3 PDF

Journal of Economics and Business 64 (2012) 117–144
Contents lists available at SciVerse ScienceDirect
Journal of Economics and Business
The dynamic adjustments of stock prices to inflation

disturbances
Victor J. Valcarcel ∗
Texas Tech University, Economics, 257 Holden Hall MS 1014, Texas Tech University, Lubbock, TX, United States
a r t i c l e i n f o a b s t r a c t
Article history: While theoretical predictions establish a strong positive relation-

Received 28 March 2011 ship between equity prices and inflation, finding substantiating
Received in revised form 17 November empirical evidence has been a difficult endeavor. Generally, the
2011
data suggests a weak negative relationship between stock prices
Accepted 20 November 2011
and inflation. Aided by two different structural VAR specifica-
tions that allow for time variation in the covariance and drift of
JEL classification:
the system, this paper finds evidence that the weakly negative
E44
correlation between stock prices and US inflation results from off-
E50
G12 setting effects of shocks to monetary policy and disturbances to
the demand for financial assets. Since the 1960s, the stock price-
Keywords: inflation correlation is estimated to be relatively more stable than
Real stock prices the volatility of either series, both of which have experienced
The Great Moderation
substantial change—albeit volatility in US economic activity is esti-
Stochastic volatility
mated to have taken place far more gradually than that of stock
Markov Chain Monte Carlo
Structural vector autoregressions prices. The volatilities of US economic activity, inflation, and stock
Parameter instability prices all rose as a result of the financial crisis and the ensuing
Structural change 2008–2009 Great Recession—with the level of inflation volatility
estimates during the Great Recession comparable to those of the
Great Inflation period of the 1970s. While it is shown that a tra-
ditional VAR approach would also predict a positive stock price
response to inflationary disturbances, our time-varying approach
enables us to uncover that during the 2008–2009 Great Recession
period a stock price increase is more pronounced following infla-
tionary shocks that stem from money supply, rather than money
demand, disturbances—in contrast to the 1980–1982 recession
where the magnitude of the stock price response to both shocks is
more similar. These conclusions are qualitatively robust to changes
in variable choice and measurement frequencies.
© 2011 Elsevier Inc. All rights reserved.
∗ Tel.: +1 806 742 2466; fax: +1 806 742 1137.

E-mail address: vic.valcarcel@ttu.edu
0148-6195/$ – see front matter © 2011 Elsevier Inc. All rights reserved.
doi:10.1016/j.jeconbus.2011.11.002
118 V.J. Valcarcel / Journal of Economics and Business 64 (2012) 117–144
1. Introduction
While asset valuations are mainly determined by the financial health and other characteristics of
individual firms, fluctuations in general economic conditions are also a principal determinant of stock
prices. If these valuations reflect the real return of holding these assets, then from money neutrality
it follows that real stock prices should not be permanently affected by inflationary shocks—whether
they come from changes in consumer preferences, business sentiment, or monetary policy. While
inflationary shocks may have little long-run impact on real stock returns, it is generally agreed that
stock prices can be sensitive to inflation in the short and medium terms (He, 2006; Lee, 2010; Rapach,
2001).
Theoretical finance models typically advocate a positive short-run response in stock prices fol-
lowing an inflationary shock that stems from monetary expansion. The Gordon Growth model, for
example, shows that stock prices are directly related to current and expected growth rates of divi-
dend returns and inversely related to the required rate of return on the equity. Thus, under this model’s
assumptions, there is a positive impact on stock prices through two channels: one, if monetary easing
stimulates the economy, it would likely have a positive impact on the growth rate of dividends. Two, a
monetary expansion that depresses bond returns would see an increased demand for equities, which
would lead the average investor to lower expected rate of returns of equities. Increased dividend
returns and decreased expected returns on investment both serve to put upward pressures on stock
prices.
However, historical evidence points to a small negative correlation between real stock returns
and inflation. A number of explanations have been advanced to account for this irregularity. One is
that, in the presence of sustained inflation, agents discount asset valuations at an artificially high
rate because such an environment makes it difficult to distinguish between real and nominal returns
when the latter includes an inflation premium (Modigliani & Cohn, 1979). Feldstein (1980) posits that
sustained increases in inflation put downward pressures on real stock prices because the tax code
exerts a distortionary effect between depreciation costs and capital gains. This observed negative
relationship between inflation and stock returns is dismissed as spurious by Fama (1981). According
to his version of the proxy-effect hypothesis (PEH) this negative correlation is induced by a positive
relationship between stock returns and expected economic activity (as proxied by inflation) and an
inverse relationship between expected economic activity and inflation.1 Another explanation for this
empirical irregularity could be attributed to the mismanagement of monetary policy in the 1960s and
1970s, when under the premise of an exploitable trade-off between inflation and unemployment, the
Federal Reserve gives into the temptation to inflate until time-consistent inflation rates were reached
(Cogley & Sargent, 2001; Sargent, 1999). The ensuing higher expectations of inflation would increase
long-term rates leading investors to more aggressively discount future dividends. On the other hand,
the subsequent contractionary actions by the central bank could have also contributed to lower stock
returns because these monetary policy actions would tend to slow down economic activity and, thus,
depress current and expected future earnings.
While there is strong evidence that stock prices are sensitive to monetary policy and price level
fluctuations, it is likely that these sensitivities are not stable over time. Under different model con-
structs, Durham (2003) and He (2006) conclude that the relationship between monetary policy and
stock market returns may be time-varying. Postwar inflation in the US is not well-characterized by
constant or smooth volatility. In fact, there is abundant empirical evidence of a substantial decline in
the volatility of most US macroeconomic aggregates. Most explanations of this volatility reduction fall
under two main categories. One view is that there might have been structural changes in the transmis-
sion mechanism due to shifts in the views or tools held by the monetary policy authority.2 A second
view proposes that there has been a reduction in the underlying heteroskedasticity of the exogenous
1
Thus, one of the missions of this paper is to estimate not only whether the stock price-inflation correlation has experi-
enced significant changes in the postwar period, but also to elucidate whether this correlation is qualitatively different when
conditioning on various macroeconomic shocks.
2
This would attribute the lower volatility levels to a “good management” of the economy by the Federal Reserve.
V.J. Valcarcel / Journal of Economics and Business 64 (2012) 117–144 119
Table 1
Relative standard deviations.
Standard deviation Relative standard deviation
Pre-1984: Post-1984 Pre-1984 Post-1984
1. Output 0.0051 0.0026 0.16 0.07

2 Stock prices 0.0326 0.0364 1.00 1.00
3. GDP deflator 0.0021 0.0010 0.06 0.03
4. PCE deflator 0.0021 0.0017 0.06 0.05
5. CPI 0.0026 0.0027 0.08 0.07
6. Core – CPI 0.0019 0.0007 0.06 0.02
Note: Output is expressed in log first differences. S&P500 composite index is deflated by GDP deflator and log differenced. Each
price measured was log transformed and differenced twice to serve as a measure of inflation growth.
shocks themselves.3 Neither explanation can be broached by a traditional structural vector autore-
gression (SVAR) approach that assumes constant autoregressive coefficients—this rules out the first
explanation—nor homoskedastic structural shocks based on a constant variance–covariance matrix of
the VAR—which would rule out the second explanation. Summers (2005), Cecchetti, Flores-Lagunes,
and Krause (2006), and Keating and Valcarcel (2011b), point out that this substantial decrease in the
volatilities of major aggregates is not an isolated phenomenon for the US but that it has been expe-
rienced by other industrialized economies as well.4 Explanations to account for the high volatility in
inflation in the 1970s and subsequent reduction thereafter can range from policy as in Clarida, Gali,
and Gertler (2000), Cogley and Sargent (CS, 2001, 2005) and Summers (2005); exogenous reductions in
the heteroskedasticity of underlying supply shocks as in Sims (1999) and Bernanke and Mihov (1998);
and improvements in inventory management as in McConnel and Perez-Quiros (MPQ, 2000).
Whatever the cause, any econometric technique that attempts to quantify the effects of monetary
and technology shocks on economic aggregates should address the possibility of time variation in the
underlying drifts and volatilities of the system. Table 1 shows standard deviations in the growth of the
US postwar GDP, various measures of inflation, and the S&P500 composite index. The year generally
associated with the structural break in volatility of economic activity is 1984 (MPQ, 2000). The table
shows that the standard deviation for the post-1984 period is about half that of the earlier period.
Measures of the growth of inflation built from the GDP deflator and core CPI see similar decreases
in the latter part of the sample. Measures from the CPI and PCE deflator see a smaller reduction
in the post-1984 period. As the third and fourth columns point out, relative to the growth of stock
prices—which experience no such decrease—the volatility of inflation and aggregate economic activity
growth has in most cases substantially reduced since 1984. Given that real stock prices do not seem
to exhibit a reduction in volatility commensurate with that of the other aggregates, it is possible that
the underlying relationship between the drifts and correlations of these variables changes over time.
In this paper, we employ recent developments in the estimation of SVAR modeling in the presence
of stochastic volatility—a time-varying structural vector autoregression approach (henceforth TV-
SVAR)—to analyze the effects of structural macroeconomic shocks on the real value of a broad index of
US stock prices. The primary objective is to measure the contribution made by identified shocks to real
stock price fluctuations and, more importantly, how these relative contributions have been changing
over time since the 1960s. This approach has certain benefits compared with more traditional SVAR
specifications. First, it allows the data to provide more information on the way parameters may evolve
over time. Second, this framework requires no assumptions about the timing, frequency, or size of
changes in impulse responses. This is potentially important in the likely event that the responses
of inflation and stock returns to the underlying structural shocks have not been constant through
3
This has been dubbed the “good luck” explanation, suggesting that the nature of negative supply shocks has been intrinsically
ameliorated since the 1980s.
4
This has been dubbed the Great Moderation. The corollary of this phenomenon is the “bad policy vs. bad luck” debate that
arises when attempting to explain the US stagflation outcomes of the 1970s.
time. Finally, this more general construct encompasses previous models that have assumed stable
parameters within certain pre-assumed subsample periods.
The rest of the paper is structured as follows: Section 2 delineates the time-varying volatility con-
struct and outlines the procedures to carry out innovation analysis. Section 3 specifies a trivariate
benchmark model (Model I), describes the identification strategy, and discusses data issues. Section 4
extends the specification presented in Section 3 to consider a five-variable approach (Model II). Sec-
tion 5 presents and contrasts evidence from both models on the volatility and correlation estimates
of the variables of interest. Section 6 discusses the evolution of the time-varying impulse responses
to the identified shocks of Model I. Section 7 presents time-varying responses from Model II. Section
8 discusses results from a further extension that substitutes inflation for inflation expectations in
Model II, and shows what responses would look like if we imposed constancy of parameters in our
specification. Section 9 concludes.
2. The time-varying SVAR model
This section describes a structural model consisting of autoregressive coefficients and a covariance
matrix that are both time-varying. The construct of the TV-SVAR follows closely those of Cogley and
Sargent (2001), Cogley and Sargent (2005), Primiceri (2005) and Gali and Gambetti (2009). Consider
the following lth order VAR process
t (L)xt = t + et (2.1)
where xt is an n-vector of I(0) endogenous (time t) variables; t (L) = 0t − 1t L − · · · − lt Ll is a lth-order
lag polynomial where 0t = In and each jt is a matrix of time-varying coefficients; and et is an n-vector
of mean-zero VAR innovations with the time-varying covariance matrix Rt . The coefficients in (2.1)
encompass all histories of the VAR parameters from 1 to the end of sample T and evolve according to5
t = t−1 + ut (2.2)
where ut is a Gaussian white noise process, with zero mean and constant covariance matrix Q, that
is independent of et at all leads and lags. I follow Aguilar and West (2000) multivariate extension
of Jacquier, Polson, and Rossi (1994) stochastic volatility construct to allow for time variation in the
underlying VAR model of (2.1).
⎛ ⎞
1 0 ... 0
⎜ .. ⎟
Let E(et et ) ≡ Rt = Ft Ht Ft where F is given by Ft = ⎜ 21t
⎜f 1 ... .⎟
⎟ and Ht =
⎝ .. . . . ..
.
⎠
. 0
fn1t . . . fnn−1t 1
⎛ ⎞
h1t 0 ... 0
⎜ . ⎟
⎜ 0 h2t . . . .. ⎟
⎜ . . ⎟
⎝ . .. ... ⎠
. 0
0 . . . . . . hnt
The diagonal elements of Ht are independent univariate stochastic processes that evolve according
to the following:
ln hjt = ln hjt−1 + t ∀j = 1, 2, . . . n (2.3)
where t ∼ iid(0, ). This specification allows us to focus on permanent shifts in the innovation
variance—such as those that are emphasized on the US economic stabilization literature (Cogley &
Sargent, 2005)—while reducing the dimensionality of the estimation procedure (Primiceri, 2005.)6
5
This is a parameterization of a more general law of motion p( t | t−1 , Q) ∝ I( t )f( t | t−1 , Q) for the posterior densities of the
states where I( t ) is an indicator function that carries out the rejection sampling mechanism necessary to rule out explosive
paths of x.
6
This presents an alternative to ARCH models. Here, the variances generated by (2.3) are unobservable components.
Stacking all the off-diagonal elements of Ft −1 into a vector t we further assume that this vector
evolves according to the following drift-less geometric random walk7
t = t−1 + t (2.4)
where t ∼ iid(0, ). All innovations are assumed to be jointly normally distributed with the fol-
lowing assumptions on the covariance matrix of the system
⎛ ⎞ ⎡ ⎤
εt I 0 0 0
⎜ ut ⎟ ⎢ 0 Q 0 0 ⎥
var ⎝ ⎠ = ⎣ ⎦
t 0 0 0
t 0 0 0
where Q, , and are positive definite matrices and I is a n × n identity matrix. None of the off-
diagonal zero restrictions are requisite for estimation.8 However, allowing for an entirely unrestricted
correlation structure among the different sources of uncertainty negates any structural interpretation
of the innovations.
The moving average representation for xt as a function of the VAR innovations is obtained by
inverting (2.1) as follows:

∞
xt =
t (L)et =
s Ls et (2.5)
s=0
Following Gali and Gambetti (2009), we assume that the innovations (et ) of the reduced–form VAR
vector are a time-varying transformation of the underlying structural shocks in the economy (εt ) as
follows:
et = ϕt εt ∀t (2.6)
where ϕt is a nonsingular matrix that satisfies ϕt ϕt = Rt . Given this normalization scheme, changes
in the contributions of different structural shocks to the volatility in innovations in the underlying
variables of interest are captured by changes in ϕt . Let the companion form of (2.1) be given by
Xt = t + t Xt−1 + et (2.7)
⎛ ⎞
1t
2t 3t . . . lt
0 ... ⎜ 1 0 ⎟
where Xt ≡ [xt , xt−1

, . . . , xt−l+1 ] , et ≡ [et , 0, . . . , 0] and t ≡ ⎜
⎟.
..
. ...⎝ 0 0
⎠
0 ... 1 0
Highly plausible and economically tractable dynamics can often be derived by imposing long-run
restrictions in a VAR setting (Keating, 1992). Therefore, we employ a Blanchard and Quah (1989)-type
decomposition as an identification strategy based on tractable and largely uncontroversial long-run
theoretical predictions.9 Consider the following k-ahead dynamic multiplier of (2.7)
∂Xt+k
= sn,n (kt ) ∀k = 1, 2, . . . (2.8)
∂et
7
Similar decompositions in the literature have specified a constant F matrix (Cogley & Sargent, 2001, 2005). A constant
matrix F would suggest that an innovation to the ith variable has a constant effect on the jth variable. Since the objective here
is to model time variation in a relatively high-dimensional system, allowing for simultaneous interactions among variables is
important.
8
Primiceri (2005) outlines a minor modification to the estimation scheme to allow for non-zero off-diagonal blocks.
9
While this identification procedure remains popular, Faust and Leeper (1997) point out that, in a low-dimensional VAR,
identified shocks must be viewed as an aggregate number of underlying shocks. They argue that if an identified structural
shock consists of two independent innovations, then the BQ approach is only valid if the macroeconomic variables in question
respond to both innovations in the same way (qualitatively). This is, in essence, a criticism of the low order bivariate construct
of the original BQ (1989). Considering higher dimensional VARs, here, allows me to address this concern. Because of the size of
the VAR specifications in this paper (a three and a five variable), the identified shocks are disaggregated into a larger number
of sensible categories than the original BQ (1989) application.
where sn,n (• ) is a function that selects the upper left n-by-n matrix from a given larger matrix (in this
case kt ). A trivial application of the chain rule renders the following k-horizon structural impulse
responses
∂Xt+k ∂Xt+k ∂et
= = sn,n (kt )ϕt ∀k = 0, 1, . . . (2.9)
∂εt ∂et ∂εt
Unlike traditional VAR models, (2.9) allows the response of a variable to a structural shock at any given
horizon to vary over time. This model is general enough to allow for feedback between disturbances
in drift and the covariance of the system even under more restricted scenarios than what we pose
here.10 Specifying a time-varying covariance matrix, however, constitutes a more general, and perhaps
more attractive, modeling strategy as it allows for the feedback between the two components to be
dynamic. Therefore, the impulse responses of (2.9) inherit the time variation imposed in both the drift
and covariance of the system.
For some purposes, we are interested in the level effects. That requires the use of cumulative
¯k=
k j
impulse responses. These are obtained as follows. First, we define . The level response
t j=0 t
of each variable to each shock after k periods is the accumulated response of the differenced series
from period zero to period k. Then, following (2.9), the accumulated responses are given by Mt,k ≡
k j ¯ k )ϕt .
j=0
sn,n ( )ϕt . Finally, from the properties of the selector function, we obtain Mt,k = sn,n (
t t
Furthermore, letting k→ ∞ allows us to define Mt ≡ sn,n ( ¯ ∞ )ϕt as a time-varying matrix of long-run
t
cumulative multipliers that indicate the long-run effect of each shock on the variable of interest.
Thus—as in standard SVAR approaches—the matrix of cumulative long-run multipliers is contingent
on a particular identification strategy. Restrictions described in Sections 3 and 4 below imply that Mt
can be obtained as the Cholesky factor of the right-hand-side of:
¯ ∞ )Rt [sn,n (
Mt Mt = sn,n ( ¯ ∞ )] (2.10)
t t
Given Mt , we can solve for ϕt and obtain the structural impulse responses of each shock occurring
at time t:
∂Xt+k
¯ ∞ ) −1 Mt ∀t,
= sn,n (kt ) sn,n ( k = 0, 1, 2 . . . (2.11)
t
∂εt
It is clear that, unlike innovation analysis from more standard SVAR models, the response of each
variable to a given shock in (2.11) is allowed to vary over time. Thus, the average effect of each shock
on each variable is not only completely unrestricted on impact, and in the short run, but, more impor-
tantly, each shock is free to have a different impact at every time period. Therefore, given that the
effects of the shocks are conditional on time, the TV-SVAR model described in this section allows for
much richer dynamics than a traditional SVAR approach.11 As mentioned in the introduction, if the
underlying structure of the economy is nonlinear,12 then forcing a traditional SVAR—that imposes con-
stant drifts and a constant covariance matrix—may tantamount to misspecification.13 The TV-SVAR
approach here imposes no such restrictions about the constancy of parameters. Another important
advantage of this approach is that it allows us to remain agnostic as to possible periods when the data
may have behaved quite differently. The traditional approach involves assuming constant parameters
across subsamples that are often assumed a priori and independently of the structural model that is
10
For example, taking a version of (2.1) with one lag and combining it with its corresponding version of (2.2), one can easily see
from the ensuing error term of the subsequent model that even in the absence of time variation in the covariance matrix, there
may be correlation between the drift and volatility terms. Whatever affects the coefficients will likely inject heteroskedasticity
and, thus, impact the forecast error variance in different periods.
11
The ensuing effects from the time-varying covariance matrix are inherently nonlinear and, therefore, in the spirit of the
family of models described in Granger (2008). Note here that with the exception of the zero restrictions of the long run cumu-
lative matrix—requisite for structural identification—the long-run effects of the shocks are nonlinear as well, insofar as they are
also time-dependent.
12
Or characterized by substantial high frequency changes in the first two moments—as it is likely the case for US inflation,
and other macro aggregates.
13
Or in a best-case scenario, to losses in efficiency.
considered. The TV-SVAR approach can easily handle changes in the underlying volatility and drifts
in the economy—whether they arise from changes in policy (Clarida et al., 2000), non-policy events
(McConnell & Perez-Quiros, 2000MPQ, 2000) or even potential changes in data quality (Keating &
Valcarcel, 2011).
Each variable considered has a time-varying moving average representation that is driven by the
underlying structural disturbances. Letting xit represent a distributed lag process for each variable i
contingent on each shock j we have

∞
j
xit = it + [Ñt,k ]i,j εt−k (2.12)
k=0
j
for i = {yt , st , pt } and εt−k = {εAS
t−k
, εPO
t−k
, εIN
t−k
} according to Model I in Section 3 below or for i = {ut ,
j
yt , st , pt , it } and εt−k = {εNRU
t−k
, εAS
t−k
, εPO
t−k
, εIN
t−k
, εMP
t−k
} according to Model II in Section 4. Here,
Ñt,k ≡ sn,n (kt ) contains the matrix of autoregressive coefficients t from the companion form of the
VAR, and the selector function sn,n , applied to kt . From (2.12) we can obtain how the time-varying
unconditional variance of xit is decomposed into the contribution of each shock to the variance of each
variable as follows:

∞
2
var t (xit ) = [sn,n (kt )ϕt ]i,j (2.13)
k=0
Similarly the time-t covariance of xit and xqt conditional on each shock j is given by

∞
covt (xit , xqt ) = [sn,n (kt )t (1)Mt ]i,j [sn,n (kt )t (1)Mt ]q,j (2.14)
k=0
where there are as many summation terms as there are j shocks. Thus, each summation term captures
the covariance between variables i and q conditional on each shock j. Therefore, the time-t uncon-
ditional covariance constitutes the aggregation of the j-contingent covariances between variables i
and q. Time-varying unconditional and conditional correlations follow tractably from the expressions
above.14
3. Model I: a trivariate benchmark specification
I consider US data from the USECON and SPM databases provided by HAVER encompassing the first
quarter of 1955 through the second quarter of 2011. The model15 is based on xt = (yt st pt ) where
yt is real output,16 pt is (different measures of) the price level,17 and st denotes real equity prices.18
14
The reduced-form model (2.1) is estimated employing the MCMC algorithms introduced by Cogley and Sargent (2001,
2005) and Primiceri (2005). Implementation of Gibbs sampling consists of four steps: step1 uses the Carter and Kohn (1994)
algorithm to draw conditional posteriors of the states (or time-varying coefficients of the reduced-form model). Steps 2 and
3 use Carter and Kohn as well as Jacquier et al. (1994) to draw conditional posteriors of the hyper-parameters of the F and H
matrices that introduce time variation in the disturbances of the measurement equation. Finally, step 4 generates posteriors of
the volatilities associated with measurement and state equations. All these posteriors are contingent on the available data and
all the remaining hyper-parameters. A detailed outline of the MCMC algorithm is available upon request.
15
Standard tests fail to reject the null of a unit root in each endogenous variable, thus all variables enter the system in first
differences.
16
Measured by the seasonally adjusted 2005 chain-weighted GDP (HAVER series GDPH).
17
Two measures of the price level are used (with nearly identical results): the GDP deflator (2005 = 100 seasonally adjusted
HAVER series JGDP) or the CPI for all urban consumers (HAVER series PCU77).
18
The S&P500 index (HAVER series PC5COM) deflated by the appropriate price index serves as the real stock price measure.
This is a commonly employed measure of stock returns (see Binswanger, 2004; Hess & Lee, 1999; Lee, 2010; Rapach, 2001) that
does not include information about dividends. Fama and French (2001) suggest that the “disappearing dividends” phenomenon
renders the long-term characteristics of cash dividends as fundamentally different from (in fact, they would seem diametrically
opposed to) stock returns. Thus, we do not consider dividend information here.
The underlying structural shocks to this economy εt = {εAS PO IN

t εt εt } —which consist of a shock in
aggregate supply (‘AS’), portfolio (‘PO’), and inflation (‘IN’), respectively—are identified according to
two well-documented macroeconomic hypotheses: one is long-run neutrality and the other is the
natural rate hypothesis.19 One of the main tenets of the natural rate hypothesis is that the single
source of non-stationarity in real output stems from disturbances to aggregate supply. We arrive at
this identification according to the following: first, we restrict exogenous changes to an inflationary
shock20 to have no long-run effect on real output. And second, unanticipated exogenous portfolio
shocks are prevented from affecting real output permanently. Thus, this is assumed to be a “pure”
portfolio shock that represents an exogenous disturbance to the demand for stocks. This could result
from a shock to transaction costs in the stock market (Tobin, 1969), or an exogenous equity-premium
shock that alters the perceived riskiness of stocks. Thus, only aggregate supply factors that disturb the
natural rate of output will have an impact in the long-run level of real output. This allows us to identify
‘AS’ shocks as the only shocks (typically related to technology) that exert long-run effects on output.
Finally, long-run neutrality restricts shocks to inflation—whether they come from monetary policy or
aggregate demand—from having any permanent impact on any real variable, such as our measures of
stock prices or output. This allows us to determine that whatever portion in the long-run movement
of real stock prices—that does not come from ‘AS’ shocks—cannot come from ‘IN’ shocks and must,
therefore, be an idiosyncratic shock to the stock market itself.21
4. Model II: a five-variable TV SVAR specification
There is a substantial literature on the response of stock prices to macroeconomic disturbances,

such as inflation, that rely on relatively low-dimensional structural models. Hess and Lee (1999),
Binswanger (2004), and Lee (2010) are three examples of bivariate SVAR specifications. There are
advantages in considering structural models that are parsimonious. However, if a low-dimensional
model is employed to characterize a system where the economy is subjected to multiple shocks, the
identified disturbances from the low order representation must be viewed as aggregates of a larger
number of underlying shocks. If the responses to these underlying shocks are qualitatively different,
then the results from the aggregated low-dimensional model may be misleading.22
Thus, in the interest of robustness, this section extends the benchmark model by considering a
five-variable specification. Because of the size of this second specification, the identified shocks are
disaggregated into a larger number of sensible categories that are defined below. This extension is
designed to address two other concerns, namely, sample selection and frequency issues. Results from
Model I—described in Section 5—show that a peak in the volatility of postwar US output growth and
inflation takes place around the Great Inflation period of the 1970s23 followed by evidence of a Great
Moderation that has been substantiated by many others. We argue earlier in this paper that the TV-
SVAR approach can easily handle periods of structural change. And the 1970s could qualify as such,
at least for the US. Thus, we want to test whether our results vary substantially when excluding this
period. Finally, it could also be argued that important dynamics in the relationship between real stock
prices and other macro aggregates could remain latent at quarterly frequencies. Thus, we want to
elucidate how our conclusions change when considering higher frequency data.
19
We arrive at the ensuing matrix of time-varying cumulative long-run multipliers by the imposition of these two long-run
identification restrictions that constitute relatively uncontroversial features of mainstream macro-monetary models.
20
Given that the goal of this paper is to shed light on the relationship between inflation and stock prices, we do not take
a stand on whether this ‘IN’ shock would come from an exogenous decrease in the policy rate or an exogenous increase in
aggregate spending.
21
Notice that this long-run structure does allow for inflationary shocks to affect the long-run levels of inflation. Furthermore,
money neutrality and the natural rate hypothesis, two well-established theories, allow us to identify the underlying structural
shocks affecting this economy under minimal restrictions. These restrictions impose no constraint on the constancy of either
the autoregressive parameters or the lower triangular elements of the covariance matrix of the TV-SVAR.
22
A similar point is raised by Faust and Leeper (1997).
23
Keating and Valcarcel (2011) report similar volatility levels in the 1970s for both variables but argue that this is not a
postwar peak. In fact, they show that the volatilities of US output growth and inflation in the 1970s are overshadowed by higher
volatility levels during, and immediately after, WWII.
Model II is based on xt = (ut yt st pt it ) . We consider monthly data24 on the Bureau of Labor
Statistics (BLS) Household Survey measure of the unemployment rate, the Dow Jones industrial aver-
age (DJIA) (properly deflated), the CPI for all urban consumers, and the three-month treasury bill rate.25
The choice for the second variable in our ordering deserves further elaboration. A common measure of
real economic activity at monthly frequencies is industrial production. However, this measure is sub-
ject to data revisions and provides low coverage—representing about 20% of the aggregate economy.
Thus, we employ the Chicago Fed Net Activity Index (CFNAI)—an index generated by the first princi-
pal component extracted from 85 separate economic series—as our measure of aggregate economic
activity. Benefits of this measure are its broad coverage, strong contemporaneous correlation to real
output, and relatively low sensitivity of the headline index to revisions in the underlying series.26
To identify the effects of shocks to each of the five variables, we use a long-run identification
scheme that follows and extends that of Model I. By ordering the unemployment rate first in a long-
run recursive structure, we posit that the only shock, in our model, that may have a permanent effect
on the unemployment rate is a shock to its natural rate (‘NRU’ shock).27 Innovations in real output
and real stock prices28 retain their interpretation from Model I, ‘AS’ and ‘PO’ shocks, respectively.
By incorporating a measure of short-term interest rates, to be included last in our ordering, the last
two shocks are identified by noting that while money supply shocks29 may have a long-run impact
on inflation, shocks to money demand should not have a permanent effect on inflation30 when the
Federal Reserve can be expected to counter such shocks in an effort to anchor the long run level of
inflation. Thus, the last two shocks in our system can be interpreted as money supply (‘MS’) and money
demand (‘MD’) shocks, respectively.
Structurally, Model II extends the specification of Model I in the previous section in a natural way.
Essentially both the aggregate supply shocks and inflationary innovations of Model I are each further
disaggregated into two types of disturbance. Thus, Model II contain two types of aggregate supply
shocks: one that results in permanent changes in both output and unemployment (the first shock in
our ordering), and one that also has a permanent effect on output but only a transitory effect on the
unemployment rate (the second in our ordering). The inflationary shocks of Model I are also further
disaggregated into underlying shocks: one that affects the long run level of inflation and short-term
rates (fourth shock in our ordering), and one that may have a permanent impact on rates but only a
transitory effect on inflation (the fifth, and last, shock identified).
All inference drawn from the analysis of second moments of the system, as described in Section 2,
allows for time variation. Thus, Section 5 reports correlation and volatility estimates and Sections 6
and 7 show results from the innovation analysis for both Model I31 and Model II.32
24
Our sample excludes the Great Inflation period but includes the Volcker Disinflation of the late 1970s, Great Moderation
period of the mid-to-late 1980s and Great Recession period of the late 2000s (1978:1–2011:7).
25
We also considered the S&P500 composite index as an alternative measure of stock prices and the Federal Funds rate as an
alternative measure of short-term interest rates with little change to our results.
26
The index is released monthly by the Federal Reserve Bank of Chicago but was originally generated by Stock and Watson
(1999), Bernanke et al. (2005), and Clark and Davig (2011) outline benefits of this series in improving the analysis of the monetary
policy transmission mechanism and inflation expectation dynamics.
27
This assumption is consistent with Phelps (1994) definition of the natural rate of unemployment as characterized by a stable
process that, even in the absence of shocks, is probably not constant over time, but converges to a unique long-run equilibrium.
The ensuing steady-state unemployment rate is contingent on other inputs of a production function, such as physical and
human capital, technology etc., that might be used to motivate a Walrasian-type equilibrium. King and Morley (2007) motivate
a similar restriction by pointing out that all that is required for a steady-state solution is identification of the aggregate impact
of the contributing structural factors and not necessarily a full specification of all the underlying determining components.
28
The second and third shocks in Model II, which were the first and second in Model I.
29
As determined by adjustments in the nominal interest rate in excess of the Federal Reserve’s reaction to fluctuations in
inflation and output.
30
As well as all the real variables included earlier on in the ordering.
31
A trivariate specification conditional on US quarterly data spanning the 1960s through 2011.
32
A five-variable specification of US monthly data from January 1978 to July 2011.
-3
a x 10 Posterior Time-Varying Standard Deviations: 1959-2011
12 Output Growth
Inflation
11
10
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010
Posterior Time-Varying Standard Deviation: 1959-2011

b
Real Stock Price Growth
0.12
0.11
0.1
0.09
0.08
0.07
0.06
0.05
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010
Fig. 1. (a) Posterior time-varying standard deviations: 1959–2011, (b) Posterior time-varying standard deviation: 1959–2011.
5. Time variation in the drift and variance of the system
Fig. 1 reports how the posterior standard deviations of the growth rates of output and the price
level (Fig. 1(a)) and real stock prices (Fig. 1(b)) have evolved from 1959 to 2011.33 The volatility of
output growth sees a 20% increase between 1960 and 1980, while the volatility in inflation more than
doubles over the same period. Consistent with the Great Moderation, both variables show a substan-
tial reduction in their standard deviations starting around 1980, which continues well into the mid
1990s—when volatility in both variables seems to settle to an all-time low (within the sample). These
volatility levels remain low for about a decade. By the mid 2000s, our estimates show that a sub-
stantial resurgence in volatility begins to take place so that, by the 2008 financial crisis, the posterior
standard deviation of inflation returns to levels comparable with those of the Great Inflation period
of the 1970s—and output volatility returns to levels not seen since before the Great Moderation.34 In
33
Estimates from Model I.
34
This finding is consistent with that of Clark (2009) who concludes that the US Great Moderation may have ended. Keating
and Valcarcel (2011a) also find that the stability gains associated with the Great Moderation have been all but lost in the US by
contrast, Fig. 1(b) shows no such similar evidence of persistent declines in the volatility of real stock
prices over the same period. It is worth mentioning that the lowest estimated levels of volatility in
this variable also take place around the mid 1990s.35
The three panels of column Fig. 2 report Model I time-varying standard deviations of output growth
(Fig. 2(a)), growth of real stock prices (Fig. 2(b)), and inflation (Fig. 2(c)) conditional on each structural
shock.36 These are reported side-by-side with each panel of column Fig. 3, which shows estimates of
Model II for similar variables.
Fig. 2(a) shows that the decrease in volatility of aggregate economic activity can be explained by a
decrease, in order of contribution, of aggregate supply, portfolio, and inflation shocks to the volatility of
output. The standard deviation of output conditional on the ‘AS’ shocks seems to decrease by roughly
50% over the same time span as the unconditional variance (1980 through 1994). Fig. 3(a) shows
that the volatility in economic activity—now measured by a different variable, at higher frequencies,
and subject to a larger number of differentiated shocks—is largely driven by similar dynamics. The
relative contributions of supply and portfolio shocks (‘AS’ and ‘PO’ in Model I and ‘S’, ‘NRU’ and ‘PO’
in Model II) explain a relatively larger portion of the volatility of output. Overall, the contributions of
nominal shocks (‘IN’ in Model I, and ‘MS’ and ‘MD’ in Model II) seem relatively marginal in explaining
higher frequency movements in short-term output volatility but do an adequate job of explaining
low-frequency movements in output volatility.
Both Model I and Model II, as described in Figs. 2(b) and 3(b), respectively, suggest that the overall
pattern in the time-varying volatility of real stock prices is fairly well-explained by the portfolio shock.
Stock price volatility contingent on ‘IN’ shocks (in Model I) or ‘MS’ and ‘MD’ shocks (in Model II) remains
much smoother and less important in explaining the unconditional volatility fluctuations of real stock
returns. Perhaps not surprisingly, this evidence suggests that it is stock market demand that drives
the overall high-frequency movements in equity prices. It is worth noting that ‘AS’ shocks in Model
I or ‘NRU’ and ‘S’ shocks in Model II have done a fairly good job of tracking peaks in the volatility of
stock returns throughout the sample. Figs. 2(b) and 3(b) are particularly interesting with regard to
two specific shocks: the ‘IN’ shock (in Model I)—suggesting that inflationary shocks are shown to have
contributed to the high stock market volatility of the mid 1970s and early 1980s, as well as the high
stock market volatility of the late 2000s—and, the ‘MS’ shock (in Model II) that had a larger contributing
role in the stock market volatility spike of the late 2000s than at any other time since the early 1980s.
Despite this contribution of inflation and money supply disturbances in these high volatility periods,
it is portfolio demand and aggregate supply shocks that seem to drive the lion’s share of the stock
market volatility over the sample.
The last two charts in each column of Figs. 2 and 3 present complementary findings. Fig. 2(c) shows
that, according to Model I, the trends in the volatility of inflation are best explained by both inflation
and supply shocks.37 According to Fig. 3(c), Model II suggests that money supply shocks are the largest
contributor of inflation volatility since the 1980s. Thus, the bulk of the volatility patterns in inflation
are explained by inflationary or money supply shocks; those of stock prices are well-explained by
shocks to equity demand; and, finally, all three shocks seem to exhibit similarities in the timing and
magnitude of volatility breaks in output, with ‘AS’ shocks as the most important contributing factor.38
Figs. 4 and 5 show Model I and Model II estimates, respectively, of the time-varying correlations
between the growth rates of stock prices and US inflation.39 Fig. 4 also shows estimates of that corre-
the time the 2008–2009 financial crisis hits. However, they show that the gains in stability achieved during the postwar period
have not yet been lost.
35
These volatility patterns, however, show some pro-cyclical behavior (high in the mid-1970s, low in the mid-1990s with
further decreases between 2004 and 2006), but unlike the standard deviations of output growth and inflation, the volatility of
real stock prices show substantial mean-reversion qualities.
36
The top solid line shows again the unconditional variances reported in Fig. 1.
37
Although the portfolio shocks track the Great Inflation period fairly well.
38
Indeed the ‘AS’ shocks seems an important contributor to the volatility of all three variables.
39
The confidence bounds represent estimates of the posterior estimates of one standard deviation around the correlation
coefficient estimates. Given the nature of this empirical approach, the confidence bounds associated with the estimates of drift
are substantially wider than those associated with the volatility estimates and, thus, they abate our inclination to draw too
strong an inference.
(a) Posterior Standard Deviation for

-3
Real GDP Growth
x 10
12 AS
PO
11
IN
10 Unconditional
2
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010
(b) Posterior Standard Deviation for

AS
0.12 PO
0.11 IN
Unconditional
0.1
0.09
0.08
0.07
0.06
0.05
0.04
0.03
0.02
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010
-3
(c) Posterior Standard Deviation for
x 10
8
AS
PO
7 IN
Unconditional
6
1
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010
Fig. 2. US conditional time-varying standard deviations: (1959–2011) Model I.

(a) Posterior Standard Deviation for

-3
Real Net Activity Growth
x 10
4.5
NRU
4 S
PO
3.5 MS
MD
3 unconditional
2.5
1.5
0.5
1985 1990 1995 2000 2005 2010
(b) Posterior Standard Deviation for

0.06
NRU
S
PO
0.05
MS
MD
unconditional
0.04
0.03
0.02
0.01
1985 1990 1995 2000 2005 2010
-3
(c) Posterior Standard Deviation for
x 10
5 NRU
S
4.5
PO
4 MS
MD
3.5 unconditional
2.5
1.5
0.5
1985 1990 1995 2000 2005 2010
Fig. 3. US conditional time-varying standard deviations: (1981–2011) Model II.

(a) Unconditional Correlation S&P500 - US CPI (b) Correlation S&P500 - US CPI: Conditioned on 'PO'
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
-0.2 -0.2
-0.4 -0.4
-0.6 -0.6
-0.8 -0.8
-1 -1
1960 1970 1980 1990 2000 2010 1960 1970 1980 1990 2000 2010
(c) Correlation S&P500 - US CPI: Conditioned on 'AS' (d) Correlation S&P500 - US CPI: Conditioned on 'IN'
1
1
0.8
0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
-0.2 -0.2
-0.4 -0.4
-0.6 -0.6
-0.8 -0.8
-1 -1
1960 1970 1980 1990 2000 2010 1960 1970 1980 1990 2000 2010
Fig. 4. US conditional time-varying correlations Model I: (1959–2011).
lation conditioned on each of the three shocks specified in Model I. Correspondingly, Fig. 5 shows the
conditional stock price-inflation correlation subject to each of the five shocks specified in Model II.40
According to Figs. 4(a) and 5(a), the correlation between the growth rates of stock prices and infla-
tion remains very small and mostly negative. This is contrary to the standard theoretical predictions
but consistent with most of the empirical literature, as mentioned in the introduction. The benefit
of our approach is that—rather than finding an estimate of that relationship that is averaged for the
whole sample—it allows us to find a correlation estimate at each period and, thus, determine whether
the relationship is changing. Indeed, we find that this correlation seems to flip back and forth from
the negative to the positive region throughout the sample, although it remains small.41 As Fig. 4(b)
depicts, when conditioning on the portfolio shock, the correlation between the growth rate of stock
prices and inflation seems mostly negative throughout the sample. This would suggest that surprises in
demand for equities—insofar as they provide a substitute for other financial assets, as well a deferred
current consumption—could serve to drive stock prices to move in the opposite direction of other
prices. Fig. 5(b) suggests that this result is qualitatively robust to different measures of stock prices,
different frequency in the data, and the introduction of more dynamics. Fig. 4(c) shows that when
conditioning on the ‘AS’ shock, the correlation between the growth rate of stock prices and inflation
is strongly negative throughout the sample. Fig. 5(c) and (d) shows that this relationship contingent
on the ‘S’ or ‘NRU’ shocks remains negative from the late 1990s to the end of the sample. However,
the evidence is less clear in the mid 1990s, when both correlations turn positive, if only briefly in the
case of the ‘NRU’ shock-contingent correlation. Even if one were to think of the ‘S’ and ‘NRU’ shocks in
Model II to be an amalgam of the ‘AS’ shock of Model I—with the ‘NRU’ being of primary importance in
the stock price-inflation relationship—in the 1990s, that relationship would seem negative according
40
Recall that Model II disaggregates the ‘AS’ shock of Model I into ‘NRU’ and ‘S’ shocks, and Model I ‘IN’ shocks into ‘MS’ and
‘MD’ disturbances.
41
We term these ünconditional correlationsïn Figs. 4(a) and 5(a), to denote that, unlike the other charts [(b)–(f)], they are not
contingent on any given shock. They are still, however, correlation estimates that are conditional on time.
(a) Unconditional Correlation DJIA - US CPI (b) Correlation DJIA - US CPI: Conditioned on 'PO'
1 1
0.5 0.5
0 0
-0.5 -0.5
-1 -1
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
(c) Correlation DJIA - US CPI: Conditioned on 'S' (d) Correlation DJIA - US CPI: Conditioned on 'NRU'
1 1
0.5 0.5
0 0
-0.5 -0.5
-1 -1
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
(e) Correlation DJIA - US CPI: Conditioned on 'MS' (f) Correlation DJIA - US CPI: Conditioned on 'MD'
1 1
0.5 0.5
0 0
-0.5 -0.5
-1 -1
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
Fig. 5. US conditional time-varying correlations Model II: (1981–2011).
to Model I and positive according to Model II. Thus, when conditioning on aggregate supply charac-
teristics, the stock return-inflation relationship seems sensitive to the frequency of the measurement.
According to Fig. 4(d), when conditioning on the ‘IN’ shock, this correlation is consistently positive.
Inspection of Fig. 5(e) suggests that a qualitatively similar prediction ensues when conditioning on the
‘MS’ shock. However, the relationship is consistently negative when conditioning on the ‘MD’ shock,
according to Fig. 5(f). This would suggest that the inflationary disturbances specified by Model I have
similar dynamics to the money supply disturbances specified in Model II. Thus, a monetary expansion
could have a positive income effect that might serve to move prices of stocks in the same direction as
inflation. Conversely, if money demand is mostly determined by agents’ preferences, it is reasonable
to expect a substitution effect between equities and monetary assets, as a function of liquidity, that
can serve to drive stock prices in the opposite direction of inflation. Overall, shocks to preferences, be
it for equities (‘PO’) or other more liquid monetary assets (‘MD’), seem to drive similar dynamics in the
stock returns-inflation relationship that are qualitatively opposite to the effects of inflationary (‘IN’)
or money supply (‘MS’) shocks.
Results from both Model I and Model II show that the relationship between inflation and stock
prices is found to be consistent with most of the empirical literature; namely, a negative and
economically weak relationship that is in stark contrast to the maintained strong and positive theo-
retical predictions. It is this weak empirical relationship that has motivated the growth of research in
this area. Importantly, results from both models show that a clear picture on that relationship emerges
when conditioning on the appropriate shocks. These estimates suggest that the ambiguity in the cor-
relation between stock prices and inflation that plagues the data could stem from the conclusion that
two disturbances that have a temporary effect on output—idiosyncratic stock market (‘PO’) shocks
and the inflationary (‘IN’), or money supply ‘MS’ shocks—seem to counterbalance each other in driv-
ing the relationship. And the effect of the disturbances that have permanent effect on output—shocks
to aggregate supply (‘AS’ or ‘S’) and the natural rate of unemployment (‘NRU’)—could be sensitive to
frequency issues.
Taken together, Figs. 2–5 carry three important implications: First, while the variances of stock
prices and inflation have changed substantially over time, these changes have been much more gradual
for inflation than stock prices. Second, the correlation between these variables is estimated to be weak
but appears to have remained relatively more stable than the variances. Lastly, and perhaps more
importantly, the weak relationship between inflation and stock prices is a function of the offsetting
effects of the underlying structural shocks.
6. The evolution of the responses to supply, portfolio, and inflation shocks: Model I
An important advantage of the TV-SVAR approach, employed here, over more traditional ones is
that it allows us to estimate the response of a given shock at each time t, rather than an averaged
response when the drift and covariance of the system are assumed constant over the whole sample or
chosen sub-samples. Fig. 6 describes results from estimated time-varying impulse response functions
in Model I.
Fig. 6(a)–(i) shows estimates from Model I on the responses for output, stock prices, and inflation to
the equivalent of a normalized one standard deviation impulse in the ‘AS’, ‘PO’, and ‘IN’ structural shocks
(column-wise) for various horizons. Each chart shows a cluster of three panels: The top panel shows
how the first two-quarter response of the relevant variable to the relevant shock has evolved over
time. The middle panel shows the second and third year (quarters 5 through 12) average responses.
The bottom panel averages the response over the fourth and fifth years post shock.42
Fig. 6(a) and (c) shows that the ‘AS’ and ‘IN’ shocks, respectively, have a positive effect on real
output. In the case of the ‘AS’ shock, the effect is permanent while the effect of the ‘IN’ shock seems
fairly persistent and substantially larger in the high inflation periods, with a substantial reduction
thereafter. Both of these responses are qualitatively consistent with standard VAR predictions. What
is more informative here, relative to more traditional models, is that the magnitude of the long-term
response of output to the ‘AS’ shock and the short-term output response to the ‘IN’ is much higher
in the late 1970s and early 1980s. The long-term ‘AS’ effects and short-term ‘IN’ effects have gradu-
ally but substantially decreased since the mid 1980s, bottoming out by the mid 1990s. But, perhaps
most interestingly, by the 2008–2009 period the magnitudes of both responses are comparable to
those found in the late 1970s period—undoubtedly driven by the higher volatility levels during that
period—shown in the previous section.
Fig. 6(b) shows that the ‘PO’ shock has a negative short-run effect on real output that again peaks
during and immediately after the Great Inflation period. This is an interesting result as one would think
that increases in the demand for stocks would be pro-cyclical. However, this feature can be reconciled
with standard theoretical predictions through the following mechanism: an exogenous decrease in the
risk premium associated with stocks could be naturally followed by a drop in demand for substitute
assets, which would undoubtedly lead to a fall in bond prices and an increase in interest rates, thereby
lowering aggregate demand. This interpretation is consistent with the evident temporary declines in
42
It is worth pointing out that these graphs are derived from the time-ordered impulse response for each year and, then,
this response is averaged over a specified horizon. Therefore, these responses constitute an “independent” point estimate for
each year. Each annual point estimate has an implied variance associated with it that was extracted from a standard bootstrap
procedure. Thus, these graphs allow us to glimpse how the average effects for the short, medium and long term might have
changed throughout the sample where each point estimate, and confidence bound associated with it, is discrete in a time sense.
(a): Output Response to 'AS' Shock (b): Output Response to 'PO' Shock (c): Output Response to 'IN' Shock
0
S-R 2 0.6
0.4
1 -0.5
0.2
0
0 -1
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010
0
M-R 2 0.6
-0.5 0.4
1 0.2
0
0 -1
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010
0
L-R 2 0.6
-0.5 0.4
1 0.2
-1 0
0
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010
(d): Stock Price Response to 'AS' Shock (e): Stock Price Response to 'PO' Shock (f): Stock Price Response to 'IN' Shock
S-R 10 8
10 6
5 5 4
2
0 0 0
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010
10 8
M-R 10 6
5 5 4
2
0 0 0
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010
10 8
L-R 10 6
5 5 4
2
0 0 0
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010
(g): Inflation Response to 'AS' Shock (h): Inflation Response to 'PO' Shock (i): Inflation Response to 'IN' Shock
0 0 0.3
S-R -0.2 -0.1 0.2
-0.2 0.1
-0.4
0
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010
0 0 0.3
M-R -0.1
-0.2 0.2
-0.2 0.1
-0.4
0
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010
0 0 0.3
L-R
-0.2 -0.1 0.2
-0.2 0.1
-0.4
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 0
1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010
Note: S-R Response refers to the first two quarter average response post shock. M-R Response to the 5th through 12th quarter and
L-R Response to the 13th to 20th quarter post shock.
Fig. 6. Time-varying responses at multiple horizons: Model I (1959–2011).
inflation following the same shock.43 Thus, taking together the evidence from Fig. 6(b) and (h), the
effects of the portfolio shock on output and inflation look like an adverse aggregate demand shock.
Fig. 6(d) and (e) show that the ‘AS’ and ‘PO’ shocks have positive and permanent effects on real stock
prices.44 The responses show considerable time variation, likely due to the inherent idiosyncratic
volatility in stock returns. According to Fig. 6(f), an exogenous inflationary expansion brings about
a short-run increase in real stock prices (though the increase diminishes considerably in the mid
1990s). This result is qualitatively consistent with Lastrapes (1998) and Rapach (2002) who use more
standard approaches. A plausible explanation is that whether this shock was a reflection of news
of an expansionary policy or exogenous increase in aggregate demand, it would lead investors to
anticipate an increase in short-run earnings while the ensuing reduction in interest rates would lower
the discount rate for future cash flows. Both of these would lead to increases in real stock prices. This
response peaks in the mid 1970s and early 1980s and decreases after that period, finally bottoming out
43
See Fig. 6(h).
44
The positive response of stock prices to a ‘PO’ shock (Fig. 6(e)) is not surprising. Fig. 6(d) shows that the time-varying
response of real stock returns to an exogenous ‘AS’ shock is always positive. The permanently higher output levels after an ‘AS’
shock described earlier (in Fig. 6(a)) likely raise real earnings, which in turn may serve to increase demand for asset claims
and, thus, put upward pressure on their returns. This result is consistent with the present value equity valuation predictions of
finance.
Table 2
Estimates of conditional volatilities for output.
Shock Standard deviation Est. % Relative standard Est. %

deviation
Pre-1990 Post-1990 Decrease Pre-1990 Post-1990 Decrease
1. Supply 0.0072 0.0054 25 0.21 0.09 57

2. Portfolio 0.0352 0.0585 (66) 1.00 1.00 –
3. Inflationary 0.0032 0.0011 65 0.09 0.02 77
Note: The DGPfor each endogenous variable takes the MA components as the true coefficients, which are contingent on the
realizations of the structural shocks for each period.
Table 3
Estimates of conditional volatilities for stock prices.
Shock Standard deviation Est. % Relative standard Est. %

deviation
1. Supply 0.0056 0.0027 52 0.08 0.05 28

2. Portfolio 0.0632 0.0513 19 1.00 1.00 –
3. Inflationary 0.0022 0.0006 72 0.03 0.01 67
Note: See footnote of Table 2.
in the mid 1990s, before stock prices become more responsive again to an inflationary shock around
the Great Recession period.
Fig. 6(g) shows that ‘AS’ shocks have a significantly negative short-run response in the US inflation
rate, especially in the mid 1970s and the 2008–2009 financial crisis. Fig. 6(h) shows a similar inflation
response—both qualitatively and in magnitude—to the ‘PO’ shocks with the possible exception at the
end of the sample when the inflation response is more muted to the ‘PO’ shocks than the ‘AS’ shocks.
Perhaps unremarkably, Fig. 6(i) shows a positive response of inflation to ‘IN’ shocks. What is more
surprising is that the magnitude of the response is as high, if not higher, in the 2008–2009 Great
Recession period than during the Great Inflation period of the 1970s. Taken together, Fig. 6(g)–(i) show
that while the 2008–2009 Great Recession may have been preceded by a financial crisis, inflation in
that period seems much more responsive to aggregate macroeconomic shocks than shocks in the stock
market.
A summary of Fig. 6 sheds light on a number of interesting relationships. First, the magnitude of
the short-run effects of all three shocks ‘AS’, ‘PO’ and ‘IN’ in Model I seem to be driven by the volatility,
which is generally high in the 1970s, reduced in the late 1980s and most of the 1990s, and higher
again during the 2008–2009 Great Recession period. Second, both ‘AS’ and ‘PO’ shocks seem to lead to
permanent increases in real stock prices.45 Indeed, stock prices respond positively to every shock spec-
ified in this model. This result is undoubtedly driven by the correlation patterns described in Fig. 4.46
Finally, and as expected, an expansionary aggregate supply shock leads to increases in real output and
decreases in the inflation rate. The increase in real output, along with expectations of boosts in real
earnings, serves to prop up real stock prices.
Tables 2–4 complement the evidence presented in Fig. 6 by reporting the conditional standard
deviations of the estimated supply, portfolio and inflation components of the growth rates of output,
stock prices, and the price level, respectively.47 Table 2 shows a decrease in the volatility of output
45
The magnitude of this positive response, however, fluctuates widely over the sample under study.
46
which can be reconciled with Fama’s (1981) PEH only if, in the correlation between real activity and stock returns, we
let inflation be a proxy for the former. Under Fama’s PEH assumptions, inflation and real activity are negatively related, while
inflation (as a proxy for real activity) and stock returns are positively correlated. These are the factors used by the PEH to explain
the absence of a strong positive correlation between stock returns and inflation as a spurious irregularity. This model, however,
shows that a lack of strong correlation is easily explained when decomposing it between the contributing shocks.
47
Results from these tables exclude the Great Recession period and its aftermath in order to test whether the estimates
experience reduction in volatility consistent with the Great Moderation.
Table 4
Estimates of conditional volatilities for inflation.
Shock Standard deviation Est. % Relative standard deviation Est. %

1. Supply 0.0024 0.0011 52 0.07 0.07 –

2. Portfolio 0.0348 0.0138 60 1.00 1.00 –
3. Inflationary 0.0038 0.0021 45 0.11 0.15 (36)
Note: See footnote of Table 2.
conditional on ‘AS’ or ‘IN’ shocks since 1990. Conversely, output volatility conditional on the ‘PO’ shock
seems to have actually increased post-1990.48 While Table 1 shows real stock prices experienced no
such volatility decrease, Table 3 shows a marked decrease in their volatility conditioning on ‘AS’ or ‘IN’
shocks. However, the stock price volatility reduction conditional on the ‘PO’ shock is much more muted.
Finally, Table 4 underscores how the substantial (nearly 50%) reduction in volatility of inflation49 can
be well-explained by a reduction in volatility of every component (52% reduction contingent on ‘AS’
shocks, 60% contingent on ‘PO’ shocks, and 45% conditional on ‘IN’ shocks).
According to these tables, we do not see evidence of a Great Moderation in stock prices commensu-
rate with that of output and inflation in the US. Model I estimates suggest that demand for equities has
continued to remain highly volatile throughout the postwar period. Further, the model estimates that
all three endogenous variables are at their most volatile when conditioned on ‘PO’ shocks throughout
the sample. Finally, the moderation in output is not explained well by stock price disturbances but by
a combination of lower volatility in output when subjected to ‘AS’ or ‘IN’ shocks.
7. The evolution of the responses to supply, portfolio, and monetary shocks: Model II
As mentioned in Section 4, Model II extends Model I in a number of different avenues involving

specification and measurement. First, we bring added variables such as the unemployment rate and
short-term rates to bear in the analysis. Second, we use alternative measures of output and stock
prices, such as the CFNAI and DJIA, instead of GDP and S&P500. Finally, we use monthly data starting
in 1978:1 rather than quarterly from 1959.50 Figs. 7–9 show time-varying responses of output, stock
prices, and inflation, respectively, to the identified shocks in Model II.51 As mentioned in Section 4,
the identification restrictions in Model II suggest that the ‘NRU’ and ‘S’ shocks could be viewed as a
disaggregation of the ‘AS’ shock of Model I. And similarly, the ‘MS’ and ‘MD’ could be interpreted as the
underlying shocks of the identified ‘IN’ disturbance of Model I. The responses of Model II, described
here, are generally consistent with those of Model I, but also provide further insights. Each of the
charts (a)–(e) in each figure is composed of four panels corresponding, from top to bottom, to the 1-,
6-, 12-, and 18-month response post-shock.
Inspection of Figs. 7–9 yields a number of important insights. The responses of output and real stock
prices to the ‘S’ shock, as described in Figs. 7(b) and 8(b), respectively, are persistently positive. The
responses of output (Fig. 7(c)) and inflation (Fig. 9(c)) to the ‘PO’ shock are negative in the short-run,
while the stock price response (Fig. 8(c)) is permanently positive. All these responses are qualitatively
consistent with those of Model I.52 The inflationary ‘IN’ shocks that were specified in Model I are
48
These results do not settle the “good luck” vs. “good policy” debate. To the contrary, they would seem to imply that the
Great Moderation (at least) in US output can be explained by a “perfect storm” marked by the reduction of both adverse supply
shocks and inflation shocks to economic activity that seems to have characterized the 1990s.
49
As measured by the GDP deflator.
50
The constraining factor for starting in 1978 is that, in the interest of robustness, we extend this model with inclusion of the
Michigan Survey of the CPI as a measure for expectations of inflation. The estimation is initialized by a training period of 36
months so estimates run from 1981:1 to 2011:7.
51
In the interest of a direct comparison with results of Model I and to save space, the unemployment and interest rate
responses—although generally consistent with textbook macro prediction—are not reported.
52
A point of distinction arises when investigating the inflation response to aggregate supply. Model I predicts a negative
inflation response to supply shocks over the sample. Of the two supply shocks allowed for in Model II, it is the ‘NRU’ shock and
not the ‘S’ shock that carries the negative relationship.
(a): Output Response to 'NRU' Shock (b): Output Response to 'S' Shock 7(c): Output Response to 'PO' Shock
0 0
1-
period -0.2 0.5
-0.1
ahead -0.4
0 -0.2
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
0 0
6-
periods -0.2 0.5
-0.1
ahead -0.4
0 -0.2
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
0 0
12-
periods -0.2 0.5
-0.1
ahead -0.4
0 -0.2
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
0 0
18-
periods -0.2 0.5
-0.1
ahead -0.4
0 -0.2
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
(d): Output Response to 'MS' Shock (e): Output Response to 'MD' Shock
0.2
1- 0.05
period 0.1
0
ahead
-0.05 0
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
0.2
6- 0.05
periods 0.1
0
ahead
-0.05 0
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
0.2
12- 0.05
periods 0.1
0
ahead
-0.05 0
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
0.2
18- 0.05
periods 0.1
0
ahead
-0.05 0
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
Fig. 7. Time-varying US output responses at multiple horizons: Model II (1981–2011).
now disaggregated in Model II into disturbances that—while having a straightforward relationship

with the inflationary ‘IN’ shocks—stem from inherently different sources; namely, policy and agents’
preferences. Thus, the responses of Model II to ‘MS’ and ‘MD’ shocks—in relation to the ‘IN’ shocks of
Model I—deserve further elaboration.
Fig. 7(d) shows a short-run positive response of output to ‘MS’ shocks in the 1980s and early-
to-mid 2000s, while Fig. 7(e) shows a monotonically positive short-run output response to the ‘MD’
shock throughout the sample. Both of these responses are generally consistent with the corresponding
output response to the ‘IN’ shock in Model I. But they also suggest that since the 1980s, inflationary
pressures that have resulted in increases in output are more likely to have come, compositionally,
from shocks to agents’ liquidity preferences than from monetary policy shocks.53
Recall that Model I predicts, not surprisingly, that inflationary ‘IN’ shocks monotonically increase
inflation from 1959 to 2011. Fig. 9(d) suggests that higher levels of inflation ensue when those infla-
tionary disturbances come from an exogenous increase in the money supply (‘MS’ shocks). However,
exogenous increases in money demand—insofar as they might serve to prop up short-term interest
rates—should have a contractionary effect on inflation. Fig. 9(e) shows evidence consistent with this
prediction. Importantly, the short-run decrease in US inflation—after an ‘MD’ shock hits—is shown to
be larger in magnitude in the 1980s than the period between the mid 1990s and mid 2000s—when
53
This does not suggest that monetary policy has been ineffectual since the 1980s. Quite the opposite, the implication is that
the Federal Reserve may have become more forward-looking since the 1980s, and better able to anchor long-run inflation (see
Clarida et al., 2000), and, thus, output volatility (at least until the Great Recession period, as this paper shows), with output
becoming generally less sensitive to changes in the money supply.
(a): Stock Price Response to 'NRU' Shock (b): Stock Price Response to 'S' Shock (c): Stock Price Response to 'PO' Shock
5 5
0
1- -2
period -4
-6
ahead -8 0 0
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
5
0 5
6- -2
periods -4
-6
ahead -8 0 0
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
5
0 5
12- -2
periods -4
ahead -6
-8 0 0
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
5
0 5
18- -2
periods -4
ahead -6
-8 0 0
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
(d): Stock Price Response to 'MS' Shock (e): Stock Price Response to 'MD' Shock
1.5 2
1- 1
period 1
0.5
ahead
0 0
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
1.5 2
6- 1
periods 1
0.5
ahead
0 0
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
1.5 2
12-
1
periods 1
0.5
ahead
0 0
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
1.5 2
18-
1
periods 1
0.5
ahead
0 0
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
Fig. 8. Time-varying stock price responses at multiple horizons: Model II (1981–2011).
US inflation volatility is at its lowest (as was shown in Fig. 1(a))—or the late 2000s—a period when
nominal rates have approached, and virtually reached, the zero bound.
Finally, Fig. 8(d) shows a short-lived, but important, positive stock price response to an ‘MS’ shock
throughout the sample. This is consistent with the corresponding stock price response to the ‘IN’ shock
of Model I. Fig. 8(e) also shows a positive, but short-lived, response to the ‘MD’ shock. If the increase in
liquidity preferences comes as a result of a perceived income effect, it could serve to drive up demand
for equities. If, on the other hand, the expansionary monetary policy shock leads investors to anticipate
expected future earnings, then this could foster higher activity in equity markets. Both of these would
then serve to increase real stock prices.
Overall, Model II responses to supply (‘NRU’ and ‘S’), portfolio (‘PO’), and monetary (‘MS’ and ‘MD’)
shocks are qualitatively consistent with those of Model I. A bit of a lateral note is that Model II predicts
that an exogenous increase in the US natural rate of unemployment serves to negatively affect output
(Fig. 7(a)), stock prices (Fig. 8(a)), and inflation (Fig. 9(a)) since the 1980s. Insofar as this type of
shock would have a negative impact on the productive capacity of the economy, it could constitute a
permanently negative income effect. Thus, these responses to the ‘NRU’ shock—although not central
to the focus of this paper—are generally consistent with standard textbook predictions.
8. Sensitivity analysis: expectations of inflation and the importance of time variation
Evidence from this paper suggests that the relationship between stock prices and US inflation has
been difficult to pin down because the dynamic responses of both variable to economic shocks have
experienced substantial variation over time and various macroeconomic shocks seem to have had
(a): Inflation Response to 'NRU' Shock (b): Inflation Response to 'S' Shock (c): Inflation Response to 'PO' Shock
0.05 0.1 0
1- 0
period -0.05 0.05 -0.1
ahead -0.1
0 -0.2
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
Response of variable 4 -- to shock in varaible 1 Response of variable 4 -- to shock in varaible 2 Response of variable 4 -- to shock in varaible 3
0.05 0.1 0
6- 0
periods -0.05 0.05 -0.1
ahead -0.1
0 -0.2
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
12- 0.05 0.1 0
0
periods -0.05 0.05 -0.1
ahead -0.1
0 -0.2
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
0.05 0.1 0
18- 0
periods -0.05 0.05 -0.1
ahead -0.1
0 -0.2
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
(d): Inflation Response to 'MS' Shock (e): Inflation Response to 'MD' Shock
0.5 0
1-
period -0.05
ahead
0 -0.1
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
0.5 Response of variable 4 -- t o shock in varaible 4 Response of variable 4 -- t o shock in varaible 5

0
6-
periods -0.05
ahead
0 -0.1
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
0
12-
periods -0.05
ahead
0 -0.1
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
0
18-
periods -0.05
ahead
0 -0.1
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
Fig. 9. Time-varying US inflation responses at multiple horizons: Model II (1981–2011).
offsetting effects on the relationship. A somewhat related, though perhaps fundamentally different
question, would be how do stock prices relate to expectations of inflation? Thus, this section presents
some comparative analysis on how our results of Model II would change if we substitute CPI inflation
by a measure of expectations of inflation, as measured by the consumer survey of expectations of
inflation conducted by the University of Michigan.54
Sections 6 and 7 describe results from Model I and Model II that show that stock prices respond to
aggregate macroeconomic disturbances in a way that is generally consistent with standard economic
theory.55 One of the contributions of this paper is in showing that the sensitivity of the stock market
to aggregate shocks has changed, in some cases substantially, over time. This would suggest the time-
varying approach is crucial in understanding the underlying relationship between stock prices and
macroeconomic factors. Thus, this section also presents results of how the predictions of Model II
would change if we “assumed away” time variation in the specification.
Fig. 11 shows Model II posterior estimates of conditional correlation between stock prices, as mea-
sured by the DJIA, and expectations of inflation, as measured by the Michigan consumer sentiment
index. To facilitate comparison, Fig. 10 describes the correlation estimates between stock prices and
inflation from Model II.56 A striking feature of this comparison is that the unconditional correlation
between stock prices and inflation is generally quite similar to that of stock prices and inflation expec-
tations from 1981 to the early 2000s. For example, the correlation between stock prices and inflation
54
Details regarding the index may be found at http://www.sca.isr.umich.edu/main.php.
55
They respond positively to any aggregate shock that may constitute a positive real income effect or an expansive monetary
effect.
56
Fig. 10 is, thus, a carbon copy of Fig. 5.
(a) Unconditional Correlation (b) Correlation DJIA -US CPI:

DJIA - US CPI Conditioned on 'PO'
1 1
0.5 0.5
0 0
-0.5 -0.5
-1 -1
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
(c) Correlation DJIA - US CPI: (d) Correlation DJIA - US CPI:

Conditioned on 'S' Conditioned on 'NRU'
1 1
0.5 0.5
0 0
-0.5 -0.5
-1 -1
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
(e) Correlation DJIA - US CPI: (f) Correlation DJIA - US CPI:

Conditioned on 'MS' Conditioned on 'MD'
1 1
0.5 0.5
0 0
-0.5 -0.5
-1 -1
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
Fig. 10. US conditional time-varying correlations for stock prices-inflation (1981–2011).
expectations is, generally, weakly negative from the Great Moderation period of the mid 1980s to the
end of the sample. Qualitatively, the stock-price-inflation correlation is also negative over the same
period until the early 2000s, when it turns positive. When conditioning on the ‘PO’, ‘MS’, and ‘MD’
shocks, the correlation between stock prices and either inflation or inflation expectations are quite
close—except, again, at the end of the period. The pattern of the inflation expectations-stock price cor-
relation look quantitatively close when conditioning on the supply side (‘S’ or ‘NRU’) shocks. This is also
the case for the correlation pattern between stock prices and inflation.57 Overall, this evidence suggests
57
However, the pattern for the inflation expectations-stock prices correlation seems more clear. In the first part of the sample,
between 1981 and the early 2000s, inflation expectations and stock prices are negatively correlated, whether we condition on
the ‘S’ or the ‘NRU’ shock before they both turn positive between the mid and late 2000s.
(a) Unconditional Correlation (b) Correlation DJIA - MICH:

DJIA -MICH Conditioned on 'PO'
1 1
0.5 0.5
0 0
-0.5 -0.5
-1 -1
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
(c) Correlation DJIA - MICH: (d) Correlation DJIA - MICHI:

Conditioned on 'S' Conditioned on 'NRU'
1 1
0.5 0.5
0 0
-0.5 -0.5
-1 -1
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
(e) Correlation DJIA - MICH: (f) Correlation DJIA - MICH:

Conditioned on 'MS' Conditioned on 'MD'
1 1
0.5 0.5
0 0
-0.5 -0.5
-1 -1
1985 1990 1995 2000 2005 2010 1985 1990 1995 2000 2005 2010
Fig. 11. US conditional time-varying correlations for stock prices-inflation expectations (1981–2011).
that the correlation between stock prices and inflation expectations is also weakly negative—albeit
more monotonically so than the stock price-inflation relationship. In this case, it seems that a strong
negative relationship when conditioning on the ‘PO’ and ‘MD’ shocks—the shocks to agents’ preferences
over money or stock holdings—seems to be dampened by the strong positive relationship when con-
ditioning on ‘MS’ shocks—shocks stemming from policy action. This is, in essence, a similar conclusion
to the relationship between inflation and stock prices from Model II or Model I.
Finally, Fig. 12 shows Model II impulse responses when averaged over the whole period. While we
lose vital information on the evolution of the response, these provide analog results to those that would
be yielded by standard SVAR approaches.58 Over the whole sample, these responses are qualitatively
58
Given that a constant parameter (linear) VAR is generally nested into a more general time-varying (nonlinear) VAR, Fig. 12
shows what the responses from Model II would look like if we averaged away all the time variation. These would be responses
yielded by a more traditional SVAR approach.
(a): 'NRU' Shock (b): 'S' Shock (c): 'PO' Shock (d): 'MS' Shock (e): 'MD' Shock
x 10
0 0 15
0.04
Output -0.02 10
0.2 -0.04 5
-0.1 0.02
-0.06 0
-0.08 -5
0 0
5 10 15 20 5 10 15 20 5 10 15 20 5 10 15 20 5 10 15 20
(f): 'NRU' Shock (g): 'S' Shock (h): 'PO' Shock (i): 'MS' Shock (j): 'MD' Shock
0
2 0.3
Stock Price 0.2 0.4
-1 1 2 0.1
0.2
0
-2 0 0 -0.1 0
5 10 15 20 5 10 15 20 5 10 15 20 5 10 15 20 5 10 15 20
(k): 'NRU' Shock (l): 'S' Shock (m): 'PO' Shock (n): 'MS' Shock (o): 'MD' Shock
0.01
0.02 0.02 0 0.2
Inflation 0
0 -0.02
0 0.1 -0.01
-0.02 -0.04 -0.02
-0.02
0
5 10 15 20 5 10 15 20 5 10 15 20 5 10 15 20 5 10 15 20
(p): 'NRU' Shock (q): 'S' Shock (r): ' PO' Shock (s): 'MS' Shock (t): 'MD' Shock
0.04 0
0 0.02 0 -0.05
Inflation 0.2
0
-0.05 -0.02 -0.1
-0.02
Expectations -0.04 -0.15
-0.1 -0.04
0
5 10 15 20 5 10 15 20 5 10 15 20 5 10 15 20 5 10 15 20
Fig. 12. US (1981–2011) average responses: Model II.
identical to those of the TV-SVAR described in Figs. 7–9. However, Sections 3 and 4 in this paper
have shown that output, real stock prices, and inflation have experienced important changes on the
dynamics of their response to the identified shocks. Importantly, these dynamics remain latent in a
model that assumes away that time variation. For example, Fig. 12(a) shows a permanently negative
response of output to an increase in the natural rate of unemployment, and Fig. 12(e) shows a positive
response of output to an exogenous increase in money demand. Neither response, however, tells us
that the magnitude of these responses seem much more important in the 1980s—a period of generally
higher volatility in output growth—than in the 1990s.
Another important example regards the stock price response to monetary shocks. Fig. 12(i) and
(j) shows that real stock prices increase following an exogenous increase in money supply or money
demand. However, these responses do not yield much insight on the timing sensitivity to the distur-
bances. It is only when inspecting the corresponding time-varying responses in Fig. 8(d) and (e) that
one can easily see that the response to money demand has been much more muted during the late
2000s at the same time that the response to money supply shocks has been much more important.59
Finally, the expectations of inflation responses to each shock are generally qualitatively consistent
with those of actual inflation. An exogenous increase in the money supply—insofar as it serves to put
downward pressures on interest rates—leads to a permanent increase in both inflation expectations
and actual inflation. Conversely, exogenous increases in money demand—with the ensuing upward
pressure on rates—lead to negative responses of inflation and inflation expectations. Interestingly, the
negative inflation response to an ‘MD’ shock is quite transitory but the same disturbance seems to
have a much more lasting effect on expectations of inflation.
9. Conclusion
While theoretical predictions establish a strong positive relationship between equity prices and
inflation, finding substantiating empirical evidence has been a difficult endeavor. The data suggests a
weak negative relationship between stock prices and inflation. Reconciling the empirical irregularity
59
This would suggest that during the Great Recession period of 2008–2009, the stock market has responded substantially to
signals of the strong accommodative stance of the Federal Reserve, and that liquidity constraints seem to have had little role in
driving stock prices in that period.
with the theory has instigated the growth of research in this area, and is one of the central motivations
of this paper. Aided by different specifications, variable choice, and measurement frequencies, this
paper shows that a clear picture emerges on the relationship between stock prices and US inflation
when we condition on appropriate shocks.
Relying on theoretically motivated and relatively uncontroversial long-run restrictions, we ana-
lyze the effects of shocks in aggregate supply, monetary factors, and market demand for equities on
the real value of a broad index of US stock prices. In light of the substantial volatility reduction of
major macroeconomic aggregates of the mid 1980s, we provide time-varying impulse response func-
tions for two different specifications. We consider a three-variable TV-SVAR of quarterly measures
of real output, real stock prices, and inflation to identified inflationary, portfolio, and aggregate sup-
ply shocks. Furthermore, we extend this benchmark model by considering a five-variable TV-SVAR of
monthly measures of US unemployment, economic activity, real stock prices, inflation (or expecta-
tions of inflation), and short-term rates. We find evidence of a weakly negative correlation between
stock prices and US inflation consistent with most of the empirical literature. Importantly, results
from both models suggest that the ambiguity in the correlation between stock prices and inflation
that plagues the data could stem from the conclusion that two disturbances that have a temporary
effect on output—idiosyncratic stock market (‘PO’) shocks and the inflationary (‘IN’), or money supply
‘MS’ shocks—seem to counterbalance each other in driving the relationship. We find that shocks that
serve to exogenously increase demand for equities (‘PO’) or liquidity (‘MD’) lead stock prices and US
inflation to move in opposite directions, whereas inflationary (‘IN’) shocks that stem from exogenous
increases in the money supply (‘MS’) lead stock prices and US inflation to move in the same direc-
tion. This would suggest that monetary factors that put downward pressures on interest rates (such
as money supply expansions) may foster activity in all markets including the stock market, driving
all prices in the same direction. On the other hand, higher demand for equities or higher demand
for liquidity—insofar as they may constitute substitution among different assets—may drive up prices
in one sector independently of the other and lead prices to move in the opposite direction. Taken
together, this is tantamount to an income and substitution effect that drives up the diametrically
opposed relationships between these two variables, essentially cancelling each other out.
An important finding of this paper is that—while the correlations between inflation and stock prices
have remained somewhat stable—both variables, in addition to output have experienced substantial
changes in volatility. Volatilities of US output and inflation decreased substantially during the Great
Moderation period while that of real stock prices did not. What is more interesting is that both models
find that the volatilities in all three variables rose (substantially in the case of US output and inflation)
during the Great Recession period of 2008–2009. The volatility of inflation during this period is esti-
mated to have surpassed that of the 1970s Great inflation period—even when the actual inflation rates
are nowhere near each other over the two periods. Monetary factors seem to be largely unimportant
in explaining the increased volatility in stock prices and output of the Great Recession period but seem
to be the primary tributary to the increases in inflation volatility at the end of the sample.
Another substantive contribution of this paper is that it shows that the effects of supply, mone-
tary, and market factors on real stock prices have changed over time—a fact that traditional models
that assume away time variation are ill-equipped to uncover, as the evidence from Fig. 12 implies. In
many cases, these changing dynamics have taken effect gradually. Thus, a regime-switching model
that assumes constant parameters across different subperiods might not be as informative as the
approaches employed is this paper. An example highlighting this advantage might be the follow-
ing: Results from Model I and Model II predict a positive response of stock prices to inflationary
disturbances—whether they come from ‘MS’ or ‘MD’ shocks. Fig. 12(i) and (j) show that a standard
approach also shows this result. However, our time-varying approach allows us to determine that
stock prices have responded much more strongly to ‘MS’ shocks during the Great Recession period
whereas ‘MD’ shocks hardly have any impact on stock prices during 2008–2009.
Most results are substantially robust to measurement and frequency issues across two different
specifications. Overall, they seem to indicate that aggregate supply, stock market demand, and inflation
are important determinants of low-frequency movements in equity returns, all exerting a positive
short-run response. However, the apparent reduction in magnitude of adverse supply shocks that
helps explain the Great Moderation in output cannot offer an explanation for the lack of moderation
in stock prices. Estimates from this paper suggest that the Great Moderation in US aggregates may be
over as a result of the financial crisis that led to the Great Recession of 2008–2009. While the salient
need to solve the “good luck” vs. “good policy” debate seems crucial to our understanding of inflation
persistence—and might yield important policy prescriptions to moderate the volatility of US economic
activity after the Great Recession—it also seems woefully inadequate in explaining stock price volatility
due to the inherently idiosyncratic volatility in equity markets. A final lesson from our results is that
in the face of higher uncertainty, the accommodative monetary policy conducted during the Great
Recession period, has likely propped up the stock market to counter the deleterious effects of higher
unemployment and lower activity at a time when disturbances in liquidity preferences seem to have
been less influential.
Acknowledgments
I would like to thank John Keating, Nathan Balke, David Rapach, Mark Wohar, Ron Gilbert, and
seminar series participants at Oklahoma State University, University of New Mexico and Texas Tech
University for helpful comments, as well as Luca Gambetti for graciously sharing some code. The
standard disclaimers apply.
References
Aguilar, O., & West, M. (2000). Bayesian dynamic factor models and portfolio allocation. Journal of Business and Economic Statistics,
18(3), 338–357.
Bernanke, B., Boivin, J., & Eliasz, P. (2005). Measuring monetary policy: A factor augmented vector autoregressive (FAVAR)
approach. Quarterly Journal of Economics, 120, 387–421.
Bernanke, B., & Mihov, I. (1998). Measuring monetary policy. Quarterly Journal of Economics, 113, 869–902.
Binswanger, M. (2004). How important are fundamentals? Evidence from a structural VAR model for the stock markets in the
US, Japan and Europe. Journal of International Financial Markets Institutions and Money, 14, 185–201.
Blanchard, O. J., & Quah, D. (1989). The dynamic effects of aggregate demand and supply disturbances. The American Economic
Review, 79(4), 655–673.
Carter, C., & Kohn, R. (1994). On gibbs sampling for state space models. Biometrika, 81(3), 541–553.
Cecchetti, S., Flores-Lagunes, A., & Krause, S. (2006). Assessing the sources of changes in the volatility of real growth. NBER
Working Paper No. 11946.
Clarida, T., Gali, J., & Gertler, M. (2000). Monetary policy rules and macroeconomic stability: Evidence and some theory. Quarterly
Journal of Economics, 115, 147–180.
Clark, T. E. (2009). Is the great moderation over? An empirical analysis. Economic Review, (4), 5–42. Federal Reserve Bank of
Kansas City
Clark, T. E., & Davig, T. (2011). Decomposing the declining volatility of long-term inflation expectations. Journal of Economic
Dynamics and Control, 35(7), 981–999.
Cogley, T., & Sargent, T. J. (2001). Evolving post World War II US inflation dynamics. NBER Macroeconomics Annual, 16, 331–373.
Cogley, T., & Sargent, T. J. (2005). Drifts and volatilities: Monetary policies and outcomes in the post WWII US. Review of Economic
Dynamics, 8, 262–302.
Durham, B. J. (2003). Monetary policy and stock price returns. Financial Analysts Journal, 59(4), 26–35.
Fama, E. F. (1981). Stock returns, real activity inflation and money. The American Economic Review, 71(4), 545–565.
Fama, E. F., & French, K. R. (2001). Disappearing dividends: Changing firm characteristics or lower propensity to pay? Journal of
Financial Economics, 60, 3–43.
Faust, J., & Leeper, E. (1997). When do long run identifying restrictions give reliable results? Journal of Business and Economic
Statistics, 15-3, 345–353.
Feldstein, M. (1980). Inflation and the stock market. The American Economic Review, 70(5), 839–847.
Gali, J., & Gambetti, L. (2009). On the sources of the great moderation. American Economic Journal: Macroeconomics, 1(1), 26–57.
Gordon, M. J. (1962). The investment, financing and valuation of the corporation. Homewood, IL: R.D. Irwin.
Granger, C. W. J. (2008). Non-linear models: Where do we go next – Time varying parameter models? Studies in Nonlinear
Dynamics and Econometrics, 12(3). Article 1.
He, L. T. (2006). Variations in effects of monetary policy on stock market returns in the past four decades. Review of Financial
Economics, 15, 331–349.
Hess, P., & Lee, B. D. (1999). Stock returns and inflation with supply and demand disturbances. The Review of Financial Studies,
12(5), 1203–1218.
Jacquier, E., Polson, N. G., & Rossi, P. E. (1994). Bayesian analysis of stochastic volatility models. Journal of Business and Economic
Statistics, 12, 371–389.
Keating, J. W. (1992). Structural approaches to vector autoregressions. Federal Reserve Bank of St Louis Review, 74(5), 37–57.
Keating, J. W., & Valcarcel, V. (2011a). The time varying effects of permanent and transitory shocks to output. University of
Kansas Working Paper series and Texas Tech University Working Papers.
Keating, J. W., & Valcarcel, V. (2011b). Time varying volatilities of output growth and inflation: A multi-country investigation.
Working Paper.
King, T., & Morley, J. (2007). In search of the natural rate of unemployment. Journal of Monetary Economics, 54, 550–564.
Lastrapes, W. D. (1998). International evidence on equity prices, interest rates and money. Journal of International Money and
Finance, 17, 377–406.
Lee, B. S. (2010). Stock returns and inflation revisited: An evaluation of the inflation illusion hypothesis. Journal of Banking and
Finance, 34, 1257–1273.
McConnell, M., & Perez-Quiros, G. (2000). Output fluctuations in the United States: What has changed since the early 1980s?
The American Economic Review, 90, 1464–1476.
Modigliani, F., & Cohn, R. A. (1979). Inflation rational valuation and the market. Financial Analysts Journal, 35, 24–44.
Phelps, E. (1994). Structural slumps: The modern equilibrium theory of unemployment interest and assets. Cambridge: Harvard
University Press.
Primiceri, G. E. (2005). Time varying structural vector autoregressions and monetary policy. Review of Economic Studies, 72(3),
821–852.
Rapach, D. E. (2001). Macro shocks and real stock prices. Journal of Economics and Business, 53, p.5–p.26.
Rapach, D. E. (2002). The long run relationship between inflation and real stock prices. Journal of Macroeconomics, 24, 331–351.
Sargent, T. J. (1999). The conquest of American inflation. Princeton: Princeton University Press.
Sims, C. A. (1999). Drifts and breaks in monetary policy. Princeton Economics Department MIMEO.
Stock, J., & Watson, M. (1999). Forecasting inflation. Journal of Monetary Economics, 44, 293–335.
Summers, P. M. (2005). What caused the great moderation? Some cross-country evidence. Economic Review, 5–32. Federal
Reserve Bank of Kansas City, issue Q III.
Tobin, J. (1969). A general equilibrium approach to monetary theory. Journal of Money, Credit, and Banking, 1(1), 15–29.

3 PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

3 PDF

Uploaded by

Copyright:

Available Formats

Journal of Economics and Business 64 (2012) 117–144

Contents lists available at SciVerse ScienceDirect

Journal of Economics and Business

The dynamic adjustments of stock prices to inﬂation

Article history: While theoretical predictions establish a strong positive relation-

∗ Tel.: +1 806 742 2466; fax: +1 806 742 1137.

Standard deviation Relative standard deviation

Pre-1984: Post-1984 Pre-1984 Post-1984

1. Output 0.0051 0.0026 0.16 0.07

2. The time-varying SVAR model

3. Model I: a trivariate benchmark speciﬁcation

The underlying structural shocks to this economy εt = {εAS PO IN

4. Model II: a ﬁve-variable TV SVAR speciﬁcation

There is a substantial literature on the response of stock prices to macroeconomic disturbances,

Posterior Time-Varying Standard Deviation: 1959-2011

5. Time variation in the drift and variance of the system

(a) Posterior Standard Deviation for

(b) Posterior Standard Deviation for

Fig. 2. US conditional time-varying standard deviations: (1959–2011) Model I.

(a) Posterior Standard Deviation for

1985 1990 1995 2000 2005 2010

(b) Posterior Standard Deviation for

1985 1990 1995 2000 2005 2010

1985 1990 1995 2000 2005 2010

Fig. 3. US conditional time-varying standard deviations: (1981–2011) Model II.

Fig. 4. US conditional time-varying correlations Model I: (1959–2011).

Fig. 5. US conditional time-varying correlations Model II: (1981–2011).

Fig. 6. Time-varying responses at multiple horizons: Model I (1959–2011).

Shock Standard deviation Est. % Relative standard Est. %

Pre-1990 Post-1990 Decrease Pre-1990 Post-1990 Decrease

1. Supply 0.0072 0.0054 25 0.21 0.09 57

Shock Standard deviation Est. % Relative standard Est. %

Pre-1990 Post-1990 Decrease Pre-1990 Post-1990 Decrease

1. Supply 0.0056 0.0027 52 0.08 0.05 28

Note: See footnote of Table 2.

Shock Standard deviation Est. % Relative standard deviation Est. %

1. Supply 0.0024 0.0011 52 0.07 0.07 –

Note: See footnote of Table 2.

As mentioned in Section 4, Model II extends Model I in a number of different avenues involving

Fig. 7. Time-varying US output responses at multiple horizons: Model II (1981–2011).

now disaggregated in Model II into disturbances that—while having a straightforward relationship

Fig. 8. Time-varying stock price responses at multiple horizons: Model II (1981–2011).

8. Sensitivity analysis: expectations of inﬂation and the importance of time variation

0.5 Response of variable 4 -- t o shock in varaible 4 Response of variable 4 -- t o shock in varaible 5

Fig. 9. Time-varying US inﬂation responses at multiple horizons: Model II (1981–2011).

(a) Unconditional Correlation (b) Correlation DJIA -US CPI:

(c) Correlation DJIA - US CPI: (d) Correlation DJIA - US CPI:

(e) Correlation DJIA - US CPI: (f) Correlation DJIA - US CPI:

Fig. 10. US conditional time-varying correlations for stock prices-inﬂation (1981–2011).

(a) Unconditional Correlation (b) Correlation DJIA - MICH:

(c) Correlation DJIA - MICH: (d) Correlation DJIA - MICHI:

(e) Correlation DJIA - MICH: (f) Correlation DJIA - MICH:

Fig. 12. US (1981–2011) average responses: Model II.

You might also like