JOURNAL OF FINANCIAL AND QUANTITATIVE ANALYSIS
COPYRIGHT 2006, SCHOOL OF BUSINESS ADMINISTRATION, UNIVERSITY OF WASHINGTON, SEATTLE, WA 98195
VOL. 41, NO. 4, DECEMBER 2006
Do Behavioral Biases Vary across Individuals?
Evidence from Individual Level 401(k) Data
Julie R. Agnew∗
This paper investigates whether some individuals are prone to behavioral biases in their
401(k) investments. Using demographic data and allocation information for over 73,000
employees, I examine two allocation biases and a participation bias. The findings suggest
that higher salaried employees tend to make significantly better choices. Participants who
earn $100,000 hold 12.7% less in company stock, are 3% less likely to follow the framing
1/n heuristic, and are37.7% more likely to participate than those earning $46,000. Women
make better choices in two of the three cases and I find evidence of mental accounting.
There is a growing literature that suggests an individual’s investment deci-
sions are affected by behavioral biases. Researchers explain financial decisions
based on behavioral theories such as excessive extrapolation, loyalty, and famil-
iarity (Benartzi (2001), Cohen (2004), and Huberman (2001)). Data from 401(k)
plans provide a fertile ground for examining these behavioral biases because par-
ticipants, who represent a diverse population of working individuals, are faced
with the same choice environment. At the 401(k) plan level, ample evidence ex-
ists that behavioral biases can be overcome or made worse by 401(k) plan design
∗Agnew, firstname.lastname@example.org, The College of William and Mary, Mason School of Busi-
ness Administration, PO Box 8795, Williamsburg, VA 23187. I thank an anonymous large benefits
provider for providing the 401(k) plan data and Stefan Bokor for his immense help in organizing the
data. I thank Pierluigi Balduzzi, Hendrik Bessembinder (the editor), John Boschen, Alicia Munnell,
Eric Jacquier, Peter Gottschalk, Shlomo Benartzi, and Jeff Pontiff (associate editor and referee) for
their careful comments and insight. In addition, Jim Poterba, Josh Rauh, Henry Richardson and Steve
Utkus were very helpful. I thank conference participants from the Retirement Research Consortium
Fourth Annual Conference, the 2002 Financial Management Association Conference, and the Frank
Batten Young Scholars Conference. In addition, I thank participants in seminars held at Boston Col-
lege, Drexel University, the Federal Reserve Board of Governors, George Washington University,
Georgia Tech, The College of William and Mary, the University of Georgia, and the University of
Richmond. I gratefully acknowledge financial support from a dissertation fellowship from the Center
for Retirement Research. Any errors are my own. The research reported herein was performed pur-
suant to a grant from the U.S. Social Security Administration (SSA) funded as part of the Retirement
Research Consortium. The opinions and conclusions are solely mine and should not be construed
as representing the opinions or policy of the SSA or any agency of the Federal Government. This
paper was previously circulated under the title “Inefficient Choices in 401(k) Plans: Evidence from
Individual Level Data.”
940Journal of Financial and Quantitative Analysis
(Benartzi (2001), Choi, Laibson, Madrian, and Metrick (2001)). Markedly less
attention has been paid to these biases at the individual level. The outstanding
questions are: does the propensity to follow behavioral biases vary across indi-
vidualsand are there commoncharacteristicsthat matter? Previousliterature does
not adequately address these questions, and this gap provides the motivation for
This paper takes a deeper look into three behavioral biases by providingnew
estimates of their severity at the individual level and by examining whether cer-
taintypesofindividualsarepronetothesebiases. I examinetwoallocationbiases,
following na¨ ıve diversification strategies and investing in company stock, and a
participationbias, optingnot toparticipatein the companysponsored401(k)plan.
Using a new dataset representing over 73,000 eligible participants, this is the first
paper to jointly model these three biases as functions of individual characteristics
and to providethe opportunityto assess the relative influence of these characteris-
tics on these choices. This paperimproveson and complementspreviousanalyses
of the two allocation biases mainly because the majority of past studies use plan
level data that can lead to aggregation bias in the results and the studies do not
address individual level heterogeneity in decisions.1This study takes advantage
of individual level allocation data. These data overcome the aggregate data prob-
lems and provide the opportunity to more strongly test new behavioral theories,
such as loyalty and its influence on company stock investment.
The principal finding suggests that higher salaried employees tend to make
significantly better choices in all three cases. Specifically, I find participants who
earn $100,000 hold 12.7% less in company stock, are 3% less likely to follow
the 1/n heuristic, and are 37.7% more likely to participate than those earning the
average wage of $46,000. Women also appear to make better choices in two of
the three cases, viz., 401(k) participation and investment in company stock. This
suggests that behavioral biases do vary across individuals and highlights that plan
level analysis will suffer from an omitted variable bias. The paper also finds more
direct evidence of mental accounting related to company stock holdings. Until
now, this theory has only been studied at the plan level.
These findings suggest that empirical research should control for individual
level heterogeneity and that more research into why salary and gender matter is
needed. Onapracticallevel,theseresults canhelpplansponsorsidentifyhighrisk
individuals with a view toward improving plan design. The results may also be
helpful in the current Social Security debate over personal accounts by providing
demographic insights into investor behavior. Finally, these results have relevancy
beyond the 401(k) literature and the broader implications are discussed in the
The paper begins with a consideration of the allocation topics because com-
pared to the participation choice, there has been relatively little research devoted
1Huberman and Jiang (2006) use individual level data in their analysis of how the number of
investment funds offered affects equity investment. They analyze how many people follow the 1/n
heuristic, but do not focus their discussion on the influence of individual characteristics. The study
also does not investigate how the mental accounting of company stock affected the 1/n strategy. Choi,
Laibson, Madrian, and Metrick’s (2004) company stock allocation study uses individual level demo-
graphic data with past company stock returns. They focus their discussion on whether participants
practice “feedback” investing and not on the role of individual characteristics on this decision.
to the allocation biases at the individual level. The first topic studied is the na¨ ıve
diversification bias. Na¨ ıve diversification strategies can result when individuals
are faced with complicated decisions that cause them to fall back on simple rules
of thumb. This paper investigates one na¨ ıve strategy called the “framing 1/n
heuristic.” This strategy is considered irrational because investors divide their
contributions evenly among the number (n) of investment options offered. They
do so regardless of the menu of investment options presented and thus are influ-
enced by the fund choices available. My paper finds that salary and employment
tenure are negatively related to this practice. I address a similar heuristic that is
considered rational, “the conditional 1/n heuristic,” and find highly compensated
individuals are 7.4% more likely to follow this rule.
Turning to the second allocation bias, company stock investment, evidence
pany stock. In fact, Hewitt Associates LLC (2003) reports from a sample of 1.5
million 401(k) plan participants in 2003 that the average company stock balance
was 41%, and the NASD recently issued a company stock warning to investors
This paper examines the links between individual characteristics, past com-
pany stock returns, and company stock allocations. The influence of past com-
pany stock performance and plan design on company stock holding is already
well documented in the literature (Benartzi (2001), Choi, Laibson, Madrian, and
Metrick (2004), Liang and Weisbenner (2002), and Sengmuller (2002)). This
study contributes to the literature by analyzing the additional link between in-
dividual characteristics and company stock holdings while controlling for past
performance. The individual level data I use offer more demographic detail than
in past research, and the returns are more precisely calculated than those used
in plan level studies. I find a one standard deviation increase in short-term re-
turns increases company stock holdings by 8% and that short-term returns matter
more than long-term returns. In addition, I find that company stock allocations
are greater for males, decrease with salary, and are higher for employees in non-
Finally, since lack of participation stands out as one of the most obvious in-
vestment mistakes an individual can make, I include a brief study of participation
choice for completeness. My findings are consistent with previous studies, and I
find that the probabilityof participatingincreases with age, job tenure, and salary.
The paper is organized as follows. Section II summarizes the dataset. Sec-
tion III describes the plan design and the asset allocation choices. Section IV
summarizes the demographic and employment characteristics of all the partici-
pants eligible to participate in the plan. Section V and VI present the empirical
results associated with na¨ ıve diversification and company stock holdings, respec-
tively. Section VII discusses the participation level in this plan, and Section VIII
presents the Conclusion.
This paper uses a detailed database supplied by an anonymous large bene-
fits provider. The cross-sectional data are from one large 401(k) plan with over
942Journal of Financial and Quantitative Analysis
73,000 eligible employees.2The plan is sponsored by a global consumer product
company and the entire sample is used to investigate the participation decision.
Thestudyoftheallocationbiases is basedonasmallersubsampleof“active”
participants. This paper defines an active participant as a plan eligible participant
who made a contribution to the plan during the first two weeks of August 1998.
The dataset includes each active participant’s contributionallocation and for most
of these individuals the actual date that this allocation was chosen. For this allo-
cation date, company stock returns over various prior periods are calculated. A
total of 28,793 participants are considered active participants.
Of these active participants, the allocation date is missing for 5,814 indi-
viduals who were enrolled in the plan prior to 1992 and did not change their
contribution allocations after this date. The data are missing because the current
ing the design of the plan prior to this date. Thus, it is possible that a different
set of fund choices may have been available prior to 1992 or that an employer
match may have been offered. Since features like these can influence choice, this
group might behave differently than participants who make allocation decisions
after 1992 (Benartzi (2001), Benartzi and Thaler (2001)).
The missing allocation decision date means that estimating company stock
returns prior to the allocation decision is not possible, nor is it possible to esti-
mate the individuals’ ages and the number of years employed at the time of the
allocation decision. Therefore, these individuals are eliminated from the sample
leaving 22,979 active participants for the analysis.3
One of the most importantfeatures of the data is detailed demographicinfor-
mation. For each eligible participant, the individual’s participation status, salary,
birth date, date of employment, compensation status, and gender are available.
Finally, three additional features of these data deserve mention. First, the
asset allocations of the contributions are broken down at the individual level. Ag-
gregatecontributionplan data can blur the results if high contributingparticipants
invest differently from low contributing participants. The effect of large contri-
bution levels is analogous to the influence of large market capitalization stocks
on a value-weighted index. Furthermore, aggregation can exaggerate very weak
relations at the individual level. This is called aggregation bias. Huberman and
Jiang (2006) use a simulation to demonstrate how aggregation bias can amplify
individual level findings in 401(k) plans. Second, these data are from one plan.
While multiple plan data are appropriate for studying across-plan variation, they
can create a potential for omitted variable bias related to plan design or plan ed-
ucational efforts. Analyzing one plan eliminates this concern. A final advantage
of the data is that allocations are based on contributions not asset balances. Asset
performance can move asset allocations based on asset balances away from the
participant’s intended allocation. Contribution allocations do not suffer from this
A disadvantage of this dataset is that information regarding participants’ as-
sets outside of the plan is not available. Another drawback of the dataset is that
2An eligible employee is an employee who may participate in the 401(k) plan if he chooses.
3Separate analyses of the allocation biases including and controlling for this subgroup were com-
pleted. These analyses are not reported here, but the findings are qualitatively the same.
it is missing some variables that have been shown to impact asset allocation de-
cisions such as marital status, education, and financial literacy (for example, see
Agnew, Balduzzi, and Sunden (2003), Sunden and Surette, (1998), and Dwyer,
Gilkeson and List (2002)).
III. Plan Design and Asset Choices
In this plan, each participant may allocate his retirement fund contributions
among four different investment vehicles: an equity income fund, an S&P 500 in-
dex fund, a guaranteed income contract fund (GIC), and company stock.4Partici-
pants have the option to change their contributionallocations daily. The company
offers no financial incentive for investing in company stock nor does it offer an
employer match. The absence of an employer match is an advantage because it
eliminates any confoundingeffects caused by the match design.
IV. Demographic and Employment Characteristics of
Panel A of Table 1 describes the demographic and employment characteris-
tics of the eligible participants. These participants may or may not contribute to
the plan. Age and time employed are measured as of August 1998, while salary
is the 1997 annual salary. In contrast, age and time employed are measured as
of the allocation decision date in the later nonparametric and regression analyses.
Individualsin this data sample are predominatelymale (78%)with an averageage
of 37 years. It is noteworthy that the participants have relatively long average job
tenures (eight years), which may indicate strong company loyalty. The median
time employed is almost five years and is approximatelyone year greater than the
1996 national median of nearly four years (CPS (1997)).
Participants in the company work in one of four divisions. A majority of the
participants (99%) work in two large consumer product manufacturing divisions,
Division 1 and Division 2. The Corporate Division employs 1% of the 401(k)
participants and 140 employees work for the Other Division.
Participants earned mean 1997 salaries of approximately $37,700. Table 2
compares the plan’s median salary by age group to the median salary of the U.S.
population, and shows that participants in this plan earn more than the general
population. However, the relation between salary and age is similar between the
Table 1, Panel B presents the same statistics as Panel A for the active partici-
pants subsample. The ratio of males to females and the distribution of employees
in each division is the same as the whole sample. However, the mean age and
4A GIC fund, sometimes called a stable value fund, is a common offering in 401(k) plans. The
fund invests in GICs that are lending contracts between insurance companies and 401(k) plans. The
401(k) plan lends the insurance company money over a fixed period of time. The insurance company
then can invest in the money in securities. In return, the insurance company pays interest to the 401(k)
plan on the loan. The insurance company guarantees the contracted interest payments and assumes
all market, credit, and reinvestment risk. The insurance company profits by generating profits greater
than the guaranteed interest it pays out to the plan. Typically, investors interested in preserving capital
and earning a steady income invest in GICs. These investments are considered low risk.
944Journal of Financial and Quantitative Analysis
Descriptive Plan Statistics
Panels A and B of Table 1 present general statistics for the eligible participants and the active only participants, respec-
tively. Panel A reports the contributionstatus (as of August 1998), gender, age in years (as of August 1998), time employed
in years (as of August 1998), division of employment, and 1997 annual salary for the sample of eligible participants to
the plan. An individual is considered eligible if he has the option to participate in the company plan. Participants are
considered contributing (active) if they contributed to the plan during the first two weeks of August 1998. Panel B presents
the same statistics for the active only subsample. This panel also includes statistics related to compensation status, which
equals yes if the individual is considered by law a highly compensated individual. Percentages may not add up to 100%
due to rounding.
Obs.% MeanStd. Min. Max.
Panel A. All Eligible Participants (active and inactive)
Participants not contributing
Total eligible participants
Age (as of 8/98)
Years Employed (as of 8/98)
1997 Annual Salary
Panel B. All Active Participants
Age (as of 8/98)
Years Employed (as of 8/98)
Highly Compensated Individual
1997 Annual Salary
Age/Salary Structure for U.S. Population and 401(k) Sample
Table 2 presents a comparison between the median salary by age group for the U.S. population at large and the 401(k)
plan participants included in the eligible participants sample and the active only subsample. The source for the U.S.
population is CPS (1997).
Median 1997 Salary
Age RangeU.S. Population 401(k) Eligible Participants 401(k) Plan Active Participants
Under 35 years
time employed of the participantsin the subsample are slightly higherthan the to-
tal sample. The median time employed is nearly two years longer than the whole
sample (6.45 years versus 4.77 years), suggesting that many new employees have
not yet joined the plan.5Another interesting difference between the two samples
is that the mean salary is close to $8,000 larger in the active sample, suggesting
that individuals with higher salaries tend to participate in the plan. This is sup-
ported in Table 2 where the median salaries for each age group in the subsample
are consistently higher than the total sample and the U.S. population. Table 1,
Panel B also presents statistics for the number of participants considered to be
highly compensatedindividuals,a legal designation based on salary and company
ownership. This status affects how much participants can contribute, but does not
restrict their allocation decisions.6In this plan, approximately 8% of the sample
is considered highly compensated.
Table 3 describes the demographic characteristics by division for the total
sample and the active subsample. The main difference between the four divisions
in both samples appears to be salary distributions. Employees of the two smallest
divisions make significantly higher salaries than employees in the other divisions.
The Corporate Division’s mean salary is approximately$97,000forthe total sam-
ple, while the Other Division’s mean salary is approximately$129,000in the total
sample. These salaries compare to approximately $40,000 and $35,000 earned in
Divisions 1 and 2, respectively. Employees in the two small divisions also earn
significantly more in the 10th and 90th percentiles of their sample. Except for
the Corporate Division, the divisions are predominately male. The groups do not
differ significantly in terms of average age or time employed. Consistent with the
earlier results, the active subsample salaries are higher than the salaries reported
in the total sample.
V. Na¨ ıve Diversification
Inthis section, threediversificationheuristicsarestudied. Thefirst, the fram-
ing 1/n heuristic, is considered a na¨ ıve strategy because individuals distribute
their contributions equally among the n choices available. As a result, allocation
decisions are influenced by available fund choices and can be considered irra-
tional. Benartzi and Thaler (2001) show that this strategy can lead to large ex
ante welfare losses when the portfolio chosen does not correspond to an individ-
5Although the database codes all the participants as immediately eligible, it is possible that there
might have been a duration of time before eligible employees could join the plan. This would result
in older ages and longer tenures for the active sample.
6Each year employers must identify employees who are considered to be highly compen-
sated. This information is used by the IRS to determine whether the 401(k) plan meets non-
discrimination tests, which are designed to insure that tax breaks derived from participating in 401(k)
plans are not limited to wealthy employees. According to the IRS website, http://www.irs.gov
/publications/p560/ch01.html, a highly compensated employee either owned more than 5% of the em-
ployer’s capital or profits at any time during the year or the preceding year, or for the preceding year
received compensation above a specified level. In 2003, the salary limit for the preceding year was
$90,000. The IRS also indicates that the employer may choose to consider those employees ranked in
the top 20% by compensation as highly compensated.
946Journal of Financial and Quantitative Analysis
Descriptive Plan Statistics by Division
Table 3 breaks down each division by demographic information for the total sample in Panel A and the active only sample
in Panel B. Panel A reports gender, 1997 annual salary, age (as of August 1998), time employed (as of August 1998), and
time enrolled in the plan (as of August 1998). In addition to those statistics, Panel B reports the percent of the sample that
is 100% invested in company stock and compensation status (HCE). HCE stands for highly compensated individual.
Panel A. All Eligible Participants (active and inactive)
Division MedianMean 10th Percentile90th Percentile
Panel B. All Active Participants
% of Division
HCEDivisionMedian Mean10th Percentile90th Percentile
ual’s risk preferences.7I also analyze the modified 1/n heuristic where individ-
uals treat company stock as a separate asset class. In the modified version of the
framing heuristic, individuals choose their company stock allocation, then divide
their remaining funds among the remaining options available. The final heuristic,
the conditional 1/n heuristic, refers to the practice of dividing allocations evenly
among the chosen funds. The number of chosen funds may be smaller than the
number of funds offered. Huberman and Jiang (2006)argue that, unlike the fram-
ing 1/n heuristic, the conditional 1/n heuristic can be rational and is consistent
with k-fund separation theories.
In this study, analyses of the heuristics are complicated by the fact that com-
panystockis anoptioninthis plan. HubermanandJiang(2006)choosetoexclude
companystockallocationsin theircalculations. As mentionedinthe Introduction,
any investment in company stock is considered inefficient in my study. Thus, an
investor who follows the conditional 1/n heuristic and includes company stock
7To illustrate, suppose that a 401(k) offers 10 investment choices that include nine equity funds
and one money market fund. An individual following the 1/n heuristic would allocate 10% of his
contributions to each fund resulting in a 90% allocation to equities. It is clear that this allocation
would not be optimal for everyone, and especially for a participant nearing retirement.
in his investment choices is not making a rational decision. Therefore, I refine
Huberman and Jiang’s (2006) definition of the conditional 1/n heuristic to only
include individuals who divide their contributions evenly (±1%) among the n
funds they choose and do not invest in company stock. In contrast, an individual
is considered to be following the framing 1/n heuristic if he puts 25% (±1%) of
his contribution into each of the four funds. These two groups are mutually ex-
clusive. Participants who follow the modified 1/n rule invest in all four funds and
divide their non-companystock options equally.
I find a small percentage of participants in the overall plan following the
framing 1/n heuristic, which is consistent with Huberman and Jiang (2006). In
fact, less than 4% follow the framing 1/n heuristic and only 5% follow the mod-
ified 1/n heuristic. On the other hand, I find nearly 8% follow the conditional
1/n rule (excluding all company stock holders and one-fund holders). I find that
most participants (35%) allocate their entire contribution to only one fund and
that a majority (66%) of those participants invest their entire contribution in com-
pany stock. If I broaden my definition of the conditional 1/n heuristic to include
one-fund investors not invested in company stock, then the percent of my sample
following this rule increases to 20%.
B.The Modified 1/n Heuristic
Using aggregate 401(k) plan data, Benartzi and Thaler (2001) find that indi-
viduals treat company stock as an asset class separate from other 401(k) invest-
ments. As a result, some participants appear to follow a modified version of the
1/n heuristic by making their company stock allocation separate from their in-
vestment in other equities. Benartzi and Thaler find that participants then split
their non-company stock investment evenly among the non-company stock op-
tions, which is a form of mental accounting (Thaler (1999)). In this paper, the
behavior is referred to as the modified 1/n heuristic.
My analysis provides stronger tests of this practice. First, the tests in this
paper are based on contribution allocations rather than asset balance allocations
and, therefore, the influence of fund performance on allocations is not a concern.
Second, the individual level data allows for the calculation of the allocation of
non-company stock holdings by individual rather than by plan, permitting the
examination of the distribution of company stock holdings across individuals and
avoiding aggregation bias (Huberman and Jiang (2006)).
The analysis begins with an examinationof the mean and medianallocations
to eachfundin Table 4. The first two columnsof Table 4 list the meanand median
allocations to each fund and the last two columns list the modified mean and
median allocations to each fund. The modified allocations are simply the percent
allocated to the particular non-company stock investment vehicle divided by the
total invested in non-company stock investment vehicles. The first subsample
includes all participants who invest in all four funds and comprises roughly 13%
of the sample. Notice that the results tend to support the modified 1/n heuristic
with modified allocations close to one-third, which equates to evenly splitting
non-company stock contributions among non-company stock assets. The same
exercise is repeated for subsamples of investors that hold three funds including
948Journal of Financial and Quantitative Analysis
company stock. The results again tend to support Benartzi and Thaler’s (2001)
assertion that some individuals treat company stock as a separate asset class and
as a result slightly modify how they follow the 1/n rule.
Asset Allocations and Modified Asset Allocations
Table 4 presents the allocations and modified asset allocations for investors who hold company stock and invest in either
two or three additional assets. The modified allocations reflect the percentage of the non-company stock holdings the
asset class represents. Percentages may not add up to 100% due to rounding.
Investment Vehicle Mean MedianMean Modified Median Modified
Invest in All Assets (3,011 obs.)
Equity Income Fund
S&P 500 Index Fund
Invest in Company Stock, Equity Income, and GIC Fund (279 obs.)
Equity Income Fund
S&P 500 Index Fund
Invest in Company Stock, Equity Income, and S&P 500 Index Fund (4,084 obs.)
Equity Income Fund
S&P 500 Index Fund
Invest in Company Stock, S&P 500 Index, and GIC Fund (560 obs.)
Equity Income Fund
S&P 500 Index Fund
Histograms provide additional detail. To illustrate, Figure 1 displays his-
tograms using the sample of participants who invest in all four funds, and shows
the frequency of company stock holdings and modified and unmodified holdings
of the three non-companystock investment vehicles.
allocation graphs (on the left) and the modified allocation graphs (on the right)
is striking. The unmodified histograms for the non-company stock funds have
several probable allocations. However, the modified frequenciesare strongly cen-
tered at 33%, suggesting that after adjusting for company stock holdings, these
individuals allocate their remaining assets evenly among the other funds in ac-
cordance with Benartzi and Thaler’s (2001) assertion that some individuals treat
company stock as a separate asset class and slightly modify how they follow the
1/n rule. Results for the other subsamples presented in Table 4 are similar and
available from the author.
C. Econometric Analysis
The final empirical question related to na¨ ıve diversification is which type
of person is most likely to follow the different heuristics? Since two strategies
are considered potentially irrational and the conditional 1/n heuristic is consid-
ered potentially rational, I would not expect the same type of individual to follow
each. To test this, four dummy variables are constructed: a framing 1/n heuris-
tic dummy, a modified 1/n heuristic dummy, a conditional 1/n heuristic dummy
These histograms display the frequency of participants’ allocations (in decimals) to each fund. The reported allocations
(unmodified) and the allocations adjusted for company stock holdings (modified) are presented. The modified allocations
represent the relative amount allocated to the non-company stock funds.
Sample Invests in All Four Funds (3,011 observations)
Unmodified Allocations Allocations Modified for Company Stock
Modified Equity Income
Modified S&P 500
excluding one-fund investors, and a conditional 1/n heuristic dummy including
one-fund investors not invested in company stock. These dummies equal one if
the individual follows the particular heuristic and zero if not.
Table 5 displays the results of a probit analysis using the 1/n dummy vari-
ables. For each 1/n variable, two regressions are run each with a different com-
pensation variable. The marginal effects of salary and employment tenure are
significant and negative for the two framing regressions. In contrast, salary and
time employed are positive for both the conditional 1/n heuristic and its broader
definition. This suggests that high salary individuals and participants with longer
job tenures are less likely to follow the potentially irrational framing 1/n rule,
while on the other hand these same individuals are more likely to follow the po-
tentially rational conditional 1/n rule. To highlight this finding, in the framing
and modified 1/n regressions an average participant who earns $100,000 would
be 3% less likely to follow the rule than an average participant earning $46,000.
Similarly, a highly compensated individual is 2% less likely to follow the rule.
Conversely, a highly compensated individual is 7.4% more likely to the follow
the broaddefinition of the conditional1/n rule. One explanationfor this behavior
is that the higher salaried individuals are more educated and therefore less likely
to rely on simple rules for investing. Regarding job tenure, it is possible that
950 Journal of Financial and Quantitative Analysis
employees’ understanding of their plan’s investment options increases with their
time on the job. Thus, this better understanding decreases the likelihood that they
will need to fall back on the 1/n rule.
Marginal Effects from Probit Regression (1/n heuristic)
Table 5 presents the marginal effects calculated from the results of a probit regression. The dependent variable equals
one if the individual follows the specific heuristic defined in the table. Male is a dummy variable equal to one if the
participant is male, zero otherwise. Salary is the annual 1997 salary (unit: $10,000). Age is the age of the participant
at the time the allocation decision is made (unit: years). The marginal effect of age takes into account a nonlinear effect
of age. Time Employed equals the time the participant has been employed at the time the allocation decision is made
(unit: years). Compensation Status is a dummy variable that equals one if the individual by law is considered highly
compensated, otherwise it equals zero. Division # and Other are dummy variables that equal one if the participant is in
that division. The Corporate Division is the omitted dummy. Robust standard errors, reported in parentheses, are adjusted
for heteroskedasticity. The pseudo R2is the log-likelihood value on a scale from zero to one, where zero corresponds to
the constant only model and one corresponds to perfect prediction (a log-likelihood of zero). **,* indicate significance at
the 1% and 5% levels, respectively.
Φ(β0+ β1Male + β2Age + β3Age2+ β4Time Employed
+ β5Salary (or Compensation) + β6Division 1 + β7Division 2 + β8Other)
The model : Prob(Y= 1)
Possibly Inefficient Decision Possibly Efficient Decision
1/n among the n
and company stock
investment is not
zero if a one-
1/n among the n
and company stock
investment is not
one if a one-
in all four
funds and non-
if . . .
25% is invested
in each of
Variables (1)(2)(1) (2)(1) (2)(1) (2)
No. of Obs.
adF/dx is for a discrete change of the dummy variable from zero to one.
VI.Company Stock Allocations
A.Cost of Holding Company Stock
I now focus on the issue of the cost of holding company stock. The well-
publicized stories of employees losing their nest eggs after Enron and WorldCom
collapsed provide strong anecdotal evidence that investing in company stock is
costly. But is it possible to actually quantify the costs of holding company stock
for average investors? Recent studies attempt to do this and agree that the cost
of holding stock is large. For example, Meulbroek (2002) calculates the cost of
holdingcompanystock as the percentof the stock market’s value that is sacrificed
by not being fully diversified. In the case where 10% of the pension is invested
in company stock, she calculates a cost of 25% for AMEX firms.8In addition,
Poterba (2003) quantifies what a log-utility investor would be willing to forego
in the value of a portfolio that is entirely invested in the S&P 500 compared to
investing in portfolios with various percentages of company stock investments.
He finds that his investorwould be indifferentbetweenforegoing57% of the S&P
500 portfolio value for a portfolio invested solely in company stock. Poterba’s
findings motivate the analysis that follows.
Consistent with anecdotal evidence, participants in the company’s 401(k)
plan show a definite tendency to invest in company stock. The overall mean allo-
cation to companystock holdingsin this plan is quite high (45%) comparedto the
10% legal maximumdefined benefit plans may hold. The large average allocation
might be partially explained by the above normal price performance of the plan’s
company stock. In this study, the company stock had an annualized stock price
return just over 20% over the 10-year period ending on December 31, 1997 com-
pared to a S&P 500’s annual return of 14.7% over the same time period. Benartzi
(2001) shows that firms with relatively high long-run returns have higher com-
pany stock allocations than poorly performing firms. I will also control for past
company stock returns in the regression analysis presented later in this section.
The general patterns of company stock allocations also deserve mention. An
interesting feature of the data is that despite the absence of restrictions on the par-
ticipants’ allocations, 75% of the allocations are clustered within one percentage
point of 0%, 25%, 50%, 75%, and 100%. Furthermore, there is a clear tendency
for many of the participants (48%) to invest either all or none of their contribu-
tions in company stock.
plan in terms of total dollars rather than percents. To calculate a rough estimate
of this number, several simplifying assumptions are made. First, each individ-
ual’s 1998 contribution amount is assumed to be equal to the number of dollars
he contributed in 1997. The 1997 contribution amount is supplied in the dataset.
This probably is a conservative estimate of 1998 contributions. However, the
alternative to annualizing the August 1998 contributions is problematic because
determining the frequency of the contributions (weekly or biweekly) is difficult.
Finally, it is assumed that the allocation percentages that the individuals chose for
their August 1998 contributions were held constant throughout 1998. Given the
documented inertia in 401(k) plans, this should be a reasonable assumption for
8This assumes that the pension assets represent 75% of the individual’s wealth and that the in-
dividual has a 15-year holding period. This figure equals 14% for NYSE firms and 36% for NASD
952Journal of Financial and Quantitative Analysis
most participants. By multiplying the estimated 1998 total dollar contributions
by the 1998 company stock percent allocation, an estimated company stock allo-
cation in dollars is calculated for each individual. Summing the estimated dollar
allocations across individuals provides an estimate of the total dollars contributed
to company stock. From this calculation, it is estimated that approximately $24.7
million was allocated to company stock. This is 39% of the estimated $63.4 mil-
lion contributed to the plan in total. This percentage is very close to the 42% that
Benartzi (2001) reports in his study of aggregate data from 103 401(k) plans.
C. Nonparametric Analysis
This section presents a nonparametric analysis of the data that will com-
plement the regression analysis to follow. Table 6 reports the company stock
allocations based on demographiccharacteristics. The non-normaldistribution of
the company stock holdings makes standard summary statistics, such as means
and standard deviations, less meaningful descriptors of the data. Therefore, in
addition to these statistics, Table 6 reports the proportion of each demographic
categorythat invests in six differentinvestmentranges: 0%, 1%–25%,26%–50%,
51%–75%, 76%–99%, and 100%. A simple test of proportions within each de-
mographic category and investment range is used to test whether a statistically
significant difference exists. If demographic characteristics do not matter, then a
statistically significant difference in proportions should not be found. For exam-
ple, under the null hypothesis gender does not matter. Therefore, the proportion
of women investing 100% of their contributions to company stock should not be
statistically different than the proportion of men investing 100% of their contri-
bution to company stock. The bold row in each category is considered the base
category and is used in each test of proportions. Table 6 reports the results of the
test of proportions.
Thefirst demographiccategorytestedis gender. Empiricalevidencesuggests
that gender may proxy for financial education or risk tolerance. For example, re-
search shows that when a measure of financial education is not available, gender
may serve as an effective proxy for it. Dwyer, Gilkeson, and List (2002) find that
womentypicallyhaveless financial knowledgethan menandthat educationaldis-
parities can substantially explain the gender differences they find in risky mutual
eral understandingof the risks associated with companystock investmentandthat
education may explain much of the variation in financial aptitude. A recent John
Hancock Financial Services’ survey (1999) highlights how individuals misread
the risks of the market. In the survey, respondents on average think that a diver-
sified stock fund is more risky than an investment in company stock. Similarly,
Benartzi (2001) reports that 83% of respondents to a Morningstar survey believe
that the overall stock market is riskier than company stock. When this sample is
limited to individuals with a high school education or less, this number increases
to 93%. Thus, while company stock investment may seem to be irrational, closer
examination reveals it may be rational given an individual’s financial knowledge.
Summary Statistics of Company Stock Holdings
Table 6 reports summary statistics for company stock holdings based on demographic characteristics. The sample is the
active only sample. In addition to the mean and median allocations, the table presents the proportion of each demographic
category invested in each of the six investment ranges. The first row of each demographic category (bold) is considered
the base category. Within each investment range and demographic category, a test of proportions is run. **,* beside
the proportions denote a statistically significant difference from the base category at the 1% and 5% levels, respectively.
Percentages may not add up to 100% due to rounding.
AllocationPercent of Sample within Each Investment Range
Category Obs. 0%1%–25% 26%–50% 51%–75% 76%–99%100% MedianMean
Sort by Gender
Highly Compensated Individual
Under 35 years
22,979 24.7% 15.3%27.7% 6.9% 2.1%23.3% 40.0%44.9%
Lack of financial knowledge may also lead to misperceptions of how 401(k)
information is used. For example, those lacking financial knowledgemay suspect
that their managers are monitoring their company stock holdings. They may fear
that a low investment in company stock signals to their employers that they “lack
commitment,” which will, in turn, harm their job prospects. As a result, their
large investment in company stock may be a rational decision based on misinfor-
If gender is a proxy for differences in financial knowledge, then men might
be expected to invest less in company stock than women. On the other hand,
empirical research finds that men are more likely to invest in riskier assets or
trade more in riskier assets than women leading to the opposite conclusion (for
9I thank Hendrik Bessembinder for this insight.
954Journal of Financial and Quantitative Analysis
example, Agnew, Balduzzi, and Sunden (2003), Barber and Odean (2001), and
Sunden and Surette (1998)).10
Thetests ofproportionssupportthelatter. Inall but twopercentranges,there
is a statistically significant (albeit economically small) difference in the propor-
tion of men investing in each investment range than women. The most significant
differencebetween the proportionof women and men investing in companystock
is at the 100%investmentrange. Observethat 24%of the menallocatetheir entire
contribution to company stock compared to 22% of the women and that this dif-
ference is significant at the 1% level. Although the medians are equal, the mean
allocation to company stock by men is 45% compared to 44% for women.
Interestingly, the gender differences obtained here are weaker than those
Clark, Goodfellow, Schieber, and Warwick (1999) find. In their study of sev-
eral 401(k) plans, men invest an average of 41% to company stock compared to
27% for women. These differences in findings could be a result of different plan
designs or varied long-run company stock performance across plans.
The next two sections of Table 6 demonstrate the influence of compensation
level, either salary or compensation status, on company stock investment. Com-
pensation is considered positively related to financial knowledge and suggests
the hypothesis that employees who earn relatively high salaries or are considered
highly compensated should hold less company stock.
Alternatively,compensationmay be a proxy for an employee’s opportunities
for stock-based compensation. Generally, greater opportunities exist for higher
salariedemployeesto receivestock-basedcompensationthanfortheir lowerwage
counterparts. This is the case in this company.11Research shows that highly
paid executives are concerned about diversifying their company stock holdings
but are often reluctant to sell their stock-based compensation. As a result they
find sophisticated ways to hedge their holdings. Results from Ofek and Yermack
(2000) suggest that executives diversify their company stock holdings through
the use of zero cost collars and equity swaps. Additional research shows that
executives with high stock ownership negate much of the impact from their stock
compensation by selling previously owned shares (Bettis, Bizjak, and Lemmon
(2001)). Given the demonstrated lengths to which these senior managers go to
of company stock in their 401(k) accounts.
The results support both theories. Table 6 shows a decrease from the low-
est wage category to the highest wage category in the proportion of individuals
allocating their entire contribution to company stock: 27% of the under $25,000
category invest their whole contribution to company stock compared to 15% of
10In terms of trading and turnover of equity investments, Barber and Odean (2001) find a significant
difference between men and women. They find that men trade 45% more than women. However,
using brokerage account data from 35,000 households, they find only a very small difference in equity
ownership as a percent of net worth between males and females. On average, they find that women in
their sample invest 13.3% of their net worth in equities compared to men who invest 13.2%.
11This company offers three plans. One plan is open to all full-time employees and the number of
options available is based on earnings. The second and third plans are targeted at middle and senior
management. The options in these plans are based on reaching performance goals. Thus, higher
salaried and middle and upper management employees have more opportunities to earn stock options
than lower salaried employees.
the $100,000 plus category. The reverse trend is observed in the proportion of in-
dividuals who invest nothing in company stock. Here, 22% of the under $25,000
category invest nothing in company stock compared to 42% of the over $100,000
group. This difference in proportions is significant at the 1% level, and supports
results from Goodfellow and Schieber’s (1997) study of 24 different plans where
low wage earners are more likely than high wage earners to hold company stock.
Table 6 also shows that highly compensated individuals make similar investment
Similar to gender, age may proxyfor risk tolerance. Many life cycle theories
predict that individuals will hold less risk in their financial portfolio as they age.
Jagannathan and Kocherlakota (1996) suggest that young investors have a long
stream of future income. As individuals age, this stream of future income short-
ens diminishing the value of their human capital. Therefore, they suggest that
individuals should offset this decline in the value of their human capital by reduc-
ing the risk of their financial portfolio. Bodie, Merton, and Samuelson’s (1992)
model leads to a similar prediction. In their model,individualscan respondto low
realized asset returns by increasing their supply of labor. However, labor flexibil-
ity generally declines with age. Therefore, similar to the previous model, older
individuals are expected to hold more conservative investments in their financial
Table 6 is consistent with the stated life cycle hypotheses. Note that age is
measured at the time the allocation decision is made. The 65 plus age category is
not discussed because it includes only six participants. Notice that as individuals
age there is a downward trend in the proportion of participants investing their
entire contributionto companystock. On the extremeends, 19% of those between
ages 55–64 invest their entire contribution to company stock compared to 24% of
the participants under 35 years old. The difference in proportions is significant at
the 1% level. This trend is reversed and significant in the proportions investing
nothing in company stock.
Time employed may also proxy for risk tolerance, as well as loyalty or fa-
miliarity. The latter two theories would predict a positive relation. The results in
Table 6 show that the percentage of participants that holds 0% in company stock
increases with the time employed. A less marked decline is observed in the 100%
category,but the proportionof eachgroupthat is 100%investedin companystock
is still relatively less than the group of employees with zero to two years of work
experience. Twenty-six percent of those with less than two years of experience
invest their entire contribution to company stock compared to 22% of those with
greater than 26 years of experience.
These findings are not consistent with loyalty and familiarity, but they are
consistent with the prediction of the Degeorge, Jenter, Moel, and Tufano (2004)
model. They use job tenure as a proxy for the firm specificity of an individual’s
human capital. These authors argue that an individual’s firm-specific human cap-
ital grows with the time he is employed by a firm. As a result, the individual’s
need to diversify away from company stock increases with his job tenure.
The results show that the employee’s company division also explains some
variation in company stock holdings. One possible explanation is that the prob-
ability of earning stock-based compensation varies with divisions. For example,
956Journal of Financial and Quantitative Analysis
a corporate division may have more employees eligible for stock-based compen-
sation than a division mostly comprised of factory workers. Thus, the expected
average allocation to company stock would be relatively lower in the corporate
division compared to the other divisions. The occupation type may also pro-
vide information about the employee’s education level beyond that obtained from
salary information. It seems likely that a corporate division may be more heavily
comprised of executives with college degrees, while a factory division may have
a higher percentageof blue-collarworkers with high school degrees. On the other
hand, the division variables may also proxy for many other unobservables so care
must be taken not to over interpret these results.
In this study, the predominant occupation does differ among divisions. A
discussion with the benefits administrator indicates that the Corporate Division
consists mainly of executives, while the employees of Division 1 and Division
2 tend to be factory workers. As predicted, Table 6 shows the Corporate Divi-
sion has the lowest proportion of individuals investing their entire contribution
in company stock and the highest proportion of individuals who invest nothing
in company stock. These results support the theory that either the executives in
the Corporate Division are limiting their company stock holdings to compensate
for stock-based compensation or they are doing so because they have a relatively
better understanding of the inherent risks of company stock investment. Another
possible theory is that the factory workers are more loyal.
D. Econometric Analysis of Company Stock Holdings
The nonparametric evidence suggests that there are relations between the
demographic variables and company stock holdings. This section will economet-
rically test for the joint effects of these factors on companystock allocations, and,
in addition, it will control for the effects of past company stock performance.
Table 7 presents a two-limit censored regression model that tests the effects
of the individual characteristics on company stock allocations.12Note, however,
that the prevalence of company stock allocations clustered at 0%, 25%, 50%,
75%, and 100% makes it possible that the errors from the two-limit censored
regression are not normally distributed. If this is true, the usual estimators based
on the log-likelihood for this regression model are inconsistent (Greene (1997)).
Therefore, as a robustness check, an ordered probit regression is also estimated
50%, 51%–75%, 76%–99%, and 100%). The ordered probit results support the
findings from the two-limit censored regression model and are available from the
Two models are estimated, each with a differentvariablemeasuringcompen-
sation, and the results are reported in Table 7. Model 1 uses salary and Model 2
uses an indicator variable that equals one if the individual is considered a highly
compensated individual and zero if not. The results suggest that men invest 3.0%
more of their contributions to company stock than women, supporting the theory
that men tend to make more risky asset allocation choices. Salary is also signif-
12Agnew, Balduzzi, and Sunden (2003) use this model to study the relation between demographic
characteristics and equity allocations in one 401(k) plan.
Two-Limit Censored Regression: Company Stock Allocations
Table 7 presents the results from a two-limit censored regression of company stock allocations (in decimals) against
participant characteristics. Male is a dummy variable equal to one if the participant is male, zero otherwise. Salary is
the annual 1997 salary (unit: $10,000). Age is the age of the participant at the time the allocation decision is made
(unit: years). Time Employed equals the time the participant has been employed at the time the allocation decision is
made (unit: years). Compensation Status is a dummy variable that equals one if the individual by law is considered highly
compensated, otherwise it equals zero. Division # and Other Division are dummy variables that equal one if the participant
is in the division. The Corporate Division is the omitted dummy. One-Year Co. Stock Return is the one year raw buy and
hold return earned prior to the allocation decision. Robust standard errors, reported in parentheses, are adjusted for
heteroskedasticity. **,* indicate significance at the 1% and 5% levels, respectively.
Independent Variables(1) (2)
One-Year Co. Stock Return
No. of Observations
icantly related to company stock holdings. The results suggest that for every ad-
ditional $10,000 in compensation company stock holdings fall by approximately
2%. In this case, salary may be a proxy for financial education or the amount
of stock-based compensation. The division of employment also has a significant
participants in Division 1 invest 11% (17%) more to company stock. Similarly,
participants in Division 2 invest 20% (26%) more to company stock. The results
supportthe hypothesisthat eithertheexecutivesinthe CorporateDivisionare lim-
iting their company stock holdings to compensate for stock-based compensation
or they are limiting their company stock holdings because they have a relatively
better understanding of the inherent risks in company stock investment. Inter-
estingly, age and time employed are not significantly related to company stock
958Journal of Financial and Quantitative Analysis
holdings in Model 1. Time employed has a very small negativeinfluence on com-
pany stock holdings in Model 2.
Finally, the results support Benartzi’s (2001) findings that past raw buy and
hold returns are positively related to company stock holdings. He finds 10-year
returnshavethe most significantinfluence,whereasI find short-termreturns(one-
year buy and hold returns) produce the most significant results (based on pseudo
R2s). I test returns overdifferentperiodsrangingfrom one to 10 years. My results
are supportedby Sengmuller(2002)who finds two- to three-yearreturns are most
closely related to company stock holdings. Most likely my ability to control for
inertia by more precisely calculating the buy and hold returns for each individual
causes the differencesin results. Sengmuller (2002)also controls for inertia in his
In this analysis, the sample returns range from a minimum of negative 19%
to a maximum of positive 78%. The average return is 28% with a standard de-
viation of 20%. The results of the regression predict that a one standard devi-
ation increase in company stock returns will increase company stock holdings
by 8.0%.13Interestingly, these results are contrary to what an optimal individual
holdingstock options would do in response to strong stock performance. An opti-
mizing individualwould reducecompanystock holdingsbecause the hedgeratios
on awarded options increase with stock returns.14
VII. Plan Participation
Finally, the analysis turns to plan participation. Choosing not to participate
in a 401(k) plan is the most obvious error an individual can make and is well
researched in the literature. The literature shows a clear link between plan level
and individual level characteristics. Munnell, Sunden, and Taylor (2001/2002)
provide a summary of the findings and the behavioral explanations behind the
results. Therefore,this section focuses primarily on how consistent my results are
with previous findings and how they compare with the other results in this paper.
In this plan, of the 73,699eligible participants in the total sample 39% made
at least one contribution during the first two weeks of August 1998. This par-
ticipation rate is low compared to other studies.15One reason might be that the
definition of active participant in this study is fairly restrictive because it lim-
its participants to those who made a contribution during the first two weeks of
August 1998. Other studies use different definitions. For example, Clark and
Schieber (1998) define an active participant as a person who makes at least one
contribution in a single year.
At the individual level, the previous literature shows that salary, age, and
time employed are related to participation. Low wage earners may be less likely
13These results are very close to Sengmuller’s (2002) finding. He finds that after controlling for the
effects of inertia, a positive one standard deviation change in one-year returns (22%) will result in an
increase in company stock inflows of five percentage points by those considered “active changers.”
14I thank the referee for making this point.
15For example, Clark and Schieber (1998) find that on average 73.5% of eligible employees partic-
ipated in their 401(k) plans in their analysis of plan data from 19 firms with 700 to 10,000 employees.
Similarly, Munnell, Sunden, and Taylor (2002) report a 72% mean participation rate among eligible
employees using the 1998 Survey of Consumer Finances data.
to participate than high wage earners due to the greater liquidity constraints they
bear, the reduced tax breaks they earn due to their lower tax brackets, and the
higher replacement rates they earn from Social Security. The positive relation
with age may be because people grow more interested in their retirement savings
as time goes by. Finally, the positive relation between participation and time
employed may be a result of individuals’ vesting schedules increasing over time
and their growing familiarity with the plan. For further discussion of why these
characteristics might matter, see Munnell, Sunden, and Taylor (2001/2002).
To test whether a common factor exists that relates to the efficiency of all
three choices and to examine whether participants in this study’s plan behave
similarly to participants in other plans, a probit regression of the participant’s
decision is modeled. Table 8 presents the findings. The findings are consistent
with previous work and, most importantly, one characteristic, salary, is related
to the efficiency of all three decisions (Tables 5, 7, and 8). In the participation
regression, salary plays a large role. For example, compared to an average partic-
ipant earning $46,000, an average participant earning $100,000 would be 37.7%
more likely to participate. This same individualwould be expected to hold 12.7%
less in company stock and would be 3.0% less likely to follow the framing 1/n
Marginal Effects from a Probit Regression: 401(k) Participation Decision
Table 8 presents the marginal effects calculated from the results of a probit regression. The dependent variable equals
one if the participant is an active participant in the plan and zero if not. Male is a dummy variable equal to one if the
participant is male, zero otherwise. Salary is the annual 1997 salary (unit: $10,000). Age is the age of the participant as of
August 1998 (unit: years). The marginal effect takes into account a nonlinear effect of age. Time Employed equals the time
the participant has been employed as of August 1998 (unit: years). Division # and Other are dummy variables that equal
one if the participant is in the division. The Corporate Division is the omitted dummy. Robust standard errors, reported in
parentheses, are adjusted for heteroskedasticity. The pseudo R2is the log-likelihood value on a scale from zero to one,
where zero corresponds to the constant only model and one corresponds to perfect prediction (a log-likelihood of zero).
**,* indicate significance at the 1% and 5% levels, respectively.
The model : Prob(Y = 1)
=Φ(β0+ β1Male + β2Age + β3Age2+ β4Time Employed + β5Salary
+ β6Division 1 + β7Division 2 + β8Other)
No. of Observations
adF/dx is for a discrete change of the dummy variable from zero to one.
960Journal of Financial and Quantitative Analysis
In addition, gender plays a statistically significant role in two of the biases—
participation and company stock allocations. In both cases, women make better
choices, with women being 4% more likely to participate and expected to hold
2% or 3% less in company stock than men.
This paperexamines the influence of individualcharacteristics on behavioral
biases in 401(k) plan allocation decisions. With over 73,000 eligible participants,
the database pertains to a single plan that consists of a diverse set of individuals
who are presentedwith similar investmentchoices. The investmentbiases studied
are the na¨ ıve 1/n heuristic and its variations, investment in company stock, and
the decision not to participate in the plan.
The goal of this paper is to determine whether the propensities to follow
biases vary across individualsand whether a commoncharacteristic can be found.
The 401(k) literature has already documented the importance of plan design, and
this papercontributestothe literaturebyhighlightingtheimportanceofindividual
The principal finding suggests that higher salaried employees tend to make
significantly better choices in all three cases. Women also appear to make bet-
ter choices in two of the three cases, viz., 401(k) participation and investment in
companystock. These findingshighlightthe needto controlfor individualhetero-
geneityin empiricalwork,andtheyindicate somedirectionsforfutureresearchas
well. For example, althoughseveral theories are presented in this paper to explain
why gender and salary matter, additional work is needed to study these theories
in detail. In particular, data on individual level financial literacy would be useful
to understand the salary results. On a practical level, these results can help plan
sponsors identify high risk individualswith a view toward improvingplan design.
Theresults may also be helpfulin the currentSocial Securitydebateoverpersonal
accounts by providing demographic insights into investor behavior.
Finally, this research has implications beyond the 401(k) literature. For ex-
ample, studies suggest that the empirical failures of the C-CAPM are a result
of including non-market participants in the sample studied (see Brav, Constan-
tinides, and Geczy (2002), Mankiw and Zeldes (1991), and Vissing-Jorgensen
(2002)). This is because the C-CAPM, just like the I-CAPM, assumes that in-
dividuals participate in the markets. The theories also assume that individuals
hold well-diversified portfolios. My study suggests that low salaried market par-
ticipants are less likely to hold well-diversified portfolios because they tend to
concentrate their assets in one security, namely, company stock. In addition, they
are more likely to follow na¨ ıve diversification strategies, which generally will not
result in a well-diversified portfolio. As a result, my findings suggest that a sam-
ple of higher income market participants is more likely to meet the diversification
conditions established by the underlyingC-CAPM and, thus, should perform bet-
ter in an empirical analysis of the model.
Agnew, J.; P. Balduzzi; and A. Sunden. “Portfolio Choice and Trading in a Large 401(k) Plan.”
American Economic Review, 93 (2003), 193–215.
Barber, B. M., and T. Odean. “Boys Will Be Boys: Gender, Overconfidence, and Common Stock
Investment.” Quarterly Journal of Economics, 116 (2001), 261–292.
Benartzi, S. “Excessive Extrapolation and the Allocation of 401(k) Accounts to Company Stock.”
Journal of Finance, 56 (2001), 1747–1764.
Benartzi, S., and R. Thaler. “Naive Diversification Strategies in Retirement Saving Plans.” American
Economic Review, 91 (2001), 79–98.
Bettis, J. C.; J. Bizjak; and M. Lemmon. “Managerial Ownership, Incentive Contracting, and the
Use of Zero-Cost Collars and Equity Swaps by Corporate Insiders.” Journal of Financial and
Quantitative Analysis, 36 (2001), 345–370.
Bodie, Z.; R. C. Merton; and W. F. Samuelson. “Labor Supply Flexibility and Portfolio Choice in a
Life Cycle Model.” Journal of Economic Dynamics and Control, 16 (1992), 427–449.
Brav, A.; G. M. Constantinides; and C. C. Geczy. “Asset Pricing with Heterogeneous Consumers and
Limited Participation: Empirical Evidence.” Journal of Political Economy, 110 (2002), 793–824.
Choi, J. J.; D. Laibson; B. Madrian; and A. Metrick. “Defined Contribution Pensions: Plan Rules,
Participant Decisions and the Path of Least Resistance.” NBER Working Paper 8655 (2001).
Choi, J. J.; D. Laibson,; B. Madrian; and A. Metrick. “Employee Investment Decisions about Com-
pany Stock.” In Pension Design and Structure: New Lessons From Behavioral Finance, O. S.
Mitchell and S. Utkus, eds. New York, NY: Oxford University Press Inc. (2004).
Clark, R. L.; G. P. Goodfellow; S. J. Schieber; and D. Warwick. “Making the Most of 401(k) Plans:
Who’s Choosing What?” In Forecasting Retirement Needs and Retirement Wealth, O. Mitchell,
P. B. Hammond, and A. M. Rappaport, eds. Philadelphia, PA: University of Pennsylvania Press
Clark, R. L., and S. Schieber. “Factors Affecting Participation Levels in 401(k) Plans.” In Living with
Defined Contribution Plans, O. S. Mitchell and S. J. Schieber, eds. Philadelphia, PA: University of
Pennsylvania Press (1998).
Cohen, L. “Loyalty Based Portfolio Choice.” Working Paper, University of Chicago (2004).
CPS. “Employee Tenure in the Mid-1990s.” CPS Publications, January 30, 1997.
Degeorge, F.; D. Jenter; A. Moel; and P. Tufano. “Selling Company Shares to Reluctant Employees:
France Telecom’s Experience.” Journal of Financial Economics, 71 (2004), 169–202.
Dwyer, P.; J. Gilkeson; and J. List. “Gender Differences in Revealed Risk Taking: Evidence from
Mutual Fund Investors.” Economic Letters, 76 (2002), 151–159.
Goodfellow, G. P., and S. J. Schieber. “Investment of Assets in Self-Directed Retirement Plans.”
In Positioning Pensions for the Twenty-First Century, M. S. Gordon, O. S. Mitchell, and M. M.
Twinney, eds. Philadelphia, PA: University of Pennsylvania Press (1997).
Greene, W. H. Econometric Analysis. Upper Saddle River, NJ: Prentice-Hall, Inc. (1997).
Hewitt Associates LLC. 2003 Hewitt Universe Benchmarks—How Well Are Employees Saving and
Investing in 401(k) Plans? (2003).
Huberman, G. “Familiarity Breeds Investment.” Review of Financial Studies, 14 (2001), 659–680.
Huberman, G., and W. Jiang. “Offering vs. Choice in 401(k) Plans: Equity Exposure and Number of
Funds.” Journal of Finance, 61 (2006), 763–801.
Jagannathan, R., and N. R. Kocherlakota. “Why Should Older People Invest Less in Stocks Than
Younger People.” Quarterly Review Federal Reserve of Minneapolis, Summer (1996), 11–23.
John Hancock Financial Services. The Sixth Defined Contribution Plan Survey (1999).
Liang, N., and S. Weisbenner. “Investor Behavior and Purchase of Company Stock in 401(k) Plans—
The Importance of Plan Design.” Working Paper, Board of Governors of the Federal Reserve
Mankiw, G. N., and S. P. Zeldes. “The Consumption of Stockholders and Nonstockholders.” Journal
of Financial Economics, 29 (1991), 97–112.
Meulbroek, L. “Company Stock in Pension Plans: How Costly Is It?” Harvard Business School
Working Paper 02-058 (2002).
Munnell, A. H.; A. Sunden; and C. Taylor. “What Determines 401(k) Participation and Contribu-
tions?” Social Security Bulletin, 64 (2001/2002), 64–75.
NASD Press Release. “NASD Warns Investors Too Much Company Stock Can Jeopardize Financial
Future: Diversification is Key to Reducing Risk,” February 17, 2005.
Ofek, E., and D. Yermack. “Taking Stock: Equity-Based Compensation and the Evolution of Man-
agerial Ownership.” Journal of Finance, 55 (2000), 1367–1384.
Poterba, J. M. “Lessons from Enron: Employer Stock and 401(k) Plans.” American Economic Review,
93 (2003), 398–404.
962Journal of Financial and Quantitative Analysis
Sengmuller, P. “Performance Predicts Asset Allocation: Company Stock in 401(k) Plans.” Working
Paper, Columbia University (2002).
Sunden, A. E., and B. J. Surette. “Gender Differences in the Allocation of Assets in Retirement
Savings Plans.” American Economic Review, 88 (1998), 207–211.
Thaler, R. H. “Mental Accounting Matters.” Journal of Behavioral Decision Making, 12 (1999),
Vissing-Jorgensen, A. “Limited Asset Market Participation and the Elasticity of Intertemporal Substi-
tution.” Journal of Political Economy, 110 (2002), 825–853.