavior and the organization of markets in
policymakers. For instance, has this increased stability been a way that has reduced the effect of given shocks on output and
associated with an alteration of the transmission of monetary inflation. For instance, Kahn, McConnell, and Perez-Quiros
policy due, for example, to changes in the behavior of (2002) argue that a more effective management of inventories
consumers, firms, or the Federal Reserve? Does monetary has been an important factor behind the reduction in the
policy still affect inflation and output as much as it did in the variability in output. On the other hand, the conduct of
1960s and 1970s? Should we expect the reduced volatility of the monetary policy may have been more responsive to
U.S. economy to last? fluctuations in inflation and output since the beginning of the
The answers to these questions depend in fact on the origin 1980s, partly compensating for the effect that shocks may have
of the changes. In particular, they require a determination of had on inflation and output (see Clarida, Gal, and Gertler
whether the reduced volatility of output and inflation is due to [2000] and Boivin and Giannoni [2002]).
smaller and less frequent disturbances, such as shocks to In this paper, we seek to understand more clearly the nature
productivity, foreign economies, fiscal policy, and monetary of the changes in the U.S. economy over the past four decades,
policy, or whether the propagation of these shocks has changed and to determine whether the reduction in output and
so that output and inflation have become less sensitive to inflation variability has been associated with a change in the
shocks. Clearly, if the main cause of the increased economic transmission of monetary policy. First, we estimate a small
stability in the past two decades is a reduction in the vector autoregression (VAR) model using U.S. data from 1960
importance of exogenous shocks, or special circumstances, to 2001 and test for its stability. Our results suggest that

Jean Boivin is an assistant professor of finance and economics at Columbia The authors thank Kenneth Kuttner for fruitful discussions, an anonymous
University; Marc Giannoni is an economist at the Federal Reserve Bank referee, and Alex Al-Haschimi and Kodjo Apedjinou for research assistance.
of New York. Jean Boivin thanks the National Science Foundation (grant SES-0001751) for
<jb903@columbia.edu> financial support. The views expressed are those of the authors and do not
<marc.giannoni@ny.frb.org> necessarily reflect the position of the Federal Reserve Bank of New York or
the Federal Reserve System.

FRBNY Economic Policy Review / May 2002 97

instability is prevalent both in the systematic part of the 2. Evidence from Reduced-Form
estimated VAR and in the variance of shocks. Second, through Vector Autoregressions
counterfactual experiments, we assess whether the reduced
variability of the economy is due principally to less important
shocks (or special events) or to changes in the propagation
mechanism of these shocks. We find that the change in the 2.1 Generalities about VARs
propagation of the shocks in the past two decades accounts for
We use a reduced-form vector autoregression (VAR) model to
roughly 40 percent of the decrease in the variance of detrended
assess whether the reduced variability of output and inflation in
output and for 60 percent of the reduction in the variance of
the past two decades is due to a reduction in the variability of
inflation.
shocks or whether it is due to a change in the propagation of
Because we find that a significant fraction of the reduced
these shocks. A reduced-form VAR, as proposed by Sims
volatility is attributable to a change in the propagation of the
(1980), is a regression of some vector of variables Y t on lags of
shocks, we impose more structure on our empirical model to
this vector. Formally, the reduced-form VAR is a system of
assess whether there has been a change in the transmission of
equations that can be written in matrix form as
monetary policy. We find that output and inflation have
displayed diminished dynamic responses to unexpected federal (1) Yt = a + A 1 Yt 1 + + Ak Yt k + u t ,
funds rate movements since the beginning of the 1980s. Some
may conclude from this observation that monetary policy must where a is a vector of constants, A 1, , A k are matrices of
therefore have had smaller effects in the past two decades. We coefficients, and u t is a vector of innovations, that is, serially
argue that while it is possible that the functioning of the uncorrelated disturbances that have zero mean and a variance-
economy has changed in such a way as to better insulate output covariance matrix E ( u t u' t ) = u .2 The evolution of the vector
and inflation from monetary policy movements, it certainly Y t , which contains the macroeconomic variables whose
does not need to be so. In fact, as we show, it is also possible that behavior we seek to understand, depends both on unexpected
the diminished estimated response of output and inflation to disturbances, u t , and on a systematic component,
monetary policy shocks reflects a monetary policy aimed at a + A 1 Y t 1 + + A k Y t k , that determines how the shocks are
stabilizing output and inflation to a larger extent. Determining propagated to the rest of the economy. The estimates of
whether the reduced response of output and inflation to a, A 1, , A k , are obtained by applying ordinary least squares
monetary policy shocks results from a change in the economy (OLS) to each part of equation 1 separately, and the estimate of
or a change in monetary policy would require a fully specified u is given by the sample covariance matrix of the OLS
structural model. An analysis of this kind is beyond the scope residuals.
of this paper, but it is one that is undertaken in Boivin and Based on the estimated reduced-form VAR for different
Giannoni (2002). samples, we can determine whether the variance-covariance
The rest of the paper is organized as follows. In Section 2, we matrix u of the shocks and the systematic componentthat
present evidence from a simple reduced-form VAR that the is, the propagation mechanismhave changed over time.
reduced variability in the U.S. economy is due at least in part to Through counterfactual experiments, we can then determine
a change in the propagation of shocks. In Section 3, we identify whether the observed reduced volatility in output and inflation
the effects of monetary policy shocks and document the has been due mainly to a change in the shocks or in the
changes in the transmission of monetary policy. In Section 4, propagation mechanism.
we discuss the implications of our empirical findings for the The specific VAR that we consider contains four variables:
effectiveness of monetary policy. Section 5 concludes. detrended output, the inflation rate, commodity price
inflation, and the federal funds rate. Clearly, this VAR provides
a very simple or simplistic description of the economy, but it
contains at least the minimum set of variables that are crucial
for any discussion of monetary policy.3 Detrended output,
which is measured as the percent deviation of real GDP from a
stochastic trend, is obtained from a high-pass filter that isolates
frequencies associated with cycles with periodicity of less than
thirty-two quarters. This variable is often referred to as the
output gap in the empirical literature.4 The inflation rate is
the annualized rate of change in the GDP deflator between two

98 Assessing Changes in the Monetary Transmission Mechanism

consecutive quarters. The commodity price measure is the Table 1
quarterly average of the monthly spot market commodity price Stability Tests on the Reduced-Form VAR
index. Commodity price inflation is added to the VAR to limit
Regressors
the extent of a price puzzle (see Sims [1992], Chari,
Christiano, and Eichenbaum [1995], and Bernanke and Mihov PCOM Detrended Fed Funds
Dependent Variable Inflation Inflation Output Rate
[1998]). The original data set runs from 1959:1 to 2001:2.5 Four
lags are included in the VAR. Because of data transformation PCOM inflation 0.003 0.271 0.098 0.064
(first-differences, high-pass filter, and lags), the analysis is Inflation 0.064 0.006 0.218 0.000
Detrended output 0.136 0.005 0.691 0.001
performed on the 1963:1-1997:4 period.6
Fed funds rate 0.072 0.075 0.006 0.086

Source: Authors calculations.

2.2 Empirical Evidence Notes: The figures are the p-values for the Andrews (1993) sup-Wald test.
Under the null of the test, the coefficients are time-invariant. The test is
applied jointly to the constant and the coefficients and the lags of the
variable corresponding to the given column. The p-values were com-
puted using the simulation approach proposed by Hansen (1997).
Stability Tests on the Reduced-Form VAR PCOM inflation denotes commodity price inflation.

The stability of parameters in estimated macroeconomic

relationships has been investigated in a number of recent
papers. The most general evidence is provided by Stock and
Watson (1996), who find widespread instability in bivariate
relationships among seventy-six macroeconomic variables. In
the VAR context, mixed results have been obtained.7 Boivin In principle, one could estimate the break date for different
(1999) argues that the different conclusions are due mainly to subsamples as a by-product of the Quandt likelihood-ratio
the small sample properties of the stability tests, and to the tests of the previous subsection. Unfortunately for our VAR,
effect of the number of parameters tested on the power of these the estimated break datesfor each combination of a
tests. He concludes that there is compelling evidence of dependent variable and lags of a regressordo not provide a
parameter instability in monetary VARs. consistent picture of the timing of the observed instability. Thus,
We now perform a similar stability investigation on the VAR on the basis of anecdotal evidence on the conduct of monetary
described in the previous section. For each equation of the policy and previous empirical studies, while making sure that the
reduced-form VAR, we test jointly for the stability of all the samples are not too small, we decided to base our benchmark
coefficients on the lags of a given variable using the Wald version comparison on the subsamples on each side of 1979:4, the date
of the Quandt (1960) likelihood-ratio test (that is, the Andrews on which Federal Reserve Chairman Paul Volcker announced a
[1993] sup-Wald test). We use a heteroskedasticity-robust shift in policy.10 As an alternative, we start the second sample
version of the test. This test, unlike the well-known Chow test, in 1984:1 to eliminate the nonborrowed reserves targeting
does not assume knowledge of the date at which the break in the experiment from the second sample. The break date of 1984:1
parameters occurs. This test is also known to have power against is also the one estimated by McConnell and Perez-Quiros
other alternatives, such as one where the coefficients are (2000) for a change in the volatility of output growth. As a
following a random walk (see Stock and Watson [1998]). result, we consider three subsamples: the first corresponds to
The p-values of the stability tests are presented in Table 1. 1963:1-1979:3, the second to 1980:1-1997:4, and the third to
Overall, the results suggest that instability is important in this 1984:1-1997:4.
VAR. Of the sixteen tests performed, six38 percentreject Using the break dates 1979:3 and 1984:1, we also test for
the null hypothesis of stability at the 5 percent level.8 We thus stability of the variance-covariance matrix u .11 Table 2
interpret these results as strong evidence of changes in the reports the p-values relative to the stability tests on the
VAR. It is important to mention that, to the extent that the true variance of the innovations u t . Of the eight tests performed,
functioning of the economy is appropriately described by a five63 percentreject the null hypothesis of equal variance
linear model, potentially omitted variables do not generate on both sides of the break date, at the 5 percent level.12 It
spurious instability; such an omission may bias the estimated appears therefore that not only the propagation of the shocks
parameters of the systematic component, but would not imply has changed over the past decades, but also that the variance of
changes across samples.9 the innovations u t has changed significantly.

FRBNY Economic Policy Review / May 2002 99

 Table 2 reduction in the variances reported in the last column should
Stability Tests on the Variance of Innovations explain most of the decrease displayed in the third column.
in the Reduced-Form VAR Overall, the results reported in Table 3 suggest that both
p-Values for Null of Stability
changes in the propagation mechanism, A, and in the variance-
covariance matrix of the innovations, u , are important in
Break Dates
explaining the reduced variability of output and inflation.
Variance of Innovation in 1979:3 1984:1 The variances of output obtained assuming unchanged A are
PCOM inflation 0.040 0.054 smaller than the ones obtained assuming unchanged u ; in
Inflation 0.014 0.023 fact, in the pre-1980 versus post-1980 comparison, 62 percent
Detrended output 0.008 0.067 of the reduction in the variance of output can be attributed to
Fed funds rate 0.040 0.117
a change in u , and the rest can be explained by a change in A
alone. However, the variances of inflation obtained assuming
Source: Authors calculations.
2
that the propagation mechanism A is unchanged are larger
Notes: The table reports p-values of the x -test for a change in the vari-
than the ones obtained assuming that the covariance of the
ance of the innovation of each equation, reported in separate rows, in
1979:3 or 1984:1. Under the null of the test, the variances of the innova- innovations u is unchanged; in fact, in the pre-1980 versus
tions are time-invariant. PCOM inflation denotes commodity price post-1980 comparison, 42 percent of the reduction in the
inflation. variance of inflation can be explained by a change in u , while
the rest is due to change in A alone. Thus, both changes in the
propagation mechanism and in the variance of the reduced-
form residuals are important in explaining the decreased
volatility of these variables.
Counterfactual Analysis with Reduced-Form VARs
So far, we have considered the residuals of the reduced-form
VAR as estimates of exogenous disturbances affecting the
Given this evidence of changes in the economy, we now
economy. This is only true, however, under very special
investigate whether the reduced volatility in output and inflation
conditions. More generally, the residuals of the reduced-form
is attributable mainly to a change in the variance-covariance
VAR (equation 1) constitute a linear combination of some
matrix of the disturbances, u , or to a change in the propagation
fundamental or structural disturbance vector t that also
of the perturbations. We do this by performing counterfactual
mixes contemporaneous responses among endogenous
experiments on the variance of output and inflation, using
variables. Suppose, for instance, that the true model of the
estimates of the reduced-form VAR over different subsamples.
economy is of the form of a structural VAR
The results are reported in Table 3. The first column reports the
variance of detrended output (or output growth) and inflation in (2) B 0 Yt = b + B 1 Yt 1 + + B p Yt k + t ,
the pre-1980 sample. The second column reports the
corresponding variances in either the post-1980 or post-1984 where B 0 is nonsingular and is normalized so as to have only 1s
sample. The third column contains the change in variances from on its main diagonal, and the variance-covariance matrix of the
one sample to the next. The fourth column reports the change in fundamental disturbances E ( t ' t ) = is assumed to be
variances obtained from the VAR, assuming that the estimated diagonal. Then, the matrices of the corresponding reduced-form
propagation mechanism, summarized by the collection of VAR (equation 1) satisfy a = B 01 b , A j = B0 1 Bj , for
matrices A = {A 1, , Ak} , has changed from the pre-1980 to the j = 1, , k , and the vector of shocks satisfies u t = B 0 1 t , so
post-1980 (or post-1984) sample, but that the covariance matrix that u = B 01 ( B 01 )' . As a result, if we estimate changes across
u of the innovations is kept constant. This provides a measure samples in the variance-covariance matrix of residuals of the
of the effect upon the variances of a change in the propagation reduced-form VAR, u , it is not possible to determine in the
mechanism in the face of unchanged disturbances. Finally, the absence of further assumptions whether this is due to a change in
last column of Table 3 contains the change in variances obtained contemporaneous relationships among variables represented by
from the VAR, assuming a constant propagation mechanism, A, B 0 , or whether it is due to a change in the variance-covariance
but assuming that the variance-covariance matrix u of the matrix of the fundamental shocks, . Therefore, it is still
innovations has changed as estimated.13 For instance, if the possible that part of the reduced volatility in output or inflation
observed diminution in the variance of inflation and output (gap attributed above to the change in u is in fact due to a change in
or growth) was explained mostly by a diminution in the the propagation of the shocks through B 0 . In contrast, if we
importance of the shocks hitting the economy, then the estimate changes in the coefficients of the matrices A 1 ,, Ak ,

100 Assessing Changes in the Monetary Transmission Mechanism

 Table 3
Counterfactual Experiments with the Reduced-Form VAR

Pre-1980 versus Post-1980

Part of the Change Due to

Variance of 1963:1-1979:3 1980:1-1997:4 Change Propagation Variance of Innovations ( u )

Detrended output 5.10 2.25 -2.85 -1.07 -1.78

Inflation 8.07 3.05 -5.02 -2.91 -2.11

Pre-1980 versus Post-1984

Part of the Change Due to

Variance of 1963:1-1979:3 1984:1-1997:4 Change Propagation Variance of Innovations ( u )

Detrended output 5.10 1.13 -3.97 -0.87 -3.10

Inflation 8.07 1.11 -6.96 -3.89 -3.07

Source: Authors calculations.

B iZZ B iZ R
then we know that the mechanism that propagates the Bi = ,
fundamental disturbances t must have changed.14 B iRZ B iRR
To isolate properly the contribution of changes in the for i = 0, , k . Noting that the scalar B 0RR = 1 , it follows
fundamental shocks, more structure needs to be added. We that our structural VAR (equation 2) can be written as
turn to this in the next section. k
(3) Z t = ( B 0ZZ ) 1 b Z + B iZZ Z t i B 0ZR Rt
i=1
k

3. Changes in the Monetary

+ B iZR R t i + tZ
i=1
Transmission Mechanism: k k
Evidence from a Structural VAR
BiRR Rt i + tR ,
R
(4) Rt = b B 0RZ Z t + B iRZ Z t i +
i=1 i=1
Having established that the mechanism that propagates the
where tZ is a vector of orthogonal disturbances and tR is a
exogenous disturbances has changed since the beginning of the
disturbance that is assumed to be orthogonal to tZ . The first
1980s, we now determine to what extent these changes have
equation describes the evolution of the nonpolicy variables of
affected the transmission of monetary policy. Many authors
have used structural VARs of the form equation 2 to describe the model in response to changes in all contemporary and past
the effects of monetary policy on key macroeconomic endogenous variables as well as unforecastable shocks. The
variables.15 While the Feds operating procedure has varied in second equation characterizes the behavior of the monetary
the past four decades, many authors have argued that the policy instrument in response to other endogenous variables,
federal funds rate has been the key policy instrument in the lagged values of the policy variable, and unforecastable shocks.
United States over most of that period (see, for example, Equations 3 and 4 are not identified without further
Bernanke and Blinder [1992] and Bernanke and Mihov assumptions. Following many papers in the literature
[1998]).16 This suggests that we split the vector Y t of including Bernanke and Blinder (1992), Rotemberg and
endogenous variables into two components: Y t = [ Z t', R t ]' , Woodford (1997), and Bernanke and Mihov (1998)we make
where R t represents the instrument of monetary policy, that is, the identifying assumption that the policy variable, Rt , affects
the fed funds rate, and Z t is a vector containing all other nonpolicy variables only with a lag of one period (assumed here
(nonpolicy) endogenous variables. Accordingly, we to be one quarter). Formally, it is assumed that the vector
decompose the matrices B i further as B 0ZR = 0 . The fed funds rate, however, is allowed to respond to

FRBNY Economic Policy Review / May 2002 101

all contemporaneous variables.17 As Z t and tR are uncorrelated Table 4
in this case, estimates of the coefficients appearing in equations 3 Variance Decomposition
and 4 are obtained by applying OLS on each equation of that
1963:1- 1980:1- 1984:1-
system separately. An estimate of var ( tR ) is given further by the
1979:3 1997:4 1997:4
sample variance of the residuals of equation 3.
Variance of monetary policy shocks 0.30 0.61 0.14
Contribution to output variance
(percent) 19 7 3
Contribution to inflation variance
3.1 On the Interpretation and the Role (percent) 14 10 6
of Monetary Policy Shocks
Source: Authors calculations.
The disturbances tR are often referred to as monetary policy
shocks. Several authors like to think of these shocks as
representing changes in monetary policy stance. Rudebusch
(1998) criticizes the use of VARs for the description of
monetary policy effects, pointing out that monetary policy equation of the form equation 4 does not provide as good a
shocks obtained from VARs typically differ substantially from description of actual monetary policy in this period.
standard interpretations of past policy actions. Specifically, he
reports that the correlation between policy shocks obtained
from VARs and those derived from federal funds futures are
generally very low. Evans and Kuttner (1998) argue instead that 3.2 Empirical Results
such a low correlation constitutes a poor measure of a VARs
performance. Furthermore, in his comments on Rudebusch,
Sims (1996) insists that VARs may well provide a correct
Variance Decomposition
description of the economys response to exogenous shocks,
even though the interpretation of the residual shocks as
Regardless of the evolution of the variance of monetary policy
historical monetary policy actions may be problematic.
shocks, the fraction of variance of output and inflation due to
So what are these monetary policy shocks? Christiano,
these shocks has decreased dramatically since the beginning of
Eichenbaum, and Evans (1999) suggest that monetary policy
the 1980s. While almost 20 percent of output variance is
shocks may reflect exogenous shocks to preferences of the
attributable to monetary policy shocks in the pre-1980 sample,
monetary authority, such as stochastic shifts in relative weights
this proportion has fallen to 3 percent in the post-1984 sample
given to unemployment versus inflation stabilization. Bernanke
(Table 4). Similarly, the variance of inflation due to such shocks
and Mihov (1998) argue that these shocks could also reflect
has fallen by half from the former sample to the latter. This
measurement error in the preliminary data. At a different level,
confirms the finding of many researchers, such as Leeper et al.
the shocks tR represent all fluctuations of Rt that are
(1996), that monetary policy shocks have accounted for very
unexplained by the systematic response to fluctuations in
little variability in output and inflation since the beginning of the
current and past Zs as well as past Rs included in the VAR. In this
1980s. Some researchers have concluded from this that
respect, the smaller the shocks (relative to the variability of the
2 monetary policy does not matter much. For many economists
fed funds rate), the higher the R statistic for this equation, and
(including us), however, monetary policy is mostly characterized
thus the better the model is able to capture the Feds actions. In
by the endogenous response to developments in the economy,
fact, the monetary policy shocks are typically small in VARs of
2 and, as we explain in Section 4, even if monetary policy shocks
the kind we estimate, as the R statistic for the interest rate
were very small (in fact, even if they were equal to zero all the
equation is above 0.96 in any of the samples considered.
time), monetary policy could still matter substantially for the
In absolute terms, the monetary policy shocks have
determination of output and inflation.
decreased dramatically since the mid-1980s. As the first line of
Table 4 reports, the variance of monetary policy shocks tR in
the post-1984 sample is only half the corresponding value for
the pre-1980 sample.18 However, the monetary policy shocks Impulse Responses to Monetary Policy Shocks
display a substantially greater variance in the post-1980 sample,
suggesting that the 1980-83 period was one of large monetary No matter the size and the interpretation of monetary policy
policy shocks. This may also indicate that an interest rate shocks, these shocks provide an exogenous source of variation

102 Assessing Changes in the Monetary Transmission Mechanism

that allows us to identify the response of the economy to fed funds rate and the associated 95 percent confidence intervals
monetary policy. Such responses will allow us to determine for all three samples.
whether and how the transmission of monetary policy has The key result from the comparison across samples is that
changed over time. Chart 1 displays the impulse response the response of detrended output and inflation is much less
functions to an unexpected 1-percentage-point increase in the pronounced and persistent since the beginning of the 1980s

Chart 1
Impulse Responses to a Monetary Shock over Different Samples

Output, 1963:1 to 1979:3 Output, 1980:1 to 1997:4 Output, 1984:1 to 1997:4

1.0 1.0 1.0

0.5 0.5 0.5

0 0 0

-0.5 -0.5 -0.5

-1.0 -1.0 -1.0

-1.5 -1.5 -1.5

Inflation, 1963:1 to 1979:3 Inflation, 1980:1 to 1997:4 Inflation, 1984:1 to 1997:4

1.0 1.0 1.0

0.5 0.5 0.5

0 0 0

-0.5 -0.5 -0.5

-1.0 -1.0 -1.0

-1.5 -1.5 -1.5

Interest Rate, 1963:1 to 1979:3 Interest Rate, 1980:1 to 1997:4 Interest Rate, 1984:1 to 1997:4
1.5 1.5 1.5

1.0 1.0 1.0

0.5 0.5 0.5

0 0 0

-0.5 -0.5 -0.5

-1.0 -1.0 -1.0

-1.5 -1.5 -1.5

0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9
Quarters Quarters Quarters

Source: Authors calculations.

FRBNY Economic Policy Review / May 2002 103

than in the pre-1980 period; the trough of the response of impulse responses for the pre-1980 and post-1980 samples;
output is more than twice as large in the pre-1980 sample as in Chart 3 displays the corresponding impulse response functions
the post-1980 or post-1984 samples. This result, which has for the pre-1980 and post-1984 samples.
already been documented in the literature (see, for example, Charts 2 and 3 show that while both sets of parameters
Gertler and Lown [2000] and Barth and Ramey [2001]), and sensibly affect the impulse response functions, most of
suggests that the effect of monetary policy shocks was stronger
before the 1980s. While this last conclusion is robust to the use
of the post-1980 or post-1984 sample in the comparison, there
are still notable differences between these two samples. In Chart 2

particular, the response of inflation appears somewhat stronger VAR-Based Counterfactual Analysis:
Pre-1980 Sample versus Post-1980 Sample
when the VAR is estimated on the post-1984 sample, and the
response of output, while of similar shape overall, becomes
(pre-80, pre-80): Pre-1980 sample
positive after six quarters, though not significantly so.19
(pre-80, post-80): Pre-1980 policy, post-1980 economy
(post-80, pre-80): Post-1980 policy, pre-1980 economy
(post-80, post-80): Post-1980 sample
Counterfactual Analysis with Structural VARs 0.5
Output

The previous section established that the economys response to

0.0
interest rate fluctuations has changed substantially over time, but
the evidence does not identify the reasons why. In fact, while the
tests of stability on the interest rate equation suggest that
-0.5
monetary policy is one potential source of this varying response
of the economy, the evidence in Table 1 is also consistent with
the presence of changes in the transmission mechanism. -1.0
We now investigate the source of change in the effect of
monetary policy by performing a counterfactual analysis on the 0.5
Inflation
structural VAR. In particular, we use two counterfactual
experiments to answer the following questions: 1) Are the
0.0
observed changes in the policy rule sufficient by themselves to
explain the evolution of the impulse response functions
reported in Chart 1? 2) Alternatively, assuming that monetary
-0.5
policy did not change, can we reproduce the evolution of the
impulse response functions through the equations of the VAR
corresponding to the nonpolicy block? -1.0
To answer these questions, let characterize monetary
1.0
policy and characterize the rest of the economy. More Interest Rate
precisely, let s be the set of estimates of the parameters of the 0.5
policy rule (equation 4) for sample s, and p be the set of
estimates of the remaining VAR parameters, that is, the 0.0
parameters of equation 3, for sample p. A combination
( s, p ) completely characterizes a set of impulse response -0.5
functions. For instance, ( pre 80, pre 80 ) corresponds to the
-1.0
impulse response functions obtained in the previous section
for the pre-1980 sample. The two counterfactual experiments -1.5
we undertake can then be expressed as ( post 80, pre 80 ) and 0 1 2 3 4 5 6 7 8 9 10
( pre 80, post 80 ) . Chart 2 displays the resulting impulse Quarters
response functions to an unexpected 1-percentage-point
increase in the fed funds rate, together with the estimated Source: Authors calculations.

104 Assessing Changes in the Monetary Transmission Mechanism

the reduced response of output and inflation in the post-1980 lines ( pre 80, pre 80 ) with ( post 80, pre 80 ) and
and post-1984 samples is accounted for by a change in the ( pre 80, post 80 ) with ( post 80, post 80 ) we note that
parameters characterizing systematic monetary policy. In fact, by the responses of output and inflation associated with the policy
changing monetary policy, , and maintaining the parameters estimated for the post-1980 sample involve considerably less
of the rest of the economy, , fixedthat is, by comparing the variation than those associated with the policy of the pre-1980
sample. In addition, by keeping the coefficients of the rest of
the economy pre 80 constant, we observe that a change in
policy from pre 80 to post 80 explains an important part of
the impulse responses ( post 80, post 80 ) obtained in the
Chart 3
second period. This is even clearer when one compares the pre-
VAR-Based Counterfactual Analysis:
1980 and post-1984 samples (Chart 3).
Pre-1980 Sample versus Post-1984 Sample
Moreover, keeping the monetary policy coefficients fixed
(pre-80, pre-80): Pre-1980 sample
and letting the rest of the economy vary, we observe that the
(pre-80, post-84): Pre-1980 policy, post-1984 economy impulse response functions are changed only little. Again, this
(post-84, pre-80): Post-1984 policy, pre-1980 economy supports the idea that most of the reduction in the response of
(post-84, post-84): Post-1984 sample output and inflation to monetary shocks is attributable to a
0.5 change in the systematic response of monetary policy.
Output
Our analysis suffers, however, from an important
shortcoming: it does not address the famous Lucas (1976)
0.0 critique. In fact, even though we have imposed some structure
on the VAR, the description of the private sector behavior, as
given by equation 3, corresponds to a reduced-form. For in
-0.5 general, firms and consumers typically care, at least in part,
about their expectations of future states of the economy.
Changes in the policy reaction function are thus likely to affect
-1.0
the way agents form their expectations of future variables, and
0.5 therefore the coefficients of the equations in the system
Inflation
(equation 3). A more structural analysis necessary to address
the Lucas critique by taking into account the effects of changes
0.0 in the policy reaction function on a system of the form
equation 3 is performed in Boivin and Giannoni (2002).

-0.5

4. On the Potency of Monetary

-1.0
Policy: A Discussion
1.0
Interest Rate The empirical evidence that we have reported suggests that the
0.5 response of output and inflation to monetary policy shocks has
decreased since 1980. Does this imply, as is sometimes argued,
0.0
that the Federal Reserve has partly lost its ability to affect the
economy? Has the economy changed in such a way that it is
-0.5
now able to insulate itself better from monetary policy
-1.0 movements? Our analysis certainly does not imply that. For as
we have argued, the change in the transmission mechanism is
-1.5 also at least partlyif not mainlydue to a change in the
0 1 2 3 4 5 6 7 8 9 10
systematic behavior of monetary policy. In fact, if monetary
Quarters
policy was responding more systematically to economic
Source: Authors calculations. conditions with the objective of minimizing the variability of

FRBNY Economic Policy Review / May 2002 105

inflation and output, we would observe precisely a reduction of characterized by an interest rate rule. For simplicity again, we
the variability of target variables due to both monetary policy consider a rule of the form
innovations and to the systematic part of monetary policy,
(7) rt = yt + t ,
hence, the transmission mechanism.
To clarify this point, we consider the following very stylized
where > 0 . The monetary authority is assumed to change its
model. In this model, output, y t , is determined by
instrument systematically in response to output fluctuations:
(5) y t = Et y t + 1 r t + t , the interest rate is raised as output is above trend, while it is
lowered as output lies below trend. Changes in the instrument
where the parameter > 0 . This equation, which is often called may also reflect monetary shocks t of the kind discussed in
the intertemporal IS equation, lies at the core of many recent Section 3 and that are supposed to be serially uncorrelated and
macroeconomic models.20 It is similar to the traditional IS uncorrelated with real demand shock t . Combining equations
equation in the sense that it relates output negatively to the real 5 and 7, we obtain
interest rate, rt .21 However, output is also affected by expected t t
-.
y t = -----------------
future output (or income), as consumers want to smooth their 1 +
consumption over time. The term t represents unforecastable
This last expression reveals that in equilibrium, output
demand shocks, such as unexpected changes in government
depends both on real demand shocks and on monetary shocks,
expenditures.22 An equation of this form can be obtained as a
as well as on the parameters and . In particular, an
log-linear approximation to an Euler equation for the optimal
unexpected unit increase in the short-term interest rate
timing across periods in a fairly large variety of models. In this
case, the parameter is associated with the elasticity of ( t = 1 ) reduces equilibrium output by ( 1 + ) .
intertemporal substitution in consumption. For simplicity, we Does a reduced output response reflect a reduced potency of
assume that the central bank conducts monetary policy by monetary policy? The answer really depends on the source of
setting the short-term real interest rate. This is clearly a the change. If the reduced output response is due to a smaller
simplification. While central banks are usually seen as affecting value of , reflecting, for instance, a lower elasticity of
short-term nominal interest rates, this assumption allows us to intertemporal substitution in consumption, then monetary
ignore the behavior of inflation, which is not relevant to our policy is less potent, as equation 6 indicates that output is less
point. (For a more general model aimed at replicating the sensitive to changes in the interest rate. In contrast, if the
actual behavior of output, inflation, and short-term nominal reduced response of output stems from an increase in , then
interest rates, see Boivin and Giannoni [2002].) Iterating this is not related to the degree of potency of monetary policy.
equation 5 forward, we obtain The reduced output response simply reflects an increased
willingness by the monetary authority to stabilize output. In
(6) y t = t r tL , the limit, policymakers could almost entirely stabilize output
by letting become very large, and hence by letting the interest
where rate respond very strongly to any fluctuation in output. Thus, a

r tL = E t rt + j reduced output response to shocks is exactly what one would
expect from a successful monetary policy that aims to stabilize
j=0
represents the long-run real interest rate.23 As equation 6 output.
makes clear, it is the long-run real interest rate that matters for
the determination of output. In this framework, short-term
interest rates affect output to the extent that their expected path
determines the long-run rate. Thus, even though a central bank 5. Conclusion
may have a direct influence only on short-term interest rates, it
is also able to affect longer term rates through its effect on In this paper, we have investigated the extent to which the
agents expectations of future short-term rates. In fact, short- reduced variability in output, inflation, and interest rates is
term rates and long-term rates must be related so as to prevent due to changes in the degree of variability of shocks, and the
arbitrage opportunities from arising. extent to which it is due to a change in the mechanism that
Following recent research on monetary policy rules (see, for propagates these shocks. We have argued that the change in
example, Taylor [1999]), we assume that monetary policy is the propagation mechanism has lowered the variability of

106 Assessing Changes in the Monetary Transmission Mechanism

macroeconomic variables to a large extent. This change has the reduced response of output and inflation to monetary
been associated further with a diminished effect of monetary shocks is due to a more stabilizing monetary policy or to a
policy shocks on output and inflation. This observation, change in the functioning of the economythe latter choice
however, does not imply that monetary policy has lost some of implying that monetary policy has lost some of its potency
its potency in the past two decades, because the change in the one needs to construct a full-fledged structural model that
transmission of monetary policy shocks could result from a can account for changes both in the behavior of private
change in the systematic behavior of the Federal Reserve. economic agents and in the policy rule chosen by the central
We did not attempt to determine the origin of the change in bank. This issue is addressed in Boivin and Giannoni (2002) in
the transmission mechanismin particular, whether the a more sophisticated and arguably more realistic model of the
change is due to parameters describing the functioning of the economy. There, we conclude that the reduced effect of
economy (such as above) or to a change in the conduct of monetary policy shocks is due mainly to a more effective
monetary policy (represented by changes in above). In fact, systematic behavior of monetary policy, and that there is little
it is not possible to disentangle these two types of changes using evidence that U.S. monetary policy has lost its potency in the
the VAR model considered in Section 3. To determine whether past two decades.

FRBNY Economic Policy Review / May 2002 107

 Endnotes

1. See, for example, McConnell and Perez-Quiros (2000) and 11. The test used is a version of the Breush-Pagan LM-test adapted for
Blanchard and Simon (2000). the present split-sample context.

2. See, for example, Sims (1980), Watson (1994), and Stock and 12. The same is true for the VAR that involves output growth instead
Watson (2001). of detrended output.

3. Bernanke and Boivin (2001) consider an empirical model that 13. Note that given the interaction between the propagation
accounts for much more information. mechanism and the variance of the errors, there is no unique way of
performing this decomposition. For instance, in the fourth column of
4. To check the robustness of our results, we have replicated all Table 3, the variance of the innovations, u , can be kept fixed at either
exercises by replacing the output gap with quarterly output growth. sample estimate. We report the change in variance evaluated at the
While we report only the results obtained from the VAR with average of the sample estimates of u . The same approach is used in
detrended output, the VAR with output growth yields very similar the last column of Table 3. More formally, we rewrite the VAR
results. (equation 1) in companion form as Y t = AY t 1 + u t , and express the
vectorized variance-covariance matrix of Y t for sample s as
5. All series are from the Standard and PoorsDRI database. The v Y, s = A s v u, s , where v Y, s vec E ( Y tY 't ) , v u, s vec E ( u tu'
t ) , and
1
mnemonics are GDPQ for real GDP, GDPD for the GDP deflator, and A s ( I A s A s ) . We then decompose the change in variance
PSCCOM for the commodity price index. between two samples as follows:

v u, 1 + v u, 2 A1 + A2
6. When output growth is considered instead of detrended output, - (A 2 A 1 ) + ------------------
v Y, 2 v Y, 1 = ------------------------ - ( v u, 2 v u, 1 ) .
2 2
the analysis is performed on the 1960:2-2001:2 period.
The figures in the fourth column of Table 3 correspond to the first
7. Bernanke, Gertler, and Watson (1997) find evidence of instability term on the right-hand side, while the numbers reported in the last
in a monetary VAR, while Bernanke and Mihov (1998) and column correspond to the second term.
Christiano, Eichenbaum, and Evans (1999) reach the opposite
conclusion. 14. Arguably, the split between the systematic and shock components
in a model of the form equation 2 is somewhat arbitrary. Indeed, if we
8. For the VAR with output growth instead of detrended output, of had a full understanding of the functioning of the economyand
the sixteen tests performed, eight50 percentreject the null considered all variables that explain the evolution of output and
hypothesis of stability at the 5 percent level. inflation as well as all variables that determine those explanatory
variables, and so onmost changes would in fact be accounted for by
9. To see this, suppose that we have a linear model, changes in the functioning of one or another part of the economy. Our
y t = x' 1t 1 + x'2t 2 + t for t = 1, , T , where the first column of understanding of the economy, however, is much more limited, so
x 1 t contains only 1s, the i s are constant, and the distribution of x it many changes are unexplained by the model and thus are represented
is time-invariant. If we omit x 2t from the regression, the OLS estimate by the disturbance vector t . Since our simple estimated VAR

of 1 is given by Tt = 1 x 1t x' 1t tT= 1 x 1t y t . The bias of 1 is

1 indicates changes over time in the propagation mechanism, we
account at least partly for changes in the economy.
given by E Tt = 1 x 1t x' 1t
1
Tt = 1 x1t x' 2t 2 . So, we see that the

bias is time-invariant, and thus misspecification by itself cannot 15. See Bernanke (1986), Bernanke and Blinder (1992), Cochrane
generate instability. (1994), Leeper, Sims, and Zha (1996), Christiano, Eichenbaum, and
Evans (1996, 1999), Bernanke and Mihov (1998), Evans and Kuttner
10. We do not include 1979:4 in the second sample to be consistent (1998), and Stock and Watson (2001), among others.
with the one used by Rotemberg and Woodford (1997), Bernanke and
Mihov (1998), and Clarida, Gal, and Gertler (2000).

108 Assessing Changes in the Monetary Transmission Mechanism

 Endnotes (Continued)

16. The federal funds rate probably provides a less adequate measure 19. A similar result is obtained by Gertler and Lown (2000) for a
of monetary policy stance for the 1979-82 period, as nonborrowed different VAR and different subsamples in the post-1980 period.
reserves were set to achieve a level of interest rates consistent with
monetary growth targets. However, Cook (1989) argues that the fed 20. See, for example, Woodford (1996, 1999), Goodfriend and King
funds rate may still provide a satisfactory indicator during this (1997), Rotemberg and Woodford (1997), Clarida, Gal, and Gertler
episode. (1999), and McCallum and Nelson (1999).

17. Note that while the identifying assumption here is not uniformly 21. All variables are expressed here in terms of percent deviations
adopted, the effects of an unexpected monetary policy shock on from their long-run values.
output and inflation described below are in line with the effects
commonly obtained. For alternative identifications of monetary 22. Technically, we assume that t satisfies E t t + j = 0 for all j > 0 .
policy, see, for instance, Bernanke (1986), Blanchard and Watson
(1986), Sims (1986), Leeper, Sims, and Zha (1996), Bernanke and 23. This sum is converging, as r t is expressed in terms of deviations
Mihov (1998), and Stock and Watson (2001). from its long-run average.

18. The results reported in the table are obtained from the VAR with
detrended output. Similar results are obtained from the VAR with
output growth.

FRBNY Economic Policy Review / May 2002 109

 References

Andrews, Donald W. K. 1993. Tests for Parameter Instability and Christiano, Lawrence J., Martin Eichenbaum, and Charles L. Evans.
Structural Change with Unknown Change Point. Econometrica 1996. Identification and the Effects of Monetary Policy Shocks.
61, no. 4 (July): 821-56. In Mario Blejer, Zvi Eckstein, Zvi Hercowitz, and Leonardo
Leiderman, eds., Financial Factors in Economic
Barth, Marvin J., and Valerie A. Ramey. 2001. The Cost Channel of Stabilization and Growth, 36-74. Cambridge: Cambridge
Monetary Transmission. NBER Macroeconomics Annual University Press.
2001.
. 1999. Monetary Policy Shocks: What Have We Learned and
Bernanke, Ben S. 1986. Alternative Explanations of the Money-Income to What End? In John B. Taylor and Michael J. Woodford, eds.,
Correlation. Carnegie-Rochester Conference Series on Handbook of Macroeconomics, 65-148. Amsterdam: Elsevier.
Public Policy 25 (autumn): 49-99.
Clarida, Richard, Jordi Gal, and Mark Gertler. 1999. The Science of
Bernanke, Ben S., and Alan S. Blinder. 1992. The Federal Funds Rate Monetary Policy: A New Keynesian Perspective. Journal of
and the Channels of Monetary Transmission. American Economic Literature 37, no. 4 (December): 1661-1707.
Economic Review 82, no. 4: 901-21.
. 2000. Monetary Policy Rules and Macroeconomic Stability:
Bernanke, Ben S., and Jean Boivin. 2001. Monetary Policy in a Data- Evidence and Some Theory. Quarterly Journal of Economics
Rich Environment. NBER Working Paper no. 8379. 115, no. 1 (February): 147-80.

Bernanke, Ben S., Mark Gertler, and Mark W. Watson. 1997. Cochrane, John H. 1994. Shocks. Carnegie-Rochester
Systematic Monetary Policy and the Effects of Oil Price Shocks. Conference Series on Public Policy 41 (December): 295-364.
Brookings Papers on Economic Activity, no. 1: 91-142.
Cook, Timothy. 1989. Determinants of the Federal Funds Rate.
Bernanke, Ben S., and Ilian Mihov. 1998. Measuring Monetary Federal Reserve Bank of Richmond Economic Review 75, no. 1
Policy. Quarterly Journal of Economics 113, no. 3: 869-902. (January-February): 3-19.

Blanchard, Olivier J., and John Simon. 2000. The Long and Large Evans, Charles L., and Kenneth N. Kuttner. 1998. Can VARs Describe
Decline in U.S. Output Volatility. Unpublished paper, Monetary Policy? In Topics in Monetary Policy Modeling.
Massachusetts Institute of Technology and Reserve Bank Basle: Bank for International Settlements.
of Australia.
Gertler, Mark, and Cara S. Lown. 2000. The Information in the High
Blanchard, Olivier J., and Mark W. Watson. 1986. Are Business Cycles Yield Bond Spread for the Business Cycle: Evidence and Some
All Alike? In Robert J. Gordon, ed., The American Business Implications. NBER Working Paper no. 7549.
Cycle: Continuity and Change, 123-56. Chicago: University
of Chicago Press. Goodfriend, Marvin, and Robert G. King. 1997. The New Neoclassical
Synthesis and the Role of Monetary Policy. In Ben S. Bernanke
Boivin, Jean. 1999. Revisiting the Evidence on the Stability of and Julio J. Rotemberg, eds., NBER Macroeconomics
Monetary VARs. Unpublished paper, Columbia University. Annual 1997, 231-83.

Boivin, Jean, and Marc Giannoni. 2002. Has Monetary Policy Hansen, Bruce E. 1997. Appropriate Asymptotic p-Values for
Become Less Powerful? Federal Reserve Bank of New York Structural-Change Tests. Journal of Business and Economic
Staff Reports, no. 144, March. Statistics 15, no. 1 (January): 60-7.

Chari, V. V., Lawrence J. Christiano, and Martin Eichenbaum. 1995. Kahn, James A., Margaret M. McConnell, and Gabriel Perez-Quiros.
Inside Money, Outside Money, and Short-Term Interest Rates. 2002. On the Causes of the Increased Stability of the U.S.
Journal of Money, Credit, and Banking 27, no. 4 (November): Economy. Federal Reserve Bank of New York Economic
1354-86. Policy Review 8, no. 1 (May): 183-202.

110 Assessing Changes in the Monetary Transmission Mechanism

 References (Continued)

Leeper, Eric M., Christopher A. Sims, and Tao Zha. 1996. What Does . 1992. Interpreting the Macroeconomic Time Series Facts:
Monetary Policy Do? Brookings Papers on Economic The Effects of Monetary Policy. European Economic Review
Activity, no. 2: 1-63. 36, no. 5 (June): 975-1000.

Lucas, Robert E., Jr. 1976. Econometric Policy Evaluation: A . 1996. Comment on Do Measures of Monetary Policy in a
Critique. Carnegie-Rochester Conference Series on Public VAR Make Sense? by Glenn Rudebusch. Unpublished paper,
Policy 1: 19-46. Princeton University.

McCallum, Bennett T., and Edward Nelson. 1999. An Optimizing Stock, James H., and Mark W. Watson. 1996. Evidence on Structural
IS-LM Specification for Monetary Policy and Business Cycle Instability in Macroeconomic Time Series Relations. Journal of
Analysis. Journal of Money, Credit, and Banking 31, no. 3 Business and Economic Statistics 14: 11-30.
(August): 296-316.
. 1998. Median Unbiased Estimation of Coefficient Variance
McConnell, Margaret M., and Gabriel Perez-Quiros. 2000. Output in a Time-Varying Parameter Model. Journal of the American
Fluctuations in the United States: What Has Changed since the Statistical Association 93: 349-58.
Early 1980s? American Economic Review 90, no. 5 (December):
1464-76. . 2001. Vector Autoregressions. Journal of Economic
Perspectives 15, no. 4 (fall): 101-16.
Quandt, Richard E. 1960. Tests of the Hypothesis That a Linear
Regression System Obeys Two Separate Regimes. Journal of Taylor, John B., ed. 1999. Monetary Policy Rules. Chicago:
the American Statistical Association 55, no. 290: 324-30. University of Chicago Press.

Rotemberg, Julio J., and Michael Woodford. 1997. An Optimization- Watson, Mark W. 1994. Vector Autoregression and Cointegration.
Based Econometric Framework for the Evaluation of Monetary In Robert Engle and Dan McFadden, eds., Handbook of
Policy. NBER Macroeconomics Annual, 297-346. Econometrics 4. Amsterdam: North-Holland.

Rudebusch, Glenn D. 1998. Do Measures of Monetary Policy in a VAR Woodford, Michael. 1996. Control of the Public Debt: A Requirement
Make Sense? International Economic Review 39, no. 4 for Price Stability? NBER Working Paper no. 5684.
(November): 907-31.
. 1999. Optimal Monetary Policy Inertia. NBER Working
Sims, Christopher A. 1980. Macroeconomics and Reality. Paper no. 7261.
Econometrica 48, no. 1 (January): 1-48.

. 1986. Are Forecasting Models Usable for Policy Analysis?

Federal Reserve Bank of Minneapolis Quarterly Review 10,
no. 1 (winter): 1-16.

The views expressed are those of the authors and do not necessarily reflect the position of the Federal Reserve Bank of New York
or the Federal Reserve System. The Federal Reserve Bank of New York provides no warranty, express or implied, as to the
accuracy, timeliness, completeness, merchantability, or fitness for any particular purpose of any information contained in
documents produced and provided by the Federal Reserve Bank of New York in any form or manner whatsoever.

FRBNY Economic Policy Review / May 2002 111