Consumption Fluctuations and Expected Returns

• DOI: http://doi.org/10.1111/jofi.12870
• Published date: 01 June 2020
• Author: VICTORIA ATANASOV,STIG V. MØLLER,RICHARD PRIESTLEY
• Date: 01 June 2020

THE JOURNAL OF FINANCE •VOL. LXXV, NO. 3 •JUNE 2020

Consumption Fluctuations and Expected Returns

VICTORIA ATANASOV, STIG V. MØLLER, and RICHARD PRIESTLEY∗

ABSTRACT

This paper introduces a novel consumption-based variable, cyclical consumption, and

examines its predictive properties for stock returns. Future expected stock returns

are high (low) when aggregate consumption falls (rises) relative to its trend and

marginal utility from current consumption is high (low). We show that the empiri-

cal evidence ties consumption decisions of agents to time variation in returns in a

manner consistent with asset pricing models based on external habit formation. The

predictive power of cyclical consumption is not confined to bad times and subsumes

the predictability of many popular forecasting variables.

IN THIS PAPER,WE TAKE a new approach to linking stock return predictability

to both bad and good economic times. Consider an economy in which investors

exhibit external habit formation as in, for example, Campbell and Cochrane

(1999), and as a result, risk premia vary over time through variation in risk

aversion. In good times, when consumption rises above its trend and hence

the marginal utility of present consumption is low, investors are willing to

give up current consumption and invest. This, in turn, forces stock prices to

increase and future expected returns to decrease. Conversely, in bad times,

when consumption falls below its trend and hence the marginal utility of

current consumption is high, expected returns in the future need to be high

in order to induce investors to postpone the valuable present consumption and

to invest and consume in the future. We conjecture that cyclical fluctuations

in aggregate consumption should be useful in picking out bad and good times

in the economy as measured from a representative agent’s point of view, and

∗Victoria Atanasov is from the Chair of Finance, University of Mannheim. Stig V. Møller is at

CREATES, Aarhus University and affiliated with the Danish Finance Institute. Richard Priestley

is at the Department of Finance, BI Norwegian Business School. We would like to thank the

editor, Stefan Nagel, an associate editor, and two anonymous referees for generous comments

that improved the paper. We also thank John Y. Campbell, John H. Cochrane, Tim A. Kroencke,

Ernst Maug, Erik Theissen, Jessica A. Wachter (WFA discussant), and participants at the AFFI

meeting 2018, the SoFiE meeting 2018, the WFA meeting 2018, the EEA-ESEM congress 2018,

and seminar participants at University of Mannheim and University of Oslo for helpful comments

and suggestions. We also thank Amit Goyal, Martin Lettau, and Robert J. Shiller for making their

data available. Weexpress our gratitude to David E. Rapach for providing his code for the bootstrap

analysis. We have read The Journal of Finance disclosure policy and have no conflicts of interest

to disclose.

DOI: 10.1111/jofi.12870

C2019 the American Finance Association

1677

1678 The Journal of Finance R

thus, informative about future excess stock returns. If this argument holds,

then we should find an inverse relation between cyclical consumption and

future expected returns in the data.

The empirical results that we present in this paper are consistent with

the idea that future expected returns are high (low) when consumption falls

below (rises above) its trend and cyclical consumption is low (high). Cyclical

fluctuations in consumption, which we refer to as cc, capture a significant

fraction of the variation in future stock market returns. The finding that

expected returns and consumption are linked is important because it suggests

that asset prices are driven by fundamental shocks that reflect changes in

marginal utility.

A second notable finding in this paper is that the predictive power of cyclical

consumption is not confined to bad times alone. Cyclical consumption provides

a consistent description of how both positive and negative macroeconomic

events, as reflected by investors’ consumption decisions, affect stock market

returns. This finding stands in stark contrast to Rapach, Strauss, and Zhou

(2010), Henkel, Martin, and Nardari (2011), Dangl and Halling (2012), and

Golez and Koudijs (2018), who document that popular predictive variables

can forecast stock returns in bad times, but find essentially no evidence of

predictability during business cycle expansions.

To extract the cyclical component of consumption, we employ the simple

and robust linear projection method of Hamilton (2018). This procedure

provides a measure of what macroeconomists often refer to as the “cyclical

component” of a time series and it has two advantages over other prominent

detrending methods. First, the procedure ensures that the cyclical component

that is identified is stationary and consistently estimated for a wide range

of nonstationary processes. Second, it produces a series that is accurately

related to the underlying economic fluctuations, as opposed to, for instance,

the popular Hodrick and Prescott (1997) filter, which can spuriously generate

dynamic relations. This feature of the Hamilton (2018) detrending procedure is

particularly appealing because it implies that any predictive ability of cyclical

consumption for stock returns is more likely to reflect actual predictability

than to be a statistical artifact of the decomposition method. When we employ

other econometric procedures to isolate the cyclical variation in consumption,

such as polynomial time trends and backward-looking moving averages, we

find even stronger evidence of predictability. We therefore view Hamilton’s

(2018) detrending procedure as providing conservative and robust evidence on

return predictability.

Our findings support theoretical explanations of asset prices that generate

time-varying expected returns, such as models with time-varying risk aver-

sion. In the external habit formation model of Campbell and Cochrane (1999),

for example, habit acts like a trend for consumption. A decline in consumption

relative to the trend, which can be thought of as bad times, leads to low stock

prices and high expected returns. Conversely, an increase in consumption

above trend, which can be thought of as good times, leads to high stock prices

and low expected returns. Under relatively mild assumptions, there exists

Consumption Fluctuations and Expected Returns 1679

a tight relation between a finite-horizon version of the surplus consumption

variable of Campbell and Cochrane (1999), which generates changes in equity

prices in the model, and cyclical consumption.

To examine the link between cyclical consumption and habit models

more formally, we simulate data from the Campbell and Cochrane (1999)

model and investigate both the extent of the model-implied predictability

and its consistency with the time-series predictability that we observe in

actual data. The simulations show that the habit model produces an inverse

relation between expected returns and cyclical consumption just as in the

data. The degree of in-sample predictability implied by the model is qual-

itatively comparable to that in the data. The out-of-sample tests reinforce

the results from in-sample regressions but typically indicate less predictable

movements in expected returns. These findings suggest that our results

can be taken as evidence of countercyclical variation in the market price of

consumption risk.

We perform a battery of robustness checks and address a number of econo-

metric concerns surrounding predictive regressions with persistent predictors

(Nelson and Kim (1993), Stambaugh (1999)). Both the IVX testing approach

of Kostakis, Magdalinos, and Stamatogiannis (2015) that accounts for the

degree of regressor persistence and an advanced bootstrap procedure that

accounts for the regressor’s time-series properties provide strong evidence of

predictability at the one-quarter horizon that extends to horizons of up to five

years. Moreover, this predictability does not vanish during the post-oil-crisis

period, a period in which standard business cycle indicators have proven

dismal as predictive variables (Welch and Goyal (2008)).

We also show that the forecasting power of fluctuations in cyclical consump-

tion is not confined to the aggregate U.S. stock market. Robust patterns of

predictability exist across industry portfolios. In addition, the strong predictive

ability of cyclical consumption extends to international equity markets. A

global measure of cyclical consumption computed as the simple average of

country-specific components captures a large part of the time variation in

future expected returns on the world market portfolio as well as on regional

portfolios such as the European portfolio, the EAFE (Europe, Australia, and

the Far East) portfolio, and the G7 portfolio.

Explaining the dynamic behavior of asset returns using aggregate con-

sumption data is a challenging task. Few studies find evidence in favor of

returns being predictable from consumption. Perhaps, the most prominent

consumption-based predictive variable is Lettau and Ludvigson’s (2001)

consumption–wealth ratio, cay. We find that the information content of

cyclical consumption is clearly above and beyond that of many well-recognized

variables, such as the consumption–wealth ratio of Lettau and Ludvigson

(2001), the labor-income-to-consumption ratio of Santos and Veronesi (2006),

and the conditional volatility of consumption of Bansal, Khatchatrian, and

Yaron (2005). In total, we consider 19 alternative economic variables that are

popular in the literature. We find that few of them have predictive power, and