Stock Market Returns and Consumption

• Date: 01 December 2020
• DOI: http://doi.org/10.1111/jofi.12968
• Author: MARCO DI MAGGIO,AMIR KERMANI,KAVEH MAJLESI
• Published date: 01 December 2020

THE JOURNAL OF FINANCE •VOL. LXXV, NO. 6 •DECEMBER 2020

Stock Market Returns and Consumption

MARCO DI MAGGIO, AMIR KERMANI, and KAVEH MAJLESI∗

ABSTRACT

This paper employs Swedish data on households’ stock holdings to investigate how

consumption responds to changes in stock market returns. We instrument the actual

capital gains and dividend payments with past portfolio weights. Unrealized capital

gains lead to a marginal propensity to consume of 23% for the bottom 50% of the

wealth distribution and about 3% for the top 30% of the wealth distribution. House-

hold consumption is significantly more responsive to dividend payouts across all parts

of the wealth distribution. Our findings are consistent with households treating cap-

ital gains and dividends as separate sources of income.

IN THE UNITED STATES,STOCKHOLDINGS REPRESENT the largest share of fi-

nancial assets on households’ balance sheets, reaching more than $32 trillion

(with about $15 trillion in nonretirement accounts), which makes them com-

parable in importance to the stock of housing wealth. Given their prominence,

movements in stock prices and dividend payments might significantly affect

households’ consumption and savings decisions. With soaring stock prices,

households’ savings rate is at a 12-year low, which raises the question of

∗Marco Di Maggio is with Harvard Business School and NBER. Amir Kermani is with UC

Berkeley and NBER. Kaveh Majlesi is with Lund University, Monash University, CEPR, IFN,

and IZA. The data used in this paper come from the Swedish Interdisciplinary Panel (SIP) ad-

ministered at the Centre for Economic Demography, Lund University, Sweden. We thank two

anonymous referees, an Associate Editor,and the Editor, Stefan Nagel, for helpful comments that

significantly improved the paper. We also thank Malcolm Baker; James Cloyne; Chris Carroll;

Francesco D’Acunto; Samuel Hartzmark; Luigi Guiso; Matti Keloharju; Ralph Koijen; David Laib-

son; Geng Li; Jonathan Parker; Luigi Pistaferri; Larry Schmidt; Paolo Sodini; David Sraer; Stijn

Van Nieuwerburgh; Gianluca Violante; Roine Vestman; Annette Vissing-Jørgensen; and seminar

participants at the 2017 meeting of the Econometric Society, New York University Conference on

Household Finance, CEPR Household Finance Workshop in Copenhagen, Helsinki Finance Sum-

mit, NBER SI Consumption Micro to Macro, MIT Sloan, UCSD, CREI and UPF, NY Fed, UC

Berkeley, University of Zurich, SFS Cavalcade, the European Finance Association, and the West-

ern Finance Association for helpful comments. We thank TerranceOdean for providing data from

a large brokerage in the United States. Leonel Drukker and Erik Grenestam provided excellent

research assistance. Kermani is grateful for research support from the Fisher center for Real Es-

tate and Urban Economics and Majlesi acknowledges research support from Jan Wallanders and

Tom Hedelius Foundation. Aside from these funding sources, all three authors have nothing to

disclose with respect to The Journal of Finance disclosure policy.

Correspondence: Amir Kermani, Haas School of Business, UC Berkeley 2220 Piedmont ave,

Berkeley,CA 94720 510-664-4132; e-mail: kermani@berkeley.edu.

DOI: 10.1111/jofi.12968

© 2020 the American Finance Association

3175

3176 The Journal of Finance®

whether stock market trends do indeed drive households’ spending habits.1

This shift away from saving, however, could leave some consumers exposed to

changes in market conditions. Furthermore, concerns about the consumption-

wealth effects of stock market returns have been the main driver of U.S.

monetary policy sensitivity to stock price movements (Cieslak and Vissing-

Jørgensen (2020)). A natural question that arises is thus how much of a decline

in aggregate consumption should we expect if stock prices take a sudden turn

for the worse as they did during past recessions?

Despite the central importance of these questions, the literature lacks a com-

prehensive study on the causal impact of changes in stock market wealth on

households’ consumption. This is due to several challenges. First, aggregate

movements in stock prices are endogenous with respect to other macroeco-

nomic shocks, such as expectations of future income growth and consumer

confidence (Beaudry and Portier (2006)). Second, due to home bias, exploit-

ing regional cross-sectional variation to control for macroeconomic fluctuations

is also not ideal. One could potentially address these challenges by exploit-

ing household-level data such as the Consumer Expenditure Survey (CEX),

but the advantages of using such data are counterbalanced by the lack of ac-

curacy in the reported measures of capital gains (Dynan and Maki (2001)).2

Furthermore, households bias their investments toward their own companies

and local industries, resulting in correlations between capital gains and other

factors that directly affect their income, which may introduce a new source

of endogeneity that is absent in the aggregate data (Benartzi (2001), Coval

and Moskowitz (2001), Mitchell and Utkus (2003), Meulbroek (2005)). Finally,

given the skewness of stockholdings, it is important to estimate the consump-

tion behavior of households at the top of the wealth distribution, which are

usually underrepresented in surveys.3

In this paper, we overcome these challenges by using highly granular

household-level data from Sweden. Due to the presence of a wealth tax, we

are able to obtain a full picture of households’ balance sheets at the end of

each year from 1999 to 2007, when the tax was repealed. We have data on

the universe of households’ portfolio holdings at the security level, as well as

information about their debt obligations and real estate transactions. To mea-

sure consumption, we follow the residual approach proposed by Koijen, Van

Nieuwerburgh, and Vestman (2015) and impute consumption as a residual

of households’ disposable income net of other transactions, and we validate

1The Commerce Department has reported that the savings rate was 2.4% of disposable house-

hold income in December 2017, the lowest rate since September 2005. The savings rate had risen

to 6.6% when the recession ended in June 2009.

2There is no direct measure of capital gain in the CEX, and capital gains are imputed based on

changes in total security holdings and the amount of sales and purchases during that year. Any

such imputation requires strong assumptions on the timing and portfolio rebalancing of house-

holds. Moreover, many households report zero capital gains in years in which the stock market

performs remarkably well.

3See Table A.1 for the distribution of stock holdings in the United States according to the

Survey of Consumer Finances.

Stock Market Returns and Consumption 3177

this measure against survey information. Koijen, Van Nieuwerburgh, and

Vestman (2015), Eika, Mogstad, and Vestad (2020), and Kolsrud, Landais, and

Spinnewijn (2019) discuss the quality of this imputed measure of consumption

based on administrative data and their comparison with survey data. These

papers show that the quality of the consumption measure based on the residual

method depends on the availability of detailed household-level asset allocation

data as well as data on housing transactions.

Even with such data, households’ portfolio choices are endogenous and might

be driven by omitted factors that also drive their consumption behavior. We

address this issue in several ways. First, we exploit the panel nature of our

data and estimate all of our regressions using first differences. This allows us

to capture time-invariant differences across households that might be corre-

lated with the level of their capital gains or dividend income. Second, we limit

the heterogeneity across households’ portfolios by estimating the marginal

propensity to consumer (MPC) separately for different parts of the wealth dis-

tribution. Third, we also exclude stockholdings in households’ own industry

from their portfolios before computing capital gains and dividends. This en-

sures that our results are driven by households’ holdings in industries other

than their own, as fluctuations in such industries are less likely to be corre-

lated with changes in households’ income.

One might still be concerned that changes in capital gains and dividend in-

come could be driven by dynamic changes in households’ portfolios. Indeed,

changes in households’ portfolios may be driven by factors such as the liquida-

tion of stock holdings due to an expenditure shock or a large durable purchase,

the very same factors that are likely responsible for household consumption.

We therefore implement a simulated IV strategy where we instrument the

variations in capital gains and dividend income with the capital gains and div-

idend income that would have accrued had the household kept its portfolio the

same as observed in previous years. Intuitively, the portfolio weights in previ-

ous years should not be determined by future shocks that drive both stock re-

turns and consumption choices. In other words, our identification comes from

the stickiness in households’ portfolios, for which we find strong evidence in

our data.

Our first main result is that the MPC out of (unrealized) capital gains for

households in the top 30% of the financial wealth distribution is about 3% and

does not exhibit significant variation between, for instance, households in the

70th to 90th percentile and households in the top 5% of the wealth distribu-

tion. In contrast, the MPC for households in the bottom half of the distribution

is significantly higher at about 23%. However, it is worth noting that these

households own less than 7% of overall stockholdings. Our estimates are ro-

bust to directly controlling for realized capital gains, which we observe for a

subsample. Intuitively, households can freely respond to changes in unrealized

capital gains by adjusting their savings decisions.4In further tests, we also

4Note that this is also why transaction costs, related to the liquidation of stock holdings, are

unlikely to drive the difference between the MPC for capital gains and dividends.