Asset Market Participation and Portfolio Choice over the Life‐Cycle

• Date: 01 April 2017
• Author: ANDREAS FAGERENG,LUIGI GUISO,CHARLES GOTTLIEB
• Published date: 01 April 2017
• DOI: http://doi.org/10.1111/jofi.12484

THE JOURNAL OF FINANCE •VOL. LXXII, NO. 2 •APRIL 2017

Asset Market Participation and Portfolio Choice

over the Life-Cycle

ANDREAS FAGERENG, CHARLES GOTTLIEB, and LUIGI GUISO∗

ABSTRACT

Using error-free data on life-cycle portfolio allocations of a large sample of Norwegian

households, we document a double adjustment as households age: a rebalancing of

the portfolio composition away from stocks as they approach retirement and stock

market exit after retirement. When structurally estimating an extended life-cycle

model, the parameter combination that best fits the data is one with a relatively large

risk aversion, a small per-period participation cost, and a yearly probability of a large

stock market loss in line with the frequency of stock market crashes in Norway.

THIS PAPER REEXAMINES EMPIRICALLY the life-cycle behavior of investors’ portfo-

lios, establishing novel features of the joint profiles of investors’ participation

in the stock market and the portfolio share invested in stocks. We estimate

the parameters of a standard life-cycle portfolio model with uninsurable labor

income, extended to incorporate costly participation and a small probability of

a stock market crash, and show that it can capture these features.

Inspired by empirical findings from novel microeconomic data on households’

portfolios, several recent studies provide new contributions to the literature on

life-cycle portfolio choice of individual investors building on the seminal contri-

butions of Mossin (1968), Samuelson (1969), and Merton (1969,1971). One key

∗Andreas Fagereng is with Statistics Norway. Charles Gottlieb is with the University of St.

Gallen (HSG-SEW) and Centre for Macroeconomics (CFM). Luigi Guiso is with the Einaudi Insti-

tute for Economics and Finance and CEPR. We would like to thank the Editor (Kenneth Singleton),

the Associate Editor, and two anonymous referees for their insightful comments. We are partic-

ularly thankful to Facundo Piguillem. Also we thank ´

Arp´

ad ´

Abrah´

am, Jerˆ

ome Adda, Francisco

Gomes, Maria Gustafsson, Elin Halvorsen, Jonathan Heathcote, Dirk Krueger, Kim Peijnenburg,

Arvid Raknerud, and Marno Verbeek for helpful discussions and suggestions. We are grateful

to Jo˜

ao Cocco for making the code of his life-cycle model available to us, and to Bernt Arne

Ødeg˚

ard for providing us with ISIN codes and end-of-period prices of the Oslo Stock Exchange. We

thank seminar participants at the University of Frankfurt, University of St. Gallen, University of

Cambridge, EIEF, University of Oxford, European University Institute, European Central Bank,

Statistics Norway, Norges Bank, NHH Bergen, ESEM 2011, SED 2012, IIPF 2013, EALE 2014,

and the 2015 WU Gutmann Symposium. We are grateful to NETSPAR for financial support. An-

dreas Fagereng thanks the Norwegian Research Council for support through grants #184563 and

#236921. Charles Gottlieb gratefully acknowledges research support from Research Center SAFE,

funded by the State of Hessen initiative for research LOEWE. No party had the right to review

the article prior to its circulation. We (the authors) did not receive any funding from interested

parties, and we also have no relevant potential conflicts of interest with this paper.

DOI: 10.1111/jofi.12484

705

706 The Journal of Finance R

finding of these new models is that the basic implication of Merton (1971)—

that the presence of human capital creates a strong incentive to invest in risky

securities when human capital is abundant, that is, early in the life-cycle, and

to rebalance away from such securities as people get older and their human

capital shrinks—carries over to more complex environments than the complete

market setting studied by Merton (1971). This implication holds in models with

labor income risk and incomplete markets as well as other realistic features

such as liquidity constraints or more general preference representations.1All

of these models consistently predict that individuals should rebalance toward

a safer portfolio as they approach retirement, the driving force being the life-

cycle pattern of human capital.2However, these features alone are not enough

to induce investors to leave the stock market altogether: many of the new mod-

els continue to predict stock market participation at all ages, in line with the

early models.

Microeconomic data on household portfolios, however, seem to depart from

these predictions. First, not only is participation in the stock market limited

at all ages, but it tends to follow a life-cycle pattern—in many instances a

hump-shaped one (Guiso, Haliassos, and Jappelli (2002)). Second, the share of

financial wealth invested in stocks (and mutual funds), the risky share, tends

to vary little with age, although the specific empirical pattern is subject to con-

troversy. Summarizing evidence from several countries, Guiso, Haliassos, and

Jappelli (2002, p. 15) argue that the age profile of the risky share is relatively

flat, albeit in some instances “there does seem to be some moderate rebalancing

of the portfolio away from risky securities” as people age. Thus, a reasonable

characterization of the empirical findings is that participation in the stock mar-

ket follows a hump-shaped profile, while the risky share varies little, if at all,

with age. The empirical finding that people do not rebalance their risky share

over the life-cycle is particularly puzzling because rebalancing is implied by an

indisputable fact of life—the decrease in the stock of human capital as people

age. It also runs counter to recent evidence that human capital drives financial

risk-taking positively (Calvet and Sodini (2014)).

While the lack of participation is a robust feature of the data, there are

at least three reasons to doubt the empirical patterns as regards age in

terms of both participation and the risky share. First, most of the available

1This is shown by several computational models of life-cycle portfolio investments that amend

the Samuelson-Merton model along one or more dimensions. See Benzoni, Collin-Dufresne, and

Goldstein (2007), Campbell and Viceira (2001), Cocco, Gomes, and Maenhout (2005), Davis, Kubler,

and Willen (2006), Gakidis (1998), Gomes and Michaelides (2003,2005), Gomes, Kotlikoff, and

Vicei ra (2008), Haliassos and Michaelides (2002), Heaton and Lucas (1997), Polkovnichenko (2007),

Storesletten, Telmer, and Yaron (2007), and Viceira(2001).

2A declining life-cycle portfolio profile may also be generated by features other than the life-

cycle of human capital, for instance, by greater labor supply flexibility when young (Bodie, Merton,

and Samuelson (1992)), by a departure from constant relative risk aversion (CRRA) utility (Gollier

and Zeckhauser (2002)), by life-cycle patterns of risk aversion and background risk, as well as by

predictability of stock returns (Kandel and Stambaugh (1995), Campbell and Viceira (1999,2002)).

While these factors may contribute to a rebalancing motive over the life-cycle, the most agreed

upon factor is the life-cycle of human capital.

Asset Market Participation and Portfolio Choice over the Life-Cycle 707

evidence comes from cross-sectional data. Inferences about the age pattern of

the portfolio must therefore be drawn from comparisons of the portfolio hold-

ings of individuals of different age, rather than of the same individual as his

age varies. Panel data may help address this issue, although adding an extra

source of variation to the data (time) also adds the need to model it. If rea-

sonable restrictions on time effects can be imposed, the effect of age can then

be distinguished from that of birth year. Second, most studies ignore the fact

that the risky share is only defined for stock market participants (or risky as-

set markets more broadly) and that participation in risky asset markets is an

endogenous choice. Thus, uncontrolled selection, if correlated with age, may be

responsible for the failure to find evidence of rebalancing of the risky share.

Third, the evidence so far is based primarily on household surveys, which are

notoriously prone to measurement problems. Most importantly, measurement

and reporting errors are likely to be correlated with age, obscuring age patterns

present in the true data.3

In this paper, we try to overcome these problems. To do so, we assemble

a new database that draws on administrative records from the Norwegian

Tax Registry (NTR). Because Norwegian households are subject to a wealth

tax, they have to report to the tax authority all their asset holdings, real and

financial, at the level of the single instrument at the end of the year. We draw

a random sample of 20% of the 1995 population of Norwegian households and

then follow these households for 15 years until 2009. This data set reports the

complete portfolio of Norwegian people and is similar in structure and content

to the one used by Calvet, Campbell, and Sodini (2007), but it spans many more

years—an essential feature when studying the life-cycle profile of portfolio

allocation. Because the data set is sourced from the tax registry, measurement

error is minimized. The main cause of nonreporting or underreporting should

stem from incentives to evade the wealth tax, but, as we argue in Section I,the

way the wealth tax is collected in Norway suggests that tax evasion is unlikely

to be an issue. Finally, since the entire population of Norwegian taxpayers

has to report to the NTR, there is little attrition in the panel apart from that

due to death, migration to another country, or divorce.

After taking into account the endogeneity of the participation decision and

modeling cohort effects, we find that both participation in the stock market and

the risky share show important life-cycle patterns. As in other studies, we also

find little asset decumulation after retirement and a hump-shaped life-cycle

profile in participation, besides limited stock market participation at all ages.

But we also find that the average risky share of those that participate varies

3Two exceptions are Ameriks and Zeldes (2002), who use a panel of TIAA-CREF contributors,

and Agnew, Balduzzi, and Sund´

en (2003), who use a four-year panel data set of people who hold

401k retirement accounts. In principle, they can distinguish between age, time, and cohort effects.

Because they use administrative data, underreporting of assets is not a major issue. However,

TIAA-CREF and the 401k retirement accounts only report assets contributed to the program,

not the complete portfolios of these individuals, and there is no obvious reason why the portfolio

allocation in pension savings should be the same as the allocation in other financial assets or follow

the same age profile (indeed it does not; see Guiso and Sodini (2013)).