household investment—the horizon effect
TRANSCRIPT
Household Investment—The Horizon Effect
PING HE1 and XIAOQING HU2*1School of Economics andManagement,TsinghuaUniversity, Beijing 100084, China2University of Illinois at Chicago, 601S. Morgan, UH 2422, Chicago, Illinois 60607, USA
ABSTRACT
Individuals tend to simplify a complex portfolio decision problem into several manageabledimensions, each of which can frame their perception of risk. We check this view by studyingthe effect of investment horizons on households’ portfolio decisions. Using the Survey ofConsumer Finances (SCF) data, we find that households allocate more of their wealth in stocksif they report longer planning horizons. The existence of foreseeable expenditure significantlychanges the dependence of risky stock investment on the planning horizon. We decompose thereported planning horizon into an objective part and a subjective mental accounting part, andfind that the mental accounting part has a greater effect on household portfolio choice. This isconsistent with the argument that individuals make investment decisions based on the horizonat which the risk is perceived rather than the horizon at which the investment reward or cash isneeded. Copyright # 2010 John Wiley & Sons, Ltd.
key words portfolio choice; time diversification
INTRODUCTION
Individuals have bounded rationality in processing information and making decisions. A
complex decision problem is often simplified. For example, in a standard investment decision
problem, investors focus on the mean and variance of risky asset returns. The way an investor
structures his decision problem can frame his perception of risk. In a dynamic setting, the time
varying risk characteristics are important inputs of an investment decision problem. Thus how
these risk characteristics affect individuals’ investment decisions is one of the central issues in
finance research. One input of an individual’s portfolio decision problem is the horizon over
which the investment is made. Some people tend to believe that there is a positive correlation
between the proportion of wealth invested in risky assets and the investment horizon. One
justification is that above-average returns tend to offset below-average returns over long
horizons. This is often referred to as ‘‘time diversification’’ in many studies. In this paper, we
Review of Behavioral Finance, 2: 81–105 (2010)
Published online 7 June 2010 in Wiley Online Library
(wileyonlinelibrary.com) DOI: 10.1002/rbf.10
�Correspondence to: X. Hu, University of Illinois at Chicago, 601 S. Morgan, UH 2422, Chicago, IL60607, USA.E-mail: [email protected]
Copyright # 2010 John Wiley & Sons, Ltd.
empirically study the impact of households’ reported planning horizons on their stock
investments, and we then further explore the source of this impact.
The Federal Reserve Board’s Survey of Consumer Finances (SCF) data contain important
information on the attitudes of US households toward financial risks. Among other things, the
SCF asks survey respondents about their most important saving and planning horizons. We
construct a ‘‘planning horizon’’ variable for each household based on its answer to this
question. First, we study the factors that determine a household’s planning horizon. Our
summary statistics show that a household’s planning horizon does not decrease uniformly
with age. Thus, we conjecture that households make ‘‘plans’’ based on their expected future
expenses. Consistent with this conjecture, we find that a household’s planning horizon is
related to the household’s foreseeable expenditures. In particular, we find that the existence of
foreseeable expenditures generally shortens a household’s planning horizon. Moreover, we
find that many other household characteristics also affect a household’s planning horizon, for
example risk aversion and education. Then, we decompose the planning horizon into two
parts, the first part depends on age and foreseeable expenditures, and the second part depends
on other household characteristics including income, wealth, education, risk aversion, marital
status, number of members in the household, health condition, and borrowing constraints. We
use this decomposition to differentiate two different concepts of horizon: the objective
investment horizon, over which the investment is made, and the subjective mental accounting
horizon, which determines at what frequency risks are perceived. This decomposition is based
on the assumption that households’ reported planning horizons reflect both their investment
horizon and their preference on risk-perceiving frequency.
We study the effect of our constructed horizon variables on household portfolio choice. We
find that households with a longer planning horizon invest more in risky assets, this result is
robust to various estimating methods. In particular, the proxy for the subjective mental
accounting horizon has a larger effect on household portfolio choice. These results are
consistent with the experimental results in Thaler et al. (1997).
Moreover, we also find that the structure of stockholdings is linked to the planning horizon.
Specifically, our results show that households invest more in indirect stockholding (in the
form of mutual fund holdings) when they report longer planning horizon. It is possible that
households invest more in risky assets because they diversify more when they have longer
horizon. However, including the structure of stockholdings as an independent variable in our
regression does not eliminate the significance of the planning horizon effect. Therefore, the
hypothesis that time diversification is driven by cross-asset diversification is rejected.
The paper proceeds as follows. In ‘Literature Review’ section we review previous studies
and related literature. In ‘Data and Summary Statistics’ section, we present a description of
the data and summary statistics. In ‘Determination of Planning Horizon’ section, we explore
the relationship between planning horizon and other household characteristics. In ‘Household
Portfolio Choice’ section, we study household portfolio choice and test the horizon effect.
‘Robustness Tests’ section provides robustness checks. We conclude in the last section.
LITERATURE REVIEW
In many studies, economists formalize the portfolio decision problem with time-separable
Von Neumann-Morgenstern utility functions as a representation of intertemporal preference.
Under this setting, Samuelson (1969) gives an example of a multi-period portfolio choice
Copyright # 2010 John Wiley & Sons, Ltd. Review of Behavioral Finance, 2, 81–105 (2010)
DOI: 10.1002/rbf.10
82 P. He and X. Hu
problem in which the proportion of wealth invested in stocks is independent of the horizon for
an individual with constant relative risk aversion (CRRA) preferences. Samuelson (1963)
points out that the rationale behind time diversification is a fallacious interpretation of the law
of large numbers.1
While Samuelson’s refutation of time diversification remains a ‘‘mathematical truth,’’
many research studies challenge the assumptions of his arguments. In terms of the investor’s
preferences, Thorley (1995) shows mathematically that if an investor has decreasing relative
risk aversion toward serially uncorrelated returns, the allocation toward risky stocks should
increasewith the investment horizon.2 In terms of the dynamic characteristics of asset returns,
Campbell and Viceira (1999) and others3 show that time-varying investment opportunities
result in an intertemporal hedging demand of stocks, and the magnitude of hedging demand
depends on an investor’s investment horizon. In addition, background income risk
characteristics also influence an individual’s life-cycle portfolio choice problem. For
instance, Viceira (2001) suggests that the optimal allocation to stocks is unambiguously
larger for employed investors (who are younger and have longer investment horizons) than for
retired investors (who are older and have shorter investment horizons) when the labor income
risk is idiosyncratic and non-tradable. He also finds that when labor income is correlated with
stock returns, time diversification appears to be optimal as long as the correlation between
labor income and stock returns is low.4
The time diversification phenomenon, or the horizon effect, is also discussed in the
behavior economics literature, in which the term ‘‘horizon’’ sometimes refers to the risk
perceiving horizon. Thaler et al. (1997) show experimentally that investors who demonstrate
myopic loss aversion are more willing to accept risks if they evaluate their investments less
often.5 Thaler et al. argue that investors who look at short-term stock returns make incorrect
decisions because they are fooled by myopic loss aversion and believe that the probability of
losses over the long run is higher than it actually is. On the other hand, Samuelson (1994)
suggests that investors who look at 30-year stock returns make incorrect decisions because
they are fooled by the illusionary happy ending and believe that the probability of losses over
the long run is zero. Nevertheless, both arguments lead to the conclusion that an investor will
allocate more of his wealth to stocks if he values them at a lower frequency, that is, his risk-
perceiving horizon is longer. In this paper, we empirically test these arguments using
household wealth and expectation data.
1Statistically, for a sequence of independent random stock returns, while the expectation of the sum increases linearlywith the number of periods, the standard deviation only increases with the square root of the number of periods.However, the size of an investor’s potential loss increases in proportion to the expected returns, thus reducing theattractiveness of the higher return-risk ratio. It is the increasing size of the potential loss that makes timediversification differ from cross-asset diversification.2Gollier (2002) provides an argument for decreasing risk aversion on wealth; he also finds that the introduction ofliquidity constraints increases an agent’s risk aversion.3See, for example, Kim and Omberg (1996), Brennan, Schwartz, and Lagnado (1997), Brandt (1999), and Barberis(2000).4Related works include Bodie, Merton, and Samuelson (1992), Jagannathan and Kocherlakota (1996), and Heatonand Lucas (1997).5The notion of myopic loss aversion is a combination of two behavioral principles, namely, loss aversion and mentalaccounting. Consistent with the prospect theory (Tversky and Kahneman, 1992), which suggests that the valuefunction has a kink at the reference point, loss aversion refers to the fact that people tend to be more sensitive todecreases in their wealth than to increases. Mental accounting, on the other hand, concerns aggregation, that is, howtransactions are grouped both cross-sectionally (whether securities are evaluated one at a time or in portfolios) andintertemporally (how often portfolios are evaluated).
Copyright # 2010 John Wiley & Sons, Ltd. Review of Behavioral Finance, 2, 81–105 (2010)
DOI: 10.1002/rbf.10
Household Investment—The Horizon Effect 83
DATA AND SUMMARY STATISTICS
Data discussionSome of the recent research on investment behavior is based on experimental data, where
participants make investment decisions in hypothetical situations with small financial
consequences (see, for example, Thaler et al. (1997)). Other studies such as Papke (1998)
explore individuals’ actual investments in 401(k) plans. An important drawback in
exclusively using 401(k) data is the lack of reliable demographic information with respect to
the individuals who invest in 401(k)s. Our analyses focus on how households’ financial
investments vary with their mental accounting and planning horizons. A panel data of
households’ wealth and future plans would be ideal for the study. However, the only
comprehensive household panel data available is the Panel Studies of Income Dynamics
(PSID). Though the PSID data contain some information on household wealth and asset
portfolios, it does not have data on ex ante expectations on future financial needs or goals.
Those variables are crucial to our study. Therefore, we use the best available alternative—the
SCF, conducted by the Federal Reserve Board. The SCF provides a detailed survey of the
asset holdings, consumer debt, pensions, income, and other demographic characteristics
of U.S. families. It also contains reasons for various financial choices and attitudes toward
financial risks. The SCF is widely considered to be the most comprehensive source of wealth
data in the United States.6 The advantages of using SCF data for our study can be summarized
as follows. First, the SCF data give a representative sample of the entire U.S. population.
Second, the SCF data contains rich information on the demographic characteristics as well as
the economic situation of individual households. Finally, and most importantly, the SCF data
disclose the information on households’ planning horizons as well as other investment-related
characteristics.
As any survey dataset, the SCF has its drawbacks. Morgan and Sonquist (1963) discussed
general problems with all survey data, such as classification issues and measurement errors.
Almost all survey samples are clustered and stratified, thus require proper application of
statistical techniques. The explanatory variables to be used in the analysis could be highly
correlated, which may reduce the power of the tests. In addition, household wealth estimates
based on the SCF were lower than independent, institution-based, estimates mainly due to an
under-sampling of wealthy households, which are believed to hold highly disproportionate
shares of many types of assets. This could affect our estimates of household portfolio choices,
though the impact is not necessarily significant. Despite of all these potential problems, the
SCF data are still the best available data for our study.
We mainly use the 2004 SCF data since it is the latest wave of SCF data for which the data
are not preliminary. By restricting our study to one specific wave of data, we can guarantee
that all investment decisions are made during approximately the same brief time period, and
thereby reduce the noise associated with changes in the economic environment. There is a
concern that using data from a survey taken just after a major equity bear market (2001–2003)
may bias the result. To make sure that our study is robust and to explore the potential impact
6SCF data exist in two versions, the raw data prepared by the University of Michigan’s Survey Research Center, andthe recoded data prepared by the Federal Reserve Board. The latter file uses a series of consistency checks andimputation procedures for missing data. Additional weights are constructed and included in the recoded data file, andnew variables are constructed not only from original variables but also by matching information from other sourcessuch as the U.S. Census. We use the latter version for our study.
Copyright # 2010 John Wiley & Sons, Ltd. Review of Behavioral Finance, 2, 81–105 (2010)
DOI: 10.1002/rbf.10
84 P. He and X. Hu
of this bear market on our results, we pool 2001 and 2004 SCF data together to analyze the
effect of changes in economic environment in some of our major empirical tests. Before any
analysis, we define the variables that we use, and present summary statistics.
Variable definitionsThe primary variable of interest in this study is the planning horizon, denoted as PLN_HRZ.
The SCF asks survey respondents about their most important saving and planning horizons.
The choices are: (1) The next few months; (2) the next year; (3) the next few years; (4) the
next five to 10 years; (5) more than 10 years. Based upon these choices, we construct
PLN_HRZ as follows:7
Reported horizon Planning horizon (Years)
Next few months 0.51 year 11–5 years 2.55–10 years 7.510þ years 10
Next, because households’ portfolio choices are related to their reported risk preferences,
we construct a household risk aversion variable, denoted as RSK_AVS, based on a household’s
self-reported attitude toward financial risks. The willingness to accept risks is categorized by
the integer numbers from one to four, where a higher number indicates more risk aversion.We
use these numbers as proxies for risk aversion. Note that these numbers only reflect the
subjective view of a household about its own risk attitude; they have nothing to do with the
risk aversion coefficients in any standard CRRA or constant absolute risk aversion (CARA)
utility function.
We expect that households with borrowing constraints have less of an incentive to take
risks, and thus invest less in stocks (see, for example, Guiso et al. (1996)). To incorporate this
concern into our study, we construct a dummy variable of borrowing constraints, or
BRW_CTR, as follows: The SCF asks whether a household had applied for credit in the past 5
years. If the household was denied credit, or approved for a smaller amount than desired, or if
the household did not apply for credit in the past 5 years because it was afraid of denial, the
dummy variable is set to one; otherwise, the dummy is set to zero.
The next two variables are measures of household wealth. Household income (INC) is the
sum of wages, salaries, income from pensions or social security, and income from private
business. This measure includes both the respondent’s and the spouse’s income. Net worth
(NTW) is defined as total assets minus total debt, where total assets are the value of all assets,
including liquid assets (such as cash, stocks, and bonds), real estate assets, the value of
pensions, and the value of all private businesses; and total debt includes liquid debt (such as
credit card balances) and mortgage debt associated with real estate assets.
To address the issue of horizon effect, or time diversification, the dependant variable of
interest is the fraction of wealth a household allocates to risky assets, which, in our study, are
stocks. We capture this quantity using the amount of wealth invested in stocks divided by
7All the major regression results in this paper are robust to different specifications of the planning horizon variable.
Copyright # 2010 John Wiley & Sons, Ltd. Review of Behavioral Finance, 2, 81–105 (2010)
DOI: 10.1002/rbf.10
Household Investment—The Horizon Effect 85
liquid assets, or ST/LIQ. Liquid assets are defined as the sum of cash, stocks, and bonds.
Another competing candidate would be the amount of wealth invested in stocks divided by net
worth, or ST/NW. In this study, we mainly discuss the results using ST/LIQ as the dependant
variable; however, we also study the case using ST/NW as a robustness check.
We prefer ST/LIQ to ST/NW for three reasons. First, the investments in stocks are deemed
to be a component of liquid assets, whereas a household’s net worth contains many other
illiquid assets. Thus, the ratio of stocks over liquid assets is more consistent with the idea of
‘‘portfolio choice’’ within the category of liquid assets, especially for individuals with mental
accounting, which is a particularly relevant concern in our study. Second, because net worth is
one of the independent variables in our regressions, the negative correlation between net
worth and ST/NW could generate unnecessary noise in the results. Third, we control for
background risks due to illiquid assets included in net worth. We discuss background risk in
more detail below.
In the literature, stockholdings divided by net worth plus capitalized human capital are
also used as a proxy for the level of risky asset holdings. However, we do not use this measure
for several reasons. First, the best way to capitalize human capital remains a subject of
controversy. Second, we do not have sufficient information on each household’s income
dynamics to accurately carry out capitalization as in Carroll and Samwick (1997). Third,
because there could be more than one working individual in a household, any capitalization
method that is appropriate to only one individual could lead to a very inaccurate estimate of
the household wealth. Fourth, in reality, the market to borrow against future income is
imperfect, and this capitalized human capital cannot be used to invest in stocks.
We also construct ratios of various other asset values to a household’s net worth, including:
real estate assets to net worth (RE/NW), business value to net worth (BUS/NW), pension value
to net worth (PEN/NW), cash to net worth, and bonds to net worth. The value of real estate
assets includes both the owner-occupied house’s value and the value of other real estate.
The value of owner-occupied housing is the self-reported market value of a homeowner’s
principle house. The value of other real estate is the sum of vacation homes and commercial
real estate. Business value is the total value of a household’s private businesses. Pension
value is total value of retirement accounts other than what is reported in other categories. Cash
is the sum of a household’s cash, checking account and savings account balances. Stocks
and bonds consist of direct and indirect holdings of equity and bonds, respectively.
Indirect holdings are assets such as mutual funds and defined contribution pensions.
Households are asked to classify their mutual funds into equity mutual funds or bond mutual
funds. If they hold balanced funds, half of the fund value is identified as stocks, and half as
bonds.
We use RE/NW, BUS/NW, and PEN/NW in our regressions to control for background risks.
Many studies show that individuals bearing more undiversifiable income risk reduce the
proportion of wealth held in risky stocks (see, for example Pratt and Zeckhauser (1987),
Kimball (1990), and Viceira (2001)). Similarly, Heaton and Lucas (2000) report a negative
effect of entrepreneurial income risk on portfolio choice, and Fratantoni (1998) shows that
committed expenditure risks associated with owner-occupied housing also have a negative
impact on relative stockholding. Labor income risk, entrepreneurial risk, and homeownership
risk are examples of background risks.
Finally, in our analysis,MRT_STS represents the marital status of a household, NOM is the
total number of people in a household, and HLTH is a dummy that equals one if the
respondent’s self-assess health (of self or spouse) is poor, and zero otherwise. Other related
Copyright # 2010 John Wiley & Sons, Ltd. Review of Behavioral Finance, 2, 81–105 (2010)
DOI: 10.1002/rbf.10
86 P. He and X. Hu
household demographic information such as age (AGE) and years of education (EDU) is also
used in our analysis.
Summary statisticsThe 2004 SCF data contain about 22,000 household observations. Since we are interested in
the variation of the composition of wealth across different planning horizons, we group
households by their planning horizons. We then report the mean and the median8 of the
relevant variables for each cohort as well as the whole sample. Table 1 presents the means and
medians for some of the variables defined in subsection ‘Variable Definitions’ above across
planning horizon cohorts as well as at the aggregate level.
Table 1 demonstrates that households with longer planning horizons are wealthier on
average than those with shorter horizons; they have higher income as well as higher net worth.
At the same time, households with longer planning horizons are less financially constrained
and less risk averse, which can be due to the fact that they are wealthier. The statistics also
suggest that on average households with longer planning horizons have more years of
education, which might be one of the reasons why they are wealthier. From the table we also
observe that relative business value (relative to net worth) is increasing in the planning
horizon, relative real estate value is decreasing in the planning horizon, and relative pension
value do not depend on the planning horizon, and cash holdings are decreasing with planning
horizons, while stock and bond holdings are increasing with planning horizons. The statistics
also suggest that more than half of the households in our sample do not hold any stocks or
bonds. Therefore, the portfolio choice puzzle as reported in Heaton and Lucas (1997), for
example, still exists in the 2004 SCF data.
DETERMINATION OF PLANNING HORIZON
One interesting finding in Table 1 is that we do not observe a smaller mean or median of age
when the planning horizon is longer. Instead, we observe a hump shape in the average age as
the planning horizon increases. In particular, there is a substantial drop in age when a
household reports a planning horizon longer than 10 years. The age can affect planning
horizon in at least two ways. On one hand, older people tend to be more patient, thus tend to
have longer planning horizons. On the other hand, as assumed in the life-cycle portfolio
choice literature, older individuals have shorter investment horizon because of shorter life
expectancy. Consistent with these two explanations, for the households with short horizons
(few months and 1 year), the mean age (46) is 3 years less than the median age (49) of the
same group. For households with medium or long horizons, the statistical median and mean
ages are similar, and longer than those within the households with short horizons. This can be
explained by a right skewed distribution in age in the short horizon group, that is, there are
many young households but also some very old ones with short horizons. Table 2 below
presents the population distributions across age and planning horizon.
8When respondents take the survey, they tend to be in the wealthy end of the distribution. The Federal Reserve Boardtherefore designs a "frequency variable" that assigns a weight to each respondent so that the resulting summarystatistics are representative of the entire U.S. population. We apply this frequency variable to calculate the mean andthe median.
Copyright # 2010 John Wiley & Sons, Ltd. Review of Behavioral Finance, 2, 81–105 (2010)
DOI: 10.1002/rbf.10
Household Investment—The Horizon Effect 87
Table
1.Summarystatistics
Planninghorizon
Few
months
Within
1year
1–5years
5–10years
10þ
years
All
Mean
Median
Mean
Median
Mean
Median
Mean
Median
Mean
Median
Mean
Median
Ageofthe
respondent
49.09
46.00
49.07
46.00
51.43
50.00
50.27
50.00
46.79
45.00
49.73
48.00
Years
ofeducation
12.38
12.00
12.77
12.00
13.42
14.00
13.91
14.00
14.32
15.00
13.38
13.00
Riskattitude
3.40
4.00
3.29
3.00
3.22
3.00
3.06
3.00
2.92
3.00
3.18
3.00
Marital
status
0.40
0.00
0.44
0.00
0.51
1.00
0.60
1.00
0.61
1.00
0.52
1.00
No.ofmem
bers
inhousehold
2.61
2.00
2.49
2.00
2.45
2.00
2.60
2.00
2.60
2.00
2.54
2.00
Borrowingconstraint
0.35
0.00
0.33
0.00
0.22
0.00
0.20
0.00
0.17
0.00
0.25
0.00
Jobincome
$27,255
$12,000
$31,725
$18,000
$38,343
$25,000
$55,572
$37,440
$70,005
$44,000
$44,141
$26,000
Net
worth
$190,076
$42,680
$252,966
$43,900
$332,272
$91,550
$625,072
$169,670
$785,999
$210,300
$432,566
$93,400
Realestate
$137,098
$50,000
$188,334
$70,000
$199,349
$103,000
$327,406
$157,000
$371,300
$177,000
$242,926
$120,000
Liquid
/totalasset
0.141
0.024
0.160
0.033
0.203
0.064
0.191
0.081
0.194
0.088
0.181
0.055
Stocks/liquid
asset
0.170
0.000
0.154
0.000
0.234
0.000
0.306
0.067
0.345
0.247
0.247
0.000
Bonds/liquid
asset
0.114
0.000
0.118
0.000
0.160
0.000
0.173
0.000
0.182
0.021
0.153
0.000
Cash/liquid
asset
0.716
1.000
0.728
1.000
0.606
0.877
0.521
0.453
0.473
0.333
0.600
0.800
Stocks/net
worth
0.070
0.000
0.053
0.000
0.098
0.000
0.107
0.005
0.136
0.028
0.095
0.000
Realestate
/net
worth
2.077
0.806
1.961
0.742
2.247
0.711
1.191
0.719
1.227
0.731
1.751
0.731
Businessvalue/
net
worth
0.055
0.000
0.032
0.000
0.044
0.000
0.057
0.000
0.071
0.000
0.052
0.000
Pension/net
worth
0.143
0.000
0.111
0.000
0.247
0.000
0.129
0.000
0.133
0.000
0.162
0.000
Note:Thistablecontainsthesummarystatistics
foroursample.Themeanandmedian(inparentheses)ofrelevantvariablesforeach
planninghorizoncohortas
wellas
thewhole
sample
arereported.Alltheratiovariablesaresignificantat
the1%
level.
Copyright # 2010 John Wiley & Sons, Ltd. Review of Behavioral Finance, 2, 81–105 (2010)
DOI: 10.1002/rbf.10
88 P. He and X. Hu
As we can see from Table 2, on average retired households have shorter horizons. For
households older than 65, the population with the planning horizon of 10þ years drops more
than those younger than 65. This is consistent with the results in Table 1. The distribution
across other planning horizons is similar across two age groups. In summary, age is not a
sufficient statistics for horizon as we do not observe a monotone relationship between age and
planning horizon.
The planning horizon might also be linked to different types of foreseeable expenditures.
Survey respondents are asked to report whether they anticipate major expenditures in the next
5–10 years. Theymay list up to six of such expenditures. In total, the SCF includes 21 types of
major expenditures, which are listed in the Appendix. The most-reported major expenditures
include a home purchase, education, health care, general support for family members or a
baby, the purchase of a car or other durable goods, business, investments, etc. Intuitively, not
only do age and foreseeable expenditures individually affect a household’s planning horizon,
but the interaction of age and foreseeable expenditures also affects planning horizon as well.
For instance, the education of children is a long-horizon plan for young households, but it
could be a short-horizon event for middle-aged households. In our 2004 SCF data, about 53%
of households report foreseeable major expenditures. Among those who report foreseeable
expenditures, about 60% report that they are currently saving for these foreseeable
expenditures. Some of the empirical exercises below are based on this particular group of
households, that is, those saving for self-reported foreseeable major expenditures in the next
5–10 years. For convenience, we define the households in this sub-sample as ‘‘savers,’’ and
we define those who report foreseeable expenditures but that are not saving for them as ‘‘non-
savers.’’ We define the remaining households as ‘‘others.’’
Before we study the effect of foreseeable expenditures on a household’s planning horizon,
we first report in Table 3 the population distribution of major foreseeable expenditures across
two age groups for those who report foreseeable expenditures, including both ‘‘savers’’ and
‘‘non-savers.’’ We group the 21 types of expenditures into seven categories. For example, the
education category includes ‘‘education for self and spouse’’ and ‘‘education for children and
grandchildren,’’ the Family/Children category includes ‘‘general support for self and
spouse,’’ ‘‘general support for children and grandchildren,’’ ‘‘general support for parents,’’
and ‘‘baby;’’ etc.
As we can see from Table 3, the distributions of major foreseeable expenditures are
different across age. More than 76% of households older than 65 claim that they anticipate
Table 2. Sample distribution by planning horizon and age
Population distribution
Age<¼ 65 Age> 65 All
Few months 18.2% 20.0% 18.54%1 Year 13.7% 15.5% 14.05%1–5 years 26.1% 34.5% 27.82%5–10 Years 27.1% 21.7% 26.00%10þ Years 15.0% 8.3% 13.59%Horizon in Yrs (Mean/Median) 4.41/2.5 3.58/2.5 4.24/2.5
Note: This table reports the population distributions across planning horizons. For each age cohort, we report the
proportion of population with certain planning horizon. All percentage values are significantly different from zero.
Copyright # 2010 John Wiley & Sons, Ltd. Review of Behavioral Finance, 2, 81–105 (2010)
DOI: 10.1002/rbf.10
Household Investment—The Horizon Effect 89
health-related expenditures, whereas only about 21% of the younger households are
concerned about health expenditures. About half of the households younger than 65 report a
home purchase and education as major foreseeable expenditures, while these expenditures are
of much less importance to older households.
We now demonstrate which household characteristics affect the planning horizon.
Specifically, we run the following regression for the full sample and for ‘‘savers:’’
PLN HRZi ¼ b0 þ bXi þX21
k¼1
gkEXPik þX21
k¼1
dkAGEi � EXPik þ "i (1)
where Xi includes log(INC), log(NTW), EDU, RSK_AVS, MRT_STS, NOM, HLTH,
BRW_CTR and AGE, and EXPik is a dummy variable indicating whether (¼1) or not
(¼0) household i has a foreseeable expenditure k. The product term AGEi�EXPik captures
the impact of the interaction between age and foreseeable expenditures on a household’s
planning horizon. We report the results in Table 4 for the full sample and for ‘‘savers.’’ Again,
we separate ‘‘savers’’ out because we expect that households who are saving for foreseeable
expenditures should be more affected by these expenditures.
In Table 4, the regression coefficients are qualitatively similar in the full sample and in the
sub-sample of savers. In the full sample regression, we find that a 1% increase in income leads
to a weak 0.05 years increase in planning horizon, and a 1% increase in net worth leads to a
0.27 years increase in planning horizon. The planning horizon of a household with one more
year of education is 0.14 years longer. More risk averse households have shorter planning
horizon. Married households have a longer planning horizon than unmarried ones, on average
by 0.72 years. One addition family member in a household reduces its planning horizon by 0.1
years. Households in poor health condition and those with borrowing constraints have shorter
planning horizons.
Many foreseeable expenditure dummies are significant in our regression despite the fact
that these expenditures are expected in 5–10 years while most households report planning
horizons much shorter than that. The existence of foreseeable expenditures has negative
impact on planning horizons. In our full sample results, the coefficient for EXP1 (for
education of children, see Appendix) is �0.88, and for AGE�EXP1 is 0.01; for a household
with median age (48 years), the joint effect is�0.40, that is, the planning horizon is shortened
Table 3. Sample distribution by type of expenditures and age
Foreseeable expenditures Age <¼ 65 (%) Age> 65 (%) All (%)
Home 46.36 11.36 42.09Education 49.64 6.29 44.36Health 21.05 76.78 27.84Family/Children 3.84 3.46 3.80Durable Good Purchases 9.01 6.08 8.65Investment 1.19 0.48 1.10Others 3.29 10.29 4.14
Note: This table reports the population distributions across major foreseeable expenditures. For each age cohort, we
report the proportion of population that reports certain foreseeable expenditures. Note that households can report up
to six types of expenditures at the same time. All percentage values are significantly different from zero.
Copyright # 2010 John Wiley & Sons, Ltd. Review of Behavioral Finance, 2, 81–105 (2010)
DOI: 10.1002/rbf.10
90 P. He and X. Hu
Table
4.Household
characteristicsandplanninghorizons
Fullsample
Subsample-savers
Variables
Coeff.
t-stat
Variable
Coeff.
t-stat
Variable
Coeff.
t-stat
Variable
Coeff.
t-stat
Variable
Coeff.
t-stat
Variable
Coeff.
t-stat
CST
1.78
6.45
EXP1
�0.88
�3.13
AGE�EXP1
0.01
2.30
CST
0.45
0.79
EXP1
0.37
0.88
AGE�EXP1
�0.02
�1.95
Log(INC)
0.05
2.54
EXP2
�0.80
�2.17
AGE�EXP2
0.00
0.20
Log(INC)
0.12
3.74
EXP2
1.16
2.19
AGE�EXP2
�0.04
�3.28
Log(NTW)
0.27
18.25
EXP3
�0.87
�2.86
AGE�EXP3
0.01
2.91
Log(NTW)
0.31
11.28
EXP3
0.23
0.49
AGE�EXP3
�0.01
�0.64
EDU
0.14
13.38
EXP4
0.37
0.79
AGE�EXP4
�0.01
�0.88
EDU
0.11
5.27
EXP4
2.30
3.48
AGE�EXP4
�0.04
�3.49
RSK_AVS
�0.28
�8.74
EXP5
8.88
3.42
AGE�EXP5
�0.18
�3.79
RSK_AVS
�0.14
�2.66
EXP5
8.13
3.20
AGE�EXP5
�0.17
�3.84
MRT_STS
0.72
11.60
EXP6
�14.56
�1.14
AGE�EXP6
0.24
1.06
MRT_STS
0.46
4.18
EXP6
�7.76
�0.55
AGE�EXP6
0.11
0.43
NOM
�0.10
�4.44
EXP7
�2.39
�1.73
AGE�EXP7
0.05
1.97
NOM
�0.03
�0.67
EXP7
�4.38
�2.41
AGE�EXP7
0.07
2.51
HLTH
�0.65
�6.16
EXP8
2.61
1.51
AGE�EXP8
�0.04
�1.24
HLTH
�0.48
�2.37
EXP8
�1.77
�0.44
AGE�EXP8
0.04
0.42
BRW_CTR
�0.54
�7.86
EXP9
�1.70
�2.16
AGE�EXP9
0.04
2.50
BRW_CTR
�0.65
�5.51
EXP9
�3.51
�2.63
AGE�EXP9
0.05
2.21
AGE
�0.03
�13.74
EXP10
13.82
2.93
AGE�EXP10
�0.49
�3.47
AGE
�0.02
�2.59
EXP10
26.02
4.14
AGE�EXP10
�0.84
�4.47
EXP11
�1.04
�4.62
AGE�EXP11
0.01
1.67
EXP11
�1.52
�3.88
AGE�EXP11
0.02
1.87
EXP12
0.60
1.24
AGE�EXP12
�0.01
�1.36
EXP12
�0.60
�0.89
AGE�EXP12
0.01
0.44
EXP13
�9.68
�2.53
AGE�EXP13
0.11
2.19
EXP13
�3.70
�0.66
AGE�EXP13
0.02
0.24
EXP14
�9.85
�4.80
AGE�EXP14
0.16
4.92
EXP14
�8.01
�3.53
AGE�EXP14
0.11
2.64
EXP15
0.37
0.56
AGE�EXP15
�0.01
�0.92
EXP15
1.56
1.77
AGE�EXP15
�0.04
�2.32
EXP16
0.54
0.57
AGE�EXP16
�0.02
�1.13
EXP16
2.27
1.95
AGE�EXP16
�0.05
�2.28
EXP17
0.30
0.17
AGE�EXP17
�0.03
�0.76
EXP17
6.90
2.24
AGE�EXP17
�0.23
�3.11
EXP18
�0.44
�0.12
AGE�EXP18
�0.03
�0.45
EXP18
�2.96
�2.09
AGE�EXP18
0.17
0.90
EXP19
�1.35
�0.57
AGEs�EXP19
0.07
1.46
EXP19
�0.71
�0.29
AGE�EXP19
0.05
0.93
EXP20
0.09
0.03
AGE�EXP20
�0.01
�0.19
EXP20
2.22
0.67
AGE�EXP20
�0.05
�0.94
Adjusted
R2
0.18
EXP21
3.64
2.17
AGE�EXP21
�0.09
�2.62
EXP21
�5.00
�1.90
AGE�EXP21
0.05
0.99
Note:This
table
documents
how
household
characteristicsaffect
theplanninghorizonin
thefullsample
andthesub-sam
ple
ofsavers.Weregress
theplanninghorizonon
household
income,
net
worth,education,risk
aversion,marital
status,
thenumber
ofhousehold
mem
bers,
healthconditions,
borrowingconstraints,age,
majorforeseeable
expenditures,andtheproductsofageandmajorforeseeable
expenditures.Wereportthecoefficientsandtheirt-statistics
usingtheOLSregression.
Copyright # 2010 John Wiley & Sons, Ltd. Review of Behavioral Finance, 2, 81–105 (2010)
DOI: 10.1002/rbf.10
Household Investment—The Horizon Effect 91
by 0.40 years. For all 21 foreseeable expenditures, the average age and expenditure effect is
�1.14 years in the full sample, and �1.34 years in the savers sample. Hence the age and
expenditure effect are quantitatively stronger in the savers sample.
To summarize, a household’s planning horizon is affected by various household
characteristics. We can divide these factors into two groups: those related to the timing of
divesting, that is, when the investment will be discontinued, and those related to subjective
mental accounting, or perception of risks. The first group of factors includes age and
foreseeable expenditures. We categorize other factors into the second group. It is easy to
understand that one’s risk aversion and self-assessed health condition are subjective
household characteristics. With respect to income, wealth, education, marital status, number
of people in the household, and borrowing constraints, we believe that these factors indirectly
affect a household’s emotional stability, patience, confidence, and other mental
characteristics.9
HOUSEHOLD PORTFOLIO CHOICE
The effect of planning horizon on portfolio choiceThe summary statistics in Table 2 suggest that the composition of wealth, as well as many
other household demographic characteristics, vary with the planning horizon. In particular,
households with longer planning horizons tend to hold more stocks and bonds and less cash;
they tend to be older, and have more years of education; they are wealthier; they own more
private businesses; and they are less financially constrained and less risk averse. Even though
the summary statistics document that households with longer planning horizons tend to hold
more stocks, it is not clear whether this result is driven by variations in other demographic
characteristics such as age, years of education, net worth, etc. To more systematically study
the correlation between stockholding and planning horizons, we run a regression that relates
the proportion of stockholdings relative to liquid assets to a number of independent
variables10. In particular, we estimate
ST
LIQ¼ b0 þ bXi þ gPLN HRNi þ "i (2)
where Xi includes log(INC), log(NTW), AGE, EDU, RSK_AVS, BRW_CTR, BUS/NW, RE/NW
and PEN/NW.
Table 5 reports regression results using the 2004 SCF data in Panel A, as well as the results
using both 2001 and 2004 SCF data in Panel B. For the results using both 2001 and 2004 SCF
data, we add a dummy variable (2001 SCF data¼ 1) to control for the business cycle effect.
We conjecture that during or right after an equity bull market, households hold relatively
more equity in their assets due to performance chasing, hence the coefficient on the 2001
dummy variable is expected to be significantly positive. In both Panel A and Panel B, the first
column of estimates presents ordinary least squares (OLS) regression results. The second
9By similar reasoning, age can also be categorized as a subjective factor, but we include it in the first group as assumedin many portfolio decision studies.10Endogeneity issue could be a problem with the OLS regression. A household’s stock investment may contribute toits total net worth; and a household who hold significant stocks may say that it has a long planning horizon. Weexplore the endogeneity issue in ‘Robustness Tests’ section, and find that the results are robust.
Copyright # 2010 John Wiley & Sons, Ltd. Review of Behavioral Finance, 2, 81–105 (2010)
DOI: 10.1002/rbf.10
92 P. He and X. Hu
column of estimates contains results of a Tobit regression based on the following latent
variable model:
Y� ¼ Xbþ "; "~Nð0; s2ÞY ¼ max f0; Y�g (3)
In our sample, more than 50% of the households have no stocks, so the data is left-
censored. One justification for using the Tobit model is the existence of some fixed costs for
entering into equity market, including mental cost.
In Panel A, both the OLS regression and the Tobit regression indicate that households with
longer planning horizons hold proportionally more stocks. For a household with 1 year longer
planning horizon, its ST/LIQ ratio is 0.34 percentage point higher (in OLS). This implies a
1.4% more investment in stocks given the mean of ST/LIQ ratio is 24.70% in the full sample.
As expected, the Tobit regression result is stronger. A household with 1 year longer planning
horizon would increase its ST/LIQ by 2.6%.
With similar calculation, we find that, with Tobit results, households with 1% more in
wealth leads to a 0.31% more in stock investment. We also observe that when households are
1 year older, they hold 0.25%more in stocks, while 1 more year of education brings additional
20.9% in stock investment! Households with one level higher risk aversion will invest 57.2%
less in stocks, and those with borrowing constraints will invest 41.8% less in stocks. Among
background risks, 1% increase in relative business value (which is an approximate 0.052%
increase in BUS/NW, as the mean of BUS/NW is 0.052 in full sample) leads to a 2.1% decrease
Table 5. Household portfolio choices and planning horizons
Panel A: 2004 SCF data Panel B: 2001 & 2004 SCF data
VariablesOLS Tobit OLS Tobit
Coeff. t-stat Coeff. p-value Coeff. t-stat Coeff. p-value
CST �0.2821 �11.57 �0.9931 0.00 �0.2532 �14.71 �0.9300 0.00Log(INC) 0.0002 0.12 �0.0148 0.00 0.0006 0.48 �0.0134 0.00Log(NTW) 0.0349 24.04 0.0762 0.00 0.0416 42.10 0.0883 0.00AGE 0.0008 4.57 0.0006 0.04 0.0002 1.77 �0.0008 0.00EDU 0.0269 26.23 0.0517 0.00 0.0220 31.61 0.0425 0.00RSK_AVS �0.0867 �28.07 �0.1414 0.00 �0.0927 �43.37 �0.1517 0.00BRW_CTR �0.0429 �6.40 �0.1032 0.00 �0.0406 �8.69 �0.0859 0.00BUS/NW �0.0599 �7.59 �0.0977 0.00 �0.0721 �12.69 �0.1302 0.00RE/NW 0.0007 2.35 0.0019 0.00 0.0000 0.88 0.0000 0.97PEN/NW �0.0006 �0.38 �0.0003 0.90 �0.0001 �0.08 0.0020 0.46PLN_HRZ 0.0034 4.90 0.0065 0.00 0.0039 8.07 0.0072 0.00DUMMY_2001 0.0122 3.80 0.0149 0.00Adjusted R2 0.28 0.31
Note: This table documents how household characteristics, including the planning horizon, affect household portfolio
choices in the full sample. We regress the proportion of stockholdings relative to liquid assets (ST/LIQ) on household
income, net worth, age, education, risk aversion, borrowing constraints, relative values of business, real estate assets,
and pension, and the planning horizon. We report the coefficients and their t-statistics or p-values for the OLS and
Tobit regressions; the adjusted R2 for the OLS regression is also reported.
Copyright # 2010 John Wiley & Sons, Ltd. Review of Behavioral Finance, 2, 81–105 (2010)
DOI: 10.1002/rbf.10
Household Investment—The Horizon Effect 93
in stock investment; 1% increase in relative real estate value leads to an increase of 1.6%
stocks; the impact of pension value is negligible.
The results in Panel B are similar, and we observe a significantly positive coefficient for the
2001 dummy, consistent with our conjecture. For a household in 1998–2000, its ST/LIQ ratio
is 1.22 percentage points higher (in OLS). This implies a 4.9% more investment in stocks
given the mean ST/LIQ ratio is 24.70% in the full sample of 2004.
It has been documented that households with low or moderate wealth do not hold any
stocks. It is not surprising, therefore, to observe that relative stockholdings increase in net
worth. The positive age effect is inconsistent with the predictions of rational portfolio choice
models such as that in Viceira (2001), who finds the opposite. However, this result does not
contradict with the results in Heaton and Lucas (2000), who document that ‘‘. . . [the] share ofstocks rises slightly with age for cohorts under age 65, and then declines significantly for
those age 65 and over. . ..’’ In our regression, pooling in age cannot reveal their finding. We
explore the age effect further in the robustness check in Section ‘Robustness Tests’, and their
results are confirmed. The effect of education suggests that more educated households are
more likely to take advantage of the equity premium. Households’ relative private business
has a strong negative effect on households’ stockholdings. This is consistent with Heaton and
Lucas (2000b), who show that entrepreneurial risks reduce the risky financial asset holdings.
The relative pension fund value is also negative but not significant. The relative real estate
value has a positive effect, probably because the real estate value here include both owner-
occupied housing and all other real estate values, hence it is a noisy measure for background
risk associated with owner-occupied housing.
In summary, our results show that controlling for other factors, a household’s planning
horizon affects its portfolio choices. The effect is both statistically and economically
significant. We also compare the effect of planning horizon on household portfolio with that
of age and education. In the full sample, the standard deviations for planning horizon, age,
and education are 3.49, 7.29, and 2.84, respectively. With the OLS coefficients, one standard
deviation change of planning horizon leads to a 4.8% change in stock investment, the same
measure is 2.4% for age, and 30.9% for education. Therefore, the planning horizon
contributes more to the variation in risky asset investment than age, but less than education.
Cross group differenceIn this subsection, we explore further the effect of planning horizon on household portfolio
choice by studying cross-group differences. Specifically, we conduct two comparisons, one
on households with foreseeable expenditures versus thosewithout, and another on households
saving for foreseeable expenditures (‘‘savers’’) versus those not saving for foreseeable
expenditures (‘‘non-savers’’). We seek to determine how the existence of foreseeable
expenditures affects the role of planning horizon on household portfolio choice. We modify
the regression in Equation (2) by allowing for different planning horizon coefficients across
the two groups, and we also test the fixed effects by allowing for different constant terms
across the two groups. The results are in Table 6.
Table 6 shows that households with foreseeable expenditure care more about planning
horizons than those without expected expenditures when they make investment decision.
Furthermore, those who are saving for the foreseeable expenditures care more about planning
horizon than who are not saving. One possible explanation is that the existence of foreseeable
expenditures enhances the awareness of planning horizon. The awareness is particularly
strong for those who are saving for the foreseeable expenditures.
Copyright # 2010 John Wiley & Sons, Ltd. Review of Behavioral Finance, 2, 81–105 (2010)
DOI: 10.1002/rbf.10
94 P. He and X. Hu
Spending horizon and risk-perceiving horizonThe classic portfolio choice models define horizon as a function of age from the perspective
of the spending horizon (or investment horizon). We do find that the reported planning
horizon has a significant impact on household portfolio choice. We have discussed that age
variation does not fully capture the difference in household planning horizon. Instead,
the planning horizon is related to various household characteristics. Another finding is that the
existence of foreseeable expenditures not only affects the planning horizon, but also affects
the dependence of portfolio choice on planning horizon. This creates a puzzle: If the existence
of foreseeable expenditures only shifts the planning horizon due to its effect on the spending
horizon, which is defined as the horizon over which consumption occurs or wealth is
maximized, then it should not affect the dependence of portfolio choice on the planning
horizon. Our findings suggest that the effect of foreseeable expenditures is more than the
effect through its impact on the objective spending horizon. We conjecture that the existence
of foreseeable expenditures also affects household mental accounting. To further explore this
subjective effect as well as the pure spending horizon effect, we decompose the planning
horizon into the following two parts using the estimated regression coefficients from
Equation (1)
SPD HRNi ¼ baAGEi þP21
k¼1
g iEXPik þP21
k¼1
diAGEi � EXPik
PEC HRNi ¼ b0 þ b�aX�ai
(4)
Table 6. Household portfolio choices and planning horizons across groups
Variablesw/ EXP (1) vs. w/o EXP (2) Savers (1) vs. non-savers (2)
Coeff. t-stat Coeff. t-stat
CST (1) �0.2851 �11.65 �0.2632 �10.50CST (2) �0.2734 �10.89 �0.2814 �11.47Log(INC) 0.0000 �0.02 �0.0005 �0.26Log(NTW) 0.0349 24.05 0.0347 23.92AGE 0.0009 4.70 0.0010 5.40EDU 0.0267 26.02 0.0264 25.80RSK_AVS �0.0869 �28.07 �0.0868 �28.11BRW_CTR �0.0423 �6.30 �0.0421 �6.29BUS/NW �0.0599 �7.59 �0.0580 �7.34RE/NW 0.0008 2.41 0.0008 2.56PEN/NW �0.0006 �0.40 �0.0009 �0.56PLN_HRZ (1) 0.0055 5.92 0.0050 4.39PLN_HRZ (2) 0.0013 1.32 0.0026 3.19Adjusted R2 0.63 0.64F-test for CST (p-value) 1.96 (0.16) 4.31 (0.04)F-test for PLN_HRZ (p-value) 10.52 (0.00) 3.13 (0.08)
Note: This table documents how the effect of planning horizon on household portfolio choice differs across groups.
We make two comparisons: Households with major foreseeable expenditures versus those without; and households
who are saving for major foreseeable expenditures versus those who are not. In our regressions, we allow for different
constant terms and different coefficients for the planning horizon across groups. We report the coefficients and their
t-statistics for the OLS regressions; the adjusted R2 and F-test results for the coefficient difference are also reported.
Copyright # 2010 John Wiley & Sons, Ltd. Review of Behavioral Finance, 2, 81–105 (2010)
DOI: 10.1002/rbf.10
Household Investment—The Horizon Effect 95
where X-a represents all the independent variables in Xi of Equation (1) other than AGE,
namely log(INC), log(NTW), EDU, RSK_AVS, MRT_STS, NOM, HLTH, and BRW_CTR.
SPD_HRZ captures the planning horizon variation due to age and foreseeable expenditures,
and PEC_HRZ captures the planning horizon variation due to other factors, which, as we
argue earlier, are categorized as subjective mental accounting factors. We use SPD_HRZ as a
proxy for the spending horizon and we use PEC_HRZ as a proxy for the risk-perceiving
horizon. The means and standard deviations of SPD_HRZ and PEC_HRZ for the full sample
and sub-samples are reported below:
Full sample Savers
Mean STD Correlation Mean STD Correlation
SPD_HRZ �1.91 0.62 0.006 �2.04 0.59 0.02PEC_HRZ 6.38 1.18 6.63 1.12
Consistent with Table 4, SPD_HRZ has a negative mean since both age and the existence
of foreseeable expenditures shorten the horizon. The standard deviation of PEC_HRZ is twice
as large as that of SPD_HRZ. At the same time, the correlation between these two horizon
variables is very low.
With the planning horizon decomposition, we rerun the regression with the full sample and
the sub-sample of savers of 2004 SCF data. The results are reported in the Panel A of Table 7.
We also include the results using both 2001 and 2004 SCF data in Panel B with a 2001 data
dummy.
Results in Table 7 show that the proxy for the risk-perceiving horizon has a more
significant effect on household portfolio choice than the proxy for the spending horizon. The
coefficient for PEC_HRZ is much larger (about 10 times in the full sample results of Panel A)
than that for SPD_HRZ; given that PEC_HRZ has a larger standard deviations, the variation of
PEC_HRZ explains a lot more of the variation in household stockholdings than does the
variation of SPD_HRZ.
The horizon effects in Panel B are similar. We only observe a significantly positive
coefficient for the 2001 dummy in the full sample, but not in the sub-sample of savers. One of
the possible reasons is that savers rationally adjust their portfolios with less performance
chasing, so their asset structure is relatively stable.
The results in Table 7 confirm that the subjective mental accounting part of the planning
horizon has more explanatory power than does the objective part. This is consistent with the
behavioral story (as in Thaler et al., 1997) that what matters is the frequency at which risks are
perceived. 11
ROBUSTNESS TESTS
In this section, we present robustness tests on the horizon effect using the ‘‘savers’’ sub-
sample. The results using the full sample are similar.
11We also conduct a cross-group comparison between those with foreseeable expenditures and those without, as wellas between savers and non-savers. The results are consistent with our general conclusion, thus omitted. Tables areavailable upon request.
Copyright # 2010 John Wiley & Sons, Ltd. Review of Behavioral Finance, 2, 81–105 (2010)
DOI: 10.1002/rbf.10
96 P. He and X. Hu
Table
7.Household
portfoliochoices,spendinghorizonsandrisk
perceivinghorizons
Panel
A:2004SCFdata
Panel
B:2001&
2004SCFdata
Fullsample
Subsample-savers
Fullsample
Subsample-savers
OLS
Tobit
OLS
Tobit
OLS
Tobit
OLS
Tobit
Variables
Coeff.
t-stat
Coeff.
p-value
Coeff.
t-stat
Coeff.
p-value
Variable
Coeff.
t-stat
Coeff.
p-value
Coeff.
t-stat
Coeff.
p-value
CST
�0.2539
�10.07
�0.9343
0.00
�0.2515
�5.78
�0.8084
0.00
CST
�0.3465
�16.47
�1.1926
0.00
�0.3645
�9.79
�0.9925
0.00
Log(INC)
�0.0039
�2.06
�0.0236
0.00
�0.0126
�4.06
�0.0247
0.00
Log(INC)
�0.0033
�2.46
�0.0239
0.00
�0.0081
�3.63
�0.0204
0.00
Log(NTW)
0.0170
6.30
0.0380
0.00
0.0205
4.31
0.0365
0.00
Log(NTW)
0.0273
14.20
0.0509
0.00
0.0298
8.77
0.0423
0.00
AGE
0.0006
2.60
0.0000
0.90
0.0005
1.42
0.0005
0.33
AGE
0.0002
1.47
�0.0010
0.00
0.0009
3.43
0.0010
0.01
EDU
0.0181
12.12
0.0323
0.00
0.0234
8.32
0.0383
0.00
EDU
0.0167
18.14
0.0281
0.00
0.0188
11.08
0.0288
0.00
RSK_AVS
�0.0691
�18.33
�0.1032
0.00
�0.0713
�10.93
�0.1000
0.00
RSK_AVS
�0.0794
�30.03
�0.1159
0.00
�0.0856
�18.72
�0.1225
0.00
BRW_CTR
�0.0086
�1.07
�0.0266
0.05
�0.0254
�1.79
�0.0607
0.00
BRW_CTR
�0.0077
�1.28
0.0016
0.87
�0.0032
�0.31
0.0031
0.84
BUS/NW
�0.0589
�7.46
�0.0966
0.00
�0.1414
�9.44
�0.2217
0.00
BUS/NW
�0.0717
�12.61
�0.1292
0.00
�0.1167
�10.99
�0.1713
0.00
RE/NW
0.0006
1.97
0.0016
0.00
0.0003
0.13
�0.0076
0.15
RE/NW
0.0000
1.04
0.0000
0.91
0.0003
0.29
�0.0132
0.00
PEN/NW
�0.0005
�0.32
�0.0003
0.93
�0.0017
�0.25
�0.0423
0.02
PEN/NW
�0.0003
�0.21
0.0013
0.64
�0.0022
�0.68
�0.0593
0.00
SPD_HRZ
�0.0081
�1.58
�0.0168
0.04
�0.0076
�1.03
�0.0051
0.62
SPD_HRZ
0.0053
1.28
0.0037
0.59
0.0094
1.52
0.0206
0.02
PEC_HRZ
0.0626
8.39
0.1360
0.00
0.0666
4.93
0.1167
0.00
PEC_HRZ
0.0514
9.35
0.1348
0.00
0.0553
5.54
0.1210
0.00
DUMMY_2001
0.0124
3.88
0.0148
0.00
0.0040
0.74
�0.0022
0.78
Adjusted
R2
0.28
0.28
Adjusted
R2
0.31
0.30
Note:Thistabledocumentstheeffectofthetwocomponentsoftheplanninghorizon,SPD_HRZandPEC_HRZ,onhouseholdportfoliochoice.Westudythiseffectinboththefull
sampleandthesub-sam
pleof‘‘savers.’’Wereportthecoefficientsandtheirt-statistics
orp-values
forboththeOLSandTobitregressions;theadjusted
R2fortheOLSregressions
arealso
reported.
Copyright # 2010 John Wiley & Sons, Ltd. Review of Behavioral Finance, 2, 81–105 (2010)
DOI: 10.1002/rbf.10
Household Investment—The Horizon Effect 97
EndogeneityIn our first robustness check, we look at the endogeneity issue in our regression defined in
Equation (2). There are two major sources for the endogeneity problem: net wealth (lg(NTW)
and the planning horizon (PLN_HRZ). When households invest more in stocks, they may
cumulate more wealth. Hence stockholdings could contribute to households’ net wealth. The
self-reported ‘‘planning horizon’’ could also depend on stockholdings, as households who
hold significant stocks may say that they have long planning horizons.12 To resolve the
endogeneity issue we identify some instrumental variables for each endogenous variable, and
use two-stage least squares (2SLS) regression method.
Specifically, we use the variable ‘‘income’’ (lg(INC)) as the instrumental variable for net
wealth (lg(NTW)). This is a natural choice since most households build up their wealth from
labor income. In addition, lg(INC) is not significant in our regression (2) as shown in Table 5, so
it is a qualified instrument variable. Using the same criteria, we find that household foreseeable
education expenditures and (EXP1,EXP2) are good candidates of instrumental variables for the
planning horizon (PLN_HRZ). These instrumental variables are both economically and
statistically important in determining households planning horizon as shown in Table 4. From
the untabulated regressions, households’ stockholdings do not depend on these variables.13
The main results remain robust in Table 8: The coefficient on the planning horizon is
positive and significant, while the proxy for the risk-perceiving horizon has more explanatory
power than the proxy for the spending horizon.14 More importantly, although the statistical
significance of horizon variables is reduced after considering the endogeneity problem, the
economic significance is substantially improved as the coefficients are much larger in
absolute value for the planning horizon and the risk perceiving horizon when the horizon is
decomposed. Hence if we endogenize household planning horizon and/or net wealth, the
results are economically stronger.
Direct/indirect holdings and cross-asset diversificationWe have shown statistically that the value of stockholdings relative to liquid assets is
increasing with the planning horizon. Next, we use our sample to analyze the effect of
planning horizon on the structure of stockholdings. The stockholdings of a household can be
divided into direct holdings and indirect holdings. Indirect holdings mainly consist of mutual
funds, and are more diversified. The structure of stockholdings reflects the degree to which a
household is diversified in its stock investments. Of particular interest is the interaction
between cross-asset diversification and time diversification. We run a new set of regressions
by replacing the dependent variable in specification (2) with direct stockholdings divided by
liquid assets (DIR/LIQ) and indirect stockholdings divided by liquid assets (IND/LIQ).15
Direct stockholdings are values of all shares of stocks in households’ portfolio. Indirect
12This is confirmed by including the ST/LIQ variable in our regression (1), which yields a positive and significantcoefficient for ST/LIQ.13We have tried different combinations of expenditure variables and the product of expenditure variable with age asinstrumental variables, the result is still robust.14There is no endogeneity problem for spending horizon and risk-perceiving horizon as they are already linearcombinations of exogenous variables by construction.15Asset risk characteristics can affect a household’s cross-asset diversification behavior. For example, Moskowitz andVissing-Jørgensen (2002) and others suggest that a low level of cross-asset diversification might be due to the positiveskewness of the underlying asset returns held by an investor. Unfortunately, we are not able to control asset riskcharacteristics in our regression due to the limitation of the data.
Copyright # 2010 John Wiley & Sons, Ltd. Review of Behavioral Finance, 2, 81–105 (2010)
DOI: 10.1002/rbf.10
98 P. He and X. Hu
stockholdings are equity holdings through mutual funds, trusts, or annuities. We report results
in Table 9.
In Table 9, wealthier households invest more in both direct and indirect stockholdings, but
the effect of wealth on direct stockholdings is larger.We argue that the rich people tend to take
more risks. At the same time, younger households tend to own more direct holdings but less
indirect holdings, or younger households tend to take more risks. Education increases
stocking holdings, but more so on indirect stockholdings. It is consistent with the idea that
cross asset diversification is better accepted by more educated households. Risk aversion has
a bigger impact on indirect holdings. This might sound a bit counter intuitive, but our
interpretation is that those who own indirect stockholding are more risk averse and more
affected by risk aversion level. Consistent with this explanation, we find that the correlation
between risk aversion and the ratio of indirect stockholdings to the sum of (directþ indirect
stockholdings) is negative.
The effects of background risks are also different for direct stockholdings and indirect
stockholdings. The existence of real estate assets shift stockholdings from direct holdings to
indirect holdings, which can be justified by risk reduction efforts from households. The
existence of pension depresses the indirect holdings because of the substitution effect
between pension and indirect stockholdings.
Our major results on horizon variables are interesting. The horizon variables are
insignificant for direct stockholdings, but are significant for indirect stockholdings. This
indicates a positive correlation between cross-asset diversification and time diversification,
that is, a household with a longer horizon tends to hold a more diversified portfolio. We
provide two interpretations for this result. First, households with shorter planning horizons
Table 8. Robustness check with endogenous independent variables
Variables
EndogenizeLog(NTW)
EndogenizePLN_HRZ
Endogenize bothLog(NTW) andPLN_HRZ
Horizondecomposition
Coeff. t-stat Coeff. t-stat Coeff. t-stat Coeff. t-stat
CST �0.3179 �8.46 �0.2101 �2.99 �0.2254 �3.85 �0.4344 �9.53Log(INC) �0.0169 �1.23Log(NTW) 0.0340 15.50 0.0112 2.51 0.0037 2.16 �0.0110 �2.32AGE 0.0010 3.19 0.0025 1.72 0.0029 2.00 0.0017 4.20EDU 0.0322 16.77 0.0236 3.00 0.0311 6.39 0.0195 8.16RSK_AVS �0.0901 �17.34 �0.0800 �7.55 �0.0916 �13.10 �0.0648 �11.14BRW_CTR �0.0633 �5.57 0.0001 0.00 �0.0445 �1.33 �0.0123 �0.97BUS/NW �0.1577 �11.02 �0.1181 �4.10 �0.1098 �3.83 �0.1262 �7.40RE/NW 0.0004 0.18 �0.0053 �1.03 �0.0025 �0.64 0.0009 0.37PEN/NW �0.0023 �0.35 0.0143 0.96 0.0049 0.46 �0.0044 �0.65PLN_HRZ 0.0050 4.13 0.0597 1.93 0.0463 1.97 �0.0052 �0.70SPD_HRZ 0.1092 14.63PEC_HRZ �0.4344 �9.53Adjusted R2 0.28 0.19 0.22 0.27
Note: This table documents the two-stage least squares regression result.We use log(INC) as the instrumental variable
for log(NTW), and/or EXP1, EXP2 for PLN_HRZ. We study the effect of horizon variables in household stock
investment decision in the savers’ sample with or without horizon decomposition. We report the coefficients and their
t-statistics for the 2SLS; the adjusted R2 for the second stage regressions are also reported.
Copyright # 2010 John Wiley & Sons, Ltd. Review of Behavioral Finance, 2, 81–105 (2010)
DOI: 10.1002/rbf.10
Household Investment—The Horizon Effect 99
are more speculative, thus taking more risk by investing in individual stocks instead of mutual
funds. Second, investing in mutual funds is associated with more initial costs such as a front-
end load and/or a minimum investment amount. The possible redemption fees may also
discourage short-term mutual fund holdings. All these costs make it worthwhile to invest in
mutual funds instead of individual stocks only when the planning horizon is not too short.
Our findings on direct and indirect stockholdings lead to the following question: Does
cross-asset diversification cause time diversification? Put differently, when households have a
longer planning horizon, do they diversify across assets? Since cross asset diversification
reduces risk and in turn leads households to invest more in stocks. We explore this question by
adding a proxy for cross-asset diversification into the regression in Equation (2). The proxy
that we use is the number of mutual funds that a household invests in. We report the results in
Table 10.
As we can see from Table 10, the number of mutual funds in which a household invests
does have a positive effect on household stockholdings, however, the horizon effect is not
affected. Therefore, the hypothesis that cross-asset diversification causes time diversification
is rejected.
Residual horizonThe decomposition of the household planning horizon into spending horizon and risk-
perceiving horizon is in no way aiming to fully and accurately separate out the subjective
factors that are affected by mental accounting. On one hand, the choice of subjective factors
that we choose is not supported by any theories. These factors can also be argued as factors
that affect household spending horizon instead. On the other hand, there are many other
factors that can affect households in reporting their planning horizon. The adjusted R2 is only
Table 9. Household direct and indirect stockholdings and planning horizons
VariablesDirect holdings Indirect holdings
Coeff. t-stat Coeff. t-stat Coeff. t-stat Coeff. t-stat
CST �0.2389 �7.25 �0.2423 �6.97 �0.0290 �0.76 �0.0093 �0.23Log(INC) 0.0002 0.09 �0.0001 �0.05 �0.0090 �3.25 �0.0125 �4.33Log(NTW) 0.0284 13.59 0.0269 7.08 0.0098 4.03 �0.0064 �1.45AGE �0.0003 �1.15 �0.0004 �1.40 0.0011 3.51 0.0009 2.75EDU 0.0081 5.27 0.0073 3.23 0.0247 13.81 0.0162 6.19RSK_AVS �0.0291 �7.01 �0.0271 �5.20 �0.0612 �12.71 �0.0442 �7.30BRW_CTR �0.0045 �0.50 �0.0016 �0.14 �0.0583 �5.52 �0.0238 �1.80BUS/NW �0.0791 �6.63 �0.0790 �6.60 �0.0658 �4.75 �0.0623 �4.49RE/NW �0.0064 �3.48 �0.0063 �3.44 0.0064 3.00 0.0066 3.10PEN/NW 0.0187 3.50 0.0185 3.46 �0.0196 �3.16 �0.0202 �3.25PLN_HRZ 0.0013 1.39 0.0038 3.40SPD_HRZ �0.0035 �0.59 – – �0.0041 �0.60PEC_HRZ 0.0069 0.64 – – 0.0597 4.77Adjusted R2 0.12 0.12 0.12 0.12
Note: This table documents the effect of the planning horizon and its two components on household direct and
indirect stockholdings. Indirect stockholdings are through mutual funds, trusts, or annuities. We study this effect in
the sub-sample of ‘‘savers.’’ We report the coefficients and their t-statistics for the OLS regressions; the adjusted R2
for the OLS regressions are also reported.
Copyright # 2010 John Wiley & Sons, Ltd. Review of Behavioral Finance, 2, 81–105 (2010)
DOI: 10.1002/rbf.10
100 P. He and X. Hu
18% for regression (1). With these concerns, we calculate the residuals from regression (1),
denoted as RES_HRZ, to check whether the residual has any explaining power in the
household portfolio variation. The standard deviations of RES_HRZ are around 3.2 in both
full sample and sub-sample of savers. The regression results are reported in Table 11.
In Table 11, the coefficients on RES_HRZ in all regressions are significant. This confirms
our conjecture that some other factors, which are not included in regression (1), can also
affect household planning horizons. To compare the effect of different constructed horizon
variables on household portfolios, we first calculate the standard deviations of each horizon
variable:
STD Full sample Savers
SPD_HRZ 0.62 0.59PEC_HRZ 1.18 1.12RES_HRZ 3.24 3.20
The product of the standard deviation and the corresponding OLS coefficient is also
calculated. The absolute value of this product is reported below, in order to compare the
relative importance of each horizon variable.
Full sample Savers
SPD_HRZ 0.005 0.005PEC_HRZ 0.074 0.077RES_HRZ 0.010 0.017
Table 10. Cross-asset diversification and time diversification
VariablesOLS(1) OLS(2)
Coeff. t-stat Coeff. t-stat
CST �0.1853 �4.52 �0.1599 �3.70Log(INC) �0.0083 �2.81 �0.0124 �4.06Log(NTW) 0.0314 12.02 0.0121 2.58AGE 0.0008 2.52 0.0006 1.80EDU 0.0306 16.04 0.0204 7.35RSK_AVS �0.0876 �17.14 �0.0675 �10.52BRW_CTR �0.0602 �5.38 �0.0193 �1.38BUS/NW �0.1301 �8.83 �0.1255 �8.50RE/NW 0.0005 0.24 0.0007 0.33PEN/NW �0.0026 �0.39 �0.0031 �0.48NMF 0.0148 14.40 0.0151 14.77PLN_HRZ 0.0038 3.24 – –SPD_HRZ – – �0.0049 �0.67PEC_HRZ – – 0.0699 5.26Adjusted R2 0.30 0.30
Note: This table documents the effect of the planning horizon and its two components on household portfolio choice
while controlling the effect of cross-asset diversification by adding the number of mutual funds (NMF) in the
regression. We study this effect in the sub-sample of ‘‘savers.’’ We report the coefficients and their t-statistics for the
OLS regressions; the adjusted R2 for the OLS regressions are also reported.
Copyright # 2010 John Wiley & Sons, Ltd. Review of Behavioral Finance, 2, 81–105 (2010)
DOI: 10.1002/rbf.10
Household Investment—The Horizon Effect 101
As we can see, the constructed subjective mental accounting variable, PEC_HRZ, has the
largest impact on household portfolio decisions compared with other horizon variables.
Although the constructed residual horizon, RES_HRZ, is statistically significant, its impact on
household portfolios is not as important as the risk perceiving factors we identify in this study.
Other robustness checksWe also conduct some other robustness checks, which can be summarized below16. We run
the regressions within two age groups: The working households, or households younger than
65; and the retired households, or households older than 65. The main results are similar to
those in Table 5, namely, the effect of the planning horizon remains significant for each age
group, and the proxy for the risk-perceiving horizon is significant but the proxy for
the spending horizon is less so. One interesting observation is that, in the younger group, the
age effect is positive and significant while in the older group, negative and insignificant. This
confirms the results in Heaton and Lucas (2000) as we pointed out earlier. This result
demonstrates that age is not a perfect proxy for investment horizon, especially for younger
households. We also find that the coefficient on borrowing constraints is large in absolute
value in older households. This implies that without labor income, borrowing constraints
become much more restrictive in limiting the risk-taking behavior for retired households.
Table 11. Robustness check using residual horizons
Full sample Subsample-savers
VariablesOLS Tobit OLS Tobit
Coeff. t-stat Coeff. p-value Coeff. t-stat Coeff. p-value
CST �0.2604 �10.31 �0.9474 0.00 �0.2579 �5.93 �0.8242 0.00Log(INC) �0.0039 �2.07 �0.0234 0.00 �0.0131 �4.20 �0.0254 0.00Log(NTW) 0.0170 6.30 0.0378 0.00 0.0195 4.11 0.0348 0.00AGE 0.0006 2.63 0.0000 0.98 0.0005 1.52 0.0006 0.26EDU 0.0182 12.17 0.0326 0.00 0.0232 8.26 0.0383 0.00RSK_AVS �0.0691 �18.33 �0.1034 0.00 �0.0712 �10.92 �0.0999 0.00BRW_CTR �0.0085 �1.06 �0.0267 0.05 �0.0233 �1.64 �0.0574 0.01BUS/NW �0.0580 �7.35 �0.0948 0.00 �0.1396 �9.33 �0.2180 0.00RE/NW 0.0006 2.00 0.0017 0.00 �0.0001 �0.04 �0.0080 0.13PEN/NW �0.0005 �0.33 �0.0002 0.93 �0.0006 �0.08 �0.0398 0.03SPD_HRZ �0.0082 �1.60 �0.0171 0.04 �0.0079 �1.07 �0.0061 0.56PEC_HRZ 0.0625 8.38 0.1354 0.00 0.0692 5.13 0.1208 0.00RES_HRZ 0.0030 4.35 0.0058 0.00 0.0052 4.30 0.0088 0.00Adjusted R2 0.28 0.28
Note: This table documents the effect of the residual component of the planning horizon on household portfolio
choice. The planning horizon is decomposed into three parts: the Spending Horizon (SPD_HRZ), the Risk Perceiving
Horizon (PEC_HRZ), and the Residual Horizon (RES_HRZ). We study this effect in both the full sample and the sub-
sample of ‘‘savers.’’ We report the coefficients and their t-statistics or p-values for the OLS and Tobit regressions; the
adjusted R2 for the OLS regressions are also reported.
16Tables and details can be obtained from the authors upon request.
Copyright # 2010 John Wiley & Sons, Ltd. Review of Behavioral Finance, 2, 81–105 (2010)
DOI: 10.1002/rbf.10
102 P. He and X. Hu
The other robustness check is that we replace the dependant variable in Equation (2) with
ST/NW, or the value of stocks divided by net worth. The main results remain robust: The
coefficient on the planning horizon is positive and significant, while the proxy for the risk-
perceiving horizon has more explanatory power than the proxy for the spending horizon.
CONCLUSION
Investors have a limited capacity to make fully rational investment decisions, and they tend to
simplify a complicated information processing and decision making problem into several
manageable dimensions. The way an investor structures his decision problem can frame his
perception of risk. In this study, we find empirical evidence that the horizon matters—
households with longer planning horizons invest more in stocks. We further explore the
interaction between a household’s spending/investment horizon and its mental accounting, or
risk-perceiving horizon. We argue that a household’s reported planning horizon reflects a
combination of its investment horizon and its risk-perceiving horizon. Our empirical results
show that the household planning horizon has a significant and positive effect on household
risk-taking behavior, and this effect is mainly driven by the mental accounting component, or
the risk perceiving horizon. Therefore, in practice individuals make investment decisions
based on the horizon at which the risk is perceived rather than the horizon at which cash is
needed. Our results are consistent with the myopic loss aversion explanation proposed by
Thaler et al. (1997), and/or the cognitive bias explanation proposed by Samuelson (1994).
An interesting question that this study raises is the following: Is the difference in relative
stockholdings across planning horizons due to the over-investment in stocks by households
with longer horizons or under-investment in stocks by households with shorter horizons?
Further research is needed to identify the answer to this question.
In addition, the perception of investment risk consists of two parts: The perception of risk
dynamics across time and the perception of the non-dynamic aspect of risks. Our study
focuses mainly on the first part of a household’s risk-perceiving behavior, though we
understand individuals may blend cross sectional risks and dynamic risks together. It is up to
future research to further study these two parts separately and/or jointly.
The authors thank Gib Bassett, Michael Brennan, Robert Chirinko, and Debbie Lucas for
their comments and suggestions. Financial supports from the University of Illinois at Chicago
and Tsinghua University are gratefully acknowledged.
REFERENCES
Barberis, N. C. (2000). Investing for the long run when returns are predictable. Journal of Finance, 55,225–264.
Bodie, Z., Merton, R. C., & Samuelson, W. F. (1992). Labor supply flexibility and portfolio choice in alife cycle model. Journal of Economic Dynamics and Control, 16, 427–449.
Brandt, M. W. (1999). Estimating portfolio and consumption choice: A conditional euler equationsapproach. Journal of Finance, 54, 1609–1646.
Brennan, M. J., Schwartz, E. S., & Lagnado, R. (1997). Strategic asset allocation. Journal of EconomicDynamics and Control, 21, 1377–1403.
Campbell, J. Y., & Viceira, L. M. (1999). Consumption and portfolio decision when expected returns aretime varying. Quarterly Journal of Economics, 114, 433–495.
Copyright # 2010 John Wiley & Sons, Ltd. Review of Behavioral Finance, 2, 81–105 (2010)
DOI: 10.1002/rbf.10
Household Investment—The Horizon Effect 103
Carroll, C. D., & Samwick, A. A. (1997). The nature and magnitude of precautionary wealth. Journal ofMonetary Economics, 40, 41–72.
Fratantoni, M. C. (1998). Homeownership and investment in risky assets. Journal of Urban Economics,44, 27–42.
Gollier, C. (2002). Time diversification, liquidity constraints, and decreasing aversion to risk on wealth.Journal of Monetary Economics, 49, 1439–1459.
Guiso, L., Jappelli, T., & Terlizzese, D. (1996). Income risk, borrowing constraints, and portfolio choice.American Economic Review, 86, 158–172.
Heaton, J., & Lucas, D. (1997). Market frictions, saving behavior and portfolio choice.MacroeconomicDynamics, 1, 76–101.
Heaton, J., & Lucas, D. (2000). Portfolio choice and asset prices: The importance of entrepreneurial risk.Journal of Finance, 55, 1163–1198.
Jagannathan, R., & Kocherlakota, N. R. (1996). Why should older people invest less in stocks thanyounger people? Federal Reserve Bank of Minneapolis Quarterly Review, 20, 11–23.
Kim, R. S., & Omberg, E. (1996). Dynamic nonmyopic portfolio behavior. Review of Financial Studies,9, 141–161.
Kimball, M. (1990). Precautionary savings in the small and in the large. Econometrica, 46, 1429–1446.Morgan, J. N., & Sonquist, J. A. (1963). Problems in the analysis of survey data, and a proposal. Journalof the American Statistical Association, 58, 415–434.
Moskowitz, T. J., & Vissing-Jørgensen, A. (2002). The returns to entrepreneurial investment: A privateequity premium puzzle? American Economic Review, 92, 745–778.
Papke, L. E. (1998). How are participants investing their accounts in participant-directed individualaccount pension plans? American Economic Review, 88, 212–216.
Pratt, J. W., & Zeckhauser, R. J. (1987). Proper risk aversion. Econometrica, 55, 143–154.Samuelson, P. (1963). Risk and uncertainty: A fallacy of large numbers. Scientia, 57, 1–6.Samuelson, P. (1969). Lifetime portfolio selection by dynamic stochastic programming. Review ofEconomics and Statistics, 51, 247–257.
Samuelson, P. (1994). Lifetime portfolio selection by dynamic stochastic programming. Journal ofPortfolio Management, 21, 15–24.
Thaler, R. H., Tversky, A., Kahneman, D., & Schwartz, A. (1997). The effect of myopia and lossaversion on risk taking: An experimental test. Quarterly Journal of Economics, 112, 647–661.
Thorley, S. R. (1995). The time-diversification controversy. Financial Analysts Journal, 51, 67–75.Tversky, A., & Kahneman, D. (1992). Advances in prospect theory: Cumulative representation ofuncertainty. Journal of Risk and Uncertainty, 5, 297–323.
Viceira, L. M. (2001). Optimal portfolio choice for long horizon investors with non-tradable laborincome. Journal of Finance, 56, 433–470.
LIST OF FORESEEABLE EXPENDITURES
EXP1: Education of children
EXP2: Education of others (including self and spouse/partner)
EXP3: Health care for self and spouse/partner
EXP4: Health Care for others (including elderly parents/disabled child)
EXP5: Health care/medical expenses—not available for whom
EXP6: General support for self/spouse/partner in retirement or old age
EXP7: General support for child/grandchild
EXP8: General support for parents
EXP9: General support for others or not available for whom
EXP10: Baby
EXP11: Purchase of new home (including vacation home)
EXP12: Purchase of car or other large durable goods
Copyright # 2010 John Wiley & Sons, Ltd. Review of Behavioral Finance, 2, 81–105 (2010)
DOI: 10.1002/rbf.10
104 P. He and X. Hu
EXP13: Burial expenses
EXP14: Taxes
EXP15: Home repairs or improvements
EXP16: Weddings, vacations, moving, and other special expenditures
EXP17: Business and investment; start/expand own business
EXP18: Investment, major purchase
EXP19: Charitable expense
EXP20: Bills and living expenses
EXP21: Other major financial obligations
Copyright # 2010 John Wiley & Sons, Ltd. Review of Behavioral Finance, 2, 81–105 (2010)
DOI: 10.1002/rbf.10
Household Investment—The Horizon Effect 105