Consumption Volatility Risk∗

Oliver Boguth Lars-Alexander Kuehn

Sauder School of Business Tepper School of Business
University of British Columbia Carnegie Mellon University

September 14, 2009

ABSTRACT
Empirically, the conditional volatility of aggregate consumption growth varies over
time. While many papers test the consumption CAPM based on realized consumption
growth, little is known about how the time-variation of consumption growth volatility
affects asset prices. We show that in a model where (i) the agent has recursive pref-
erences, (ii) the conditional first and second moments of consumption growth follow a
Markov chain and (iii) the state of the economy is latent, the perception about condi-
tional moments of consumption growth affect excess returns. In the data, we find that
the perceived consumption volatility is a priced source of risk and exposure to it strongly
negatively predicts future returns in the cross-section. These results suggest that the rep-
resentative agent has an elasticity of intertemporal substitution greater than unity. In
the time-series, changes in beliefs about the volatility state strongly forecast aggregate
quarterly excess returns.

JEL Classification: G12, G17, E44.

Keywords: Asset pricing, consumption volatility, cross-section of returns, volatility risk,

recursive preferences.

∗
We thank Murray Carlson, Alexander David, Adlai Fisher, Lorenzo Garlappi, Burton Hollifield, Lars
Lochstoer (WFA discussant), Lukas Schmid (EFA discussant), Monika Piazzesi, Lu Zhang and seminar
participants at UBC, the 2009 North American Summer Meetings of the Econometric Society, 2009 WFA
meeting, 2009 CEPR Gerzensee Summer Symposium on Financial Markets for valuable comments and
2009 EFA meeting. Contact information, Boguth: 2053 Main Mall, Vancouver BC, Canada V6T 1Z2,
[email protected]; Kuehn: 5000 Forbes Avenue, Pittsburgh, PA 15213, [email protected].
I. Introduction

It is well known that the volatility of macroeconomic quantities, such as consumption and
output, vary over time.1 While many papers test the consumption CAPM based on realized
consumption growth, e.g., Lettau and Ludvigson (2001b), Parker and Julliard (2005) and
Yogo (2006), little is known about how the time-variation of consumption growth volatility
affects both the cross-section and time-series of stock returns. The goal of this paper is to fill
this void from a theoretical as well as empirical perspective. Specifically, we are interested
in whether innovations to the conditional volatility of consumption growth are a priced risk
factor.2
This research question poses several challenges: First, a natural candidate to model con-
sumption volatility is the ARCH model proposed by Engle (1982) and its various generaliza-
tions. Asset pricing theory, however, states that only innovations are priced and in a GARCH
model the volatility process has no separate innovations relative to the main process. In par-
ticular, Restoy and Weil (2004) show that a GARCH consumption model does not give rise to
a volatility risk factor in an equilibrium model with Epstein and Zin (1989) utility. Second,
while consumption growth rates are observable, the conditional volatility is latent and has to
be estimated from the data. Lastly, aggregate consumption is measured with error thereby
making statistical inference more difficult (Breeden, Gibbons, and Litzenberger (1989) and
Wilcox (1992)).
In our model, which follows the work of Bansal and Yaron (2004) and Kandel and Stam-
baugh (1991) and uses the same building blocks as in Lettau, Ludvigson, and Wachter (2008),
the representative agent has recursive Epstein and Zin (1989) preferences and the conditional
first and second moments of consumption growth follow independent two-state Markov chains.
Recursive preferences imply that the agent cares not only about shocks to current consump-
tion growth but also about changes to the distribution of future consumption growth, which
in the model is driven by a persistent conditional mean and volatility. Since the state of
the economy is unobservable, the agent uses Bayesian updating to form beliefs about the
1
For instance, refer to Cecchetti and Mark (1990), Kandel and Stambaugh (1990), Bonomo and Garcia
(1994), Kim and Nelson (1999), or Whitelaw (2000).
2
Other recent contributions testing the consumption CAPM include Campbell (1996), Aı̈t-Sahalia, Parker,
and Yogo (2004), Campbell and Vuolteenaho (2004), Bansal, Dittmar, and Lundblad (2005), Lustig and
Nieuwerburgh (2005), and Jagannathan and Wang (2007).

1
state, similar to David (1997) and Veronesi (1999). Consequently, the agent’s perception
about the conditional first and second moment of consumption growth are priced. Given that
both risk aversion and the elasticity of intertemporal substitution (EIS) are greater than one,
changes about the perceived conditional mean carry a positive price of risk and changes of
the perceived conditional consumption volatility a negative one.
The economic mechanism underlying our model is the following. When risk aversion and
EIS are greater than unity, the intertemporal substitution effect dominates the wealth effect.
As a result, the demand for the risky asset and thus the wealth-consumption ratio increases
with expected consumption growth and decreases with consumption growth volatility. This
effect also implies a positive price of risk for the first moment and negative one for the
second moment of consumption growth. Intuitively, consider an asset that comoves negatively
with future consumption growth. Its payoff is high (low) when investors learn that future
consumption growth is low (high). Investors will demand a low return from this asset as it
is a welcome insurance against future bad times. Similarly, consider an asset that comoves
highly with future consumption volatility. This asset has high (low) payoffs when investors
learn that future consumption is very (little) volatile. This asset serves as insurance against
uncertain times and thus has a lower required return.
To empirically test this intuition, we follow Hamilton (1989) and estimate a Markov chain
process for first and second moments of consumptions growth. Bayesian updating provides
beliefs about the states for mean and volatility. To obtain time-varying risk loadings with
respect to innovations in the perceived conditional first and second moment of consumption
growth, we run rolling quarterly time-series regressions of individual stock returns on con-
sumption growth as well as innovations in beliefs for mean and volatility. In cross-sectional
Fama and MacBeth (1973) regressions, we find that loadings on innovations in the perceived
mean consumption growth do not help to explain future returns. Loadings on consump-
tion growth volatility, however, significantly negatively forecast cross-sectional differences in
returns.
Potentially, inference based on Fama-MacBeth regressions is misleading since it is not
feasible to keep track of standard errors across estimation stages. Yet we obtain similar evi-
dence in a more conservative approach by sorting stocks into portfolios based on consumption
volatility loadings. We observe that stocks which covary more with consumption volatility

2
have lower future returns. A volatility risk (VR) factor, which is the return of holding a long
position in the value-weighted quintile of stocks with high volatility risk and a short posi-
tion in low volatility risk, has an average return of −5% per year. Importantly, consumption
volatility risk quintiles do not display variation in the average book-to-market characteristic.
Nevertheless, the loadings of the 25 size and value portfolios of Fama and French (1992) on
the VR factor monotonically decrease in the book-to-market ratio.
The main implication of the model is that consumption volatility is a priced risk factor.
In order to test this hypothesis, we perform two stage regressions of excess returns on log
consumption growth, changes in the perceived mean and volatility of consumption and the
VR factor. Crucially, the coefficient on the innovation in the perceived consumption volatility
and VR are negative implying that the representative agent has EIS greater than one. This
finding contributes to a long standing debate in the literature on the magnitude of the EIS.
Early evidence suggests that the EIS is smaller than one, e.g., Hall (1988) and Campbell and
Mankiw (1989). More recently, Attanasio and Weber (1993), Vissing-Jorgensen (2002) and
Vissing-Jorgensen and Attanasio (2003) find the opposite.
We also augment the CAPM and Fama-French 3-factor model with the VR factor. In
particular, the VR factor shows up strongly and significant in addition to the market and the
three Fama and French (1993) factors. Adding the VR factor to specifications that already
contain HML, the overall fit of the regression as measured by the R2 statistic improves only
marginally. At the same time, replacing HML by VR has little effect on the goodness of
the model. We thus conclude that HML and VR are substitutes in the cross-sectional pricing
relation. But in contrast to HML, the volatility risk factor has a clear economic interpretation.
Another implication of our model is the predictability of the equity premium in the time-
series. In states with low conditional mean or high conditional volatility of consumption
growth, the model predicts a high equity premium when the representative agent has risk
aversion and EIS greater than unity. We show in a predictive regression that innovations
to consumption volatility are a significant and robust predictor of one-quarter ahead equity
returns. The R2 of this univariate predictive regression is almost 5% and a one standard
deviation increase of the perceived consumption volatility results in a 1.8% rise of the quarterly
equity premium. Both values are very close to the predictive power of the consumption-wealth
ratio cay of Lettau and Ludvigson (2001a), the best known macroeconomic predictor of the

3
short horizon equity premium. In our model, changes in consumption volatility enter the
pricing kernel only because they affect the consumption-wealth ratio. Thus, one might expect
that measures of the consumption-wealth ratio such as cay already contain information about
the volatility state. Empirically, this is not the case. Both variables are virtually uncorrelated
and both remain strong and robust predictors in multivariate settings. The in-sample R2 of
regressions containing both changes in volatility and cay exceed 10%.
In the literature, it is common to measure consumption risk by using non-durable plus
service consumption. This assumption is usually justified with a felicity function which is
separable across goods. With Epstein-Zin utility, however, felicity can be separable across
goods, but due to the time-nonseparability of the time-aggregator, other goods still matter
for asset pricing because they enter the pricing kernel via the wealth-consumption ratio. The
wealth-consumption ratio can be a function, for instance, of human capital (e.g. Jagannathan
and Wang (1996), Lettau and Ludvigson (2001b), and Santos and Veronesi (2006)), durable
goods (e.g. Yogo (2006)) or housing consumption (e.g. Piazzesi, Schneider, and Tuzel (2007)).
If the wealth-consumption were observable, it would subsume all these variables. But the
wealth-consumption ratio is unobservable.3 The contribution of this paper is to show that
the conditional volatility of consumption growth is a significant determinant of the wealth-
consumption ratio by documenting that it is priced in the cross-section and time-series after
controlling for other factors.

Related Literature

Pindyck (1984) and Poterba and Summers (1986) are among the first to show that a decrease
in prices is generally associated with an increase in future volatility, the so-called leverage
or volatility feedback effect. Similarly, Campbell and Hentschel (1992) and Glosten, Jagan-
nathan, and Runkle (1993) look at the relation between market returns and market volatility
in the time-series. More recently, Ang, Hodrick, Xing, and Zhang (2006) use a nonparametric
measure of market volatility, namely, the option implied volatility index (VIX), to show that
innovations in aggregate market volatility carry a negative price of risk in the cross-section.
Adrian and Rosenberg (2008) use a GARCH inspired model to decompose market volatility
into short and long run components and show how each of the two components affects the
3
One of the first papers which tries to estimate the wealth-consumption ratio is Lettau and Ludvigson
(2001a). A more recent contribution is Lustig, Van Nieuwerburgh, and Verdelhan (2008).

4
cross-section of asset prices.
All of the papers mentioned above use some measure of stock market volatility, and can
therefore be interpreted as extensions of a market based CAPM. We differ from the existing
research in two important dimensions. First, we extract aggregate volatility from consumption
data and not from financial data. While the quality of consumption data is considerably worse
than the quality of financial data, we are able to robustly show that consumption volatility
risk is priced. Second, we explicitly model the fact that conditional moments of consumption
growth are unobservable and investors learn about them.
Considerable less research has been done on the pricing implications of volatility in a
consumption-based model. Notable exceptions are Jacobs and Wang (2004) and Balduzzi
and Yao (2007), who use survey data to estimate the variability of idiosyncratic consumption
across households. They find that exposure to idiosyncratic consumption risk bears a negative
risk premium for the 25 Fama-French portfolios. Parker and Julliard (2005) empirically
measure a version of long-run risk as the covariance between one-period asset returns and long-
horizon movements in the pricing kernel. Their ultimate consumption risk measure performs
favorably in explaining the return differences of the 25 Fama-French portfolios. Similarly,
Tedongap (2007) estimates conditional consumption volatility as a GARCH process and finds
that value stocks covary more negatively with changes in consumption volatility over long
horizons, thus requiring high average returns. In contrast to Tedongap (2007), we extract
innovations to beliefs about consumption volatility, whereas a GARCH model does not allow
that. Tedongap (2007) obtains significant results only at long horizons since GARCH models
account for innovations to volatility only through realized data.
Motivated by the long-run risk model of Bansal and Yaron (2004), there exist important
papers which study the relation between aggregate volatility and prices. Notably, Bansal,
Khatchatrian, and Yaron (2005) find that the conditional consumption volatility predicts
aggregate valuation ratios. Bansal, Kiku, and Yaron (2007) estimate the long-run risk model
using the cross-section of returns. Following the model, they estimate consumption volatility
as an affine function of the observable aggregate price-dividend ratio and short-term interest
rate. Their Table IV indicates that consumption volatility plays a minor role in explaining
the size and value spread relative to shocks to expected consumption growth. In contrast,
we filter consumption volatility directly from consumption data without the use of financial

5
data. We also find that consumption volatility is a dominant contributor to risk premia in
the cross-section.
Drechsler and Yaron (2008) extend the long-run risk model for jumps in consumption
growth and volatility. Their model generates a variance premium and return predictability
which is consistent with the data. Bansal and Shaliastovich (2008) find evidence that measures
of investors uncertainty about their estimate of future growth contain information about
large moves in returns at frequencies of about 18 months. They explain this regularity with
a recursive-utility based model in which investors learn about latent expected consumption
growth from signals with time-varying precision. Bollerslev, Tauchen, and Zhou (2008) study
the asset pricing implication when the variance of stochastic volatility is stochastic.4
More closely related are Calvet and Fisher (2007) and Lettau, Ludvigson, and Wachter
(2008). Calvet and Fisher (2007) study the asset pricing implications of multi-fractal Markov
switching in a recursive preference model at the aggregate level. Similarly, Lettau, Ludvig-
son, and Wachter (2008) estimate a Markov model with learning to show that the decline
in consumption volatility–also referred to as the “Great Moderation”–can explain the high
observed stock market returns in the 1990s and the following decline in equity risk premium.
We extend their work by studying the cross-section and time-series of returns.
The remainder of the paper is organized as follows: In Section 2, we derive the asset
pricing implication of a recursive preference model where the agent does not observe the state
of the economy. This section motivates our empirical analysis of Sections 3–5. In Section
3, we test whether consumption growth and its conditional moments forecast returns in the
cross-section. We run Fama-MacBeth regressions and form portfolios based on consumption
volatility loadings. In Section 4, we test whether consumption growth and its conditional
moments as well as the VR factor are priced risk factors. Section 5 contains time-series
predictability tests and Section 6 concludes.

II. Model

In this section, we derive the asset pricing implications of a model where the representative
agent has recursive preferences and the state of the economy is unobservable. A crucial
4
Other papers building on the long-run risk framework of Bansal and Yaron (2004) include Bhamra, Kuehn,
and Strebulaev (2007), Hansen, Heaton, and Li (2008) and Bansal, Dittmar, and Kiku (2009).

6
implication of recursive preferences is that the agent cares not only about shocks to current
consumption growth but also about news regarding the distribution of future consumption
growth. In our model, future consumption growth is influenced by time-variation of the
conditional mean and volatility of consumption growth which is unobservable to the agent.
This latent nature implies the agent forms beliefs about the conditional first and second
moments of consumption growth and, most importantly, changes in the perceived first and
second moments of consumption growth are priced.

A. Consumption

We assume that the conditional first and second moments of consumption growth follow
a Markov chain. Specifically, let ∆ct+1 denote log consumption growth, µt its conditional
expectation and σt its conditional volatility, then log consumption growth follows

∆ct+1 = µt + σt t+1 t ∼ N (0, 1) (1)

with iid innovations t . For tractability in the empirical estimation, we assume two states for
the mean and two for the volatility which are denoted by µt ∈ {µl , µh } and σt ∈ {σl , σh }.
The conditional first and second moments of consumption growth follow Markov chains with
transition matrix P µ and P σ , respectively, given by
   
µ µ
pll 1 − pll pll
σ 1 − pll
σ
Pµ =   Pσ =   (2)
µ µ
1 − phh phh 1 − phh
σ phh
σ

To keep the number of parameters to be estimated manageable, we impose that mean and
volatility states switch independently. Thus, the joint transition matrix is the product
of the marginal probabilities for mean and volatility and the 16-element matrix can be
fully characterized by 4 parameters. Importantly, the assumption of independent switch-
ing probabilities does not imply that mean and volatility or the beliefs thereof are indepen-
dent. Since we assume two drift and two volatility states, there are four states in total,
{(µl , σl ), (µl , σh ), (µh , σl ), (µh , σh )}, denoted by st = 1, ..., 4. Our specification follows Kandel
and Stambaugh (1990), Kim and Nelson (1999), and Lettau, Ludvigson, and Wachter (2008).
In contrast to Bansal and Yaron (2004) and Kandel and Stambaugh (1991), we assume that
the representative agent does not observe the state of the economy. Instead, she must infer it
from observable consumption data. This assumption ensures that the empirical exercise is in

7
line with the model. The inferences at date t about the underlying state is captured by the
posterior probability of being in each state based on the available data Yt . Let ξt+1|t denote
the posterior belief vector of size 4 × 1:

ξt|t−1  ηt
ξt+1|t = P 0 (3)
1 (ξt|t−1  ηt )
0

where  
f (∆ct |µt−1 = µl , σt−1 = σl , Yt−1 )
 
 f (∆ct |µt−1 = µl , σt−1 = σh , Yt−1 )
 
ηt = 


 f (∆ct |µt−1 = µh , σt−1 = σl , Yt−1 )
 

 
f (∆ct |µt−1 = µh , σt−1 = σh , Yt−1 )
is a vector of Gaussian likelihood functions and P = P µ ⊗ P σ is the transition matrix.

B. Recursive Utility

The representative household maximizes recursive utility over consumption following Kreps
and Porteus (1978), Epstein and Zin (1989), and Weil (1989):
  ρ/(1−γ) 1/ρ
Ut = (1 − β)Ctρ +β 1−γ
Et [Ut+1 ] (4)

where Ct denotes consumption, β ∈ (0, 1) the rate of time preference, ρ = 1 − 1/ψ and ψ
the elasticity of intertemporal substitution (EIS), and γ relative risk aversion. Implicit in the
utility function (4) is a constant elasticity of substitution (CES) time aggregator and CES
power utility certainty equivalent.
Epstein-Zin preferences provide a separation of the elasticity of intertemporal substitution
and relative risk aversion. These two concepts are inversely related when the agent has power
utility. Intuitively, the EIS measures the agents willingness to postpone consumption over
time, a notion well-defined even under certainty. Relative risk aversion measures the agents
aversion to atemporal risk across states.5
We know from Epstein and Zin (1989) that the Euler equation for any return Ri,t+1 can
be stated as " −(1−θ)  −γ #
P Ct+1 + 1 Ct+1

Et β θ Ri,t+1 = 1 (5)
P Ct Ct
5
Recursive preferences also imply preference for early or late resolution of uncertainty. This feature, however,
is not important for this paper since the agent cannot choose between consumption lotteries which differ in
the timing of the resolution of uncertainty.

8
 1−γ
where θ = 1−1/ψ and P Ct = Pt /Ct denotes the wealth-consumption ratio. For the empirical
exercise, it is useful to study the log-linearized pricing kernel. A log-linear approximation of
the pricing kernel implicit in (5) is

mt+1 ≈ (θ ln β − (1 − θ)k0 ) − γ∆ct+1 − (1 − θ)(pct+1 − k1 pct ) (6)

where pct = ln(Pt /Ct ) denotes the log wealth-consumption ratio and k0 , k1 are constants. The
value of k1 is given by k1 = P C/(P C − 1) > 1 and P C is the mean wealth-consumption ratio.
The Epstein-Zin pricing kernel is thus a function of consumption growth and the level of the
log wealth-consumption ratio. Alternatively, the pricing kernel can approximately be stated in
terms of changes of the log wealth-consumption ratio if k1 is close to one. Lustig, Van Nieuwer-
burgh, and Verdelhan (2008) estimate the unconditional quarterly wealth-consumption ratio
to be almost 351 implying that k1 = 1.003. Consequently, the log pricing kernel can be closely
approximated by:

mt+1 ≈ (θ ln β − (1 − θ)k0 ) − γ∆ct+1 − (1 − θ)∆pct+1 (7)

The log pricing kernel (7) implies that excess returns are determined as covariance between
returns and log consumption growth as well as the covariance between returns and changes
of the log wealth-consumption ratio:

Et [Ri,t+1
e
] ≈ −Covt (Ri,t+1 , mt+1 ) = γCovt (Ri,t+1 , ∆ct+1 ) + (1 − θ)Covt (Ri,t+1 , ∆pct+1 ) (8)

In an endowment model which is solely driven by iid shocks, the wealth-consumption

ratio is constant. In our model, however, the first and second moments of consumption
growth follow a Markov chain. The unobservability of the state implies that the agent’s
beliefs characterize the state of the economy. Consequently, the wealth-consumption ratio is
a function of the agent’s beliefs about the state of the economy, i.e., P Ct = P C(ξt+1|t ).
For the pricing of the return on the consumption claim, Euler equation (5) simplifies to
" 1−γ #
θ Ct+1

P Ct = Et β (P Ct+1 + 1)
θ θ
(9)
Ct

Based on the law of iterated expectations, equation (9) implies that

4
X
P Ctθ = ξt+1|t (i)P Ct,i
θ
(10)
i=1

9
where ξt+1|t (i) is i-the element of ξt+1|t and
" 1−γ  #
Ct+1

P Ct,i
θ
= E β θ (P Ct+1 + 1)θ  st+1 = i, ξt+1|t (11)

Ct 

Equation (10) says that the agent forms a belief-weighted average of the state- and belief-
conditioned wealth-consumption ratios (11).
In order to study how the wealth-consumption ratio changes with beliefs about the state,
we further define the posterior belief that the mean state is high tomorrow by

bµ,t = P (µt+1 = µh |Ft ) = ξt+1|t (3) + ξt+1|t (4) (12)

and the posterior belief that the volatility state is high tomorrow by

bσ,t = P (σt+1 = σh |Ft ) = ξt+1|t (2) + ξt+1|t (4) (13)

conditional on the current information set Ft . The univariate effects of changing beliefs
about the volatility (mean) while holding the the mean (volatility) constant can locally be
approximated. Given the volatility, changes of the log wealth-consumption ratio are

1 P Cµ=µ
θ
h ,σ
− P Cµ=µ
θ
l ,σ
∆pct+1 ≈ ∆bµ,t+1 (14)
θ bµ,t P Cµ=µ
θ
h ,σ
+ (1 − bµ,t ) P Cµ=µ
θ
l ,σ

where P Cµ=µh ,σ denotes the wealth-consumption ratio when expected consumption growth
is high and the consumption volatility is constant, a similar definition applies for the other
wealth-consumption ratios. Analogous, given the mean, changes of the log wealth-consumption
ratio are
1 P Cµ,σ=σ
θ − P Cµ,σ=σ
θ
∆pct+1 ≈ ∆bσ,t+1 h l
(15)
θ bσ,t P Cµ,σ=σ
θ
h
+ (1 − bσ,t ) P C θ
µ,σ=σl
Changes in the log wealth-consumption ratio are thus locally proportional to changes in beliefs.
From an empirical asset pricing perspective, this finding implies that changes in beliefs are
priced in the time-series and cross-section since they affect the wealth-consumption ratio,
according to Equation (8).
This implication does not necessarily follow from an equilibrium model where the condi-
tional consumption volatility follows a GARCH process. In a GARCH model, the conditional
volatility is a function of lagged volatility and lagged squared residuals of the consumption
process. Thus, a GARCH process is not driven by separate innovations relative to the con-
sumption process. Consequently, Restoy (1991) and Restoy and Weil (2004) have shown that

10
a GARCH consumption model does not give rise to a priced risk factor in a log-linearized ap-
proximation to an equilibrium model.6 Specifically, Equation (4.5) in Restoy and Weil (2004)
states that the covariance of any stock with the wealth-consumption ratio is proportional
to its covariance with consumption growth. Volatility, which affects the wealth-consumption
ratio, therefore can have pricing implications as it determines the loading on the consumption
growth factor, but it does not give rise to a second priced risk factor. Restoy and Weil con-
tinue to say: ”This result embodies the fundamental insight that, for a GARCH(1,1) process,
returns are only able to predict future conditional means of consumption growth but carry
no information about the future conditional variances”.

C. Estimation

To estimate the model, we obtain data on quarterly per capita consumption from the NIPA
tables as the sum of nondurables and services. In accordance with the observation that the
consumption behavior in the United States in the years following World War II is systemati-
cally different from later years, we restrict our time-series from the first quarter of 1955 until
the fourth quarter of 2008. The choice of 1955 allows sufficient consumption observations to
ensure that the impact of prior beliefs on posteriors has vanished by the time we start the
portfolio analysis in 1964.
The resulting parameter estimates of the Markov chain are reported in Table I, Panels
A and B. Expected consumption growth is always positive and about twice as high in the
high state relative to the low state (µl = 0.37%, µh = 0.78% quarterly). State-conditioned
consumption volatilities are σl = 0.21% and σh = 0.48%. The probability of remaining in a
given regime for the mean is 0.93 in the low state and 0.89 in the high state. The volatility
regimes are somewhat more persistent, with probabilities of 0.96 and 0.95, respectively. Our
estimates differ from the ones presented by Lettau, Ludvigson, and Wachter (2008), who
estimate volatility in both states to be more persistent (0.991 and 0.994). In addition to small
differences in the time-series, their consumption measure is total consumption, while we use
the common definition of consumption as nondurables plus services.
We assume independent switching in mean and volatility states. This assumption greatly
reduces the number of parameters to be estimated and thus improves estimation precision.
6
In empirical tests of equilibrium models, GARCH-inspired processes have been used by Adrian and Rosen-
berg (2008) and Tedongap (2007) to motivate additional factors in the cross section.

11
Yet this assumption does not appear to be a significant restriction of consumption data. If
the true (unobservable) correlation between regime switches were very different from zero, we
would expect to see a large correlation in our estimated posterior beliefs. Panel C of Table I
shows that this correlation is around 8% which is small enough to suggest that the assumption
of independence is not a contradiction of the data.
Figure 1 shows the filtered beliefs for the regimes. Panel A depicts the belief dynamics
for mean consumption growth bµ,t and Panel B for the standard deviation bσ,t . These graphs
visually confirm that the mean regimes are less persistent than the standard deviation states.
In particular, the parameter estimates for the Markov chain imply that mean states last
for 3.4 years whereas volatility states last for 11.2 years on average. Further, a decline in
consumption volatility from the 1990s onwards, as pointed out by Kim and Nelson (1999),
is easily observable. Importantly, there is significant variation in beliefs about the volatility
state in addition to decline in the early 1990s.

D. Implications

Based on the parameters estimated from consumption data, we solve the model numerically
to study its properties. In the following, we are interested in how the perception about the
first and second moments of consumption growth affect the wealth-consumption ratio. To
this end, we define expected consumption growth and expected consumption volatility as a
belief weighted average:

µ̂t = bµ,t µh + (1 − bµ,t )µl σ̂t = bσ,t σh + (1 − bσ,t )σl (16)

In Figures 2 and 3, we plot the wealth-consumption ratio as a function of the perceived

conditional first µ̂t (left graph) and second σ̂t (right graph) moments of consumption growth
for the benchmark calibration of Table I when the agent has a high EIS of 1.5 (Figure 2) and
low EIS of 0.5 (Figure 3). We further calibrate the model to a quarterly rate of time preference
of β = 0.995 and risk aversion of γ = 30. Risk aversion of 30 seems unrealistically high. This
section, however, is meant to yield qualitative guidance and not quantitative results.
To gain a better understanding of the economics, it is convenient to recall the Gordon
growth model. Under the assumption that discount and growth rates are constant, the Gordon
growth model states that the wealth-consumption ratio is negatively related to the risk-free
rate rf and risk premium rE and positively to the growth rate g, i.e., P C = 1/(rf + rE − g).

12
 In Figure 2, the wealth-consumption ratio is increasing in the perceived mean and de-
creasing in the perceived volatility of consumption growth when the EIS equals 1.5. In Figure
3, the opposite is true when the EIS equals 0.5. Interestingly, both, Figures 2 and 3 suggest
that the log wealth-consumption ratio is approximately affine in the perceived first and second
moment of consumption growth for our parameterization, confirming the validity of Equations
(14) and (15).
The sign change in the slope of the wealth-consumption ratio with respect to expected
consumption growth is driven by two opposing effects. On the one hand, a higher perception
about the growth rate increases the wealth-consumption ratio as in the Gordon growth model.
On the other hand, in equilibrium, an increase in expected consumption growth also raises
the risk-free rate since the riskless asset becomes less attractive relative to the risky asset.
This second effect lowers the wealth-consumption ratio. When the EIS is greater than unity,
the first effect (intertemporal substitution effect) dominates the second effect (wealth effect).
As a result, the demand for the risky asset and thus the wealth-consumption ratio rises with
the perceived expected growth rate of consumption.
Similarly, the sign change in the slope of the wealth-consumption ratio with respect to
expected consumption growth volatility (Figure 2 versus 3) is also driven by two opposing
effects. On the one hand, a higher perceived conditional consumption volatility increases the
risk premium which lowers the wealth-consumption ratio as in the Gordon growth model. On
the other hand, in equilibrium, an increase in expected consumption growth volatility also
reduces the risk-free rate since the riskless asset becomes more attractive relative to the risky
asset. This effect increases the wealth-consumption ratio. If γ > 1, the first effect dominates
the second one when the EIS is greater than one.
In order to test this intuition in the cross-section of returns, it is convenient to restate the
fundamental asset pricing equation (8) in terms of betas:7

Et [Ri,t+1
e
] ≈ βc,t
i
λc,t + βpc,t
i
λpc,t ≈ βc,t
i
λc,t + βµ,t
i
λµ,t + βσ,t
i
λσ,t (17)

where βc,t
i , β i , β i , β i denote risk loadings of asset i at date t with respect to consumption
pc,t µ,t σ,t
7
Equation (10) states that variations in the wealth-consumption ratio depend on the beliefs about four states,
three of which are linearly independent. In an exact implementation of the model, the wealth-consumption
ratio is thus a nonlinear function of three variables. We reduce it to two in order to achieve a more meaningful
economic interpretation for mean and volatility states. Note, however, that the empirical results are not
affected by this assumption.

13
growth, the wealth-consumption ratio and the conditional first and second moment of con-
sumption growth and λc,t , λpc,t , λµ,t , λσ,t are the respective market prices of risk. The main
cross-sectional implications of the model are the following. Assuming that both risk aver-
sion and EIS are greater than one, the agent requires lower expected excess returns for stock
which load less (low betas) on expected consumption growth and more (high betas) on future
consumption growth volatility.
The economic intuition is straightforward. Equations (8) and (17) imply that the market
prices of the conditional first and second moments of consumption growth are their respective
conditional variances, each multiplied by a constant.8 The sign of both constants depends on
(1 − θ), coming from Equation (8), and on the slope of the wealth-consumption ratio with
respect to the conditional mean and volatility of consumption growth. Assuming both risk
aversion and EIS greater than one, (1 − θ) is positive and the wealth-consumption ratio is
upward-sloping in the conditional mean and downward-sloping in the conditional volatility
of consumption growth (Figure 2). Consequently, the market price of expected consumption
growth is positive, i.e. λµ,t > 0, and the market price of conditional consumption volatility is
negative, i.e. λσ,t < 0.
Moreover, the positive sign of λµ,t and negative one of λσ,t also implies that investors
require higher compensation for holding stocks which load strongly (high beta) on expected
consumption growth and less compensation for stocks which load strongly (high beta) on
consumption growth volatility. Intuitively, assets, which comove negatively with future con-
sumption growth, have high payoffs when investors learn that future consumption growth is
low. These assets thus provide insurance against future bad times. Similarly, assets, which
comove highly with future consumption volatility, have high payoffs when investors learn that
future consumption is very volatile. These assets serve as insurance against uncertain times
and thus have lower required returns.
8
Intuitively, the log wealth-consumption is approximately affine in the perceived first and second mo-
ments of consumption growth which expressed in changes can be stated as ∆pct ≈ A∆µt + B∆σt . Sub-
stituting this expression into (8), one obtains (17) and λµ,t = A(1 − θ)Vart (∆µt+1 )/Et [Mt+1 ] and λσ,t =
B(1 − θ)Vart (∆σt+1 )/Et [Mt+1 ].

14
III. Cross-Sectional Return Predictability

The goal of this section is to demonstrate that loadings on the perceived (filtered) conditional
consumption volatility forecast returns. To this end, we first run quarterly time-series regres-
sions to obtain loadings on risk factors. Next, we test using both Fama-MacBeth regressions
and portfolio sorts whether these risk loadings forecast returns. Our main finding is that
innovations in consumption volatility risk is a strong and robust predictor of future returns,
while exposure to consumption growth and changes in expected consumption growth risk do
not help to explain the cross-section of asset prices.

A. Data

Our sample consists of all common stocks (shrcd = 10 or 11) on CRSP that are traded on the
NYSE or AMEX (exchcd = 1 or 2). While the results are generally robust to the inclusion of
NASDAQ stocks, this restriction mitigates concerns that only a small fraction of total market
capitalization has a large impact on the portfolio analysis. To obtain valid risk measurements
for a given quarter, the asset is required to have at least 60 months of prior data and at
least 16 out of 20 valid quarterly returns. Since we use size and book-to-market ratio as
characteristics, we require market capitalization to be available in December that occurs 7
to 18 months prior to the test month as well as book value of equity from COMPUSTAT in
the corresponding year. The choice of the long delay is motivated by the portfolio formation
strategies in Fama and French (1992), who want to ensure that the variables are publicly
available when they are used in the study. Due to limited availability of book values in earlier
years, we begin the empirical exercise in January 1964. The first time-series regression to
estimate risk loadings thus covers the time span from 1959 to 1963. We end our analysis in
December 2008.

B. Consumption Volatility Risk and Stock Returns

Our first set of empirical results is based on time-series regressions of individual securities
onto log consumption growth, the perceived conditional mean and volatility of consumption
growth as well as the excess return on the market. This specification is a generalization of
the pricing restriction (8), where changes in the wealth-consumption ratio can be linearly
approximated as shown in Equations (14) and (15). To be conservative, we also include the

15
market return in the regression to control for time effects so that consumption-based betas
purely load on cross-sectional differences in returns.
In particular, for each security, we estimate factor loadings in each quarter t∗ using the
previous 20 quarterly observations from

Rti − Rtf = αti∗ + βc,t

i
∗ ∆ct + βµ,t∗ ∆µ̂t + βσ,t∗ ∆σ̂t + t
i i i
(18)

where Rtf denotes the risk-free rate and ∆ct consumption growth for t ∈ {t∗ −19, t∗ }. Further,
∆µ̂t and ∆σ̂t are innovations in the perceived conditional moments of consumption growth.
These are defined as the differences in believed moments before and after consumption is
realized:
∆µ̂t = µ̂t − µ̂t−1 ∆σ̂t = σ̂t − σ̂t−1 , (19)

where µ̂t and σ̂t are the perceived conditional first and second moments of consumption
growth, respectively, defined in Equation (16).
The estimated parameters from Equation (18) allow us to evaluate the cross-sectional
predictive power of these loadings in two different ways. First, we use cross-sectional regres-
sions as in Fama and MacBeth (1973) to investigate if the factor loadings help to predict
cross-sectional variation in returns. Second, we form portfolios based on the estimated risk
exposures and analyze their properties in the time-series.

C. Fama-MacBeth Regressions

We now investigate the predictive power of the estimated loadings from model (18) by cross-
sectionally regressing the returns in month s + 1 of each asset onto its latest available risk
loadings as well as size and value characteristics:

Rs+1
i
= γ0,s+1 + γ1,s+1 β̂c,t
i
∗ + γ2,s+1 β̂µ,t∗ + γ3,s+1 β̂σ,t∗
i i
(20)

+γ4,s+1 M Eti∗ + γ5,s+1 BMti∗ + ηs+1

i

The explanatory variables are normalized each quarter so they are centered around zero with
unit variance. Equation (20) states that each set of three monthly regressions in one quarter
will share the same predictor variables. For example, the returns in each of the months
April, May, and June are regressed onto the risk loadings estimated from the window ending
in the first quarter of the same year. We are interested whether the factor loadings have

16
any predictive power for the cross-sectional variation of returns or, equivalently, whether
γk,s , k = 1, ..., 3 are on average different from zero.
The results of the Fama-MacBeth regressions are presented in Table II. Model specifi-
cations I-III present univariate effects of each risk loading. Consistent with prior research
(Mankiw and Shapiro (1986), Lettau and Ludvigson (2001b)), an asset’s contemporaneous
short horizon loading on consumption growth does not help to predict cross-sectional differ-
ences in returns. The average coefficient of γ̄1 is small and insignificant. The same holds
for innovations in the belief about the expected growth rate, γ̄2 . Exposure to consumption
volatility risk, however, as measured by γ̄3 , shows up strongly negative and significant. Stocks
that comove highly with changes in consumption volatility underperform their peers in the
future. Specification IV is the full model. Now, both the loading on consumption growth
and consumption volatility risk are significant. In regression VI and V, we add two char-
acteristics known to predict stock returns, namely, the market capitalization (γ4 ) and the
ratio of book value of equity to market value (γ5 ), to confirm that the predictive power of
consumption volatility is not already captured by these predictors. The absolute value of the
point estimate γ̄3 is slightly reduced by the addition of the two characteristics, but it remains
significant.
What do these findings mean? The novel implications of our model are that beliefs about
mean and volatility states of consumption growth are priced sources of risk. As a result,
exposure to these sources should be associated with a spread in future returns. The sign of
the risk premium associated with each of these two factors depends on preference parameters.
In the case where both risk aversion and EIS are greater than unity, the model predicts that
returns are positively related to βµ,t and negatively to βσ,t . An EIS lower than one yields
the opposite predictions. We do not find convincing evidence that exposure to fluctuations in
expected consumption growth predicts returns. Yet the coefficient on consumption volatility
is strongly negative. This finding is consistent with the model only if risk aversion and EIS
of the representative agent is greater than one.
There are two potential problems in our analysis. First, when comparing the influence of
the predictor variables, it is important to note the possibility of an error-in-variable problem.
The first four variables are estimates from first stage regressions and are thus noisy. More-
over, while researchers often treat ∆ct as observable, the consumption time-series actually

17
is measured with significant noise (Breeden, Gibbons, and Litzenberger (1989) and Wilcox
(1992)). Both ∆µt and ∆σt are estimates and themselves depend on the imposed model for
consumption growth dynamics. The t-statistics reported are obtained from standard OLS
theory and are likely based on standard errors that are too small. However, the necessary
adjustment would require to keep track of standard errors across all stages of estimation and
is thus not feasible in this case. In the next section, we use a different methodology to confirm
that our results are robust and not due to misrepresented standard errors.
Second, imposing a Markov model on consumption growth leads to a collinearity problem
in the first stage regression (18). Rational Bayesian updating leads investors to lower their
belief about the expected consumption growth rate whenever the realized growth rate is lower
than their prior belief. Since the distance between the two mean states is small relative to the
total variation of consumption growth, this leads to a high correlation between ∆ct and ∆µt .
We estimate it to be 0.60 over the entire sample, and it ranges between 0.39 and 0.93 in 5-year
subperiods. To avoid this collinearity, we omit either variable in the first stage regression (18).
In untabulated results we confirm that the cross-sectional impact of volatility beliefs remains
robust.
To reduce noise in measured consumption, we also estimate a very basic consumption
tracking portfolio similar to Vassalou (2003) which also attenuates collinearity. To this end,
we regress consumption growth on the market return, HML and SMB. The fitted values
estimated from this regression can be interpreted as returns on a financial portfolio. This
additional step does not only reduce the collinearity identified above, but also mitigates the
concerns about measurement error in consumption data. A projection of consumption growth
onto the space of financial payoffs eliminates possible noise in consumption data which is
orthogonal to financial markets. At the same time, valuable information that is contained
in the consumption series but not in financial data might be lost. In unreported results, we
confirm previous findings.

D. Portfolio Sorts

An alternative approach to utilize the estimated risk loadings from regression (18) is to group
the estimates cross-sectionally and form portfolios. This approach has several important
advantages relative to Fama-McBeth regressions. First, the error-in-variable problem that led

18
to underestimated standard errors in the regression approach now leads to conservatism in
the statistical inferences. When variables are measured with noise, the portfolio sorts will be
less accurate as some stocks will be assigned to the wrong portfolio. Under the assumption
of cross-sectional predictive power, this leads to smaller return differences across portfolios.
Since the statistical inference is based solely on portfolio returns, the measurement error
ultimately leads to a decrease in statistical significance.
Second, forming portfolios allows to identify a non-linear relation between risk and ex-
pected excess returns. While the pricing kernel in equation (7) indicates that expected excess
returns are approximately log-linear in the wealth-consumption ratio, the wealth-consumption
ratio itself is not linear in beliefs about the mean or volatility states. Lastly, the portfolio
approach results in a time-series of returns, which allows a further analysis of the relation
between this strategy and other known risk factors.
At the end of each quarter, we sort all stocks in our sample into portfolios based on their
estimated risk loadings from the time-series regression (18). Table III reports the average
returns of equally-weighted (EW) and value-weighted (VW) quintiles as well as a long-short
strategy that each months invests $1 into quintile 5 (high risk) and sells $1 of quintile 1 (low
risk).
In Panel A, portfolios are formed based on loadings with respect to consumptions growth
i . Neither weighting scheme results in a measurable return dispersion across portfolios.
β̂c,t
A similar result obtains by forming portfolio based on changes in beliefs about expected
consumption growth β̂µ,t
i (Panel B). In contrast, consumption volatility risk β̂σ,t
i shows up

strongly negatively (Panel C). An equally-weighted strategy results in a return of the long-
short portfolio of −0.32% monthly. The value weighted return is even larger (in absolute
value) with −0.43% per month or in excess of −5% annually. In both cases, average returns
across quintiles are monotonically decreasing. The large difference in returns is thus not
driven by extreme observations in quintiles 1 and 5. Overall, the results in this table confirm
the findings from the cross-sectional regressions in Table II.
Cross-sectional differences in returns might not be surprising if consumption volatility
betas covary with other variables known to predict returns. Crucially, Table IV shows that
this is not the case for the firm characteristics size and book-to-market. In Panel A, we again
report average returns for each consumption volatility exposure quintile and its average beta.

19
Panel B reports firm characteristics for each portfolio. Since market capitalization is non-
stationary, and the value characteristic varies dramatically over time, we compute size- and
value deciles for each stock at each month and take the average over these deciles within each
portfolio. The table reports time-series means of portfolio characteristics. For market equity,
we observe that the two extreme quintiles are composed of somewhat smaller than average
stocks. This effect often shows up when ranking stocks by a covariance measure. Returns of
small stocks are on average more volatile and risk estimates are therefore more likely to be
very large or very small. However, there is no difference in size rank between quintile 1 and 5.
Most importantly, there is no variation in the book-to-market ratio across portfolios. Thus,
consumption risk portfolios do not load on firm characteristics which are known to predict
future returns.
A number of so-called anomalies are confined to a small subsets of stocks, often just
to small companies or illiquid stocks (e.g. Fama and French (2008), Avramov, Chordia,
Jostova, and Philipov (2007)). In Tables V and VI, stocks are independently sorted into three
portfolios based on β̂σ,t
i , and into two portfolios based on market capitalization (Table V) or

book-to-market ratio (Table VI). The number of portfolios for each variable follows Fama and
French (1993) and trades off the desire to obtain sufficient dispersion along each dimension
while keeping the number of stocks in each portfolio large enough to minimize idiosyncratic
risk. The bivariate sort in Table V shows that consumption volatility risk is consistently
present and strong for both equal and value-weighted strategies with return differences ranging
from −0.20% to −0.35% monthly. Interestingly, the effect is stronger for big than for small
companies since returns of smaller stocks have a larger idiosyncratic component, and thus the
risk estimates from the first stage regression are less precise. With these findings, there is no
reason to believe that the predictive power of consumption volatility risk is associated with
possible mispricing or slow information diffusion in small stocks. Similarly, Table VI confirms
that consumption volatility risk is also present within book-to-market groups.

IV. Consumption Volatility Risk Pricing

Building on the findings of the previous section, we now investigate the pricing implications of
beliefs about consumption moments cross-sectionally. Following the suggestions of Lewellen,
Nagel, and Shanken (2008), we use not only the 25 Fama and French (1992) portfolios as test

20
assets but we also employ the 5 value-weighted consumption volatility risk portfolios of Table
III, Panel C, as test assets.
The 25 Fama-French portfolios have been shown to challenge the single factor CAPM.
Fama and French (1993) propose two additional factors that help to explain the return differ-
ences: size (SMB) and value (HML). While a convincing, unified economic interpretation for
these factors is still outstanding, from an econometric view, the three factor model has been
proven very successful.
We show that changes in beliefs about consumption volatility carry a negative price of risk,
while changes in beliefs about the mean state do not contribute to explaining the cross-section
of returns. To circumvent possible econometric problems related to multicollinearity in the
independent variables, we also form a long-short portfolio based on consumption volatility risk
(VR) and demonstrate that it shows up strongly and significantly as a priced factor in cross-
sectional regressions. While the VR portfolio only modestly correlates with HML, both factors
are substitutes in the pricing relation. This evidence provides an economic interpretation for
the risk associated with the HML factor.

A. Factor Pricing with Consumption Data

Equation (17) states that, in a log-linear approximation, expected excess returns depend on
consumption growth and the perceived first and second moments of consumption growth. We
evaluate the performance of our model in two stages. First, for each 25 Fama-French portfolio,
we obtain risk loadings from the time-series regression

Rti − Rtf = αi + βci ∆ct + βµi ∆µ̂t + βσi ∆σ̂t + it i = 1, ..., 25 (21)

In the second stage, we estimate the prices of risk by a cross-sectional regression of returns
onto the loadings from the first stage.
Results from the second stage regression are summarized in Table VII. For each factor,
the table reports point estimates for the prices of risk and associated t-statistics, which are
adjusted for estimation error in the first stage as proposed by Shanken (1992) and are robust
to heteroscedasticity and autocorrelation as in Newey and West (1987) with 4 quarterly lags.
In addition, the following regression statistics are shown: The second stage R2 and adjusted
R2 as well as the model J-test (χ2 statistic) with its associated p-value (in percent). Return
observations are at a quarterly frequency and the factors used are log consumption growth

21
(∆ct ), changes in beliefs about the conditional mean of consumption growth (∆µ̂t ), as well
as changes in beliefs about consumption growth volatility (∆σ̂t ). A fourth factor, which is
the return of a long-short portfolio that buys assets with high consumption volatility risk and
sells assets with low consumption volatility risk, is also considered and denoted by VR.
Regression I shows results for the standard consumption CAPM. Confirming prior re-
search, the market price of consumption risk is insignificant and an R2 of about 5% indicates
that the C-CAPM performs very poorly in pricing the chosen set of test assets. Regressions II
- IV show the incremental effects of adding beliefs about conditional moments of consumption
growth. Even though results in the previous section have shown that exposure to ∆µ̂t does
not provide any return predictability, it is possible that ∆µ̂t carries a contemporaneous risk
premium. This is ruled out by regression II. The estimate for the price of risk on ∆µ̂t is
zero. In contrast, regression III shows that consumption volatility is a significant factor in the
cross-section and, as the previous section suggests, the estimate for the price of volatility risk
is negative. Regression IV reports the full three factor model (21). Similar to the previous
findings, both ∆ct and ∆µ̂t are insignificant. The price of volatility risk, ∆σ̂t , is negative,
and strongly significant with a t-statistic of about 2.
The full specification, however, suffers from a multicollinearity problem. While all explana-
tory variables are statistically insignificant, the model cannot be rejected, and the associated
p-value of 51 percent is overwhelmingly large. An F -test of the joint significance confirms
this intuition. The hypothesis that first stage risk loadings for ∆ct , ∆µ̂t , and ∆σ̂t are jointly
equal to zero in the second stage is rejected at any significance level.9
While this evidence presents strong indication of multicollinearity, econometric theory is
mostly silent about how to deal with it. We mitigate this concern by forming a consumption
volatility risk (VR) portfolio as a proxy for ∆σ̂t . The VR factor is a zero investment strat-
egy that is long in the value-weighted quintile with the highest exposure and short in the
value-weighted quintile with the lowest exposure to innovations in beliefs about consumption
volatility as measured by β̂σ,t
i in Table III, Panel D. The estimated risk loadings are obtained

from 20-quarter rolling regressions that end before the portfolio formation and are thus, con-
9
Other indicators of multicollinearity are the Variance Inflation Factor (VIF) or the condition number of
the matrix to be inverted (X 0 X). As a rule of thumb, multicollinearity is often considered problematic if the
VIF is greater than 5 or if the condition number is greater than 20. In the given problem, both number are
close to the corresponding values. The VIF is 4.25 and the condition number of the matrix (X 0 X) in the
second stage is 16.

22
ditional on consumption dynamics, publicly available information. We choose the univariate
β̂σ,t
i dispersion as basis for the factor since the consumption volatility exposure is unrelated

to size and value characteristics, as documented in Table IV.10

Regressions V and VI show a significantly negative price of risk for the VR factor, while
beliefs about the mean consumption growth continue to be insignificant. The price for a unit
of VR risk is estimated to be about 2.5% per quarter. The specifications with VR perform
well. The consumption CAPM augmented with the VR factor (Model VI) achieves second
stage R2 in excess of 30 percent, compared to 5 percent for the consumption CAPM.
The predictions of our theory in Section II depend on the preference parameters of the
representative agent. Under the common assumption of risk aversion greater than unity, an
EIS greater than one yields a negative price of risk for consumption volatility. While prior
research often finds a negative price of risk for market volatility (Ang, Hodrick, Xing, and
Zhang (2006), Adrian and Rosenberg (2008)), only a general equilibrium consumption-based
model allows us to draw conclusions about preference parameters. The estimated prices of
risk for both ∆σ̂t and its mimicking VR portfolio are significantly negative, thus, suggesting
an EIS greater than unity for the representative agent.
The second stage pricing errors are depicted in Figure 4. Each panel plots average quar-
terly excess returns against the model predicted excess returns for a given set of explanatory
variables. If the model correctly prices assets and there are no errors induced from estimation
or small sample size, all asset returns should line up exactly on the diagonal line. The top
left graph depicts the consumption CAPM (regression I in Table VII). Visually, this graph
confirms that the consumption CAPM does not work well in pricing the 25 Fama-French
portfolios. While the portfolios vary drastically in their average returns, the model predicted
returns are all very close together, resulting in a narrow cloud. In the top right graph, we aug-
ment the consumption CAPM with ∆σ̂t (regression III in Table VII). Adding ∆σ̂t as pricing
factor helps to achieve more variation in predicted returns, but the pricing errors remain fairly
large. In the two bottom graphs, we depict the full model with ∆σ̂t (bottom left, regression
IV), and the VR factor (bottom right, regression V). Both graphs confirm that in the full
model, either with ∆σ̂t or its mimicking portfolio, pricing errors are small and loadings on
10
We do not form a factor based on loadings on expected consumption growth (∆µ̂t ) since the spread between
the high and low quintile is on average close to zero. Consequently, theory predicts that its market price of
risk should be zero too.

23
risk factors successfully explain average excess returns.

B. Factor Pricing with Portfolio Returns

To relate the pricing implications of consumption volatility risk to the existing literature,
we now study market based rather than consumption based models. In particular, we are
interested whether our VR factor is a substitute for Fama-French’s HML factor, thereby
providing a macroeconomic interpretation for HML. Even though VR is independent of book-
to-market characteristics and comoves only modestly with HML, we find that substituting
HML by VR in the Fama-French three factor model results in similar pricing and leaves
pricing errors unaffected.
Summary statistics for the VR portfolio are given in Table VIII. The VR portfolio has a
mean return of −0.43% and a standard deviation of 3.25% per month. Its standard deviation
is lower than the market volatility, but comparable to the ones of the Fama-French factors.
The monthly Sharpe ratio (in absolute value) of 0.13 is larger in magnitude than the Sharpe
ratio of SMB (0.08) and close to the Sharpe ratio of HML (0.15). The correlation matrix of
the pricing factors (Panel B) shows that the VR portfolio returns are uncorrelated with the
market and SMB. Importantly, the correlation with the HML factor is moderate at −0.23
even though the VR portfolio is neutral with respect to the book-to-market characteristic
(see Panel B of Table IV). Overall, the correlations of VR with all other factors are smaller
than the pairwise correlations between the Fama French factors. Parameter estimates from
a time-series regression of the VR factor onto the other factors are reported in Panel C. The
CAPM (regression II) does not explain the returns of the VR portfolio. In regression III, the
Fama French factors attenuate the estimated intercept α̂ slightly towards zero, but it remains
large and significant. This attenuation is solely driven by HML and both the market and
SMB have insignificant coefficients.
Using the same econometric methods as in the previous section, we now obtain risk load-
ings for each of the 25 Fama-French portfolios from time-series regressions. In the second
stage, we estimate the prices of risk by a cross-sectional regression of returns onto the load-
ings from the first stage. Table IX reports factor loadings from first stage regressions of excess
returns on the market excess return (RM,t
E ) and the VR factor

Rti − Rtf = αi + βM
i
RM,t
E
+ βVi R V Rt + it i = 1, ..., 25 (22)

24
The risk estimates for both factors vary considerably across portfolios. Panel A confirms
previous findings about market factor loadings. Along the size dimension, market betas
decrease in size. While this is generally consistent with a risk based explanation of the size
effect, the dispersion in betas is not sufficient to explain the large dispersion in returns. Along
the value sorted portfolios, risk estimates actually decrease in the book-to-market ratio, while
returns increase. This well known finding challenges a risk based explanation and contradicts
the CAPM.
Risk exposures to the volatility risk factor are shown in Panel B. There is little variation
in risk loadings βVi R across size portfolios. Small stocks have an average loading of −0.02,
and the loading monotonically decreases to −0.09 for large stocks. The dispersion in the risk
loadings along the value dimension is much larger and decreases monotonically from 0.08 for
growth stocks to −0.14 for value stocks. A low risk exposure is consistent with high expected
returns for value stocks since the price of VR risk is negative. Loading on the VR factor
therefore suggest a risk based explanation of the value anomaly.
Results from second stage regressions are reported in Table X. For each factor, the table
presents point estimates for prices of risk. The associated t-statistics are based on standard
errors that are Shanken (1992) and Newey and West (1987) adjusted. In addition, the fol-
lowing regression statistics are shown: The second stage R2 and adjusted R2 as well as the
model J-test (χ2 statistic) with its associated p-value (in percent). Regressions I and IV show
the results for the benchmark models, the market CAPM (I) and the Fama-French model
(IV). The CAPM does a very poor job in explaining the cross-section of returns. The point
estimate for the market risk premium is negative and the regression R2 is only 7%. The three
factor model reduces the pricing errors significantly and yields an R2 of 77%. The estimated
market risk premium remains negative and the model is still rejected as indicated by the high
χ2 statistic.
The remaining regressions show various combinations of the Fama-French factors with VR.
In all specifications, the estimates for the price of a unit VR risk are significant and negative,
ranging from −0.39% to −0.67% monthly. To gauge the economic impact of VR on the cross-
section of returns, we multiply average VR loadings of the value quintiles with the market
price of VR risk. Panel B of Table IX shows that the average risk loadings monotonically
decrease from 0.08 to −0.14 along the book-to-market characteristic, yielding a differential

25
of −0.22. Assuming a conservative price of VR risk (regression VI), this translates into an
annual expected return differences between value and growth stocks of 1%.
In regression III, the factors are the market portfolio and VR as in Equation (22). This
specification yields considerable improvements over the one factor market model and results
in a similar fit relative to a model that contains the market and HML (regression VII).
Interestingly, although VR is based on consumption data, a three factor model based on the
market, SMB and VR (regression V) produces an R2 of 73% which is slightly lower than the
R2 of the Fama-French model. Augmenting the Fama-French three-factor model with VR
as a fourth factor (regression VI) does not lead to an improvement in the model’s ability to
price the cross-section. In summary, replacing HML by VR does not deteriorate the model’s
performance, while including both VR and HML as factors does not seem to improve the
model fit. Hence, HML and VR are substitutes in cross-sectional pricing for the 25 Fama
French portfolios. In contrast to HML, however, the consumption volatility risk portfolio has
a clear economic interpretation.
Adrian and Rosenberg (2008) perform a similar analysis. They decompose stock market
volatility into two components, which differ in persistence, and estimate them with a GARCH
inspired model. In contrast, our VR portfolio is based on a Markov model for low-frequency
consumption data. Interestingly, their short-run volatility component has similar pricing im-
plications as VR, whereas their long-run component performs worse than VR. However, the
persistence of their short-run volatility component is 0.327 for daily data while our consump-
tion volatility regimes last on average for several years. The VR factor thus has a much
different and macroeconomically more meaningful interpretation.
Figure 5 replicates Figure 4 for market based pricing models. The top left graph depicts
the CAPM (regression I in Table X). The remaining graphs show the CAPM augmented with
the volatility risk factor (top right graph, regression III), the Fama-French three factor model
(bottom left graph, regression IV), and a three factor model that uses VR instead of HML
(bottom right graph, regression V). Visually, these graphs confirm that simply adding VR
to the market factor improves the model fit dramatically. Both the Fama-French model and
the three factor VR model, though statistically still rejected, seem to price the 25 portfolios
equally well.

26
V. Time Series Predictability

In the previous sections, we showed that loadings on consumption growth volatility predict
returns cross-sectional and consumption growth volatility is a priced risk factor. The model
also predicts that the first and second moments of consumption growth forecast aggregate
returns in the time-series. As explained in Section II, the model implies a negative relation
between expected excess returns and expected consumption growth and a positive relation
between expected excess returns and consumption growth volatility when both risk aversion
and EIS are greater than unity. Noting that the wealth-consumption ratio is inversely related
to expected excess returns, this effect can be seen in Figure 2. The opposite holds when risk
aversion and EIS are smaller than unity (see Figure 3).
We find that changes in consumption growth volatility are a strong and robust predictor
of short horizon market returns whereas expected consumption growth is not a significant pre-
dictor variable. Specifically, higher consumption growth volatility is associated with higher
future excess returns. This positive relation is consistent with previous results where loadings
on consumption growth volatility are negatively related to returns. In a univariate regres-
sion of future returns on the consumption growth volatility, the regression R2 is of similar
magnitude as the one obtained in a univariate regression using the consumption-wealth ratio
variable cay of Lettau and Ludvigson (2001a).
Table XI reports time-series regressions of quarterly excess market returns onto lagged
predictor variables. Regressions I-III represent standard benchmark models. As predictor
variables, we use the Treasury bill rate relative to its recent average as proposed by Hodrick
(1992), the term spread, the default spread, and the dividend yield.11 All of those have been
shown to predict stock returns at various horizons.12 Lettau and Ludvigson (2001a) use the
household budget constraint to motivate the variable cay and show that it works exceptionally
well at short horizon forecasts.13
In regressions IV and V, we study the predictive power of our two consumption state vari-
11
Data are from FRED. The relative Treasury bill rate is the yield on a 90 day T-bill less its past 12
months moving average. The term spread is the difference between yields of long (10 year) and short (1 year)
government bonds, and the default spread is the yield difference between Baa rated and Aaa rated corporate
bonds. The dividend yield is computed from CRSP as the ratio of gross cum-dividend index returns to gross
ex-dividend returns.
12
See, for example Rozeff (1984), Campbell and Shiller (1988), Fama and Schwert (1977), Keim and Stam-
baugh (1986), Campbell (1987), and Campbell and Thompson (2008).
13
The variable cay was downloaded from Martin Lettau’s homepage on Oct. 18, 2008.

27
ables. Similar to the cross-sectional results in the previous sections, we find that beliefs about
expected consumption growth do not predict stock returns, while changes in the volatility
beliefs show up economically and statistically significant and yield an regression R2 of al-
most 5% in a univariate regression. The R2 of cay in the univariate regression II is slightly
higher at about 6%. The economic impact of consumption volatility risk is large. A one stan-
dard deviation increase in ∆σ̂t results in an increase in the expected risk premium of 1.8%
quarterly.14 Regressions VI to VIII demonstrate that the marginal impact of ∆σ̂t remains
strong and significant even after controlling for all other predictors including cay. This is
surprising since in our model, changes in consumption volatility enter the pricing kernel only
because they affect the consumption-wealth ratio. Thus, one might expect that measures of
the consumption-wealth ratio such as cay already contain information about the volatility
state. In the data, however, this is not the case suggesting that cay is an imperfect measure
of the wealth-consumption ratio.
It is well known that parameter estimates and t-statistics are potentially biased in predic-
tive regressions. Persistence in stock returns, caused e.g. by using overlapping observations,
leads to biased standard errors, which is examined in Hodrick (1992). In our setup, we do not
use overlapping observations and the autocorrelation in market returns is very small. Another
bias arises when the predictor variable is persistent and its innovations are correlated with
future returns, as discussed in Stambaugh (1999), Lewellen (2004), Boudoukh, Richardson,
and Whitelaw (2006) and Ang and Bekaert (2007). Especially when price ratios are used as
predictors, this bias shows up strongly and conventional tests will reject the null hypothesis
too frequently. To gain a better understanding, consider the following setup

rt = α + βxt−1 + rt

xt = φ + ρxt−1 + xt

where rt denotes returns and xt a predictor variable. Lewellen (2004) shows that β estimates
are biased by γ(ρ̂ − ρ) where γ = Cov(rt , xt )/Var(xt ) when rt is correlated with xt . When
the dividend yield is used as predictor, for instance, an increase in price leads to a positive
realized return as well as a decrease in the dividend yield. Consequently, rt is correlated with
xt . Lewellen (2004) reports an auto-correlation of 0.997 and Corr(rt , xt ) = −0.96 for the
14
Note that ∆σt has a standard deviation of around 0.0003.

28
dividend yield as predictor, invaliding standard estimates and tests. For the variable ∆σ̂t ,
this bias is less of a concern since it is not a price scaled variable. For our one period forecasts,
we estimate Corr(∆σ̂t , ∆σ̂t−1 ) = 0.006 and Corr(∆σ̂t , εt ) = 0.032, which is too small to bias
statistical inference.
We acknowledge that the predictive results presented have limitations. First, they are in
sample results. We leave evaluating the out of sample power for future work. Second, there
is a look-ahead bias in ∆σ̂t . In estimating the Markov chain for consumption growth, beliefs
are updated according to Bayes’ rule and therefore are not forward looking. The parameter
estimates, however, are obtained by maximum likelihood, employing the full sample. In
particular, investors in the early sample period know of the possibility that at one point in
the future consumption volatility might switch to a state much lower than what had been
experienced in the past. This is similar to the critique by Brennan and Xia (2005), who point
out that estimating cay over the entire sample induces a look ahead bias and a simple linear
time trend would work as well as cay. Their criticism does not apply to our results since we
use changes in beliefs as predictor which do not have a trend. Third, aggregate consumption
data is not publicly available at the end of a quarter. Instead, initial estimates are published
within the following month and they are subject to revisions for up to three years. Hence, we
cannot conclude that it is possible to implement our predictability results in practice. Yet we
succeed in identifying a new source of aggregate risk.

VI. Conclusion

When consumption growth is not i.i.d. over time and the representative household has recur-
sive preferences, the wealth-consumption ratio is time-varying and enters the pricing kernel as
a second factor (Epstein and Zin (1989), Weil (1989)). We generalize Bansal and Yaron (2004)
to account for the latent nature of the conditional first and second moments of consumption
growth. In the model, we identify innovations in beliefs about the conditional mean and
volatility of consumption growth as two state variables that affect the wealth-consumption
ratio and thus asset prices.
To test these predictions, we estimate a Markov model with two states for the conditional
mean and two states for the conditional volatility of consumption growth, as in Kandel and
Stambaugh (1990) and Lettau, Ludvigson, and Wachter (2008). Using the estimated beliefs

29
from the Markov model, we empirically test the pricing implications for the cross-section
and time-series of stock returns. We find that an asset’s exposure to changes in beliefs about
consumption volatility significantly forecast returns, while exposure to changes in beliefs about
expected consumption growth do not.
In cross-sectional pricing tests using the 25 Fama-French size and value portfolios as well
as 5 consumption volatility risk portfolios, both, changes in volatility and a portfolio, which is
long assets with high volatility exposure and short assets with low volatility exposure, show up
as priced factors. The effects are robust to a variety of pricing models, including augmented
versions of the consumptiom CAPM, the market CAPM, and the Fama-French three factor
model. The comparison of pricing errors across different model specifications indicates that
the volatility risk factor has similar pricing implications to the value factor HML, even though,
the volatility risk factor is neutral with respect to the value characteristic.
In time-series tests, we find that innovations to beliefs about the volatility state forecast
the equity premium. A one standard deviation increase in perceived volatility is followed
by an increase of the equity returns of 1.8% quarterly. In a univariate regression, changes
about perceived consumption volatility achieve an R2 of about 5% which is comparable to
predictive power of Lettau and Ludvigson (2001a) cay variable. The signs of the coefficients
on consumption volatility in time-series and cross-sectional regressions indicate that the rep-
resentative agent has risk aversion and elasticity of intertemporal substitution greater than
unity.

30
References
Adrian, Tobias, and Joshua Rosenberg, 2008, Stock returns and volatility: Pricing the short-run and long-run
components of market risk, Journal of Finance 63.

Aı̈t-Sahalia, Yacine, Jonathan A. Parker, and Motohiro Yogo, 2004, Luxury goods and the equity premium,
Journal of Finance 59, 2959–3004.

Ang, Andrew, and Geert Bekaert, 2007, Stock return predictability: Is it there?, Review of Financial Studies
20, 651–707.

Ang, Andrew, Robert J. Hodrick, Yuhang H. Xing, and Xiaoyan Y. Zhang, 2006, The cross-section of volatility
and expected returns, Journal of Finance 61, 259–299.

Attanasio, O. P., and G. Weber, 1993, Consumption growth, the interest-rate and aggregation, Review of
Economic Studies 60, 631–649.

Avramov, Doron, Tarun Chordia, Gergana Jostova, and Alexander Philipov, 2007, Momentum and credit
rating, Journal of Finance 62, 2503–2520.

Balduzzi, Pierluigi, and Tong Yao, 2007, Testing heterogeneous-agent models: an alternative aggregation
approach, Journal of Monetary Economics 54, 369–412.

Bansal, Ravi, Robert F. Dittmar, and Dana Kiku, 2009, Cointegration and consumption risks in asset returns,
Forthcoming, Review of Financial Studies.

Bansal, Ravi, Robert F. Dittmar, and Christian T. Lundblad, 2005, Consumption, dividends, and the cross
section of equity returns, Journal of Finance 60, 1639–1672.

Bansal, Ravi, Varoujan Khatchatrian, and Amir Yaron, 2005, Interpretable asset markets?, European Economic
Review 49, 531–560.

Bansal, Ravi, Dana Kiku, and Amir Yaron, 2007, Risks for the long run: Estimation and inference, Working
Paper.

Bansal, Ravi, and Ivan Shaliastovich, 2008, Confidence risk and asset prices, Working Paper.

Bansal, Ravi, and Amir Yaron, 2004, Risks for the long run: A potential resolution of asset pricing puzzles,
Journal of Finance 59, 1481–1509.

Bhamra, Harjoat S., Lars-Alexander Kuehn, and Ilya A. Strebulaev, 2007, The levered equity risk premium
and credit spreads: A unified framework, University of British Columbia Working Paper.

Bollerslev, Tim, Georg Tauchen, and Hao Zhou, 2008, Expected stock returns and variance risk premia,
Working Paper.

Bonomo, Marco, and Rene Garcia, 1994, Can a well-fitted equilibrium asset-pricing model produce mean
reversion, Journal of Applied Econometrics 9, 19–29.

Boudoukh, Jacob, Matthew Richardson, and Robert F. Whitelaw, 2006, The myth of long-horizon predictabil-
ity, Forthcoming, Review of Financial Studies.

Breeden, Douglas T., Michael R. Gibbons, and R. H. Litzenberger, 1989, Empirical tests of the consumption-
oriented capm, Journal of Finance 44, 231–262.

Brennan, Michael J., and Yihong Xia, 2005, tay’s as good as cay, Finance Research Letters 2, 1–14.

31
Calvet, Laurent, and Adlai Fisher, 2007, Multifrequency news and stock returns, Journal of Financial Eco-
nomics 86, 178–212.

Campbell, John Y., 1987, Stock returns and the term structure, Journal of Financial Economics 18, 373–399.

, 1996, Understanding risk and return, Journal of Political Economy 104, 298–345.

, and Ludger Hentschel, 1992, No news is good-news - an asymmetric model of changing volatility in
stock returns, Journal of Financial Economics 31, 281–318.

Campbell, John Y., and Gregory Mankiw, 1989, Consumption, income, and interest rates: Reinterpreting the
time series evidence, in O. Blanchard Fisher, and S., ed.: National Bureau Economic Research Macroeco-
nomics Annual: 1989 . pp. 185–216 (MIT Press: Cambridge, MA).

Campbell, John Y., and Robert Y. Shiller, 1988, The dividend-price ratio and expectations of future dividends
and discount rates, Review of Financial Studies 1, 195–228.

Campbell, John Y., and Samuel B. Thompson, 2008, Predicting excess stock returns out of sample: Can
anything beat the historical average?, Review of Financial Studies 21, 1509–1531.

Campbell, John Y., and Tuomo Vuolteenaho, 2004, Bad beta, good beta, American Economic Review 94,
1249–1275.

Cecchetti, Stephen G., and Nelson C. Mark, 1990, Evaluating empirical tests of asset pricing models: Alter-
native interpretations, American Economic Review 80, 48–51.

David, Alexander, 1997, Fluctuating confidence in stock markets: Implications for returns and volatility,
Journal of Financial and Quantitative Analysis 32, 427–462.

Drechsler, Itamar, and Amir Yaron, 2008, What’s vol got to do with it, Working Paper.

Engle, Robert F., 1982, Autoregressive conditional heteroscedasticity with estimates of the variance of united
kingdom inflation, Econometrica 50, 987–1007.

Epstein, Larry, and Stanley Zin, 1989, Substitution, risk aversion, and the temporal behavior of consumption
and asset returns: A theoretical framework, Econometrica 57, 937–969.

Fama, Eugene F., and Kenneth R. French, 1992, The cross-section of expected stock returns, Journal of Finance
47, 427–465.

, 1993, Common risk-factors in the returns on stocks and bonds, Journal of Financial Economics 33,
3–56.

, 2008, Dissecting anomalies, Journal of Finance 63, 1653–1678.

Fama, Eugene F., and James D. MacBeth, 1973, Risk, return, and equilibrium - empirical tests, Journal of
Political Economy 81, 607–636.

Fama, Eugene F., and G. William Schwert, 1977, Asset returns and inflation, Journal of Financial Economics
5, 115–146.

Glosten, Lawrence R., Ravi Jagannathan, and David E. Runkle, 1993, On the relation between the expected
value and the volatility of the nominal excess return on stocks, Journal of Finance 48, 1779–1801.

Hall, Robert E., 1988, Intertemporal substitution in consumption, Journal of Political Economy 96, 339–57.

Hamilton, James D., 1989, A new approach to the economic-analysis of nonstationary time-series and the
business-cycle, Econometrica 57, 357–384.

32
 , 1994, Time Series Analysis (Princeton University Press).

Hansen, Lars Peter, John C. Heaton, and Nan Li, 2008, Consumption strikes back? Measuring long-run risk,
Journal of Political Economy 116, 260–302.

Hodrick, R. J., 1992, Dividend yields and expected stock returns - alternative procedures for inference and
measurement, Review of Financial Studies 5, 357–386.

Jacobs, Kris, and Kevin Q. Wang, 2004, Idiosyncratic consumption risk and the cross section of asset returns,
Journal of Finance 59, 2211–2252.

Jagannathan, Ravi, and Yong Wang, 2007, Lazy investors, discretionary consumption, and the cross-section of
stock returns, Journal of Finance 62, 1623–1661.

Jagannathan, Ravi, and Zhenyu Wang, 1996, The conditional CAPM and the cross-section of expected returns,
Journal of Finance 51, 3–53.

Kandel, Shmuel, and Robert F. Stambaugh, 1990, Expectations and volatility of consumption and asset returns,
Review of Financial Studies 3, 207–32.

, 1991, Asset returns and intertemporal preferences, Journal of Monetary Economics 27, 39–71.

Keim, Donald B., and Robert F. Stambaugh, 1986, Predicting returns in the stock and bond markets, Journal
of Financial Economics 17, 357–390.

Kim, Chang-Jin, and Charles R. Nelson, 1999, Has the us economy become more stable? a bayesian approach
based on a markov-switching model of the business cycle, Review of Economics and Statistics 81, 608–616.

Kreps, David M., and Evan L. Porteus, 1978, Temporal resolution of uncertainty and dynamic choice theory,
Econometrica 46, 185–200.

Lettau, Martin, and Sydney Ludvigson, 2001a, Consumption, aggregate wealth, and expected stock returns,
Journal of Finance 56, 815–849.

, 2001b, Resurrecting the (C)CAPM: A cross-sectional test when risk premia are time-varying, Journal
of Political Economy 109, 1238–1287.

, and Jessica Wachter, 2008, The declining equity premium: What role does macroeconomic risk play?,
Review of Financial Studies 21, 1607–1652.

Lewellen, Jonathan, 2004, Predicting returns with financial ratios, Journal of Financial Economics 74, 209–235.

, Stefan Nagel, and Jay Shanken, 2008, A skeptical appraisal of asset pricing tests, Journal of Financial
Economics.

Lustig, Hanno, and Stijn Van Nieuwerburgh, 2005, Housing collateral, consumption insurance and risk premia:
An empirical perspective, Journal of Finance 60, 1167–1219.

Lustig, Hanno, Stijn Van Nieuwerburgh, and Adrien Verdelhan, 2008, The wealth-consumption ratio, Working
Paper.

Mankiw, N. Gregory, and Matthew D. Shapiro, 1986, Risk and return: Consumption beta versus market beta,
Review of Economics and Statistics 68, 452–459.

Newey, Whitney K., and Kenneth D. West, 1987, A simple, positive semidefinite, heteroskedasticity and
autocorrelation consistent covariance-matrix, Econometrica 55, 703–708.

33
Parker, Jonathan A., and Christian Julliard, 2005, Consumption risk and the cross section of expected returns,
Journal of Political Economy 113, 185–222.

Piazzesi, Monika, Martin Schneider, and Selale Tuzel, 2007, Housing, consumption, and asset pricing, Journal
of Financial Economics 83, 531–569.

Pindyck, Robert S., 1984, Risk, inflation, and the stock market, The American Economic Review 74, 335–351.

Poterba, James M., and Lawrence H. Summers, 1986, The persistence of volatility and stock market fluctua-
tions, The American Economic Review 76, 1142–1151.

Restoy, Fernando, 1991, Essays in intertemporal asset pricing, PhD dissertation, Harvard University.

, and Philippe Weil, 2004, Approximate equilibrium asset prices, Working Paper.

Rozeff, Michael S., 1984, Dividend yields are equity risk premiums, Journal of Portfolio Management 11, 68–75.

Santos, Tano, and Pietro Veronesi, 2006, Labor income and predictable stock returns, Review of Financial
Studies 19, 1–44.

Shanken, Jay, 1992, On the estimation of beta-pricing models, Review of Financial Studies 5, 1–33.

Stambaugh, Robert F., 1999, Predictive regressions, Journal of Financial Economics 54, 375–421.

Tedongap, Romeo, 2007, Consumption volatility and the cross-section of stock returns, Working Paper.

Vassalou, Maria, 2003, News related to future gdp growth as a risk factor in equity returns, Journal of Financial
Economics 68, 47–73.

Veronesi, Pietro, 1999, Stock market overreaction to bad news in good times: A rational expectations equilib-
rium model, Review of Financial Studies 12, 975–1007.

Vissing-Jorgensen, A., 2002, Limited asset market participation and the elasticity of intertemporal substitution,
Journal of Political Economy 110, 825–853.

, and O. P. Attanasio, 2003, Stock-market participation, intertemporal substitution, and risk-aversion,

American Economic Review 93, 383–391.

Weil, Philippe, 1989, The equity premium puzzle and the risk-free rate puzzle, Journal of Monetary Economics
24, 401–421.

Whitelaw, Robert F., 2000, Stock market risk and return: An equilibrium approach, Review of Financial
Studies 13, 521–47.

Wilcox, David W., 1992, The construction of u.s. consumption data: Some facts and their implications for
empirical work, The American Economic Review 82, 922–941.

Yogo, Motohiro, 2006, A consumption-based explanation of expected stock returns, Journal of Finance 61,
539–580.

34
 Figure 1. Bayesian Beliefs about the Mean and Volatility State
This figure displays the estimated Bayesian belief process for the state of the conditional mean
(top figure) and conditional volatility (bottom figure) of consumption growth. The estimation
procedure follows Hamilton (1994). We use quarterly per capita real non-durable plus service
consumption.

Posterior Probability of High Mean State

1

0.8

0.6

0.4

0.2

0
1950 1960 1970 1980 1990 2000 2010

Posterior Probability of High Volatility State

1

0.8

0.6

0.4

0.2

0
1950 1960 1970 1980 1990 2000 2010

35
 Figure 2. Wealth-Consumption Ratio for a high EIS Agent
This figure shows the wealth-consumption ratio as a function of the perceived conditional first
µ̂t (left graph) and second σ̂t (right graph) moment of consumption growth for the benchmark
calibration when the agent has a high EIS of 1.5. Further, relative risk aversion equals 30 and
the quarterly rate of time preference is 0.995.

l ow σ t l ow µ t
5.682 hi gh σ t 5.682 hi gh µ t

5.68 5.68

5.678 5.678
Log PC−ratio

Log PC−ratio

5.676 5.676

5.674 5.674

5.672 5.672

5.67 5.67
0.4 0.5 0.6 0.7 0.25 0.3 0.35 0.4 0.45
Perceived mean (%) Perceived volatility (%)

36
 Figure 3. Wealth-Consumption Ratio for a low EIS Agent
This figure shows the wealth-consumption ratio as a function of the perceived conditional first
µ̂t (left graph) and second σ̂t (right graph) moment of consumption growth for the benchmark
calibration when the agent has a low EIS of 0.5. Further, relative risk aversion equals 30 and
the quarterly rate of time preference is 0.995.

l ow σ t l ow µ t
hi gh σ t hi gh µ t
4.63 4.63

4.625 4.625
Log PC−ratio

Log PC−ratio

4.62 4.62

4.615 4.615

4.61 4.61

4.605 4.605

0.4 0.5 0.6 0.7 0.25 0.3 0.35 0.4 0.45

Perceived mean (%) Perceived volatility (%)

37
 Figure 4. Pricing Errors of the Consumption-Based Model
This figure depicts average quarterly excess returns of the 25 Fama-French portfolios (black
dots) and 5 volatility risk portfolios (red stars) against model predicted excess returns. In
all graphs, consumption growth (∆ct ) is included as explanatory factor. In the top-right and
bottom-left graph, we add changes in the perceived second (∆σ̂t ) and first (∆µ̂t ) moments of
consumption growth. In the bottom-right graph, we replace beliefs about the volatility state
with the VR factor in the full model.

∆c ∆c , ∆σ
4 4
Realized excess return (%)

Realized excess return (%)

3 3

2 2

1 1

0 0
0 1 2 3 4 0 1 2 3 4
Predicted excess return (%) Predicted excess return (%)
∆c , ∆µ , ∆σ ∆c , ∆µ , VR
4 4
Realized excess return (%)

Realized excess return (%)

3 3

2 2

1 1

0 0
0 1 2 3 4 0 1 2 3 4
Predicted excess return (%) Predicted excess return (%)

38
 Figure 5. Pricing Errors of the Market-Based Model
This figure depicts average quarterly excess returns of the 25 Fama-French portfolios (black
dots) and 5 volatility risk portfolios (red stars) against model predicted excess returns. The
top-left graph represents the standard CAPM and the bottom-left graph the Fama-French
three factor model. In the top-right graph, we add the VR factor to the CAPM and, in the
bottom-right graph, we replace the HML factor with VR in the Fama-French model.

RE
M
E
RM , VR
Realized excess return (%)

Realized excess return (%)

1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
0 0.5 1 0 0.5 1
Predicted excess return (%) Predicted excess return (%)
REM , SMB, HML REM , SMB, VR
Realized excess return (%)

Realized excess return (%)

1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
0 0.5 1 0 0.5 1
Predicted excess return (%) Predicted excess return (%)

39
 Table I
Markov Model of Consumption Growth

This table reports parameter estimates of the Markov model for log consumption growth

∆ct+1 = µt + σt t+1 t ∼ N (0, 1)

where µt ∈ {µl , µh } and σt ∈ {σl , σh } follow independent Markov processes with transition
matrices P µ and P σ respectively.

pµll 1 − pµll pσll 1 − pσll

   
P =µ
P =
σ
1 − pµhh pµhh 1 − pσhh pσhh

Standard errors are reported in parenthesis. Panel C shows the correlation between the filtered
beliefs for each state.

Panel A: Parameter Estimates

µl µh σl σh
0.3668 0.7800 0.2112 0.4772
(0.0376) (0.0600) (0.0279) (0.0532)

Panel B: Marginal Transition Probabilities

pllµ phh
µ pllσ phh
σ
0.9304 0.8929 0.9617 0.9452
(0.0353) (0.0597) (0.0458) (0.0399)

Panel C: Correlation of Beliefs

bµ bσ
bµ 1
bσ 0.0821 1

40
 Table II
Fama-MacBeth Regressions

This table reports cross-sectional regressions of monthly returns on estimated risk loadings
and characteristics. Time-varying risk loadings are obtained from 5-year rolling time-series
regressions of individual excess returns on the market excess return, log consumption growth,
and changes in the perceived conditional mean and volatility of consumption growth using
quarterly data. In the cross-section, we regress monthly future returns onto the loadings of
log consumption growth (β̂c,t
i ), changes in the perceived conditional mean (β̂ i ) and volatility
µ,t
i ) of consumption growth as well as market capitalization (M E i ) and book-to-market ratio
(β̂σ,t t
(BMti ). Both characteristics are measured in December which is 7 to 18 months prior the test
month. All explanatory variables are normalized so they are centered around zero with unit
variance. Reported are time-series averages of the second stage coefficients with associated
t-statistics. The sample period is from January 1964 to December 2008. t-statistics are in
parenthesis.

Fama-MacBeth Regressions
β̂c,t
i β̂µ,t
i β̂σ,t
i M Eti BMti

I 0.043
(0.85)
II 0.000
(0.00)
III -0.134
(-3.23)
IV 0.159 0.033 -0.151
(1.98) (0.57) (-2.45)
V 0.178 0.044 -0.113 -0.057 0.490
(2.36) (0.82) (-2.10) (-1.74) (2.54)

41
 Table III
Portfolios formed on Risk Exposure

Each quarter stocks are assigned into quintiles based on loadings from time series regres-
sions (18) of individual excess returns on log consumption growth (β̂c,t
i ) and changes in the

perceived first (β̂µ,t ) and second moment (β̂σ,t ) of consumption growth. This table reports
i i

average equally-weighted (EW) and value-weighted (VW) returns of these portfolios and their
associated t-statistics.

Panel A: Univariate sorts based on β̂c,t

i

β̂c,t
i low med high high - low

EW 1.08 1.17 1.15 1.18 1.24 0.16

(4.22) (5.53) (5.47) (5.02) (4.19) (1.14)
VW 0.84 0.88 0.80 0.84 0.89 0.06
(3.96) (4.87) (4.40) (4.27) (3.46) (0.30)

Panel B: Univariate sorts based on β̂µ,t

i

β̂µ,t
i low med high high - low

EW 1.25 1.10 1.14 1.12 1.19 -0.06

(4.69) (5.17) (5.42) (4.89) (4.20) (-0.48)
VW 0.99 0.80 0.83 0.83 0.96 -0.03
(4.19) (4.31) (4.65) (4.29) (3.74) (-0.14)

Panel C: Univariate sorts based on β̂σ,t

i

β̂σ,t
i low med high high - low

EW 1.29 1.20 1.17 1.14 1.06 -0.24

(4.58) (5.24) (5.46) (5.23) (4.08) (-1.90)
VW 1.09 0.91 0.84 0.84 0.65 -0.44
(4.61) (4.69) (4.64) (4.29) (2.67) (-2.43)

42
 Table IV
Characteristics of Volatility Risk Exposure Portfolios

This table reports various characteristics and risk measures for the quintile portfolios based
on consumption volatility loadings (as in Table III, Panel D). Panel A shows average returns
and consumption volatility betas. Panel B reports the average value and mean decile rank
for size and book-to-market characteristics of each portfolio.

Panel A: Univariate sorts based on β̂σ,t

i

low med high high - low

RET 1.18 0.99 0.96 0.99 0.75 -0.43

(5.33) (5.36) (5.30) (5.18) (3.27) (-3.02)
β̂σ /100 -3.70 -1.18 0.01 1.24 4.17 7.87

Panel B: Characteristics of sorts based on β̂σ,t

i

low med high high - low

ME 1100.91 1789.75 1867.59 1470.50 794.74 -306.18

ME RANK 4.04 5.31 5.55 5.25 4.03 0.01

BM 1.00 0.94 0.94 0.94 0.97 -0.03

BM RANK 5.48 5.46 5.51 5.48 5.44 0.04

43
 Table V
Portfolios Formed on Consumption Volatility Risk and Market Capitalization

This table reports average returns of equally-weighted (Panel A) and value-weighted (Panel
B) portfolios formed independently on consumption volatility exposure (low 30%, medium
40%, and high 30%) and market capitalization (small 50% and big 50%).

Panel A: Equally-Weighted Returns

Low Med High H-L

Small 1.57 1.50 1.37 -0.20

(5.35) (6.05) (4.93) (-2.12)
Big 1.24 1.15 0.99 -0.24
(5.33) (5.79) (4.20) (-2.75)
S-B 0.33 0.35 0.37
(2.09) (2.78) (2.51)

Panel B: Value-Weighted Returns

Low Med High H-L

Small 1.47 1.43 1.27 -0.20

(5.45) (6.02) (4.74) (-1.85)
Big 1.21 1.08 0.96 -0.25
(5.57) (5.73) (4.24) (-2.73)
S-B 0.26 0.34 0.32
(1.95) (3.09) (2.35)

44
 Table VI
Portfolios Formed on Consumption Volatility Risk and Book-to-Market Ratio

This table reports average returns of equally-weighted (Panel A) and value-weighted (Panel
B) portfolios formed independently on consumption volatility exposure (low 30%, medium
40%, and high 30%) and the ratio of book equity to market equity (low 50% and high 50%).

Panel A: Equally-Weighted Returns

Low Med High H-L

Low BM 1.19 1.04 0.84 -0.35

(4.56) (4.76) (3.34) (-3.36)
High BM 1.52 1.35 1.31 -0.22
(6.13) (6.73) (5.47) (-2.55)
H-L 0.34 0.31 0.47
(3.63) (3.77) (4.58)

Panel B: Value-Weighted Returns

Low Med High H-L

Low BM 1.03 0.91 0.80 -0.24

(4.93) (4.80) (3.66) (-1.85)
High BM 1.29 1.06 0.99 -0.30
(6.29) (6.19) (4.65) (-2.15)
H-L 0.26 0.15 0.20
(2.17) (1.44) (1.46)

45
 Table VII
Volatility Risk Pricing

This table reports second stage regressions to estimate market prices to risk. ∆ct denotes log
consumption growth, ∆µ̂t and ∆σ̂t are changes in filtered beliefs about the first and second
moment of consumption growth. The VR factor is the return of holding a long position in
the value-weighted quintile of stocks with high volatility risk (β̂σ,t
i ) and a short position in

low volatility risk, as reported in Panel D of Table III. Test assets are the value weighted 25
Fama-French size and book-to-market portfolios as well as the 5 value weighted consumption
volatility risk portfolios as in Panel C of Table III. The data covers January 1964 until
December 2008. The t-statistics are corrected for estimation error in the first stage as proposed
by Shanken (1992) and are Newey and West (1987) adjusted to account for heteroskedasticity
and autocorrelation. For each specification, we report the R2 , adjusted R2 in parenthesis,
regression J-statistic (χ2 ) with the associated p-value (in %), mean absolute pricing error
(MAPE) and root square mean square error (RMSE).

CONST. ∆ct ∆µ̂t ∆σ̂t VR R2 χ2 MAPE

(t-stat) (t-stat) (t-stat) (t-stat) (t-stat) (adj. R2 ) (p-val) (RMSE)

I 1.21 0.14 4.82 103.32 0.61

(1.60) (0.73) (1.42) (0.00) (0.74)
II 1.10 0.28 -0.00 8.94 75.92 0.59
(1.33) (1.67) (-0.13) (2.19) (0.00) (0.72)
III 1.06 0.10 -0.04 10.20 44.77 0.59
(0.91) (0.35) (-1.86) (3.55) (2.33) (0.72)
IV 0.88 0.28 -0.02 -0.06 17.32 26.09 0.56
(0.62) (1.00) (-0.54) (-1.98) (7.78) (51.35) (0.69)
V 1.11 0.21 -0.01 -2.48 37.58 68.18 0.49
(1.32) (1.29) (-0.41) (-3.45) (30.38) (0.00) (0.60)
VI 1.23 0.05 -2.52 32.50 91.08 0.51
(1.58) (0.28) (-3.61) (27.50) (0.00) (0.62)

46
 Table VIII
Volatility Risk Factor

This table provides descriptive statistics of the volatility risk (VR) portfolio. The VR portfolio
is the return of holding a long position in the value-weighted quintile of stocks with high
volatility risk (β̂σ,t
i ) and a short position in low volatility risk, as reported in Panel C of Table

III. The time-series for the VR portfolio covers January 1964 until December 2008. Panel A
shows the mean, standard deviation and the Sharpe Ratio of the VR portfolio as well as the
three Fama-French portfolios. Panel B presents the correlation matrix of the factor returns.
Panel C reports parameter estimates from time-series regressions of the VR portfolio on the
market and the three Fama-French factors with Newey and West (1987) adjusted standard
errors.

Panel A: Summary Statistics

VR MKTMRF SMB HML

N 540 540 540 540

MEAN (%) -0.44 0.37 0.26 0.43
STD (%) 4.23 4.47 3.20 2.90
SR -0.10 0.08 0.08 0.15

Panel B: Correlations
VR MKTMRF SMB HML

VR 1
MKTMRF 0.05 1
SMB 0.08 0.30 1
HML -0.23 -0.38 -0.26 1

Panel C: Time-Series Regressions

α̂ β̂M KT M RF β̂SM B β̂HM L R2

I -0.0044 0
(-2.48)
II -0.0045 0.0433 0.0021
(-2.52) (0.58)
III -0.0027 -0.0556 0.0424 -0.3609 0.0576
(-1.43) (-0.72) ( 0.24) (-2.70)

47
 Table IX
Factor Exposures of the 25 Fama-French Portfolios

This table reports factor loadings of the 25 Fama-French portfolios with the market return
and volatility risk (VR) portfolio. The VR portfolio is the return of holding a long position in
the value-weighted quintile of stocks with high volatility risk (β̂σ,t
i ) and a short position in low

volatility risk, as reported in Panel C of Table III. Test assets are the value-weighted 25 Fama-
French portfolios constructed from independent quintile sorts based on market capitalization
(S1: small, S5: big) and book-to-market ratio (BM1: low, BM5: high). The time series starts
in January 1964 and ends in December 2008.

Panel A: Loading on the Market Factor

BM1 BM2 BM3 BM4 BM5 Average

S1 1.45 1.24 1.09 1.01 1.04 1.17

(29.17) (28.57) (30.45) (29.51) (27.77)
S2 1.44 1.18 1.05 0.99 1.06 1.14
(38.58) (39.11) (37.90) (35.53) (32.25)
S3 1.36 1.12 0.99 0.92 1.01 1.08
(43.52) (48.90) (42.72) (37.74) (33.24)
S4 1.25 1.08 0.99 0.93 1.01 1.05
(53.66) (56.88) (46.68) (40.40) (34.84)
S5 1.00 0.96 0.86 0.79 0.83 0.89
(57.97) (57.31) (42.01) (33.95) (27.53)

Average 1.30 1.12 1.00 0.93 0.99

Panel B: Loading on the Volatility Risk Factor

BM1 BM2 BM3 BM4 BM5 Average

S1 0.14 0.04 -0.06 -0.10 -0.13 -0.02

(2.14) (0.74) (-1.18) (-2.24) (-2.50)
S2 0.11 -0.04 -0.11 -0.09 -0.09 -0.04
(2.17) (-0.91) (-2.97) (-2.43) (-2.11)
S3 0.08 -0.10 -0.14 -0.12 -0.14 -0.08
(1.84) (-3.13) (-4.47) (-3.62) (-3.48)
S4 0.09 -0.10 -0.12 -0.11 -0.20 -0.09
(2.83) (-3.87) (-4.31) (-3.57) (-5.16)
S5 0.00 -0.10 -0.05 -0.16 -0.15 -0.09
(0.00) (-4.40) (-1.86) (-4.98) (-3.59)

Average 0.08 -0.06 -0.10 -0.12 -0.14

48
 Table X
Volatility Risk Pricing Factor

This table reports statistics from two-pass regressions to document the pricing implications
of the volatility risk (VR) model. The VR factor is the return of holding a long position in
the value-weighted quintile of stocks with high volatility risk (β̂σ,t
i ) and a short position in

low volatility risk, as reported in Panel D of Table III. Test assets are the value weighted 25
Fama-French size and book-to-market portfolios as well as the 5 value weighted consumption
volatility risk portfolios as in Panel C of Table III. The data covers January 1964 until
December 2008. The t-statistics are corrected for estimation error in the first stage as proposed
by Shanken (1992) and are Newey and West (1987) adjusted with 6 lags to account for
heteroskedasticity and autocorrelation. For each specification, we report the R2 , adjusted R2
in parenthesis, regression J-statistic (χ2 ) with the associated p-value (in %), mean absolute
pricing error (MAPE) and root square mean square error (RMSE).

CONST RME SMB HML VR R2 χ2 MAPE

(t-stat) (t-stat) (t-stat) (t-stat) (t-stat) (adj. R2 ) (p-val) (RMSE)

I 1.02 -0.41 7.34 99.01 0.20

(2.64) (-0.95) (4.03) (0.00) (0.23)
II 0.58 -0.67 19.42 94.86 0.19
(2.50) (-2.42) (16.54) (0.00) (0.22)
III 0.71 -0.16 -0.62 19.96 93.12 0.19
(1.81) (-0.35) (-3.14) (14.03) (0.00) (0.22)
IV 0.95 -0.56 0.22 0.48 76.93 76.68 0.08
(3.24) (-1.59) (1.50) (3.24) (74.27) (0.00) (0.12)
V 1.88 -1.44 0.19 -0.49 72.67 68.14 0.10
(4.11) (-3.04) (1.31) (-2.59) (69.52) (0.00) (0.13)
VI 1.00 -0.63 0.22 0.46 -0.40 79.33 71.59 0.08
(3.40) (-1.76) (1.49) (3.13) (-2.18) (76.02) (0.00) (0.11)
VII -0.56 0.97 0.44 51.21 88.93 0.14
(-1.02) (1.62) (2.95) (47.59) (0.00) (0.17)
VIII -0.56 0.96 0.42 -0.39 52.32 86.51 0.14
(-1.00) (1.60) (2.87) (-2.14) (46.81) (0.00) (0.17)

49
 Table XI
Market Predictability in the Time-Series

This table reports time-series regressions of excess returns on lagged predictor variables. The
market return is the value-weighted CRSP index less the 90 day T-bill rate. Predictor variables
are the lagged market return (RM,t ), the 90 day T-bill less its 12 months moving average (T-
bill), the difference between yields of long and short government bonds (Term), the yield
difference between Baa rated and Aaa rated corporate bonds (Default), the dividend yield
(DY), the consumption-wealth ratio of Lettau and Ludvigson (2001a) (cay), and changes in
beliefs about the first (∆µ̂t ) and second moments (∆σ̂t ) of consumption growth. The sample
period includes the first quarter of 1957 until the fourth quarter of 2006.

RM,t T-bill Term Default DY cay ∆µ̂t ∆σ̂t R2 (%)

I -0.02 -0.01 0.01 0.00 0.93 4.58

(-0.28) (-1.57) (0.83) (-0.16) (1.39)
II 0.05 1.51 5.97
(0.66) (3.50)
III 0.01 -0.01 0.00 0.01 0.05 1.44 8.50
(0.18) (-1.64) (-0.43) (0.66) (0.07) (2.87)
IV 0.03 0.17 0.11
(0.44) (0.02)
V 0.04 59.94 4.71
(0.55) (3.09)
VI -0.01 -0.01 0.01 -0.01 1.03 60.02 9.13
(-0.20) (-1.44) (1.12) (-0.41) (1.57) (3.11)
VII 0.05 1.50 59.35 10.48
(0.76) (3.55) (3.14)
VIII 0.02 -0.01 0.00 0.01 0.20 1.36 56.91 12.57
(0.23) (-1.50) (-0.12) (0.39) (0.28) (2.75) (2.99)

50