Contents lists available at ScienceDirect

Journal of Financial Economics

Journal of Financial Economics 102 (2011) 1–27

0304-40

doi:10.1

$ We

Associa

Garrick

Karolyi,

(AFA se

Tobacm

dation c

of Bosto

Associa

Francisc

Ivkovic

in our

financia

All rem

this pap

High-Exn Corr

E-m

journal homepage: www.elsevier.com/locate/jfec

Behavioral biases of mutual fund investors$

Warren Bailey a,n, Alok Kumar b, David Ng c,d

a Cornell University, Johnson Graduate School of Management, USAb University of Miami, School of Business Administration, USAc Cornell University, USAd University of Pennsylvania, Wharton School, USA

a r t i c l e i n f o

Article history:

Received 19 October 2009

Received in revised form

6 July 2010

Accepted 9 July 2010Available online 27 May 2011

JEL classification:

G11

D03

D14

Keywords:

Individual investors

Mutual funds

Trend chasing

Behavioral biases

Factor analysis

5X/$ - see front matter & 2011 Elsevier B.V.

016/j.jfineco.2011.05.002

thank an anonymous referee, Malcolm Baker

tion discussant), Nick Barberis, Robert Battal

Blalock, Charles Chang, Susan Christoffersen,

George Korniotis, Lisa Kramer, Charles Lee,

ssion chair), J. Spencer Martin, Jay Ritter,

an, Jeff Wurgler, and seminar participants at

onference (Frankfurt), Cornell University, Fe

n, Ohio State’s Alumni Summer conference

tion meetings, McGill University, and 2009

o) for comments and helpful discussions. W

and Lu Zheng for providing data for identifyin

sample. We are grateful to the BSI Gamm

l support. Taehoon Lim provided excellent r

aining errors and omissions are our own. Ear

er were entitled ‘‘Why Do Individual Investo

pense Funds Instead of Index Funds?’’

esponding author.

ail address: Wbb1@cornell.edu (W. Bailey).

a b s t r a c t

We examine the effect of behavioral biases on the mutual fund choices of a large sample

of US discount brokerage investors using new measures of attention to news, tax

awareness, and fund-level familiarity bias, in addition to behavioral and demographic

characteristics of earlier studies. Behaviorally biased investors typically make poor

decisions about fund style and expenses, trading frequency, and timing, resulting in

poor performance. Furthermore, trend chasing appears related to behavioral biases,

rather than to rationally inferring managerial skill from past performance. Factor

analysis suggests that biased investors often conform to stereotypes that can be

characterized as Gambler, Smart, Overconfident, Narrow Framer, and Mature.

& 2011 Elsevier B.V. All rights reserved.

All rights reserved.

(American Finance

io, Zahi Ben-David,

Josh Coval, Andrew

Ulrike Malmendier

Rene Stulz, Jeremy

BSI Gamma Foun-

deral Reserve Bank

, Northern Finance

AFA meetings (San

e also thank Zoran

g the mutual funds

a Foundation for

esearch assistance.

ly presentations of

rs Hold Stocks and

1. Introduction

Previous studies of behavioral biases in the investmentdecisions of individual investors focus on the selection ofindividual stocks. Odean (1998, 1999), Barber and Odean(2001), and other empirical studies show that the stock-picking decisions of individual investors exhibit a varietyof behavioral biases. However, little work has been doneto link the decision-making biases of individuals to theirmutual fund investments. Understanding the role ofbehavioral biases in individual mutual fund decisions isimportant for several reasons.

First, individual investors increasingly use mutualfunds to invest in the equity market instead of tradingindividual stocks. French (2008, p. 1539) reports: ‘‘Indivi-duals hold 47.9% of the market in 1980 and only 21.5% in2007. This decline is matched by an increase in theholdings of open-end mutual funds, from 4.6% in 1980to 32.4% in 2007.’’ Hence, it is increasingly important to

W. Bailey et al. / Journal of Financial Economics 102 (2011) 1–272

understand how individual investors hold and trademutual funds.

Second, even though direct stock trading by indivi-duals has declined, their mutual fund investment deci-sions can affect stock returns indirectly. Coval andStafford (2007) argue that large flows force some mutualfunds to trade heavily, causing price pressure for secu-rities held across many funds. Previous papers show thatmutual fund flows affect individual stock returns. Gruber(1996) and Zheng (1999) find that fund flows are followedby positive short-term fund returns, perhaps due to amomentum effect. Frazzini and Lamont (2008) show thatmutual fund flows appear to be ‘‘dumb money’’: Fundinflows are associated with low future returns, whileoutflows are associated with high future returns.

Third, the manner in which individuals employ mutualfunds cuts right to the heart of basic principles of financialmanagement. Traditional portfolio choice models imply asimple investment strategy based on well-diversified, lowexpense mutual funds and minimal portfolio rebalancing.Index funds, and other equity funds with low fees and lowturnover, are cheap, convenient vehicles for individualinvestors to implement such a strategy. The extent towhich individuals adhere to these principles in their useof mutual funds is an important measure of the rationalityand effectiveness with which investors approach capitalmarkets.

The purpose of our paper is to test whether behavioralbiases explain why the use of mutual funds varies sub-stantially across individual investors and often departsfrom the simple strategies suggested by classic theories.The growing literature on behavioral finance has uncov-ered a variety of decision-making biases in how investorsuse individual common stocks. These behavioral forcesshould also have an impact on whether a particularinvestor uses mutual funds and whether she uses themeffectively.

The mutual fund literature has already revealed twospecific anomalies. First, individual investors buy fundswith high fees. Gruber (1996) and Barber, Odean andZheng (2005) show that many individual investors holdsignificant positions in high expense mutual funds. Evenmore puzzling is the finding of Elton, Gruber and Busse(2004) that substantial amounts have gone into index fundscharging high fees (over 2% per year) for passive holdings ofbroad indexes such as the Standard & Poor’s (S&P) 500.Second, individual investors chase returns. Sirri and Tufano(1998), Bergstresser and Poterba (2002), and Sapp andTiwari (2004) find that fund flows tend to chase fundswith high past returns. This could be fostered by Morning-star’s practice of rating funds based on past returns(Del Guercio and Tkac, 2008).

Several explanations have been offered for these twoanomalies. Carlin (2008) explains participation in high feeindex funds using a model with search costs. Choi,Laibson and Madrian (2010) interpret their experimentson Wharton School master of business administrationstudents and participation in high fee funds as consistentwith behavioral biases. Return-chasing has been ascribedto an agency problem that induces fund managers to alterthe riskiness of the fund to maximize investment flows

instead of risk-adjusted expected returns (Chevalier andEllison, 1997). It could also reflect inferring managerialskill from past returns (Sirri and Tufano, 1998; Gruber,1996; Berk and Green, 2004). However, with the excep-tion of the experimental data used by Choi, Laibson andMadrian (2010), these authors study aggregate fundflows, not individual investor behavior.

In contrast to previous studies, we link the decision-making biases of particular individual investors to theirindividual history of mutual fund investing using adatabase of tens of thousands of brokerage records of USindividual investors. The key to our experiment is theuse of individual investor records of stock holdings andtrading to estimate the behavioral bias proxies that pre-vious authors have used to explain how investors tradeindividual stocks. These individual behavioral bias proxiesare, in turn, related to the mutual fund holdings and tradingof those individuals in a variety of empirical specificationsthat reveal different facets of mutual fund investorbehavior.

We can easily imagine behavioral biases affectingmutual fund selection. For example, the disposition effect(selling winners too quickly and holding losers too long)could lead some investors to overestimate expectedholding periods and mistakenly select high front-end loadfunds. Investors with narrow framing bias (buying andselling individual assets without considering total portfo-lio effects), overconfidence (frequent trading plus poorperformance), or a preference for speculative stocks couldselect funds that facilitate aggressive switching acrossasset classes without considering higher fees. Local bias(preference for stocks of companies geographically closeto home) could induce the selection of locally managedmutual funds without regard to cost or expected perfor-mance. Investors who view their portfolios in terms oflayers that serve different purposes (Shefrin and Statman,2000) could demonstrate different behavior in their use ofindividual stocks versus mutual funds. For example, ifmutual funds are viewed as substantially safer thanselecting individual stocks on their own, investors couldlet their guard down and spend less time assessing fundperformance and costs. Regardless of the type of beha-vioral bias, poor decisions about timing, holding periods,and choice of funds can combine with the substantialvariety in mutual fund fee structures to yield poorperformance.

To examine the interactions and consequences ofmutual fund choices and behavioral biases, we adopttwo empirical viewpoints. First, we present tests acrossindividual investors. Estimates of several dimensions ofbehavioral bias for each individual in our sample are usedto explain individual investor choices across index funds,other types of mutual funds, and individual stocks. Wealso test whether behavioral biases influence associationsbetween trading decisions and recent fund performancebecause those biases could cause some investors tomisuse performance information.

Second, we present tests across different types offunds. We summarize individual investor holding periodsand returns across mutual funds classified by fee struc-ture and by the extent of several behavioral biases of each

W. Bailey et al. / Journal of Financial Economics 102 (2011) 1–27 3

fund’s investors. Behaviorally biased investors could clus-ter in particular types of funds and demonstrate poorperformance or very frequent trading. Furthermore, thefund industry’s offerings could include some fundsdesigned to attract and perhaps even exploit such inves-tors. A large and growing number of mutual funds offer avariety of themes and fee structures to US individualinvestors. Even across relatively generic index funds,many competing products offer a wide range of feestructures and resultant performance (Elton, Gruber andBusse (2004); Hortacsu and Syverson, 2004). It is plausi-ble that different types of funds attract different clienteles(Nanda, Wang, and Zheng, 2009), and some funds couldhave been designed specifically with behaviorally biasedclienteles in mind.1

A handful of previous papers have examined specificdimensions of the mutual fund choices of individualinvestors. Barber, Odean and Zheng (2005) find thatinvestors are more sensitive to salient fees such asfront-end loads, but not as sensitive to hidden manage-ment fees. Christoffersen, Evans and Musto (2006)consider how fund managers respond to the preferencesof their investors. Malloy and Zhu (2004) show thatinvestors who reside in less affluent and less educatedneighborhoods tend to select high expense funds. Zhu(2005) shows that busy investors are more likely to investin funds instead of individual stocks. Huang, Wei and Yan(2007) characterize the effect of the information environ-ment on the associations between fund flows and pastperformance. Bergstresser, Chalmers and Tufano (2009)study whether mutual fund brokers help educate inves-tors and attenuate their behavioral biases, but they con-clude that brokers do not deliver tangible benefits for thefees they earn. Ivkovic and Weisbenner (2009) examineaggregate individual investor fund flows for tax effects.

Our paper offers several substantial contributions.First, unlike earlier studies, we examine a combinationof behavioral factors, plus controls for other likely influ-ences on portfolio selection, to reveal the interactionsbetween investor decisions, the characteristics of themutual funds they select, and the consequences forportfolio performance. Second, because we employproxies for a number of dimensions of investor behaviorin our tests, we are also able to study the associationsbetween different investor characteristics. In particular,applying factor analysis to the correlation structure of ourinvestor characteristics reveals interesting overlapsamong biases and other characteristics, and it permitsus to identify and profile five investor stereotypes that welabel Gambler, Smart, Overconfident, Narrow Framer, and

1 Some evidence exists that skilled capital market participants

outsmart individual mutual fund investors. Money market funds appear

to raise fees to exploit investors who are insensitive to fees and

performance (Christoffersen and Musto, 2002). Weak associations

between equity fund fees and performance could also reflect such

behavior (Gil-Bazo and Ruiz-Verdu, 2009). Corporations are aware of

patterns in mutual fund inflows and outflows and attempt to exploit

them in timing equity issues (Frazzini and Lamont, 2008). Mutual fund

inflows are attracted to seemingly high performance assessed against

benchmarks that funds specify but which do not match fund styles

(Sensoy, 2009).

Mature. Third, our tests take the viewpoints of both theinvestor, who could ignore or misuse mutual funds, andthe mutual fund industry, which could design some of itsproducts to exploit the poor decision-making skills ofsome investors. Last, we extend the empirical behavioralliterature beyond the choice of individual stocks todecisions about professionally managed portfolios.

A summary of our results is as follows. We find thatsophisticated investors (better informed, higher income,older, and more experienced) investors make good use ofmutual funds, holding a high proportion of funds for longperiods, avoiding high expense funds, and experiencingrelatively good performance. However, investors withstrong behavioral biases or lack of attention to firm-specific or macro-economic news are less likely to holdmutual funds or select mutual funds for the wrongreasons. When they do buy mutual funds, they tradethem frequently, tend to time their buys and sells badly,and prefer high expense funds and active funds ratherthan index funds. We also find that biased investors aremore likely to chase fund performance, casting doubt onthe idea that trend chasing reflects rational fund selectiondecisions.

Evidently, these decisions are suboptimal because theyare associated with lower overall returns. For instance,top-quintile narrow-framing investors have averagemutual fund returns that are 2.16% lower than those inthe bottom quintile, and top-quintile disposition effectinvestors have average returns that are 0.89% lower thanthose in the bottom quintile. In contrast, behavioral biasesdo not appear to affect the performance of index fundholdings.

Thus, our behavioral bias and news inattentivenessproxies, though crude, demonstrate that behavioraleffects are at work in the mutual fund decisions of manyinvestors and take a toll on performance. Furthermore,the bias and inattention to news proxies are themselvescorrelated in interesting ways that allow us to identifyand study stereotypical investors. The five factors identi-fied using factor analysis can explain over 75% of thevariance of the behavioral factors and other investorcharacteristics. The intuitive combinations of investorcharacteristics that comprise these five factors relate tomutual fund trading habits and performance in an inter-esting and consistent manner.

The rest of the paper is organized as follows. Section 2describes our explanatory variables and test specifica-tions. Section 3 describes the individual investor databaseand other data sources. We present our empirical resultsin Sections 4 and 5, and we conclude in Section 6 with abrief discussion.

2. Measuring investor characteristics

Our main objective is to relate mutual fund use andperformance to behavioral factors that vary across oursample of investors. We begin by using each sampleinvestor’s record of common stock holdings and tradingto estimate a set of variables that proxy for the behaviorsevident in each investor’s common stock portfolio. Recog-nizing that behavioral factors are unlikely to be the only

(footnote continued)

larger number of stocks with capital gains and capital losses. Thus, use of

the original measure of the Disposition Effect in cross-sectional analysis is

likely to induce mechanical associations with variables that are corre-

lated with portfolio size and trading frequency. Similar issues apply to

the Narrow Framing measure because the trade clustering measure used

to proxy for narrow framing is correlated with portfolio size, number of

stocks, and trading frequency. Further, there might be a mechanically

induced relation between proxies for Narrow Framing and Disposition

Effect. To minimize the potential influences of portfolio size, number of

stocks, and trading frequency, we compute peer group-adjusted proxies

of both Disposition Effect and Narrow Framing biases. Our stock-level and

fund-level local bias measures are adjusted with the means for the

market. This does not affect estimation because the same constant is

applied to all investors, but this allows us to think about an investor’s

portfolio characteristics relative to a typical investor.3 We measure the performance and turnover from the stock hold-

W. Bailey et al. / Journal of Financial Economics 102 (2011) 1–274

determinant of mutual fund choices, we also constructcontrols for other drivers of mutual fund decisions sug-gested by the mutual fund and behavioral finance litera-tures. We use these variables in a variety of tests acrossindividual investors and then across types of mutualfunds. Detailed descriptions of behavioral factors, otherinvestor characteristics, and references to supportingpapers can be found in Appendix A.

2.1. Behavioral bias proxies

We begin by estimating Disposition Effect and Narrow

Framing, two mental accounting biases that have beenexplored extensively in the behavioral finance literature.The Disposition Effect is the propensity of an investor tosell winners too early and hold losers too long. As detailedin Appendix A, we measure each investor’s peer group-adjusted disposition effect by comparing each investor’sactual propensity to realize gains versus losses witha peer group’s propensity to realize gains and losses.A positive value of our disposition effect proxy indicatesthat the investor sells a greater proportion of winners anda relatively smaller proportion of losers.

Disposition Effect could be related to tax incentives. Forexample, selling winners but retaining losers is particu-larly costly for high-income US individuals. In contrast,realizing losses in December instead of other monthscould represent a sophisticated tax minimization strategy.To distinguish disposition effect from tax loss selling, weconstruct a disposition effect times high income interac-tion variable (DEnHigh Income) and a disposition effecttimes no December tax loss selling interaction variable(DEnNo December Tax Loss Selling). Selling winners toosoon and holding losers too long is particularly costly forhigher-income investors because they face higher mar-ginal tax rates. Similarly, a cleaner measure of dispositioneffect could be isolated by identifying individuals whoappear entirely unaware of the tax consequences of theirtrades. Therefore, both of these interaction terms areintended to isolate cleaner and severe facets of thedisposition effect.

Our second bias proxy, Narrow Framing, is the propen-sity of an investor to select investments individually,instead of considering the broad impact on her portfolio.Intuitively, the time interval between two consecutivedecisions reflects the decision frame, with temporallyseparated decisions more likely to be framed narrowlythan simultaneous decisions. Hence, investors who exe-cute less-clustered trades are more likely to be usingnarrower decision frames. The appendix describes howeach investor’s trade clustering measure is peer-groupadjusted for portfolio size, number of stocks, and tradingfrequency. A low trade clustering measure indicates aninvestor who is more likely to use a narrow viewpoint inmaking investment choices.2

2 Odean (1998) computes Disposition Effect as the proportion of

losses realized minus the proportion of gains realized and notes that this

measure is sensitive to portfolio size and trading frequency. For

example, proportions are likely to be smaller for investors who hold

larger portfolios and trade frequently because those portfolios contain a

Another important concept from the empirical beha-vioral finance literature is overconfidence, an investor’spropensity to trade frequently but unsuccessfully. OurOverconfidence Dummy variable is set to one for investorsin the highest portfolio turnover quintile and lowestperformance quintile for their individual common stocktrading.3 Because male investors typically exhibit over-confidence, we use a male dummy as an additional proxyfor overconfidence.

Next, we compute a proxy for familiarity, as articu-lated by Merton (1987) and Huberman (2001).4 Specifi-cally, the Local Bias of an investor’s common stockportfolio equals the mean distance between her homezip code and the headquarters’ zip codes of companies inher portfolio minus the mean distance to the companies’headquarters in the market portfolio. Fund Level Local Bias

equals the mean distance between the investor’s homezip code and the headquarters of the mutual funds in herportfolio, minus the same measure aggregated across allfunds held by all investors in the sample.

We measure each investor’s preference for gamblingand speculation. Following Kumar (2009), Lottery Stocks

Preference is the investor’s mean portfolio weight (relativeto the weight in the market portfolio) assigned to stocksthat have low prices, high idiosyncratic volatility, andhigh idiosyncratic skewness.

Last, we construct two indicators of whether a parti-cular investor appears to ignore potentially relevanteconomic news. One variable captures inattention toearnings news, and the other captures inattention tomacroeconomic news. Both measures are computed usingeach individual’s record of individual stock trades usingthe formula 1�(number of investor trades around theevent)/(total number of investor trades), where ‘‘around’’the event is defined as days t�1, t, and tþ1, where t is theearnings announcement date. To compute Inattention to

Earnings News, earnings announcements for each stockheld by the individual are collected from the InstitutionalBrokers’ Estimate System, I/B/E/S. To compute Inattention

ings of the investors for the entire period. We also construct an

alternative measure for performance and turnover using the first year

of investors’ record. The results are very similar.4 A related concept is home bias, the tendency for some investors to

under-diversify their portfolios internationally. See Bailey, Kumar and

Ng (2008) for evidence that home bias could have its origins in

behavioral biases.

6 Options and short sale dummies could proxy for skill and experi-

W. Bailey et al. / Journal of Financial Economics 102 (2011) 1–27 5

to Macroeconomic News, we collect dates of federal fundstarget rate (www.federalreserve.gov/releases/h15/update/)changes, Non-Farm Payroll Reports (www.bls.gov/bls/archived_sched.htm), and producer price index releases (www.bls.gov/ppi/ppirel.pdf).5

The measures we construct are only proxies for beha-vioral biases. They do not correspond exactly to thedefinitions of decision-making biases in the psychologyliterature. Nonetheless, at the very least, these measuresare indicators of suboptimal stock investment decisions.They reflect portfolio management mistakes and allow usto measure associations between an individual’s propen-sity to make such mistakes, his use of mutual funds, andthe consequences for portfolio performance.

Furthermore, there are other ways to think about thebehavioral bias proxies and our results. What we call‘‘behavioral bias proxies’’ could simply represent eachinvestor’s ‘‘financial literacy.’’ Put another way, it is costlyto continually acquire the skills and information neededto make successful investment decisions. While basicnotions of portfolio management suggest that a simplebuy-and-hold use of index funds is a sensible way toavoid incurring such costs, ‘‘bounded rationality’’ couldlead some investors to other decisions. For example, aninvestor could display narrow framing bias if he elects notto incur the cost of thinking more carefully about invest-ment decisions.

Aside from recognizing that each investor couldrationally strike a different balance between the costsand benefits of becoming a better investor, we must alsoconsider preferences. While a preference for lottery-typestocks sounds suboptimal and is associated with under-performance, it could simply represent skewness prefer-ence in the investor’s objective function.

Finally, some behavioral bias proxies could representfrictions in the investment process. For example, ouroverconfidence proxy identifies investors whose indivi-dual stock portfolio is high on turnover and low on return.While this could represent investors who are irrationallyaggressive, it could also reflect a combination of smallportfolio size, commission costs, and other frictions. Witha portfolio of only a few stocks, rebalancing by tradingjust one stock yields high turnover, and even overconfi-dence if performance is poor. If such small investorsrecognize that mutual funds are particularly advanta-geous, this could even induce a correlation betweenoverconfidence and the propensity to use mutual funds.Our inclusion of portfolio size as a control variable in ourregressions might not completely correct for such effects.

2.2. Control variables

Though we focus on the behavioral forces for whichSection 2.1 describes proxies, we also control for other factorsthat are likely to influence mutual fund choices. Specifically,we consider a set of demographic characteristics, which

5 Subsequent results shed light on whether inattention is a bias or

part of a sensible passive strategy. For example, Barber and Odean

(2008) find no evidence that trading based on other measures of news

arrival is beneficial.

includes Age, Marital Status (a dummy set to one for marriedinvestors), Family Size (number of family members in thehousehold), Professional Dummy (a dummy set to zero forthe investor in a blue collar profession, one otherwise), andRetired Dummy (a dummy set to one if the investor is retired).These factors could proxy for forces, such as the availability oftime to study investments (Zhu, 2005) that can affectportfolio selection.

Other control variables are more directly related toeach individual’s investment activities. Stock portfoliodiversification is measured as the negative of Normalized

Portfolio Variance (that is, the variance of the portfolio ofindividual domestic securities divided by the averagevariance of the individual common stocks in the portfo-lio). Investors who demonstrate awareness of the value ofdiversification in their portfolio of individual stocks arelikely to extend that insight into their choice of mutualfunds. Income (the total annual household income) andPortfolio Size (the sample-period natural log of the averagemarket capitalization of the investor’s common stockportfolio) identify investors who are more likely to under-stand the basic precepts of portfolio management and,therefore, tend to select index funds or other low expensefunds and hold them for relatively long periods. Invest-

ment Experience (years since the brokerage account wasopen) and a dummy for residence in a Financial Center

could indicate more experienced investors with easieraccess to information and opinions about investments(Christoffersen and Sarkissian, 2009). The Options Dummy

equals one if the investor executes at least one optiontrade during the sample period. The Short Sale Dummy

equals one if the investor executes at least one short tradeduring the sample period.6 Stock Portfolio Performance (theintercept from the market model time series regressionwith the monthly common stock portfolio return asdependent variable) could identify particularly skillful,successful investors. Success could originate from a vari-ety of strategies, ranging from selecting individual stocksto timing the market.7 No December Tax Loss Selling equalsone minus the ratio of realized losses in December to bothrealized and paper losses in December. Holds Tax-Deferred

Account is a dummy variable equal to one if the investorholds an Individual Retirement Account (IRA) or Keoghaccount at the brokerage. Stock Portfolio Beta, Size, Value,and Momentum Factor loadings are computed with marketor four-factor regressions using monthly returns.

3. Data and summary statistics

Having outlined the behavioral proxies and controlvariables that support our study of multiple dimensions ofinvestors’ mutual fund decisions, we now describe thedata sets needed for the empirical tests.

ence, or they could reflect a tendency to speculate. See Campbell (2006)

on the correlation between investor sophistication and investment

mistakes.7 For example, an informed investor could optimally focus on only a

few stocks (Goetzmann and Kumar, 2008; Ivkovic, Sialm, and

Weisbenner, 2008; Van Nieuwerburgh and Veldkamp, 2010).

Table 1Summary statistics on mutual fund investments of individual investors.

This table summarizes the stock and mutual fund investment activities of our sample individual investors. The individual investor data are from a large

US discount brokerage house for the 1991–1996 period. The median numbers are indicated in parentheses. We identify a total of 136 index funds that

were available to our sample of investors during this time period. The Center for Research in Securities Prices (CRSP) universe of individual stocks

available during this time period is about 12,000.

Statistic Equity funds Index funds Stocks

Number of assets 1,492 33 10,877

Sample-period trades

Number of investors with trades 32,122 5,594 62,387

Number of buys 405,376 (67.03%) 15,354 (73.66%) 1,015,735 (54.76%)

Number of sells 199,365 (32.97%) 5,491 (26.34%) 839,041 (45.24%)

Mean (median) number of trades 19 (6) 4 (2) 30 (11)

Mean buy trade quantity 2,787 470 634

Mean buy trade size $9,929 $6,879 $11,251

Mean sell trade quantity 4,226 964 694

Mean sell trade size $15,744 $13,244 $13,684

End-of-month positions

Number of investors with positions 29,381 4,432 59,387

Mean (median) portfolio size $39,986 ($12,827) $13,659 ($5,200) $35,629 ($13,869)

Mean (median) number of assets 3.51 (2) 1.37 (1) 3.89 (3)

W. Bailey et al. / Journal of Financial Economics 102 (2011) 1–276

3.1. Data sources

Our primary database is a 6-year (January 1991–November 1996) panel of trades and monthly portfoliopositions of individual investors with accounts at a majorUS discount broker.8 The database has been used by anumber of other authors including Odean (1998) andBarber and Odean (2000). The database indicates theend-of-month portfolios of all investors, records all tradesby these investors, and supplies demographic information(measured as of June 1997 and supplied to the brokeragehouse by Infobase) such as age, occupation, income, self-reported net worth, gender, marital status, and zip code.9

We obtain the zip codes of the headquarters of a subset ofmutual fund families from Professors Josh Coval andZoran Ivkovic. We supplement this data set with addi-tional information from the Lionshare database, 1996Nelson’s Directory of Investment Managers, and Googlesearches.

We also obtain data from several standard sources. Foreach common stock and mutual fund in our sample, weobtain monthly returns data from the Center for Researchin Security Prices (CRSP). We also use the CRSP mutualfund database to obtain information on fund character-istics such as the expense ratio and front-end load. Finally,we obtain the monthly time series of the three Fama-French factors and the momentum factor from ProfessorKenneth French’s data library.10

8 The brokerage firm has not made more recent data available. The

time period covered largely excludes such phenomena as exchange-traded

funds (ETFs) and high-frequency online day trading by individuals.9 Each demographic variable is available for only a subset of the

investors in the sample. For instance, both age and income are available

for only 31,260 investors. Consequently, the number of observations in

each cross-sectional regression depends upon the subset of demographic

variables included.10 The data library is available at http://mba.tuck.dartmouth.edu/

pages/faculty/ken.french/.

3.2. Summary statistics

Table 1 provides summary statistics on individual inves-tor trading and holding of mutual funds and, for compar-ison, individual stocks. Sample investors traded or held1,492 different equity mutual funds (of which 33 are indexfunds) and close to 11,000 stocks. A total of 32,122 investorshave executed at least one mutual fund trade and 29,381have held equity mutual funds at least once. Among these,only 5,594 have executed at least one index fund trade and4,432 have held index funds at least once. The balance ofbuys and sells suggests that, in contrast to individual stocks,mutual fund investors tend to buy and hold funds, ratherthan buying and selling more actively as with individualstocks. Trade sizes and quantities are typically modest.

The mean (median) number of equity funds in a typicalmutual fund portfolio is 3.51 (2.0) and number of tradesexecuted is 19 (6.0). The mean (median) number of indexfunds held is 1.37 (1.0) and number of trades executed is 4(2.0). In contrast, a typical investor holds 3.89 individualstocks (median is three) and executes 30 (median is 11) stocktrades.

Beyond what is reported in the table, the proportion ofmutual funds in a typical equity portfolio that includesmutual funds is 23.78%.11 This proportion increasesslightly with equity portfolio size to about 26% in thehighest size decile portfolios. The proportion of indexfunds in the aggregate mutual fund portfolio is low,varying between 5.30% and 8.39%, with a mean of only6.54%. Nevertheless, among the investors who hold indexfunds, the proportion of index funds in the mutual fundportfolio is about 38%. Furthermore, there is much

11 If we include all investors, not just those who hold mutual funds,

this proportion is only 13.49%. Consistent with the common industry

trend, it has grown steadily from 7.63% in January 1991 to 16.58% in

November 1996. About 10% of all investors hold only mutual funds in

their equity portfolios while about 17% hold more than three-fourths in

equity mutual funds.

ral

me

asu

res.

,54

2in

div

idu

als

wh

oh

av

etr

ad

ed

ind

ivid

ua

lst

ock

sd

uri

ng

the

sam

ple

pe

rio

d.A

ny

corr

ela

tio

nco

effi

cie

nt

wit

ht-

sta

tist

icg

rea

ter

tha

no

re

qu

al

to2

.57

6is

pre

sen

ted

inb

old

typ

e

ifica

nce

.A

llse

rie

sa

rew

inso

rize

da

tth

e1

%le

ve

l,a

nd

resu

lts

thro

ug

ho

ut

the

pa

pe

ra

rev

ery

sim

ila

rfo

rw

inso

rizi

ng

at

the

5%

lev

el.

siti

on

Na

rro

w

Fra

min

g

Ov

erco

nfi

den

ce

Du

mm

y

Ma

le

Du

mm

y

Loca

l

Bia

s

Lott

ery

Sto

cks

Pre

fere

nce

Ina

tten

tio

n

toE

arn

ing

s

New

s

Ina

tten

tio

nto

Ma

cro

eco

no

mic

New

s

Fun

d-L

evel

Loca

lB

ias

Fun

d-L

evel

Ina

tten

tio

n

DEn

No

Dec

emb

er

Sell

ing

DEn

Hig

h

Inco

me

0.2

30

0.0

13

0.0

08

0.0

06

0.0

44

0.0

38

�0

.01

1�

0.0

06

0.0

04

0.9

22

0.4

81

1.0

00

0.0

80

0.0

12

0.0

07

0.0

81

0.0

82

�0

.01

0�

0.0

11

0.0

05

0.2

00

0.1

21

0.0

80

1.0

00

0.0

19

�0

.03

90

.06

20

.01

5�

0.0

10

�0

.00

7�

0.0

12

0.0

01

0.0

06

0.0

12

0.0

19

1.0

00

0.0

10

0.0

06

�0

.00

40

.00

20

.00

90

.00

40

.00

60

.01

6

0.0

07

�0

.03

90

.01

01

.00

0�

0.0

41

0.0

11

�0

.00

8�

0.0

24

�0

.00

20

.00

70

.00

9

0.0

81

0.0

62

0.0

06

�0

.04

11

.00

0�

0.0

65

0.0

21

�0

.01

3�

0.0

02

0.0

37

0.0

31

0.0

82

0.0

15

�0

.00

40

.01

1�

0.0

65

1.0

00

�0

.06

00

.00

5�

0.0

01

0.0

28

0.0

23

11

�0

.01

0�

0.0

10

0.0

02

�0

.00

80

.02

1�

0.0

60

1.0

00

0.0

03

0.0

43

�0

.00

8�

0.0

09

06

�0

.01

1�

0.0

07

0.0

09

�0

.02

4�

0.0

13

0.0

05

0.0

03

1.0

00

�0

.01

3�

0.0

04

�0

.01

0

0.0

05

�0

.01

20

.00

4�

0.0

02�

0.0

02

�0

.00

10

.04

3�

0.0

13

1.0

00

0.0

02

0.0

12

0.2

00

0.0

01

0.0

06

0.0

07

0.0

37

0.0

28

�0

.00

8�

0.0

04

0.0

02

1.0

00

0.4

44

0.1

21

0.0

06

0.0

16

0.0

09

0.0

31

0.0

23

�0

.00

9�

0.0

10

0.0

12

0.4

44

1.0

00

W. Bailey et al. / Journal of Financial Economics 102 (2011) 1–27 7

evidence that our sample of brokerage records representstypical US individual investors.12

In addition to detailed descriptions of each investorcharacteristic variable, the appendix includes univariatesummary statistics on those variables.13 Features of thedata are noteworthy. For example, some of the behavioralbias proxies are skewed to the left (Disposition Effect,Narrow Framing) and others are skewed right with largepositive outliers (Lottery Stock Preference). The median ageof our sample investors is about 50 years, median incomeis $87,500 per year, and median family size is two. Almost90% of the accounts are held by males. The average(median) market risk-adjusted return on an investor’sportfolio of individual stocks is an unflattering �0.378%(�0.278%) per month and ranges from a minimum of�11.474% to a maximum of 6.437%. The median indivi-dual stock portfolio beta is a surprisingly high 1.157.

4. Empirical results

We begin by examining our behavioral bias and newsinattention proxies in more detail and, in particular, look forintuition from the associations among these proxies andwith other investor characteristics. Next, we study mutualfund participation and fund selection decisions across oursample investors. We then arrange information about thesedecisions by type of fund, not by individuals. In these tests,we examine the fees and expenses of funds chosen by theinvestors in our sample and whether there are associationswith turnover, performance, and behavioral biases. We alsoinvestigate whether investors’ trend-chasing behavior isinfluenced by their behavioral biases. Further tests sum-marize the impact of individual investors’ mutual fundinvestment decisions on portfolio performance. Last, wereport the results of various robustness checks.

4.1. Associations between investor characteristics

The recent behavioral finance literature has proposed anumber of behavioral factors. However, previous paperstypically focus on only one behavioral factor. One of ourcontributions is to examine different behavioral factorsjointly and to measure how they relate to each other andto other investor characteristics.

Table 2 presents correlations among the behavioralbiases that we measure. A number of statistically signifi-cant associations are evident. Disposition Effect, Narrow

Framing, Lottery Stocks Preference, and Inattention to Earn-

ings News often appear in the same individuals. Theseindividuals time their trades poorly, make decisions inisolation, buy speculative stocks, and ignore firm-specific

Ta

ble

2C

ross

-co

rre

lati

on

so

fth

eb

eh

av

io

Co

mp

uta

tio

ns

are

ba

sed

on

21

toin

dic

ate

stro

ng

sta

tist

ica

lsi

gn

Mea

sure

Dis

po

Eff

ect

Dis

po

siti

on

Eff

ect

1.0

00

Na

rro

wFr

am

ing

0.2

30

Ov

erco

nfi

den

ceD

um

my

0.0

13

Ma

leD

um

my

0.0

08

Loca

lB

ias

0.0

06

Lott

ery

Sto

cks

Pre

fere

nce

0.0

44

Ina

tten

tio

nto

Ea

rnin

gs

New

s

0.0

38

Ina

tten

tio

nto

Ma

cro

eco

no

mic

New

s

�0

.0

Fun

d-L

evel

Loca

lB

ias

�0

.0

Fun

d-L

evel

Ina

tten

tio

n0

.00

4

DEn

No

Dec

emb

erSe

llin

g0

.92

2D

En

Hig

hIn

com

e0

.48

1

12 Ivkovic et al. (2005) find the distribution of stock holding periods

is very similar across our sample and the general population reflected in

tax returns. Zhu (2005), Goetzmann and Kumar (2008), and Ivkovic,

Sialm and Weisbenner (2008) confirm that our sample closely resembles

the general US individual investor population. Bailey, Kumar and Ng

(2008) show similarities with the Census Bureau’s 1995 Survey of

Income and Program Participation and the Federal Reserve Board’s

Survey of Consumer Finances of 1992 and 1995.13 These statistics are computed prior to 1% winsorizing which is

employed throughout the balance of the paper.

W. Bailey et al. / Journal of Financial Economics 102 (2011) 1–278

information. Although uncorrelated with Disposition Effect,Overconfidence Dummy is significantly positively corre-lated with Narrow Framing, Male Dummy, and Lottery

Stocks Preference, suggesting a class of particularly aggres-sive investors prone to speculation. Some correlations forLocal Bias suggest a cautious investor type (negativecorrelation with Overconfidence Dummy and Lottery Stocks

Preference). Inattention to Macroeconomic News is nega-tively correlated with Inattention to Earnings News, sug-gesting that some individuals invest on a top down basisand look at broad news, while ignoring firm-specific news.

To save space, we do not report correlations among theother investor characteristics or between the behavioralbiases and the other characteristics (they are availableupon request). We summarize these correlations as fol-lows. Many of the other investor characteristics are relatedin sensible ways. For example, Age is positively correlatedwith Marital Status, Retired Dummy, Investment Experience,and Stock Portfolio Size. Income is positively correlatedwith Family Size, Professional Dummy, and Financial Center

Dummy. The use of options or short sales is correlated withInvestment Experience and Financial Center Dummy. Finan-cial sophistication is evident in correlations among Invest-

ment Experience, Options Dummy, Short Sale Dummy,Stock Portfolio Diversification, and tax minimization.A number of correlations are unexpected, such as noassociation between Investment Experience and Stock Port-

folio Performance and negative association between Stock

Portfolio Diversification and Stock Portfolio Performance.High loadings of individual stock portfolios on market,size, value, and momentum factors are associated withpoor performance.

The (unreported) correlations between the behavioralbias variables and the other investor characteristics beginto suggest links between investment decision-makingbiases and more fundamental individual characteristics.For example, it is sensible that maturity and intelligence(represented by Age, Income, Professional Dummy, andRetired Dummy) are typically uncorrelated or even nega-tively correlated with biases. Narrow Framing is more likelyfor young, relatively low-income investors, which is con-sistent with the findings of Kumar and Lim (2008). Lotterystock preference is associated with growth and valuestocks [as proxied by SMB (small minus big) and HML(high minus low) factor exposures] and poor performance.Among the biases, only Local Bias is positively correlatedwith Stock Portfolio Performance, suggesting that familiaritybias is not necessarily detrimental. As we would predictgiven its definition, Narrow Framing tends to be negativelycorrelated with Stock Portfolio Diversification.

While it is difficult to comprehensively grasp hundredsof individual cross correlations, some hint at effectiveinvesting, some suggest cautious behavior, and many implythat poor decision making leads to inferior stock portfolioperformance. To highlight these associations in a moreformal and dramatic manner, Table 3 presents the resultsof factor analysis applied to the observed characteristics ofthe 21,542 investors in the database who traded individualstocks during the sample period.

The first factor explains 21.8% of the variance ofthe investor characteristics. This factor has substantial

positive loadings on Disposition Effect, Narrow Framing,and, especially, Lottery Stocks Preference. This suggeststhat this factor reflects investors with substantial beha-vioral biases, particularly a taste for risky stocks. We labelthis factor Gambler. Negative loadings on Age, Income,Professional Dummy, Retired Dummy, Investment Experi-

ence, and Portfolio Size suggest that Gambler is relativelyyoung, poor, unsophisticated, and inexperienced. Thenegative loading on Stock Portfolio Diversification indicatesa tendency to plunge rather than spread risk. This isconsistent with models (Mitton and Vorkink, 2007;Barberis and Huang, 2008) in which some investors takeundiversified positions in skewed securities that appeal totheir preferences. The loadings on risk factors indicate anappetite for high beta stocks, small stocks, value stocks,and trading against momentum. The negative loading onStock Portfolio Performance suggests that Gambler typi-cally suffers poor performance. This is consistent with theempirical finding in Kumar (2009) that investors withhigh Lottery Stocks Preference often select small valuestocks that do not perform well.

The second factor explains 18.1% of the variation of theinvestor characteristics. In contrast to Gambler, this factorrepresents investors who seem to do everything right andearn good returns from individual stocks as a consequence.We label this factor Smart. Smart displays negative loadingson several behavioral biases and has high income, profes-sional status, and long investment experience. Smart’slarge, diverse individual stock portfolio has relatively mod-est loadings on market, size, value, and momentum risksand reflects the value of December tax loss selling. Amongthe first five factors, Smart is the most likely to maintain atax-deferred brokerage account. This combination of goodcharacteristics yields relatively high individual stock port-folio performance. Smart is likely to use short-selling,implying sophistication in investment tactics.

The third factor explains 15.3% of the investor character-istics and puts cumulative variance explained above 55%.We label this factor Overconfident given the large positiveloading on Overconfidence Dummy (which, by construction,is consistent with the large negative loading on Stock

Portfolio Performance). Overconfident is typically male,inclined to Lottery Stocks Preference, single, not retired, andinexperienced with investments. An association betweenmale gender and overconfident investing mirrors the find-ings of Barber and Odean (2001). Overconfident’s individualstock portfolio is poorly diversified and has a large loadingon market risk. The use of options is associated with thisineffective decision maker, unlike the use of short saleswhich is associated with the successful Smart investor.

The fourth factor explains 12.3% of the investor char-acteristics. We label it Narrow Framer given its particu-larly large loading on that bias. With significant positiveloadings on three biases, youth, and low income, poorStock Portfolio Diversification, and weak Stock Portfolio

Performance, Narrow Framer is reminiscent of the Gamblerand Overconfident stereotypes. Similar to the findings inKumar and Lim (2008), Narrow Framer exhibits strongerdisposition effect and hold less diversified portfolios.Narrow Framer does seem aware of tax issues, given thenegative loading on No December Tax Loss Selling, perhaps

Table 3Factor analysis for the behavioral measures and other investor characteristics.

Computations are based on 21,542 individuals who have traded individual stocks during the sample period. The ‘‘varimax’’ method is run for ten

factors but only the first five are reported given variance explained.

Factor

Variable Gambler Smart Overconfident Narrow Framer Mature

Factor CharacteristicsEigenvalue 2.288 1.894 1.607 1.286 1.071

Variance explained 0.218 0.181 0.153 0.123 0.102

Cumulative variance explained 0.218 0.399 0.552 0.675 0.777

Factor LoadingsDisposition Effect 0.189 �0.213 0.055 0.253 �0.302

Narrow Framing 0.216 �0.101 0.095 0.588 �0.221

Overconfidence Dummy 0.055 �0.058 0.472 0.090 �0.232

Male Dummy 0.021 0.004 0.202 �0.001 �0.013

Local Bias �0.044 �0.206 �0.02 0.005 �0.020

Lottery Stocks Preference 0.563 �0.202 0.143 �0.011 �0.243

Inattention to Earnings News �0.058 �0.011 0.090 0.196 �0.013

Inattention to Macroeconomic News 0.029 �0.007 �0.052 �0.011 �0.008

Fund-Level Local Bias �0.020 �0.033 0.000 0.032 �0.016

Fund-Level Inattention 0.005 �0.017 0.001 �0.004 �0.010

DEnNo December Tax Loss Selling 0.023 �0.028 0.028 0.015 �0.015

DEnHigh Income 0.028 �0.027 0.037 0.022 �0.019

Age �0.335 0.067 �0.026 �0.202 0.458

Income �0.404 0.020 �0.005 �0.167 �0.126

High Income �0.027 0.196 �0.010 0.004 �0.085

Marital Status �0.032 �0.033 �0.255 �0.001 0.054

Family Size 0.008 �0.04 �0.023 �0.001 �0.104

Professional Dummy �0.155 0.332 �0.002 0.000 �0.589

Retired Dummy �0.342 0.055 �0.331 0.008 0.890

Investment Experience �0.333 0.509 �0.221 �0.015 0.292

Financial Center Dummy 0.045 0.005 0.032 �0.007 �0.059

Options Dummy 0.066 0.094 0.301 0.010 �0.029

Short Sale Dummy 0.014 0.332 0.014 0.012 �0.011

Stock Portfolio Diversification. �0.323 0.723 �0.333 �0.41 0.403

Stock Portfolio Size �0.202 0.407 0.080 �0.303 0.552

Stock Portfolio Performance �0.454 0.354 �0.828 �0.236 �0.020

No December Tax Loss Selling 0.006 �0.498 0.088 �0.398 �0.311

Holds Tax-Deferred Account �0.004 0.202 0.027 �0.005 �0.311

Market Factor Exposure 0.471 0.091 0.556 0.220 �0.046

SMB Factor Exposure 0.806 0.125 0.150 �0.023 �0.044

HML Factor Exposure 0.594 0.121 0.213 �0.110 0.059

UMD Factor Exposure �0.555 0.087 �0.045 �0.072 0.010

W. Bailey et al. / Journal of Financial Economics 102 (2011) 1–27 9

because he or she carefully accounts for each stock, thoughseparately.

The fifth factor explains 10.2% of variance and, given thatit is the last factor with eigenvalue above one and putscumulative variance explained above 75%, it is the finalfactor for which we offer detailed interpretation.14 Giventhat this factor has a high loading on Age, Retired Dummy,and Investment Experience, a negative loading on behavioralbiases, a large, well-diversified portfolio, and an under-standing of tax-timing, we label it Mature. Unlike Smart,Mature’s individual stock portfolio performance is notextraordinary, but it successfully avoids the cost of obviousbiases and mistakes. Caution is also reflected in Mature’srelatively modest loadings on market, size, value, andmomentum risks. Mature is less likely to hold a tax-deferredaccount, perhaps because such accounts must be drawn

14 Given that we use factor analysis instead of principal compo-

nents, a cutoff of one for the eigenvalue is conservative. Information on

the sixth through tenth factors is unreported but available on request.

down upon approaching retirement or are less valuable torelatively low income investors. Many of the characteristicsof Mature parallel what Korniotis and Kumar (2011) reportfor older investors. To reconcile generally unbiased decisionmaking with mediocre performance, they suggest that agingis associated with deterioration in cognitive skills

We recognize that the labels we have placed on thefirst five factors are at best speculative. Nonetheless, theclusters of characteristics they identify across tens ofthousands of individual US investors are intuitive. Theyvalidate the behavioral biases and other investor char-acteristics that the empirical behavioral finance literaturehas developed. We employ these biases, and the factorswe have extracted, in subsequent tests to understand howbehavioral biases affect the use of equity mutual funds.

4.2. Participation in open end mutual funds: logit

regression estimates

Our next set of tests examines investors’ mutual fundparticipation decisions. We estimate logit regressions in

W. Bailey et al. / Journal of Financial Economics 102 (2011) 1–2710

which the dependent variable is the fund participationdummy, which equals one for an investor who invests inmutual funds at least once during the sample period. Themain independent variables of interest are the behavioralbias proxies, inattention measures, and tax-related inter-actives. The logit regression estimates are presented inthe first four specifications of Table 4. The independentvariables are standardized so that coefficient estimatescan be easily compared within and across specifications.15

In specifications 1 and 2 of Table 4, we explain themutual fund participation dummy with behavioral biasproxies. Specification 2 also includes the control variablespreviously described. Consistent with the presence ofbehavioral biases, negative slopes on Disposition Effect,Narrow Framing, Lottery Stocks Preference, and Inattention

to Earnings News indicate that investors who score high onthese characteristics are less likely to invest in equitymutual funds. The negative slope on the interactive termfor Disposition Effect and No December Tax Loss Selling

indicates that investors prone to both the Disposition

Effect and no attention to tax issues are even less likelyto invest in equity mutual funds. Somewhat surprisingly,we find that overconfident investors (that is, those whotrade stocks more frequently, yet earn lower returns) aremore likely to invest in mutual funds. This could reflectoverconfidence in their ability to identify good funds.16

In economic terms, the logit regression estimates indi-cate that the propensity to invest in mutual funds declinesby 3.15% (0.126�25), 3.90%, 4.67%, and 0.95% when thelevel of disposition effect, narrow framing, lottery prefer-ence, or inattention to earnings news increases by onestandard deviation, respectively.17 The absolute size of slopecoefficients is the largest for Lottery Stocks Preference,suggesting that the propensity to pick individual stocks ismost likely to divert investment away from sensible strate-gies involving mutual funds. The finding for Lottery Stocks

Preference is particularly significant as, unlike some of ourother factors as discussed in Section 2.1, it is hard tocharacterize this factor as anything other than behavioralor, at best, skewness preference.

These findings are robust to the inclusion of thecontrol variables. Moreover, the estimated slopes on thecontrol variables are intuitive. We find that investors whoearn higher income, work as a professional, do not livenear a financial center, are sufficiently sophisticated touse options, or who appear to value diversification in theirstock portfolios are also more likely to invest in mutualfunds. Those who ignore tax loss selling of their individualstocks or load high on market, size, or value risks are lesslikely to hold equity mutual funds.

Specifications 3 and 4 repeat the tests described pre-viously but for the index fund participation dummy, whichis set to one only for those investors who invest in indexfunds at least once during the sample period. The decisionto participate in index funds could be different from the

15 To alleviate concerns about multi-collinearity, we check the

variance inflation factor (VIF) for each explanatory variable.16 Subsequent tests address this potentially puzzling finding.17 Following Wooldridge (2003), we use a factor of 25% to interpret

the logit regression results.

decision to participate in mutual funds generally. Theevidence on behavioral biases and index funds in Specifi-cations 3 and 4 largely echoes what we find for mutualfunds generally in Specifications 1 and 2. Investors whoscore high on Disposition Effect, Narrow Framing, Inatten-

tion to Earnings News, and Disposition Effect interacted withNo December Tax Loss Selling are more likely to avoid indexfunds. Once again, the importance of the propensity totrade risky individual stocks is evident: The strong aver-sion to mutual funds for those with Lottery Stocks Pre-

ference is heightened for index funds. The associationbetween overconfidence and mutual fund investmentdisappears, perhaps indicating that overconfident inves-tors confine themselves to actively managed funds.

Again, these findings are robust to the inclusion of thecontrol variables. The estimates of the coefficients on thecontrol variables also suggest that older investors, higherincome investors, those with smaller stock portfolios,those who appear to value diversification, those whoare cognizant of tax issues, those who do not live neara financial center, and those who avoid individual stockswith high loadings on market and size risks aremore likely to value index funds. Thus, the clienteleof index funds differs somewhat from the clientele ofother mutual funds. However, behavioral biases appear tohave a significant influence on the use of equity mutualfunds regardless of type. In the following sections, weconduct additional tests to refine and extend thesefindings.

4.3. Extent of fund investment: cross-sectional

regression estimates

In our third set of tests, we estimate cross-sectionalregressions with portfolio weights in mutual funds asdependent variables. Similar to the participation regres-sions, the independent variables are the behavioral factorsthat we focus on, plus control variables. One concern insuch regressions is that the cross-correlation of individualsin decision making could inflate the statistical significanceof our regressions. For instance, some segment of investorscould select very similar portfolios of funds and havecorrelated preferences for active, small cap, and industryfunds. As a result, their fund choices could be correlated.

We take the following steps to address such concernsfor each of our cross-sectional regressions. First, clusteredstandard errors are intended to correct for correlation ofresiduals within each cluster (Petersen, 2009), though thismethod assumes independence across groups.18

We do not know the exact nature of any cross-sectional dependence of returns residuals. Therefore, wetry two different forms of clustered standard errors, byzip code (treating each investor within a zip code asone observation) and by peer group (same quintile ofportfolio size, trading frequency, and number of stocks).19

18 Kumar (2009) uses a similar method to account for potential

cross-sectional dependence in performance across investors.19 The results with peer group-clustered standard errors are very

similar. For brevity, we report the results with zip code-clustered standard

errors only.

Table 4Investor characteristics and mutual fund participation decisions and stock versus funds allocation

The first four specifications in the table are logit regressions. In Specifications 1 and 2, the dependent variable is one for investors who hold or trade mutual funds at least once during the sample period. In

Specifications 3 and 4, the dependent variable in the logit regression is one for investors who hold or trade index funds at least once during the sample period. Specifications 5 and 6 are cross-sectional

regression estimates in which the proportion of mutual funds in the equity portfolio is the dependent variable. In Specification 5, the dependent variable is the mean weight of mutual funds in the total equity

(stocks and mutual funds) portfolio. In Specification 6, the dependent variable is the mean weight of index funds only. The dependent variable is multiplied by one hundred. Independent variables are defined in

Appendix A, and a constant term is included. They are standardized so coefficients can be compared within or across specifications. There is one observation per investor. An intercept is included but not

reported. Robust zip code clustered standard errors are used to obtain the t-statistics. The statistically significant coefficient estimates are indicated in bold font. The individual investor data are from a large US

discount brokerage house for the 1991–1996 period.

Mutual fund participation dummy (LOGIT) Mutual fund portfolio weight

All mutual funds Index funds only Mutual fund weight Index fund weight

(1) (2) (3) (4) (5) (6)

Independent Variable Coefficient z-value Coefficient z-value Coefficient z-value Coefficient z-value Coefficient t-statistic Coefficient t-statistic

Behavioral Bias ProxiesDisposition Effect �0.126 �3.37 �0.092 �2.78 �0.106 �3.12 �0.096 �2.67 �1.081 �3.11 �0.569 �1.77

Narrow Framing �0.156 �5.91 �0.106 �4.39 �0.104 �4.54 �0.092 �3.67 �1.936 �7.95 �1.122 �3.36

Overconfidence Dummy 0.057 2.19 0.060 3.30 �0.005 �0.63 �0.004 �0.62 0.790 3.67 �0.800 �2.22

Male Dummy 0.017 1.04 �0.017 �0.52 �0.032 �1.38 �0.016 �1.11 0.288 1.35 �0.311 �0.74

Local Bias �0.014 �1.16 �0.014 �1.31 �0.013 �0.41 �0.014 �0.33 �0.242 �1.01 �0.172 �1.13

Lottery Stocks Preference �0.187 �8.91 �0.170 �6.29 �0.239 �10.14 �0.230 �9.10 �1.319 �5.35 �0.911 �3.04

Inattention to Earnings News �0.038 �2.17 �0.044 �2.11 �0.047 �2.49 �0.057 �2.11 �0.580 �2.30 �0.690 �2.55

Inattention to Macroeconomic News �0.019 �1.18 �0.013 �1.09 �0.014 �1.14 �0.013 �1.03 �0.452 �1.78 �0.206 �1.42

DEnHigh Income �0.021 �1.73 �0.018 �1.68 �0.014 �1.42 �0.013 �1.21 �0.401 �1.60 �0.388 �1.66

DEnNo December Tax Loss Selling �0.091 �2.98 �0.081 �2.11 �0.074 �3.14 �0.069 �2.84 �0.327 �3.10 �0.430 �2.74

Control VariablesAge 0.022 1.19 0.186 4.01 0.488 1.60 1.355 3.25

Income 0.035 2.26 0.046 1.77 0.767 2.60 0.838 2.18

High Income Dummy 0.050 2.90 0.084 3.01 0.588 2.11 0.438 2.18

Marital Status 0.006 1.44 0.021 1.30 0.727 2.01 �0.101 �0.43

Family Size 0.024 0.70 0.003 0.31 �0.208 �0.70 �0.055 �0.22

Professional Dummy 0.032 1.99 0.030 1.20 0.454 1.86 �0.200 �1.11

Retired Dummy �0.009 �0.22 0.028 1.55 �0.071 �0.41 1.011 2.91

Investment Experience 0.029 1.51 0.028 1.40 0.122 0.40 0.533 3.01

Financial Center Dummy �0.084 �3.98 �0.067 �2.11 �1.034 �3.44 �0.960 �3.11

Options Dummy 0.066 3.01 0.016 1.11 0.101 1.53 �0.188 �0.76

Short Sale Dummy 0.033 1.17 0.025 1.55 0.717 2.01 �0.142 �0.61

Stock Portfolio Diversification 0.158 6.80 0.273 7.16 0.940 3.11 0.767 3.30

Stock Portfolio Size �0.022 �0.98 �0.160 �2.11 �1.399 �10.02 �0.594 �2.55

Stock Portfolio Performance 0.035 1.70 0.020 1.40 0.105 0.36 �0.409 �2.21

No Dec Tax Loss Selling �0.047 �2.52 �0.036 �1.95 �1.013 �2.06 �1.322 �3.44

Holds Tax-Deferred Account 0.135 9.08 0.105 7.11 2.452 7.46 0.650 2.29

Market Factor Exposure �0.041 �2.60 �0.031 �3.01 �0.980 �2.93 �0.148 �1.70

SMB Factor Exposure �0.168 �5.91 �0.055 �4.53 �0.937 �3.72 �0.242 �1.77

HML Factor Exposure �0.038 �2.27 �0.009 �1.20 �0.392 �2.93 �0.375 �2.19

UMD Factor Exposure �0.017 �1.68 �0.010 �2.09 �0.462 �2.09 �0.400 �2.13

Pseudo R2 0.038 0.092 0.027 0.074 0.104 0.126

Number of Observations 22,984 21,542 22,984 21,542 21,542 21,542

W.

Ba

ileyet

al.

/Jo

urn

al

of

Fina

ncia

lE

con

om

ics1

02

(20

11

)1

–2

71

1

20 See Capon, Fitzsimons and Prince (1996) for survey evidence that

about 39% of mutual fund investors were unaware of the load charged

by the funds they held.21 Starks and Yates (2008) investigate a related familiarity-based

hypothesis and find that individuals often cluster their choice of funds

within the same family of funds.

W. Bailey et al. / Journal of Financial Economics 102 (2011) 1–2712

Second, we construct risk-adjusted returns to remove themarket-wide movement in returns that is common to allinvestors.

Specifications 5 and 6 of Table 4 present the regressionestimates. In Specification 5, the dependent variable is themean weight assigned to mutual funds in an investor’sequity portfolio. The results parallel the findings from theparticipation regressions reported in Table 3. Individualswho score high on Disposition Effect, Narrow Framing,Lottery Stocks Preference, Inattention to Earnings News, orinteraction between Disposition Effect and No December Tax

Loss Selling typically put a smaller fraction of their portfo-lio in mutual funds, while overconfident investors typi-cally allocate a larger proportion of their equity portfolioto mutual funds. In economic terms, a one standarddeviation increase in narrow framing propensity is asso-ciated with a 1.94% lower allocation to mutual funds. Theestimates of other statistically significant behavioral biasproxies are also economically significant. The estimates forthe coefficients on the control variables show that inves-tors who have higher income, are married, do not live neara financial center, understand short selling and diversifi-cation, have relatively small stock portfolios, understandtax issues, and have relatively low loadings on risk factorsin their stock portfolios typically hold a higher proportionin mutual funds. Thus, similar forces drive the decision toparticipate in mutual funds and the extent of thatparticipation.

In Specification 6, the dependent variable is the meanweight assigned to index funds. The cross-sectionalregression results with index fund weight reinforce thefindings from the index fund participation regressions.Investors with stronger behavioral biases typically allo-cate a smaller proportion of their equity portfolio to indexfunds, although the effect of overconfidence flips betweenSpecifications 5 and 6. Even though overconfident inves-tors allocate a slightly larger weight to mutual funds, theyallocate a smaller proportion of their equity portfolio toindex funds. Thus, such investors focus more on activelymanaged funds. The extent to which index funds areheld goes up as individual stock portfolio performancegoes down.

4.4. Behavioral biases and preference for certain types of

mutual funds

To better understand investor preferences for differenttypes of funds, we examine three additional characteris-tics of investors’ mutual fund portfolios. Table 5, Panel Apresents the cross-sectional estimates. In Specifications1–3, the dependent variable is the mean expense ratio,the mean front-end load, and the mean fund turnover,respectively, for each individual’s mutual fund portfolio.Specification 1 shows that investors with stronger Dis-

position Effect, Narrow Framing, Overconfidence Dummy,Lottery Stocks Preference, Inattention to Earnings News, andinteraction between Disposition Effect and No December

Tax Loss Selling tend to select mutual funds with higherexpense ratios. Specification 2 examines front-end loadsand confirms that the same set of biases that driveinvestors to higher expense funds is also associated with

choosing mutual funds with higher front end loads.Specification 3 shows that individuals who are overconfi-dent, male, have lottery stocks preference, display inat-tention to earnings news, and have positive loading onmeasures of particularly severe disposition effect (Dispo-

sition EffectnHigh Income and Disposition EffectnNo Decem-

ber Tax Loss Selling) tend to invest in funds with higherturnover.

If we assume that funds with higher expense ratios,higher front-end loads, and high levels of turnover arepoor choices, our evidence indicates that investors whodemonstrate poor decision making with individual stocksalso appear to make poor decisions about mutual funds.The slope coefficients on behavioral factors in Specifica-tion 2 are particularly large, suggesting that behavioralbiases are important in driving investors into high frontend load funds. The slope coefficients on the controlvariables indicate that younger, poorer, less experienced,and less tax-savvy investors are more likely to elect theseapparently poor choices.

4.5. A closer look at fund-level local bias and inattention

Why do some investors go against common wisdomand hold high front-end load funds? One possibility isthat they are unaware of the load.20 Alternatively, someinvestors could be more willing to pay a high load forfunds they are familiar with. In particular, they could havemore awareness of funds headquartered in their geo-graphic area, perhaps due to localized marketing efforts.21

As a result, they are willing to pay high fees for suchfunds. To investigate this thesis, we test whether inves-tors with high Fund-Level Local Bias are more likely to holdfunds with high fees and expenses. Having employedproxies for local bias and inattention to news based ontrading of individual stocks, we also investigate whethersome investors concentrate their equity mutual fundtrades around news. In Table 5, Panel B we introduceour Fund-Level Local Bias measure into the cross-sectionalregression specification. This variable is distinct fromindividual equity local bias in that it measures thegeographical proximity between an investor’s home andthe headquarters of mutual funds held by the investor,not the proximity of the headquarters of an individuallisted company. We also introduce Fund-Level Inattention

to the cross sectional regressions. This variable measureseach individual’s propensity to trade mutual funds aroundmacroeconomic news events as 1�(number of mutualfund trades around the event)/(total number of mutualfund trades).

The estimates in Specifications 1–3 show that inves-tors with stronger Fund-Level Local Bias tend to selectmutual funds with higher expense ratios, front end loads,and turnover, even after controlling for other behavioral

Table 5Characteristics of investors and the funds they select.

This table reports cross-sectional regression estimates in which three different mutual fund portfolio characteristics are employed as dependent

variables. In Panel A, Specifications 1–3, the mean expense ratio, the mean front-end load, and the mean turnover of the funds in the mutual fund

portfolio is the dependent variable, respectively. In all specifications, the dependent variable is multiplied by one hundred. There is one observation per

investor. Independent variables are defined in Appendix A, and an intercept term is included but not reported. In Panel B, we consider two additional

independent variables. Zip code clustered standard errors are used to obtain the t-statistics. The statistically significant coefficient estimates are indicated

in bold font. There is one observation per individual.

(1) Expense ratio (2) Front-end load (3) Fund turnover

Independent Variable Coefficient t-statistic Coefficient t-statistic Coefficient t-statistic

Panel A: Mutual fund portfolio characteristic regression estimates

Behavioral Bias ProxiesDisposition Effect 0.012 3.02 0.033 3.11 0.004 0.55

Narrow Framing 0.019 3.55 0.041 2.42 0.004 1.01

Overconfidence Dummy 0.020 3.11 0.029 2.50 0.022 2.67

Male Dummy 0.005 1.05 0.012 1.22 0.018 2.19

Local Bias 0.003 0.18 0.021 1.70 �0.009 �1.40

Lottery Stocks Preference 0.024 3.95 0.033 2.29 0.017 2.66

Inattention to Earnings News 0.013 2.33 0.022 2.65 0.019 2.08

Inattention to Macroeconomic News 0.005 1.13 0.017 1.54 0.003 0.34

DEnHigh Income 0.007 1.60 0.011 1.69 0.021 2.80

DEnNo December Tax Loss Selling 0.024 3.60 0.026 2.44 0.025 3.51

Control VariablesAge �0.014 �2.30 �0.030 �1.65 �0.037 �2.98

Income 0.007 1.51 0.011 1.00 0.023 1.71

High Income Dummy 0.003 0.90 0.015 0.69 �0.034 �2.30

Marital Status �0.005 �1.70 �0.008 �0.50 0.005 0.55

Family Size 0.004 0.80 0.012 1.01 0.015 0.70

Professional Dummy 0.007 1.11 0.024 1.22 �0.021 �2.05

Retired Dummy �0.015 �2.30 0.012 0.56 �0.017 �1.81

Investment Experience �0.014 �2.59 �0.025 �2.51 �0.033 �3.00

Financial Center Dummy �0.008 �1.41 �0.004 �1.22 �0.027 �2.67

Options Dummy 0.002 0.35 0.012 1.33 0.019 2.75

Short Sale Dummy �0.003 �1.13 �0.014 �0.99 0.013 1.99

Stock Portfolio Diversification �0.001 �0.11 0.003 0.90 0.013 1.09

Stock Portfolio Size 0.001 0.17 0.008 0.45 0.005 0.60

Stock Portfolio Performance �0.006 �1.54 �0.004 �0.26 0.009 0.91

No December Tax Loss Selling 0.015 2.52 0.031 2.89 0.034 2.99

Holds Tax-Deferred Account �0.022 �4.81 �0.013 �3.53 �0.020 �3.86

Market Factor Exposure 0.010 2.56 0.016 2.65 0.022 2.97

SMB Factor Exposure 0.018 3.17 0.012 2.26 0.024 3.39

HML Factor Exposure 0.003 0.93 0.012 2.39 0.001 0.23

UMD Factor Exposure 0.019 3.72 0.024 3.34 0.031 3.52

Adjusted R2 0.071 0.054 0.066

Number of Observations 21,542 21,542 21,542

Panel B: Regression estimates with the fund-level local bias and inattentiveness measures

Behavioral Bias ProxiesDisposition Effect 0.013 2.99 0.032 3.07 0.002 0.21

Narrow Framing 0.016 3.44 0.045 2.44 0.003 0.50

Overconfidence Dummy 0.018 3.24 0.025 2.32 0.022 2.43

Male Dummy 0.004 0.87 0.012 1.21 0.016 2.31

Local Bias 0.004 0.29 0.020 1.68 �0.011 �1.52

Lottery Stocks Preference 0.022 4.12 0.028 2.49 0.017 2.74

Inattention to Earnings News 0.011 2.19 0.022 2.49 0.015 2.58

Inattention to Macroeconomic News 0.008 1.85 0.019 2.68 0.002 0.22

DEnHigh Income 0.007 1.63 0.016 2.03 0.017 2.29

DEnNo December Tax Loss Selling 0.022 3.49 0.022 2.36 0.023 3.44

Fund-level bias proxies

Fund-Level Local Bias 0.024 5.43 0.050 3.59 0.036 4.65

Fund-Level Inattention to Macro News 0.018 2.37 0.017 2.11 0.002 0.54

Control VariablesAge �0.016 �2.42 �0.028 �1.33 �0.034 �2.03

Income 0.007 1.23 0.011 1.05 0.021 1.60

High Income Dummy 0.003 0.50 0.015 0.65 �0.030 �2.13

Marital Status �0.006 �1.70 �0.007 �0.70 0.005 0.51

Family Size 0.005 1.02 0.018 1.03 0.014 0.74

Professional Dummy 0.005 1.06 0.017 0.98 �0.025 �2.05

Retired Dummy �0.014 �1.93 0.011 0.55 �0.018 �1.83

W. Bailey et al. / Journal of Financial Economics 102 (2011) 1–27 13

Table 5 (continued )

(1) Expense ratio (2) Front-end load (3) Fund turnover

Independent Variable Coefficient t-statistic Coefficient t-statistic Coefficient t-statistic

Investment Experience �0.015 �2.61 �0.016 �2.12 �0.025 �2.62

Financial Center Dummy �0.004 �1.06 �0.005 �1.35 �0.024 �2.63

Options Dummy 0.002 0.30 0.011 1.35 0.021 3.22

Short Sale Dummy �0.008 �1.53 �0.018 �1.39 0.013 2.02

Stock Portfolio Diversification 0.004 0.76 0.004 0.91 0.011 1.06

Stock Portfolio Size 0.002 0.38 0.011 1.01 0.007 0.60

Stock Portfolio Performance �0.008 �1.51 �0.007 �0.30 0.008 0.74

No December Tax Loss Selling 0.013 2.33 0.030 2.80 0.029 2.71

Holds Tax-Deferred Account �0.020 �4.71 �0.012 �3.33 �0.021 �3.67

Market Factor Exposure 0.011 2.55 0.015 2.78 0.023 2.73

SMB Factor Exposure 0.018 3.11 0.012 2.21 0.023 3.32

HML Factor Exposure 0.003 0.90 0.013 2.43 0.001 0.21

UMD Factor Exposure 0.021 3.70 0.026 3.54 0.030 3.50

Adjusted R2 0.072 0.056 0.069

Number of Observations 21,542 21,542 21,542

W. Bailey et al. / Journal of Financial Economics 102 (2011) 1–2714

biases. Indeed, Fund-Level Local Bias emerges as the vari-able with the largest economic and statistical significancecompared with all other behavioral biases. Intriguingly,further correlation analysis (unreported but availableupon request) shows that Fund-Level Local Bias is nega-tively correlated with age and positively correlated withRetired Dummy and Stock Portfolio Size. This again suggestslocalized marketing efforts: Older investors are typicallycleverer and avoid Fund-Level Local Bias, but retiredinvestors with large portfolios could be subjected torecommendations or marketing efforts from brokers,bankers, and social peers.

Thus, investors who exhibit a stronger preference tohold local funds, which could be thought of as a famil-iarity effect, are more likely to buy funds with high fees,expenses, and turnover. Furthermore, Fund-Level Inatten-

tion is positive and significant in two of the threespecifications, those for expense ratios and front endloads. Investors who pay less attention to news seem toselect funds that impose higher expenses and loads onthemselves. These findings suggest that behavioral biasescan combine with ignorance to yield costly suboptimalmutual fund investment decisions.

4.6. Behavioral biases and trend chasing behavior

Our next set of tests examines whether behavioralbiases also play an important role in explaining individualinvestors’ trend-chasing behavior. Many explanationshave previously been proposed for this robust patternobserved in mutual fund flow data. Chevalier and Ellison(1997) show that agency problems induce fund managersto alter the riskiness of the fund to maximize investmentflows instead of risk-adjusted expected returns. Sirri andTufano (1998) and Gruber (1996) propose that investorsinfer managerial skill from past returns. Berk and Green(2004) feature investors who infer managerial skill frompast returns and, therefore, chase returns. However, fundmanagers facing decreasing returns to scale in their activeportfolios no longer outperform the index when morefunds flow in, and, as a consequence, past performance

does not predict future returns. Instead of analyzingaggregate flows, our data allow us to study the relationbetween behavioral tendencies and trend-chasing beha-vior at the individual investor level.

Table 6 examines trend chasing in individual mutualfund portfolios. For each mutual fund purchase, wecompute the return prior to the purchase, which is thenaveraged for each individual. Specification 1 uses 1-yearpast returns as dependent variable, and Specification 2uses the 2-year past returns. The results from bothspecifications show that investors with certain behavioralbiases, or inattention to macro news, tend to buy fundswith more positive recent returns. Although the disposi-tion effect does not seem to be associated with trendchasing, the coefficients on the Disposition EffectnHigh

Income and Disposition EffectnNo December Tax Loss Selling

are strongly significantly positive. Among the coefficientson the control variables, some evidence shows thatsophisticated investors (those who are professionals, livenear a financial center, trade options, or have well-diversified, well-performing individual stock portfolios)are less likely to engage in trend chasing. As was foundpreviously (Table 5) for the propensity to select high-costmutual funds, the size of slope coefficients suggest thatOverconfidence Dummy and Lottery Stocks Preference areamong the strongest predictors of whether a particularinvestor will trend-chase with mutual funds.

This evidence suggests that trend chasing is not arational strategy. This interpretation is supported by theempirical results of previous authors concerning mutualfund flows and subsequent returns on individual stocksheld by the funds. Frazzini and Lamont (2008) findrelatively poor monthly returns on portfolios of individualstocks held disproportionately heavily by mutual fundsthat experience high inflows over the previous 6 monthsto 3 years. We find that it is more behaviorally biasedindividuals who are responsible for trend-chasing inflows.Thus, some of what they describe as the ‘‘dumb money’’effect must be ascribed to a subset of investors who wehave also identified as making poor decisions with theirindividual stock portfolios.

Table 6Returns chasing and fund selection.

This table reports cross-sectional regression estimates with two different mutual fund portfolio performance measures as dependent variables, the

12-month past return and the 24-month past return. There is one observation per investor. The independent variables include behavioral bias proxies,

control variables, and an intercept term that is included but unreported. Independent variables are defined in Appendix A. Investors with fewer than 12

months of data are excluded. Zip code clustered standard errors are used to obtain the t-statistics. The statistically significant coefficient estimates are

indicated in bold font. The individual investor data are from a large US discount brokerage house for the 1991–1996 period.

(1) 12-Month past return (2) 24-Month past return

Independent Variable Coefficient t-statistic Coefficient t-statistic

Behavioral Bias ProxiesDisposition Effect �0.022 �0.25 0.087 0.34

Narrow Framing 0.644 4.35 0.764 3.45

Overconfidence Dummy 1.370 5.04 1.604 6.87

Male Dummy 0.062 0.46 0.258 2.51

Local Bias 0.154 1.09 0.034 0.33

Lottery Stocks Preference 0.978 6.39 1.196 5.75

Inattention to Earnings News 0.199 1.62 0.291 1.85

Inattention to Macroeconomic News 0.581 2.18 0.492 2.84

DEnHigh Income 0.353 2.05 0.508 2.57

DEnNo December Tax Loss Selling 0.480 2.16 0.390 2.06

Control VariablesAge �0.329 �1.66 �0.886 �2.11

Income �0.427 �1.83 �0.542 �1.62

High Income Dummy �0.061 �0.58 �0.196 �1.49

Marital Status �0.077 �0.81 0.443 1.52

Family Size �0.149 �1.22 �0.609 �1.76

Professional Dummy �0.629 �2.32 �1.052 �2.92

Retired Dummy �0.487 �2.91 �0.152 �1.92

Investment Experience 0.057 0.30 �0.468 �1.98

Financial Center Dummy �0.404 �2.13 �0.510 �1.63

Options Dummy �0.347 �2.71 �0.492 �2.63

Short Sale Dummy �0.024 �0.12 0.070 0.46

Stock Portfolio Diversification �0.093 �0.46 �0.492 �2.02

Stock Portfolio Size �0.139 �0.68 �0.103 �0.71

Stock Portfolio Performance �0.558 �3.46 �0.768 �2.13

No December Tax Loss Selling 0.380 1.92 0.407 2.93

Holds Tax-Deferred Account �0.062 �0.61 �0.168 �2.38

Market Factor Exposure 0.862 3.87 0.565 3.54

SMB Factor Exposure 0.393 2.64 0.485 3.82

HML Factor Exposure 0.068 0.67 0.151 2.15

UMD Factor Exposure 0.228 2.40 0.264 2.51

Adjusted R2 0.091 0.076

Number of Observations 21,542 21,542

W. Bailey et al. / Journal of Financial Economics 102 (2011) 1–27 15

The disposition effect result merits further discussion.In the classic form of this bias, investors sell well-performing individual stocks too quickly and hold poor-performing stocks too long. Trend chasing by individualswho invest in mutual funds is broadly contradictory to adisposition effect in individual stocks: Trend-chasers seekand then hold good performers, instead of selling themquickly. Our disposition effect interactive terms isolateinvestors who display a disposition effect that is likely tobe particularly severe, and both terms earn a stronglysignificantly positive slope coefficient in the regressionsof Table 6. Thus, individuals who display particularlydamaging forms of the disposition effect in their indivi-dual stock portfolios tend to contradict themselves bydisplaying trend chasing in their mutual fund choices.This implies that behavioral biases do not just vary acrossindividuals but also across the components within aparticular investor’s portfolio, with professionally mana-ged assets handled in a radically different manner thanindividual stocks. This could be consistent with the idea

that investors decompose their portfolios into layers thatserve different purposes (Shefrin and Statman, 2000).

Overall, our cross-sectional regression estimatesreported in Tables 4–6 confirm that investors who aremore behaviorally biased on any of several dimensions ordo not pay attention to salient news are more likely todisplay poor mutual fund investment decisions. Theytypically have a greater proportion of their equity invest-ment in individual stocks, not mutual funds, suggestingthat they do not value diversification. When they buyfunds, they prefer actively managed funds to index funds,tend to buy funds with high fees and loads, and chasefunds with high recent returns. The strength of one of oursimplest behavioral bias measures, Lottery Stocks Prefer-

ence, is particularly compelling.The missing link in our evidence and interpretations to

this point is more explicit evidence on performance.While it appears that behavioral biases and ignoring newslead to poor choices, we must also show the consequencesfor performance. For example, individual investors

W. Bailey et al. / Journal of Financial Economics 102 (2011) 1–2716

typically avoid high front-end load funds (Barber, Odeanand Zheng, 2005), but some investors could be able todiscriminate between good and bad quality front-endload funds and enjoy superior portfolio performance fromthose high load funds that they do elect to hold. Thus, ournext task is to examine the performance of investors’mutual fund portfolios.

4.7. Performance of mutual fund portfolios

We again estimate cross-sectional regressions with thesame behavioral proxies and controls as explanatoryvariables. Table 7, Panel A studies mutual fund perfor-mance for each investor’s actual holdings. The dependentvariables are four measures of the sample period perfor-mance of each investor’s mutual fund portfolio, the rawperformance measure (mean monthly portfolio return),the net-of-expenses performance measure (the netmonthly return), the Sharpe ratio, and the market modelalpha. We again use zip code clustered standard errors tocompute the t-statistics because performance estimatesare unlikely to be independent.22

Specification 1 explains the mean monthly return. Dis-

position Effect, Narrow Framing, Overconfidence Dummy,Lottery Stock Preference, and both measures of inattentionto news are associated with lower performance. For exam-ple, mean monthly return is lower by �0.041 per monthfor each standard deviation of increase in narrow framing.Because the highest and lowest quintiles of narrow framingdiffer by 4.3 standard deviations, this implies a 2.12% peryear lower return for highest quintile narrow framinginvestors compared with those in the lowest quintile.Similarly, highest quintile disposition effect investors havereturns 1.34% lower than those in the lowest quintile.23

Thus, our behavioral proxies detect poor decision-makingskills that reduce portfolio performance.

Among the control variables, investment experience issignificant, and the positive slope makes sense. The use ofoptions or short sales is associated with better mutualfund performance, which is consistent with those vari-ables reflecting skill or financial sophistication. Specifica-tion (2) examines net monthly returns and shows similarassociations between behavioral biases and performance.

Specification 3 examines the Sharpe ratio. We againfind broadly similar associations with the behavioral biasproxies, inattention measures, and control variables.Narrow Framing, Overconfidence Dummy, and, to a lesserextent, Disposition Effect are associated with lower per-formance. Results are similar when we account for poten-tial cross-sectional dependence in performance inducedby market-wide factors and consider a risk-adjustedperformance measure as the dependent variable (Specifi-cation 4). Collectively, the evidence in Table 7, Panel Ashows that behavioral biases measured from individualstock selection are also associated with lower raw and

22 As before, other forms of standard error clustering yield very

similar results.23 Given that the highest and lowest quintiles of disposition effect

differ by 4.13 standard deviations, their yearly performance difference is

1.34% (�0.027% times 12 times 4.13).

risk-adjusted returns from mutual funds. Thus, poordecision making in one domain appears to spill over intothe performance experienced with other classes ofinvestments.

While Table 7, Panel A describes the actual realizedreturns of individual investors based on their total hold-ings at the end of each month, Panel B studies perfor-mance based on investor trades under both actual andhypothetical holding periods computed using daily fundreturns data from Morningstar.24 Specifications 1 and 2study actual holding period returns from trades. Theyconfirm that investors with higher values on most of ourbehavioral bias proxies and inattention to news measureshave significantly lower holding period returns andshorter holding period, in contrast to the buy-and-holdstrategies prescribed by standard portfolio theory. Local

Bias is associated with longer holding periods. Correlationanalysis (unreported but available upon request) indicatesthat Local Bias is associated with poor diversification andmediocre performance in the individual stock portfolio,but Specification 2 shows us that it could also yieldsensible low turnover of mutual fund holdings.

Specifications 3 and 4 adopt the alternative viewpointof returns based on actual trades but standardizedhypothetical holding periods. Following Odean (1999)and Kumar and Lee (2006), we calculate the subsequentk-month returns following each buy trade averaged overthe trading history of an individual and subtract thesubsequent k-month returns following each sell tradeaveraged over the trading history. The summary statisticson 1- and 12-month post-trade buy–sell return differen-tials show that investors who score high on most of ourbehavioral and inattention proxies have lower post-tradebuy–sell returns differentials. In other words, investorswith strong behavioral biases tend to time their buys andsells poorly, and they experience inferior performancerelative to less-biased investors. The results are especiallysignificant for 12-month returns differentials.

Table 8 features interactions between investor portfo-lio characteristics and fund characteristics to explainperformance. Individual household mutual fund perfor-mance is regressed on the behavioral biases and inatten-tion measures previously employed, characteristics of theindividual’s mutual fund portfolio (the weight ofthe portfolio held in mutual funds, and the averages ofthe expense ratio, 12-B-1 fee, and front-end load onthe funds held), interactive terms that combine beha-vioral and portfolio characteristics, and (unreported)control variables.

The results confirm the negative impact of dispositioneffect, narrow framing, overconfidence, lottery stockspreference, and inattention to news on performance asdocumented previously. Among the mutual fund portfoliocharacteristics, investors with higher weight on mutualfunds tend to enjoy superior fund performance, which isconsistent with classic notions of portfolio management.

24 Partial sales are excluded from our calculations. Unlike Panel A,

these calculations exclude any funds that were held prior to the start of

our sample period.

Table 7Investor characteristics and performance of mutual fund investments.

This table reports cross-sectional regression estimates to explain two measures of mutual fund portfolio performance, position-based performance

measures in Panel A and trade-based performance measures in Panel B. In Panel A, the dependent variables are (1) the mean monthly percent return (in

percentage terms), (2) the net monthly return which equals the mean monthly return minus expenses (but not loads), (3) the Sharpe ratio of net returns

multiplied by one hundred, and (4) the monthly market model alpha. In Panel B, the dependent variables in Specifications 1 to 4 are the mean annualized

holding period return, the mean holding period, the one-month post trade buy–sell return differential, and the 12-month post trade buy–sell return

differential (PTBSD), respectively. Independent variables are defined in appendix. A constant term is included. Investors with fewer than 12 months of

data are excluded. Zip code clustered standard errors are used to obtain the t-statistics. The statistically significant coefficient estimates are indicated in

bold font. The individual investor data are from a large US discount brokerage house from 1991 to 1996.

Panel A: Position-Based Mutual Fund Portfolio Performance Regression Estimates

(1) Mean monthly returns (2) Net monthly returns (3) Net Sharpe ratio�100 (4) Market model alpha

Independent Variable Coefficient t-statistic Coefficient t-statistic Coefficient t-statistic Coefficient t-statistic

Behavioral Bias ProxiesDisposition Effect �0.027 �2.14 �0.028 �2.16 �0.554 �1.94 �0.027 �1.96

Narrow Framing �0.041 �2.94 �0.047 �2.75 �1.550 �3.83 �0.050 �2.65

Overconfidence Dummy �0.025 �2.17 �0.031 �2.60 �1.502 �2.39 �0.033 �2.06

Male Dummy �0.009 �1.05 �0.012 �1.42 �1.503 �1.90 �0.039 �2.29

Local Bias 0.010 1.20 0.010 1.30 0.113 0.42 0.006 0.88

Lottery Stocks Preference �0.026 �2.99 �0.024 �2.78 �1.249 �1.97 �0.059 �3.26

Inattention to Earnings News �0.014 �2.18 �0.015 �2.48 �1.041 �2.13 �0.026 �2.82

Inattention to Macroeconomic News �0.022 �2.22 �0.024 �2.45 �1.010 �2.03 �0.019 �1.92

DEnHigh Income �0.009 �1.26 �0.012 �1.26 0.143 0.19 0.006 0.80

DEnNo December Tax Loss Selling �0.016 �1.66 �0.016 �1.62 �0.155 �0.51 �0.028 �2.58

Control VariablesAge �0.007 �1.59 �0.008 �1.45 0.448 0.65 �0.003 �0.72

Income 0.002 0.22 0.003 0.18 �0.581 �0.88 �0.011 �0.31

High Income Dummy 0.026 1.63 0.030 1.89 0.200 1.04 0.026 0.48

Marital Status 0.004 0.34 0.009 0.59 0.218 0.36 �0.003 �0.08

Family Size �0.004 �0.34 �0.004 �0.40 �0.679 �0.97 �0.039 �1.91

Professional Dummy �0.002 �0.12 �0.017 �0.87 �0.308 �0.40 0.002 0.43

Retired Dummy 0.003 0.21 0.003 0.18 0.046 0.08 �0.034 �1.54

Investment Experience 0.028 3.15 0.026 2.89 1.991 2.72 0.051 2.35

Financial Center Dummy �0.001 �0.08 �0.011 �0.88 �0.510 �0.85 �0.014 �1.42

Options Dummy 0.034 2.51 0.043 1.79 1.517 2.62 0.062 3.12

Short Sale Dummy 0.051 2.36 0.021 1.55 0.978 1.64 0.035 1.33

Stock Portfolio Diversification 0.028 1.83 0.024 1.57 0.105 0.18 �0.013 �0.94

Stock Portfolio Size 0.023 1.39 0.023 1.43 1.288 2.03 �0.011 �0.28

Stock Portfolio Performance �0.032 �1.29 �0.033 �2.07 �0.672 �1.17 0.001 0.31

No December Tax Loss Selling 0.003 0.29 0.002 0.18 �0.614 �1.66 �0.018 �1.47

Holds Tax-Deferred Account �0.003 �0.59 �0.003 �0.58 0.168 1.02 �0.017 �1.62

Market Factor Exposure 0.021 2.78 0.019 2.43 0.595 3.08 0.009 0.51

SMB Factor Exposure 0.012 1.38 0.004 0.66 0.670 3.51 0.038 2.71

HML Factor Exposure 0.007 1.35 0.006 1.22 0.025 0.14 0.014 1.39

UMD Factor Exposure 0.031 3.35 0.024 2.98 0.449 2.75 0.016 1.68

Adjusted R2 0.042 0.043 0.037 0.029

Number of Observations 21,542 21,542 21,542 20,142

Panel B: Trade-Based Mutual Fund Portfolio Performance

(1) Holding period return (2) Holding period (3) One-month PTBSD (4) 12-month PTBSD

Independent Variable Coefficient t-statistic Coefficient t-statistic Coefficient t-statistic Coefficient t-statistic

Intercept 0.418 12.57 444.20 17.85 0.042 0.97 �2.526 �14.34

Behavioral Bias ProxiesDisposition Effect �0.059 �2.46 �25.15 �4.26 �0.070 �2.48 �0.295 �2.26

Narrow Framing �0.054 �3.40 �15.16 �2.97 �0.096 �3.48 �0.462 �3.26

Overconfidence Dummy �0.066 �3.06 �42.61 �7.75 �0.090 �2.17 �0.569 �2.99

Male Dummy 0.018 0.53 �3.36 �0.56 �0.018 �1.09 �0.114 �1.63

Local Bias �0.005 �0.14 10.02 2.67 0.013 0.69 �0.356 �2.27

Lottery Stocks Preference �0.037 �2.27 �24.17 �2.36 �0.058 �2.27 �0.551 �2.34

Inattention to Earnings News �0.010 �1.26 �4.02 �1.62 �0.033 �2.66 �0.474 �2.36

Inattention to Macroeconomic News �0.048 �2.29 �11.71 �2.85 �0.029 �2.60 �0.526 �2.69

DEnHigh Income �0.065 �3.17 �19.66 �2.67 �0.061 �2.55 �0.485 �2.07

DEnNo December Tax Loss Selling �0.037 �1.99 �15.69 �2.89 0.002 0.36 �0.079 �1.39

Control VariablesAge �0.049 �2.46 18.65 1.99 �0.054 �2.06 �0.417 �2.23

Income 0.015 0.22 �2.27 �0.25 �0.005 �0.15 �0.084 �0.72

High Income Dummy 0.014 0.94 �4.58 �1.39 �0.024 �1.63 �0.257 �1.26

W. Bailey et al. / Journal of Financial Economics 102 (2011) 1–27 17

Table 7 (continued )

Panel B: Trade-Based Mutual Fund Portfolio Performance

(1) Holding period return (2) Holding period (3) One-month PTBSD (4) 12-month PTBSD

Independent Variable Coefficient t-statistic Coefficient t-statistic Coefficient t-statistic Coefficient t-statistic

Marital Status 0.025 1.09 5.02 0.59 0.008 0.46 0.183 1.61

Family Size 0.016 1.82 �2.65 �0.27 0.007 0.17 0.061 0.20

Professional Dummy �0.010 �0.18 �4.49 �0.42 0.013 0.88 �0.191 �1.27

Retired Dummy �0.038 �1.68 27.61 3.33 �0.037 �2.17 �0.197 �2.19

Investment Experience 0.038 2.21 8.67 1.58 0.089 3.50 0.587 3.14

Financial Center Dummy 0.006 0.31 �16.56 �2.03 0.020 1.31 �0.051 �0.31

Options Dummy 0.040 1.97 �33.55 �5.56 0.050 2.14 0.302 2.71

Short Sale Dummy 0.034 1.82 �12.02 �2.01 0.041 1.86 0.181 1.98

Stock Portfolio Diversification 0.013 0.73 41.47 3.88 0.016 1.01 0.091 1.33

Stock Portfolio Size 0.026 1.70 3.33 0.55 0.022 1.55 0.071 0.95

Stock Portfolio Performance �0.042 �2.50 �25.00 �4.92 �0.017 �1.05 �0.409 �2.65

No December Tax Loss Selling �0.015 �1.32 45.25 3.35 �0.074 �2.77 �0.223 �1.85

Holds Tax-Deferred Account 0.023 1.95 7.11 1.53 �0.013 �0.38 0.035 0.25

Market Factor Exposure 0.004 0.16 �26.14 �3.53 0.047 2.09 0.261 1.78

SMB Factor Exposure 0.035 1.98 �23.73 �3.86 0.013 0.95 0.457 3.13

HML Factor Exposure �0.015 �1.21 �2.36 �0.51 0.025 1.65 0.234 1.68

UMD Factor Exposure 0.024 1.11 16.83 2.99 0.017 0.88 0.095 0.76

Adjusted R2 0.045 0.093 0.040 0.064

Number of Observations 15,210 15,210 18,002 18,002

25 This supports the notion that individual investors should be

encouraged to make good decisions, as with retirement savings plan

(Benartzi and Thaler, 2007).

W. Bailey et al. / Journal of Financial Economics 102 (2011) 1–2718

Investors with higher weight on expenses, 12-B-1 market-ing fees, and front-end load funds typically experienceinferior fund performance.

Among the interactive terms, we see particularly poorperformance for high disposition effect investors whoselect funds with high 12-B-1 marketing fees or highfront-end loads. This also appears to be the case forinvestors with strong framing effects or overconfidence.The coefficients for interactives for high inattention andfees are uniformly significantly negative. Thus, investorswith particularly high behavioral biases who choose toremain poorly informed could make particularly poorchoices, stumbling into mutual funds with high expenseratios, high 12-B-1 marketing fees, or front-end loads.This echoes the finding in Table 5 that behavioral biasesare particularly powerful in pulling investors into highfront end load funds. This is also consistent with thepossibility that the mutual fund industry positions certainproducts to exploit particularly biased individuals.

In unreported results, we examine the performancedifferences among investors who use index funds. We donot find significant associations between the performanceof individual index fund portfolios and individual beha-vioral biases. We consider different types of tests, includ-ing univariate sorts and multivariate regressions with andwithout controls or interaction terms. All our resultsconsistently show that behavioral biases do not affectthe performance of investors’ index fund portfolios. Thisevidence indicates that investors can protect themselvesfrom their own worst impulses by holding index funds andreinforces the classic intuition that most individual inves-tors perform better if they stick to well-diversified indexfunds. Our findings also echo Korniotis and Kumar(forthcoming) who show that the performance differencebetween smart and dumb investors is insignificant whenboth hold well-diversified stock portfolios, but it is highly

significant for those that choose concentrated portfolios,with smart investors outperforming by a wide margin.25

4.8. Aggregating the behavioral bias proxies and

other characteristics

Next, we measure the combined effects of investorcharacteristics using both the factors constructed fromthe behavioral bias proxies and other investor character-istics and an equally weighted index that combines thebehavioral bias proxies. Panel A of Table 9 summarizesregressions similar to those of Tables 4–7 but replaces theindividual investor characteristics with the first fivefactors resulting from factor analysis described inSection 4.1.

The first two columns study the first factor, which wepreviously labeled Gambler. The evidence in the tableconfirms this characterization. Gambler represents indi-viduals who are less likely to use mutual funds, tend toselect high expense funds, are more likely to trend-chase,and suffer significantly inferior mutual fund portfolioperformance as a consequence. Put another way, Gambleremploys mutual funds less than he probably should, but,when he does, he makes poor use of them.

We previously identified the second factor as Smart,given that the individual stock portfolio of this stereotypeavoids biases and displays relatively good performance.The evidence in Panel A of Table 9 suggests that Smart’sbeneficial behavior extends to his use of mutual funds.The signs and significance of regression coefficients indi-cate that the Smart stereotype is more likely to use

Table 8Behavioral biases, mutual fund portfolio characteristics, and portfolio performance.

This table reports cross-sectional regression estimates with two different mutual fund portfolio performance measures as dependent variables: (1) the

mean net monthly percent return and (2) the Sharpe ratio computed using net returns multiplied by 100. There is one observation per investor. The

independent variables are behavioral bias proxies and inattention measures, mutual fund characteristics, bias-load interaction terms, and control

variables. Independent variables are defined in Appendix A. Mutual fund characteristics include the initial weight assigned to mutual funds in the equity

(stocks and mutual funds) portfolio and three expense measures of the mutual fund portfolio: the sample period mean expense ratio, the sample period

mean 12-B-1 fee, and the sample period mean front-end load. Bias-load interaction terms equal the multiplication between each of three behavioral bias

measures and each of three mutual fund expense ratio measures. The three behavioral bias measures are high disposition effect, strong framing effects,

and overconfidence. The inattention measure is the equally weighted average of the two stock-level inattention measures. The three expense ratio

measures are high expense ratios, high 12-B-1 fees, and high front-end loads. The mutual fund portfolio weight is measured at the time an investor enters

the sample or invests in mutual funds for the first time. High and low dummy variables are defined using the highest and the lowest quintile of the

respective variable. Investors with fewer than 12 months of data are excluded. Zip code clustered standard errors are used to obtain the t-statistics. The

statistically significant coefficient estimates are indicated in bold font. The individual investor data are from a large US discount brokerage house for the

1991–1996 period.

(1) Net monthly return (2) Net Sharpe ratio � 100

Independent Variable Coefficient t-statistic Coefficient t-statistic

Intercept 1.320 13.14 40.156 21.80

Behavioral Bias ProxiesDisposition Effect �0.025 �2.01 �0.556 �1.98

Narrow Framing �0.043 �2.90 �1.565 �3.67

Overconfidence Dummy �0.021 �2.00 �1.446 �2.24

Male Dummy �0.010 �1.12 �1.498 �1.81

Local Bias 0.009 0.56 0.100 0.22

Lottery Stocks Preference �0.033 �3.11 �1.301 �1.99

Inattention to Earnings News �0.015 �2.11 �1.114 �2.34

Inattention to Macroeconomic News �0.025 �2.34 �0.989 �2.00

DEnHigh Income �0.011 �1.55 0.101 0.16

DEnNo December Tax Loss Selling �0.015 �1.52 �0.151 �0.59

Mutual Fund Portfolio CharacteristicsInitial Weight in Mutual Funds 0.044 3.71 1.721 3.65

Mutual Fund Portfolio Expense Ratio �0.010 �0.74 �0.730 �3.44

Mutual Fund Portfolio 12-B-1 Fee 0.051 3.55 �2.234 �4.36

Mutual Fund Portfolio Front-End Load �0.048 �3.97 �2.142 �5.02

Bias-load interaction terms

High Disposition EffectnHigh Expense Ratio �0.004 �0.40 �0.487 �1.49

High Disposition EffectnHigh 12-B-1 Fee �0.025 �2.99 �2.356 �7.15

High Disposition EffectnHigh Front-End Load �0.054 �5.09 �2.381 �8.58

Strong Framing EffectsnHigh Expesne Ratio �0.008 �0.86 �0.268 �0.79

Strong Framing EffectsnHigh 12-B-1 Fee �0.015 �1.91 �2.298 �6.93

Strong Framing EffectsnHigh Front-End Load �0.055 �7.09 �2.464 �8.71

OverconfidentnHigh Expense Ratio �0.006 �0.71 �0.544 �1.66

OverconfidentnHigh 12-B-1 Fee �0.024 �3.88 �2.433 �7.39

OverconfidentnHigh Front-End Load �0.052 �6.86 �2.312 �8.35

High InattentionnHigh Expense Ratio �0.017 �3.88 �1.119 �2.83

High InattentionnHigh 12-B-1 Fee �0.022 �2.46 �2.106 �4.73

High InattentionnHigh Front-End Load �0.063 �4.97 �0.927 �2.21

Control VariablesCoefficient estimates have been suppressed.

Adjusted R2 0.055 0.051

Number of Observations 21,542 21,542

W. Bailey et al. / Journal of Financial Economics 102 (2011) 1–27 19

mutual funds, more likely to use funds with low expenseratios or loads, less likely to trend-chase, and enjoyssignificantly positive mutual fund performance based onall eight of the performance measures we examine.

We previously labeled the third factor Overconfidentbased on trading of individual equities and other char-acteristics. The evidence on Overconfident’s mutual fundportfolio confirms our impression that this stereotypeis a poor decision-maker. Overconfident avoids participa-tion in mutual funds and trend-chases to an evengreater degree than Gambler, and he also tends to selecthigh expense, high load, and high turnover funds.Whether Overconfident’s mutual fund performance is

even worse than Gambler’s varies across our eight per-formance measures.

We labeled the fourth factor Narrow Framer. NarrowFramer’s mutual fund participation is about as bad asGambler’s, though not as bad as Overconfident’s. Smallholdings of mutual funds, selection of high expense funds,trend chasing, and consequent poor performance arealso evident, though milder than for Gambler andOverconfident.

Finally, the mutual fund use and performance repre-sented by the fifth factor, Mature, mirrors what wereported earlier for Mature’s individual stock portfolio.To Mature’s credit, he participates and holds mutual funds

27 Among 158,031 accounts in our sample there are 64,416 IRA and

1,299 Keogh accounts. A typical household holds multiple accounts. Out

W. Bailey et al. / Journal of Financial Economics 102 (2011) 1–2720

to a greater extent than our other stereotypes and avoidshigh-expense funds and trend chasing to an even greaterextent than Smart. However, other elements of Mature’sdecision making about mutual funds yield significantnegatives on four of our eight performance measures.This finding is consistent with the evidence in Korniotisand Kumar (2011), who show that older investors aremore likely to follow common investing rules but employthem less effectively and subsequently experience worseportfolio performance.

One interesting observation from Panel A of Table 9concerns the use of index funds. Unsurprisingly, Gambler,Overconfident, and Narrow Framer score negatively onboth index fund participation and holdings. Their lack ofinterest in these useful and prudent funds is consistentwith a pattern of bad decision making in their use of otherfunds and individual stocks. Mature seems to participatein index funds as frequently as Smart and holds an evengreater proportion of such funds than Smart. However,this is not enough to overcome Mature’s other decision-making problems and yield positive performance.

As an alternative to the five named factors from factoranalysis, Panel B of Table 9 presents similar results basedon an equally weighted behavioral index.26 Specifically,we normalize each behavioral factor to have a mean ofzero and a standard deviation of one, then average thesenormalized behavioral proxies for each individual in thesample. The table shows that in all cases the bias index isstatistically significant and, more important, economicallysignificant. In the discussion that follows, we infer thedecisions of investors in the lowest and the highest biasquintiles.

The average behavioral bias index values of investorsin the extreme bias quintiles are �0.709 and 0.627. Thestandard deviation of the behavioral bias measure is0.491, which indicates that the low and high behavioralbias quintiles are 2.721 standard deviations away fromeach other. In the participation regressions, the bias indexestimates indicate that an investor who moves from thelowest to highest bias quintile reduces the probability ofinvesting in mutual funds by �0.439�2.721¼1.189%,while the propensity to invest in index funds drops by1.933%. In the holdings regressions, we find that movingacross the extreme bias quintiles reduces the weightassigned to mutual funds by 2.038%. This effect is evenstronger (5.254%) for index funds. The other regressionssummarized in Panel B of Table 9 paint a similar picture.Behavioral biases are associated with selecting higherexpense funds, trend chasing with funds, and significantunder-performance from fund holdings. In economicterms, the combined effects of all behavioral biases aremoderately to strongly significant.

of 77,995 households in the sample, 43,706 hold at least one retirement

account.28 See Sialm and Starks (forthcoming) for evidence that funds

directed at taxable investors appear more tax-efficient than funds

directed at retirement accounts. This approach is distinct from our use

of the holds tax deferred account dummy in earlier regressions, which

identifies all accounts, regular or tax-deferred, held by someone who

5. Additional diagnostics

In this section, we discuss additional tests that aug-ment our main results by examining their robustness,

26 This includes the five basic biases and the two inattention

measures but excludes the two tax interactives.

considering alternative explanations for our findings, andoffering additional evidence on the most biased investors.

5.1. ‘‘Play money’’ accounts?

In our first set of additional considerations, we testwhether our results are driven primarily by a ‘‘playmoney’’ effect. We compute the average portfoliosize-to-annual income ratio for each investor, excludinginvestors in the lowest quintile. Unreported results indi-cate that our findings remain qualitatively similar evenwhen we exclude investors who hold portfolios thatare small relative to their annual income. For example,the coefficient estimate of the bias index in Table 4,Column 1 is �0.749 (t-statistic¼�5.49) for the fullsample and �0.755 (t-statistic¼�5.88) for the subsam-ple that excludes potential play money. This evidenceindicates that our results are unlikely to be induced by asubset of investors who maintain a small portfolio andtrade it for irrational or frivolous reasons.

5.2. Mutual fund decisions for retirement accounts

Many investors in our sample hold personal retire-ment accounts. About 42% of the accounts in our sampleare retirement accounts (IRA or Keogh).27 Thus, weexamine whether investors’ mutual fund choices varybetween retirement and non retirement accounts.28 It isplausible that the adverse effects of behavioral biases onmutual fund decisions are mainly concentrated in nonretirement accounts. We could view a retirement accountas the opposite of a play money account and predict thatit is managed in a more conservative manner. We define a‘‘taxable account only’’ dummy, which is set to zero forinvestors who hold only retirement accounts in theirequity portfolios and one otherwise. We include thisdummy variable as an additional independent variablein our regression specifications.29

We find that investors do not exhibit a greater pro-pensity to hold mutual funds in their retirement accounts.The taxable account only dummy has an insignificantcoefficient estimate (�0.003 with z-statistic of �0.25).No evidence exists of a stronger propensity to hold indexfunds for investors who hold retirement accounts. Thetaxable account only dummy has a coefficient estimate of�0.011 and z-statistic of �1.19. Even among investorswho choose to hold mutual funds, no evidence shows thatthey allocate a larger proportion of their equity portfolio

holds at least one tax-deferred account.29 All results are qualitatively similar when reestimated over two

subsamples: investors who hold only retirement accounts and investors

who hold retirement and non-retirement accounts.

Table 9Associations between aggregated behavioral biases and other characteristics, fund decisions, and consequences.

Panel A reports the combined effect of multiple bias proxies on mutual fund decisions using the five most important factors from factor analysis of the behavioral bias proxies and other investor

characteristics. Panel B reports the combined effect of multiple bias proxies on mutual fund decisions using an equally weighted index of the behavioral bias proxies. The behavioral factors are defined in

Appendix A and the factor analysis is detailed in Table 3. This table summarizes estimates of the regressions of Tables 4–7 in which the behavioral proxies and other investor characteristics are replaced with

the five most important factors from factor analysis. For brevity, only the coefficient estimates for the variable of interest are reported. PTBSD equals the post-trade buy–sell return differential.

Panel A: Estimates when the dependent variable is a factor of the behavioral bias proxies and other investor characteristics

Gambler factor Smart factor Overconfident factor Narrow Framer factor Mature factor Adjusted

R2

Number of

Observations

Regression Type Coefficient t- or z-

statistic

Coefficient t- or z-

statistic

Coefficient t- or z-

statistic

Coefficient t- or z-

statistic

Coefficient t- or z-

statistic

Participation (Table 4)

All mutual funds: column 2 �0.339 �3.77 0.125 2.24 �0.722 �3.75 �0.350 �2.73 0.258 3.11 0.059 21,542

Index funds only: column 4 �0.229 �2.93 0.171 2.59 �0.402 �2.68 �0.311 �2.60 0.174 2.81 0.051 21,542

Holdings (Table 4)

Weight in all mutual funds: column 5 �2.827 �3.02 1.764 1.42 �3.193 �3.75 �1.981 �2.73 2.541 3.55 0.049 21,542

Weight in index funds only: column 6 �1.901 �2.71 1.166 1.81 �2.792 �3.18 �1.591 �2.76 2.407 2.76 0.055 21,542

Portfolio characteristics (Table 5)

Expense ratio: column 1 0.209 5.15 �0.014 �2.88 0.079 4.46 0.027 2.13 �0.111 �5.42 0.038 21,542

Front-end load: column 2 0.132 2.72 �0.017 �2.15 0.082 2.21 0.063 1.91 �0.085 �3.11 0.031 21,542

Fund turnover: column 3 0.114 3.71 �0.029 �2.31 0.145 3.59 0.033 1.22 �0.148 �4.68 0.043 21,542

Trend chasing (Table 6)

12-Month lagged fund performance:

column 1

1.180 2.93 �0.096 �1.23 1.729 3.02 0.863 1.97 �1.367 �2.65 0.071 21,542

24-Month lagged fund performance:

column 2

1.156 2.61 �0.532 �2.33 1.941 2.98 1.167 2.57 �2.079 �3.14 0.050 21,542

Portfolio performance (Table 7)

Mean monthly returns: Panel A,

column 1

�0.109 �2.65 0.085 2.33 �0.111 �2.82 �0.066 �1.92 �0.025 �1.82 0.028 21,542

Net monthly returns: Panel A, Column 2 �0.122 �2.60 0.076 2.51 �0.189 �3.77 �0.058 �1.98 �0.019 �1.31 0.026 21,542

Net Sharpe ratio: Panel A, column 3 �2.568 �3.20 2.853 3.12 �1.110 �2.04 �2.109 �3.08 0.664 1.12 0.024 21,542

Four-factor alpha: Panel A, Column 4 �0.164 �2.84 0.123 2.34 �0.092 �2.78 �0.058 �2.88 �0.026 �1.26 0.021 21,542

Holding period returns: Panel B,

column 1

�0.095 �3.08 0.059 2.42 �0.074 �2.90 �0.078 �2.94 �0.055 �2.40 0.033 15,210

Holding period: Panel B, column 2 �27.904 �3.07 18.514 2.38 �15.504 �2.27 �8.211 �2.58 21.675 �2.99 0.065 15,210

One-month PTBSD: Panel B, column 3 �0.152 �3.49 0.125 3.48 �0.093 �3.36 �0.051 �2.77 �0.076 �3.21 0.025 15,210

One-year PTBSD: Panel B, column 4 �0.916 �3.37 0.936 3.70 �0.691 �3.28 �0.544 �2.82 �0.722 �2.87 0.050 15,210

Panel B: Estimates when dependent variable is equally weighted index of behavioral bias proxies

Regression Type Coefficient t- or z-statistic Adjusted R2 Number of Observations

Participation (Table 4)

All mutual funds: column 2 �0.439 �7.11 0.033 21,542

Index funds only: column 4 �0.719 �7.41 0.065 21,542

Holdings (Table 4)

Weight in all mutual funds: column 5 �0.744 �5.44 0.068 21,542

Weight in index funds only: column 6 �1.933 �4.72 0.142 21,542

Portfolio characteristics (Table 5)

Expense ratio: column 1 0.032 4.13 0.055 21,542

W.

Ba

ileyet

al.

/Jo

urn

al

of

Fina

ncia

lE

con

om

ics1

02

(20

11

)1

–2

72

1

Ta

ble

9(c

on

tin

ued

)

Pa

nel

B:

Est

ima

tes

wh

end

epen

den

tv

ari

ab

leis

equ

all

yw

eig

hte

din

dex

of

beh

av

iora

lb

ias

pro

xies

Re

gre

ssio

nT

yp

eC

oe

ffici

en

tt-

or

z-st

ati

stic

Ad

just

ed

R2

Nu

mb

er

of

Ob

serv

ati

on

s

Fro

nt-

en

dlo

ad

:co

lum

n2

0.0

33

3.5

50

.04

42

1,5

42

Fun

dtu

rno

ve

r:co

lum

n3

0.0

16

2.0

10

.05

32

1,5

42

Tre

nd

cha

sin

g(T

ab

le6

)

12

mo

nth

lag

ge

dfu

nd

pe

rfo

rma

nce

:co

lum

n1

1.4

41

4.9

00

.08

32

1,5

42

24

mo

nth

lag

ge

dfu

nd

pe

rfo

rma

nce

:co

lum

n2

1.2

76

3.5

50

.06

52

1,5

42

Po

rtfo

lio

pe

rfo

rma

nce

(Ta

ble

7)

Me

an

mo

nth

lyre

turn

s:P

an

el

A,

colu

mn

1�

0.0

52

�3

.71

0.0

38

21

,54

2

Ne

tm

on

thly

retu

rns:

Pa

ne

lA

,co

lum

n2

�0

.06

2�

3.4

70

.04

22

1,5

42

Ne

tS

ha

rpe

rati

o:

Pa

ne

lA

,co

lum

n3

�2

.49

9�

3.8

50

.03

12

1,5

42

Fou

r-fa

cto

ra

lph

a:

Pa

ne

lA

,C

olu

mn

4�

0.0

55

�3

.39

0.0

28

20

,14

2

Ho

ldin

gp

eri

od

retu

rns:

Pa

ne

lB

,C

olu

mn

1�

0.0

63

�5

.16

0.0

33

15

,21

0

Ho

ldin

gp

eri

od

:P

an

el

B,

Co

lum

n2

�2

1.1

75

�4

.44

0.0

80

15

,21

0

On

e-m

on

thP

TB

SD

:P

an

el

B,

colu

mn

3�

0.3

81

�4

.12

0.0

29

15

,21

0

On

e-y

ea

rP

TB

SD

:P

an

el

B,

colu

mn

4�

0.6

22

�4

.23

0.0

48

15

,21

0

W. Bailey et al. / Journal of Financial Economics 102 (2011) 1–2722

to mutual funds. The taxable account only dummy hasstatistically insignificant estimates in all specifications.

Examining the characteristics of funds in the portfoliosof investors who hold only retirement accounts, we findthat they do not have lower expense ratios, lower frontend loads, or lower turnover. Moreover, there is a greatertendency to engage in trend chasing among these inves-tors. When we reestimate the trend chasing regressions ofTable 6 with the taxable account only dummy variable, ithas a significantly positive coefficient estimate (coeffi-cient estimate¼0.029, t-statistic¼2.99).

To examine whether ‘‘retirement accounts only’’ inves-tors exhibit better performance, we reestimate all theperformance regressions with the taxable accountdummy as an additional independent variable. In allspecifications, this dummy variable has an insignificantcoefficient estimate. Overall, we do not find evidence ofsuperior mutual fund decisions when investors holdretirement accounts. The adverse effects of behavioralbiases on mutual fund decisions are similar across bothretirement and non retirement accounts. Thus, behavio-rally biased investors do not manage retirement fundsany more carefully than their regular accounts.

5.3. How do the most severely biased investors use

mutual funds?

Next, we consider whether the most severely behavio-rally biased investors tend to concentrate in particulartypes of funds, how often they trade those funds, andwhat consequences for performance result. We summar-ize unreported (but available on request) evidence onholdings, holding periods, and returns for the mutualfunds owned by quintiles of investors who score higheston disposition effect, narrow framing, overconfidence,local bias, preference for lottery stocks, and inattentionto news. Our primary prediction is that severely biasedinvestors are more likely to select higher expense fundsand avoid index funds. We also expect the strongestDisposition Effect and Overconfidence Dummy investorsto turn their mutual fund holdings over relativelyfrequently.

There is much evidence to support such conjectures.For example, front-load funds are 27.15% of the mutualfund holdings of typical investors, but we observe statis-tically significantly greater front-load fund holdings forthe highest Disposition Effect (31.05%), Narrow Framing

(26.69%), and Overconfidence Dummy (30.81%) cohorts.The mutual fund holdings of the highest Local Bias andInattention Bias investors have, on average, about 2% lessfront load funds than typical investors. Holding periodsfor front end load funds are, on average, significantly lowfor highest Disposition Effect (215 days) and Overconfi-

dence Dummy (233 days) investors and are significantlyhigh for highest Narrow Framing (306 days), Local Bias

(323 days), and Inattention Bias (327 day) investors.Somewhat similar, but weaker, results are observed forholdings of back-end load funds and in comparing hold-ings of index funds and other funds.

Table A1Brief description of behavioral proxies and other investor characteristics.

Variable Description References Calculation

Disposition Effect Investor’s propensity to sell

winners too early and hold losers

too long. Measured by the

proportion of gains realized

minus proportion of losses

realized, adjusted for the peer

group’s disposition effect.

Shefrin and Statman

(1985), Odean (1998), and

Kumar and Lim (2008).

Proportion of gains realized (PGR)¼realized

gains/(realized gainsþpaper gains). Proportion

of losses realized (PLR)¼realized losses/

(realized lossesþpaper losses). A peer group of

an investor is defined as those in the same

quintile of portfolio size, trading frequency and

number of stocks. Adjusted PGR¼PGR of an

investor – mean PGR of peer group. Adjusted

PLR¼PLR of an investor – mean PLR in her peer

group. Adjusted disposition effect¼adjusted

PGR – adjusted PLR.

Narrow Framing Investor’s propensity to select

investments individually instead

of considering the broad impact

on her portfolio. Measured by the

degree of trade clustering,

adjusted for the peer group’s

framing propensity.

Kahneman and Lovallo

(1993), Kahneman (2003),

and Kumar and Lim

(2008).

Trade clustering¼1 – (number of trades/number

of trading days). A peer group of an investor is

defined as those in the same quintile of portfolio

size, trading frequency, and number of stocks.

Adjusted trade clustering¼trade clustering –

mean trade clustering of the peer group.

Overconfidence Investor’s propensity to trade

frequently but unsuccessfully.

Measured with a dummy

variable.

Odean (1999) and Barber

and Odean (2001).

Dummy variable equal to one for investors in

the highest portfolio turnover quintile and

lowest performance quintile for their individual

common stock trading and zero otherwise. Also

captured by a gender dummy variable equal to

one if the investor is male.

Local Bias Investor’s propensity to select

stocks with headquarters close to

his geographical location.

Huberman (2001), Coval

and Moskowitz (1999),

Grinblatt and Keloharju

(2001), Zhu (2003), and

Ivkovic and Weisbenner

(2005).

Local bias of an investor’s common stock

portfolio¼mean distance between her home zip

code and the headquarters’ zip codes of

companies in her portfolio – mean distance

between home zip code and the headquarters’

zip codes of companies in the market portfolio.

Lottery Stock Preference Investor’s propensity to select

stocks with lottery-like features

(low price, volatile returns, and

skewed returns).

Barberis and Huang

(2008), and Kumar (2009).

Investor’s mean portfolio weight (relative to the

weight in the market portfolio) assigned to

stocks that have bottom quintile prices, top

quintile idiosyncratic volatility, and top quintile

idiosyncratic skewness.

Inattention to Earnings

News

Degree to which investor does

not trade a particular individual

stock around earnings news.

New in this paper. 1 – (number of investor trades around the

event)/(total number of investor trades) on days

t–1, t, and tþ1 where t is the date of quarterly

earnings announcement from the Institutional

Brokers’ Estimate System (I/B/E/S). Only trades

around each firm’s own earnings news are

considered.

Inattention to

Macroeconomic News

Degree to which investor does

not trade any individual stocks

around macroeconomic news

events.

New in this paper. 1 – (number of investor trades around the

event)/(total number of investor trades) on days

t–1, t, and tþ1 where t is the date of federal

funds target rate changes, Non-Farm Payroll

reports, and producer price index

announcements.

Fund-Level Local Bias Investor’s propensity to select

funds with headquarters close to

his geographical location.

New in this paper. Equals mean distance between the investor’s

home zip code and the headquarters of the

mutual funds in his portfolio – the same mean

distance averaged across all investors in the

sample.

Fund Level Inattention Individual’s propensity to trade

mutual funds around

macroeconomic news events.

New in this paper. Equals 1 - (number of mutual fund trades

around the event)/(total number of mutual fund

trades).

DEnNo December Tax

Loss Selling

Extent of Disposition Effect for

investor who ignores tax loss

selling.

New in this paper. Disposition Effect times dummy variable equal to

one for investor with no December tax loss

selling.

DEnHigh Income Extent of Disposition Effect for

investor with high income.

New in this paper. Disposition Effect times High Income Dummy.

Age Age of the investor. Self-reported. Age of the investor.

Income Income of the investor. Self-reported. Annual income of investor.

High Income Dummy Affluence of the household Graham and Kumar

(2006).

Equals one if the investor’s average income

exceeds $125,000 and zero otherwise.

Marital Status Marital status of the investor. Self-reported. Equals one if the investor is married and zero

otherwise.

Family Size Family size. Self-reported. Number of family members in the household.

Professional Dummy Self-reported.

W. Bailey et al. / Journal of Financial Economics 102 (2011) 1–27 23

Table A1 (continued )

Variable Description References Calculation

A indicator whether an investor

is a white collar or blue collar

worker.

Equals zero for investor in a blue collar

profession and one otherwise.

Retired Dummy Retirement status of investor. Self-reported. Equals one if the investor is retired and zero

otherwise.

Investment Experience Investment experience of

investor.

Self-reported. Years since the brokerage account was open.

Financial Center Dummy An indicator whether an investor

lives near a financial center.

Based on self reported

address.

Equals one if the zip code of the investor’s

address is close to a metropolitan area and zero

otherwise.

Options Dummy An indicator for whether the

investor has ever traded an

option in the investment account.

Based on investment

record.

Equals one if the investor executes at least one

option trade during the sample period and zero

otherwise.

Short Sale Dummy An indicator for whether the

investor has ever shorted a stock

in the investment account.

Based on investment

record.

Equals one if the investor executes at least one

short trade during the sample period and zero

otherwise.

Stock Portfolio

Diversification

The extent to which the stock

portfolio of the investor is

diversified.

Based on investment

record.

Negative of Normalized Portfolio Variance, that is,

the variance of the portfolio of individual

domestic securities divided by the average

variance of the individual common stocks in the

portfolio.

Stock Portfolio Size The size of the investor’s

portfolio.

Based on investment

record.

Sample-period average market capitalization of

the investor’s common stock portfolio.

Stock Portfolio

Performance

Risk-adjusted excess returns of

the investor’s stock portfolios.

Based on investment

record.

The intercept, alpha, from the Capital Asset

Pricing Model regression with the monthly

common stock portfolio return as dependent

variable.

No December Tax Loss

Selling

An indicator if the investor fails

to realized losses of his stock

trade in December

Based on investment

record.

1– proportion of realized losses in December¼1 –

(realized losses in December/number of paper

losses)

Holds Tax-Deferred

Account

An indicator for whether the

investor holds a tax deferred

account in the brokerage.

Based on investment

record.

Equals one if the investor holds an Individual

Retirement Account (IRA) or Keogh account in

the brokerage.

Stock Portfolio Market

Factor (Beta) Exposure

The beta of the investor’s stock

portfolio.

Based on investment

record.

The loading of the stock portfolio on the market

factor in a four-factor regression model with

market, size, value, and momentum factors. All

four factors come from Ken French’s website,

(mba.tuck.dartmouth.edu/pages/faculty/

ken.french/).

Stock Portfolio SMB

Factor (Size) Exposure

The loading of the stock portfolio

on the small-minus-big factor

(SMB) in a four-factor model

regression.

Based on investment

record.

The loading of the stock portfolio on the size

(SMB) factor in a four-factor regression model

with market, size, value, and momentum

factors. All four factors come from Ken French’s

website.

Stock Portfolio HML

Factor (Value) Exposure

The loading of the stock portfolio

on the high-minus-low book-to-

market factor (HML) in a four-

factor model regression.

Based on investment

record.

The loading of the stock portfolio on the value

(HML) factor in a four-factor regression model

with market, size, value, and momentum

factors. All four factors come from Ken French’s

website.

Stock Portfolio UMD

Factor (Momentum)

Exposure

The loading of the stock portfolio

on the up-minus-down factor

(UMD) in a four-factor model

regression.

Based on investment

record.

The loading of the stock portfolio on the

momentum (UMD) factor in a four-factor

regression model with market, size, value, and

momentum factors. All four factors come from

Ken French’s website.

W. Bailey et al. / Journal of Financial Economics 102 (2011) 1–2724

6. Summary and conclusions

Using thousands of brokerage accounts of US indivi-dual investors, we show that behavioral factors influencethe decisions of individual investors to hold individualstocks as opposed to mutual funds, including passiveindex funds. As might be expected, investors with higherincome, relatively higher educational level, and greaterinvestment experience are more likely to use mutualfunds and benefit from their choices. However, investors

with strong behavioral biases tend to gravitate towardindividual stocks and avoid low expense index funds.When they do invest in mutual funds, they tend toselect high expense funds, trade funds frequently,avoid index funds, and time their buys and sells poorly,thereby damaging their portfolio’s performance. Theyalso exhibit stronger trend-chasing behavior, suggestingthat trend chasing by mutual fund investors is not theresult of rationally inferring managerial skill from pastperformance.

Table A2Univariate summary statistics on investor characteristics (21,542 observations).

Variable Mean Standard

deviation

Minimum 10th

percentile

25th

percentile

Median 75th

percentile

90th

percentile

Maximum

Disposition Effect 3.719 112.197 �100.00 �100.00 �11.111 12.609 66.667 100.000 100.000

Narrow Framing 0.010 0.155 �0.683 �0.207 �0.081 0.038 0.131 0.181 0.440

Overconfidence Dummy 0.090 0.287 0.000 0.000 0.000 0.000 0.000 0.000 1.000

Male Dummy 0.898 0.282 0.000 0.899 1.000 1.000 1.000 1.000 1.000

Local Bias 0.273 0.395 �1.323 �0.204 0.058 0.272 0.542 0.773 0.996

Lottery Stocks Preference 12.025 17.206 0.000 0.000 0.000 4.265 18.510 33.644 100.000

Inattention to Earnings News 0.057 0.061 0.000 0.000 0.000 0.048 0.087 0.133 0.500

Inattention to Macroeconomic

News

0.301 0.143 0.000 0.133 0.214 0.292 0.375 0.476 1.000

Fund Level Local Bias 0.000 0.703 �1.249 �0.854 �0.468 �0.97 0.394 1.015 4.171

Fund Level Inattention 0.303 0.107 0.000 0.250 0.304 0.304 0.304 0.333 1.000

Age 50.429 11.537 18.000 36.000 42.000 52.000 56.000 68.000 94.000

Income 89.358 60.381 7.500 35.000 62.500 87.500 112.500 250.000 250.000

High Income Dummy 0.241 0.427 0.000 0.000 0.000 0.000 0.000 1.000 1.000

Marital Status 0.736 0.386 0.000 0.000 0.736 1.000 1.000 1.000 1.000

Family Size 2.814 1.417 1.000 1.000 2.000 3.000 4.000 5.000 10.000

Professional Dummy 0.610 0.336 0.000 0.000 0.610 1.000 1.000 1.000 1.000

Retired Dummy 0.166 0.256 0.000 0.000 0.000 0.000 0.166 0.166 1.000

Investment Experience 9.809 3.190 5.255 5.880 6.915 9.630 12.019 13.964 22.373

Financial Center Dummy 0.327 0.469 0.000 0.000 0.000 0.000 1.000 1.000 1.000

Options Dummy 0.124 0.330 0.000 0.000 0.000 0.000 0.000 1.000 1.000

Short Sale Dummy 0.138 0.345 0.000 0.000 0.000 0.000 0.000 1.000 1.000

Stock Portfolio Diversification �0.422 0.135 �0.966 �0.598 �0.514 �0.422 �0.323 �0.245 0.000

Stock Portfolio Size 36.410 98.119 0.001 4.255 7.824 15.326 32.277 71.899 4079.582

Ln(Stock Portfolio Size) 2.797 1.159 �7.082 1.448 2.057 2.729 3.474 4.275 8.314

Stock Portfolio Performance �0.378 1.460 �11.474 �2.111 �1.116 �0.278 0.468 1.253 6.437

No December Tax Loss Selling 0.818 0.386 0.000 0.000 1.000 1.000 1.000 1.000 1.000

Holds Tax-Deferred Account 0.490 0.500 0.000 0.000 0.000 0.000 1.000 1.000 1.000

Market Factor Exposure 1.196 0.557 �1.911 0.555 0.850 1.157 1.521 1.895 3.901

SMB Factor Exposure 0.853 1.028 �2.163 �0.268 0.098 0.675 1.410 2.257 7.810

HML Factor Exposure 0.182 0.838 �3.258 �0.797 �0.359 0.119 0.647 1.269 5.279

UMD Factor Exposure �0.331 0.667 �3.898 �1.182 �0.704 �0.267 0.089 0.410 2.986

W. Bailey et al. / Journal of Financial Economics 102 (2011) 1–27 25

When we use factor analysis to characterize associa-tions among investor characteristics, we find interestingand intuitive patterns along multiple dimensions of biasand other characteristics that often crop up in the sameindividual. There is consistency across the behavioralbiases, other characteristics, use of individual stocks, useof mutual funds, and resultant performance that ourGambler, Smart, Overconfident, Narrow Framer, andMature stereotypes display.

Our evidence on behavioral biases and mutual fundclienteles provides a new perspective on puzzles inmutual fund investment presented by previous authors.Several authors trace the mutual fund decisions of indi-vidual investors to such factors as excess focus on front-end loads, advertising, search costs, and complexity offund features intended to exploit consumers.30 Our evi-dence shows that investors who score high on behavioralbiases tend to invest in funds with higher expense ratiosand loads. They experience poor investment performanceas a result.

In his American Finance Association presidentialaddress, Martin Gruber (1996) notes several puzzling

30 See Barber, Odean and Zheng (2005), Hortacsu and Syverson

(2004), and Carlin (2008).

aspects of individual portfolio allocation decisions. Hespeaks of ‘‘sophisticated’’ investors who make decisionsbased on performance and ‘‘disadvantaged’’ investors whoare susceptible to sales pressure or constrained by tax orinstitutional issues. In his presidential address, JohnCampbell (2006) suggests that naıve investors couldsubsidize sophisticated investors in financial productssuch as mortgages. Our results echo the spirit of theseideas. A complex set of factors, some rational and somebehavioral, appear to drive investors’ stocks versus fundsdecisions and their mutual fund choices after they decideto invest in mutual funds. Some types of investors appearto make effective choices that enhance portfolio perfor-mance, while others do not.

Given the misuse of equity mutual funds, a publiccampaign to increase awareness of basic investmentprinciples and the benefits and pitfalls of equity mutualfunds is likely to help many types of individual investorsmake better decisions. Furthermore, the lack of attentionto low cost or index funds suggests more explicit dis-closure of fund expenses and turnover, perhaps even asprominent as the health warnings now displayed onpackets of cigarettes. Finally, the reliance of mutual fundinvestors on broker-supplied information at the time afund is selected and on delegated investment decisionsafterward suggests that even more explicit disclosure of

W. Bailey et al. / Journal of Financial Economics 102 (2011) 1–2726

fund characteristics be imposed on brokerage firms andfund managers.

Appendix A

Descriptions of behavioral proxies and other investorcharacteristics can be found in Tables A1. Summarystatistics on investor characteristics can be found inTables A2.

References

Bailey, W., Kumar, A., Ng, D., 2008. Foreign investments of US individualinvestors: causes and consequences. Management Science 54,443–459.

Barber, B.M., Odean, T., 2000. Trading is hazardous to your wealth: thecommon stock investment performance of individual investors.Journal of Finance 55, 773–806.

Barber, B.M., Odean, T., 2001. Boys will be boys: gender, over-confidence,and common stock investment. Quarterly Journal of Economics 116,261–292.

Barber, B.M., Odean, T., 2008. All that glitters: the effect of attention andnews on the buying behavior of individual and institutional inves-tors. Review of Financial Studies 21, 785–818.

Barber, B.M., Odean, T., Zheng, L., 2005. Out of sight, out of mind: theeffects of expenses on mutual fund flows. Journal of Business 78,2095–2120.

Barberis, N., Huang, M., 2008. Stocks as lotteries: the implications ofprobability weighting for security prices. American EconomicReview 98, 2066–2100.

Benartzi, S., Thaler, R.H., 2007. Heuristics and biases in retirementsavings behavior. Journal of Economic Perspectives 21, 81–104.

Bergstresser, D., Chalmers, J., Tufano, P., 2009. Assessing the costs andbenefits of brokers in the mutual fund industry. Review of FinancialStudies 22, 4129–4156.

Bergstresser, D., Poterba, J., 2002. Do after-tax returns affect mutual fundinflows? Journal of Financial Economics 63, 381–414.

Berk, J., Green, R., 2004. Mutual fund flows and performance in rationalmarkets. Journal of Political Economy 112, 1269–1295.

Campbell, J.Y., 2006. Household finance. Journal of Finance 61,1553–1604.

Capon, N., Fitzsimons, G.J., Prince, R., 1996. An individual level analysisof the mutual fund investment decision. Journal of Financial ServicesResearch 10, 59–82.

Carlin, B.I., 2008. Strategic price complexity in retail financial markets.Journal of Financial Economics 91, 278–287.

Chevalier, J., Ellison, G., 1997. Risk taking by mutual funds as a responseto incentives. Journal of Political Economy 105, 1167–1200.

Choi, J., Laibson, D., Madrian, B., 2010. Why does the law of one pricefail? An experiment on index mutual funds. Review of FinancialStudies 23, 1405–1432.

Christoffersen, S., Musto, D., 2002. Demand curves and the pricing ofmoney management. Review of Financial Studies 15, 1499–1524.

Christoffersen, S., Sarkissian, S., 2009. City size and fund performance.Journal of Financial Economics 92, 252–275.

Christoffersen, S., Evans, R., Musto, D., 2006. The economics of mutualfund brokerage: evidence from cross section of investment channels.Working Paper, University of Pennsylvania, Wharton School,Philadelphia, PA, unpublished.

Coval, J.D., Moskowitz, T., 1999. Home bias at home: local equitypreference in domestic portfolio. Journal of Finance 54, 2045–2073.

Coval, J.D., Stafford, E., 2007. Asset fire sales (and purchases) in equitymarkets. Journal of Financial Economics 86, 479–512.

Del Guercio, D., Tkac, P., 2008. Star power: the effect of Morningstarratings on mutual fund flows. Journal of Financial and QuantitativeAnalysis 43, 907–936.

Elton, E.J., Gruber, M.J., Busse, J.A., 2004. Are investors rational? Choicesamong index funds. Journal of Finance 59, 261–288.

Frazzini, A., Lamont, O., 2008. Dumb money: mutual fund flows and thecross-section of stock returns. Journal of Financial Economics 88,299–322.

French, K.R., 2008. The cost of active investing. Journal of Finance 63,1537–1574.

Gil-Bazo, J., Ruiz-Verdu, P., 2009. The relations between price andperformance in the mutual fund industry. Journal of Finance 64,2153–2184.

Goetzmann, W.N., Kumar, A., 2008. Equity portfolio diversification.Review of Finance 12, 433–463.

Graham, J., Kumar, A., 2006. Do dividend clienteles exist? Evidence ondividend preferences of retail investors. Journal of Finance 61,1305–1336.

Grinblatt, M., Keloharju, M., 2001. How distance, language, andculture influence stockholdings and trade. Journal of Finance 56,1053–1073.

Gruber, M.J., 1996. Another puzzle: the growth in actively managedmutual funds. Journal of Finance 51, 783–810.

Hortacsu, A., Syverson, C., 2004. Product differentiation, search costs,and competition in the mutual fund industry: a case study of S&P500 index funds. Quarterly Journal of Economics 119, 457–488.

Huang, J., Wei, K.D., Yan, H., 2007. Participation costs and the sensitivityof fund flows to past performance. Journal of Finance 63, 1273–1311.

Huberman, G., 2001. Familiarity breeds investment. Review of FinancialStudies 14, 659–680.

Ivkovic, Z., Poterba, J., Weisbenner, S., 2005. Tax-motivated trading byindividual investors. American Economic Review 95, 1605–1630.

Ivkovic, Z., Sialm, C., Weisbenner, S., 2008. Portfolio concentration andthe performance of individual investors. Journal of Financial andQuantitative Analysis 43, 613–656.

Ivkovic, Z., Weisbenner, S., 2005. Local does as local is: Informationcontent of the geography of individual investors’ common stockinvestments. Journal of Finance 60, 267–306.

Ivkovic, Z., Weisbenner, S., 2009. Individual investor mutual fund flows.Journal of Financial Economics 92, 223–237.

Kahneman, D., 2003. Maps of bounded rationality: Psychology forbehavioral economics. American Economic Review 93, 1449–1475.

Kahneman, D., Lovallo, D., 1993. Timid choices and bold forecasts: acognitive perspective on risk taking. Management Science 39, 17–31.

Korniotis, G.M., Kumar, A., Do portfolio distortions reflect superiorinformation or psychological biases? Journal of Financial and Quan-titative Analysis, forthcoming.

Korniotis, G.M., Kumar, A., 2011. Do older investors make betterinvestment decisions? Review of Economics and Statistics 93,244–265.

Kumar, A., 2009. Who gambles in the stock market? Journal of Finance64, 1889–1933.

Kumar, A., Lee, C.M., 2006. Retail investor sentiment and return comove-ments. Journal of Finance 61, 2451–2486.

Kumar, A., Lim, S.S., 2008. How do decision frames influence the stockinvestment choices of individual investors? Management Science 54,1052–1064.

Malloy, C., Zhu, N., 2004. Mutual fund choices and investor demo-graphics. Working Paper, University of California, Davis, CA,unpublished.

Merton, R.C., 1987. A simple model of capital market equilibrium withincomplete information. Journal of Finance 42, 483–510.

Mitton, T., Vorkink, K., 2007. Equilibrium underdiversification and thepreference for skewness. Review of Financial Studies 20, 1255–1288.

Nanda, V., Wang, Z.J., Zheng, L., 2009. The ABCs of mutual funds: on theintroduction of multiple class shares. Journal of Financial Interme-diation 18, 329–361.

Odean, T., 1998. Are investors reluctant to realize their losses? Journal ofFinance 53, 1775–1798.

Odean, T., 1999. Do investors trade too much? American EconomicReview 89, 1279–1298.

Petersen, M.A., 2009. Estimating standard errors in finance panel datasets: comparing approaches. Review of Financial Studies 22,435–480.

Sapp, T., Tiwari, A., 2004. Does stock return momentum explain the‘smart money’ effect? Journal of Finance 59, 2605–2622.

Sensoy, B., 2009. Performance evaluation and self designated benchmarkindexes in the mutual fund industry. Journal of Financial Economics92, 25–39.

Sialm, C., Starks, L. Mutual fund tax clienteles. Journal of Finance,forthcoming.

Sirri, E., Tufano, P., 1998. Costly search and mutual fund flows. Journal ofFinance 53, 1589–1622.

Shefrin, H., Statman, M., 2000. Behavioral portfolio theory. Journal ofFinancial and Quantitative Analysis 35, 127–151.

Shefrin, H., Statman, M., 1985. The disposition to sell winners too earlyand rise losers too long. Journal of Finance 40, 777–790.

Starks, L.T., Yates, M., 2008. Reputation and mutual fund choice. WorkingPaper (June), University of Texas at Austin.

W. Bailey et al. / Journal of Financial Economics 102 (2011) 1–27 27

Van Nieuwerburgh, S., Veldkamp, L., 2010. Information acquisition andunderdiversification. Review of Economic Studies 77, 779–805.

Wooldridge, J., 2003. Introductory Econometrics: A Modern Approach.Thompson Southwestern, Mason, OH.

Zheng, L., 1999. Is money smart? A study of mutual fund investor’s fundselection ability. Journal of Finance 54, 901–932.

Zhu, N., 2003. The local bias of individual investors. Working Paper,University of California at Davis.

Zhu, N., 2005. Search costs and household choice between direct anddelegated portfolio management. Working Paper, University of

California at Davis.