THE JOURNAL OF FINANCE • VOL. LXIV, NO.

4 • AUGUST 2009

Who Gambles in the Stock Market?

ALOK KUMAR∗

ABSTRACT
This study shows that the propensity to gamble and investment decisions are cor-
related. At the aggregate level, individual investors prefer stocks with lottery fea-
tures, and like lottery demand, the demand for lottery-type stocks increases during
economic downturns. In the cross-section, socioeconomic factors that induce greater
expenditure in lotteries are associated with greater investment in lottery-type stocks.
Further, lottery investment levels are higher in regions with favorable lottery envi-
ronments. Because lottery-type stocks underperform, gambling-related underperfor-
mance is greater among low-income investors who excessively overweight lottery-type
stocks. These results indicate that state lotteries and lottery-type stocks attract very
similar socioeconomic clienteles.

THE DESIRE TO GAMBLE IS DEEP-ROOTED in the human psyche. This fascination with
games of chance can be traced back at least a few centuries. A complex set of bio-
logical, psychological, religious, and socioeconomic factors jointly determines an
individual’s propensity to gamble (e.g., France (1902), Brenner (1983), Walker
(1992)). In this study, I investigate the extent to which people’s overall attitudes
toward gambling inf luence their stock investment decisions.
Previous studies have emphasized the potential role of gambling in invest-
ment decisions (e.g., Friedman and Savage (1948), Markowitz (1952), Shiller
(1989, 2000), Shefrin and Statman (2000), Statman (2002), Barberis and Huang
∗ Alok Kumar is at the McCombs School of Business, University of Texas at Austin. I would like
to thank two anonymous referees; an anonymous associate editor; Lucy Ackert; Warren Bailey;
Brad Barber; Nick Barberis; Robert Battalio; Garrick Blalock; Markus Brunnermeier; Sudheer
Chava; Vidhi Chhaochharia; Lauren Cohen; Shane Corwin; Josh Coval; Henrik Cronqvist; Steve
Figlewski; Margaret Forster; Amit Goyal; Bing Han; Cam Harvey (the editor); David Hirshleifer;
Scott Irwin; Narasimhan Jegadeesh; Danling Jiang; George Korniotis; Lisa Kramer; Charles Lee;
Chris Malloy; Bill McDonald; Victor McGee; Stefan Nagel; Terrance Odean; Jerry Parwada; Allen
Poteshman; Stefan Ruenzi; Kevin Scanlon; Paul Schultz; Mark Seasholes; Devin Shanthikumar;
Bob Shiller; Sophie Shive; Kent Womack; Jeff Wurgler; Wei Xiong; Lei Yu; Eduardo Zambrano;
Ning Zhu; and seminar participants at the Spring 2005 NBER Behavioral Finance Group Meeting,
University of Notre Dame, 2005 EFA Meeting, 2006 AFA Meeting, Ohio State University, University
of Texas at Austin, University of California at Los Angeles, Tuck School at Dartmouth, Columbia
University, and University of North Carolina at Chapel Hill for helpful discussions and valuable
comments. In addition, I would like to thank Nick Crain, Jeremy Page, and Margaret Zhu for
excellent research assistance; Itamar Simonson for making the investor data available to me;
Brad Barber and Terrance Odean for answering numerous questions about the investor database;
and Garrick Blalock for providing the state lottery expenditure data. I am grateful to Thomson
Financial for access to its Institutional Brokers Estimate System (I/B/E/S), provided as part of a
broad academic program to encourage earnings expectations research. Of course, I am responsible
for all remaining errors and omissions.

1889
1890 The Journal of FinanceR

(2008)). For instance, Markowitz (1952) conjectures that some investors might
prefer to “take large chances of a small loss for a small chance of a large gain.”
Barberis and Huang (2008) posit that investors might overweight low proba-
bility events and exhibit a preference for stocks with positive skewness.
In spite of its intuitive appeal, it has been difficult to gather direct evidence of
gambling-motivated investment decisions for at least two reasons. First, peo-
ple’s gambling preferences and portfolio decisions are not directly observed.
Second, a precise and well-established definition of stocks that might be per-
ceived as instruments for gambling does not exist.
In this paper, I use individual investors’ socioeconomic characteristics to infer
their gambling preferences and attempt to detect traces of gambling in their
stock investment decisions. Specifically, I conjecture that people’s gambling
propensity, as ref lected by their socioeconomic characteristics, predicts gam-
bling behavior in other settings, including the stock market. This conjecture is
motivated by recent research in behavioral economics that demonstrates that
people’s risk-taking propensity in one setting predicts risky behavior in other
settings (e.g., Barsky et al. (1997)).
I consider the most common form of gambling (state lotteries), where the
identities of gamblers can be identified with greater ease and precision, and
identify the salient socioeconomic characteristics of people who exhibit a strong
propensity to play state lotteries. The extant evidence from lottery studies in-
dicates that the heaviest lottery players are poor, young, and relatively less
educated, single men, who live in urban areas and belong to specific minority
(African-American and Hispanic) and religious (Catholic) groups. Therefore,
a direct implication of my main conjecture is that investors with these spe-
cific characteristics also invest disproportionately more in stocks with lottery
features.
To formally define lottery-type stocks, I examine the salient features of state
lotteries and also seek guidance from recent theoretical studies that attempt to
characterize lottery-type stocks. Lottery tickets have very low prices relative to
the highest potential payoff (i.e., the size of the jackpot); they have low negative
expected returns; their payoffs are very risky (i.e., the prize distribution has
extremely high variance); and, most importantly, they have an extremely small
probability of a huge reward (i.e., they have positively skewed payoffs). In sum,
for a very low cost, lottery tickets offer a tiny probability of a huge reward and
a large probability of a small loss, where the probabilities of winning and losing
are fixed and known in advance.
While any specific stock is unlikely to possess the extreme characteristics of
state lotteries, particularly the huge reward to cost ratio, some stocks might
share these features qualitatively. To identify those stocks that could be per-
ceived as lotteries, I consider three characteristics: (i) stock-specific or idiosyn-
cratic volatility, (ii) stock-specific or idiosyncratic skewness, and (iii) stock price.
As with lotteries, if investors are searching for “cheap bets,” they are likely
to find low-priced stocks attractive. Within the set of low-priced stocks, they
are likely to find stocks with high stock-specific skewness more attractive. And
among the set of stocks that have low prices and high idiosyncratic skewness,
 Who Gambles in the Stock Market? 1891

stocks with greater idiosyncratic volatility are more likely to be perceived as lot-
teries because the level of idiosyncratic volatility could inf luence the estimates
of idiosyncratic skewness. When volatility is high, investors might believe that
the extreme return events observed in the past are more likely to be realized
again. In contrast, if a low price-high skewness stock has low idiosyncratic
volatility, the extreme return events observed in the past might be perceived
as outliers, and the re-occurrence of that event is likely to be assigned a con-
siderably lower probability.
With this motivation, I assume that individual investors perceive low-priced
stocks with high idiosyncratic volatility and high idiosyncratic skewness as lot-
teries. Therefore, I use this empirical definition of lottery-type stocks to gather
evidence of gambling-induced stock investment decisions among individual
investors.
The empirical investigation is organized around four distinct themes. First,
I compare the aggregate stock preferences of individual and institutional in-
vestors and examine whether individual investors exhibit a stronger prefer-
ence for stocks with lottery features. Next, I investigate whether individual
investors’ preferences for lottery-type stocks are stronger among socioeconomic
groups that are known to exhibit strong preferences for state lotteries. I also
directly examine whether investment levels in lottery-type stocks are higher in
regions with more favorable lottery environments.1 Third, I examine whether,
similar to the demand for lotteries, the aggregate individual investor demand
for lottery-type stocks increases during bad economic times. Finally, I examine
whether investment in lottery-type stocks has an adverse inf luence on port-
folio performance. In particular, I investigate whether, like state lotteries, in-
vestment in lottery-type stocks is regressive, where low-income investors lose
proportionately more from their gambling-motivated investments.
The main data set for my empirical analysis is a 6-year panel of portfolio
holdings and trades of a group of individual investors at a large U.S. discount
brokerage house. Using this data set, I show that individual investors exhibit a
strong preference for stocks with lottery features, whereas institutions exhibit
a relative aversion for those stocks. Individual investors’ preferences for lottery-
type stocks are distinct from their known preferences for small-cap stocks, value
stocks, dividend paying stocks, and “attention grabbing” stocks (e.g., Barber and
Odean (2000, 2001, 2008), Graham and Kumar (2006)). Over time, similar to
lottery demand, individual investors’ aggregate demand for lottery-type stocks
increases when economic conditions worsen. These aggregate-level results in-
dicate that, similar to state lotteries, lottery-type stocks are more attractive to
a relatively less sophisticated individual investor clientele.
Examining cross-sectional differences within the individual investor cate-
gory, I find that socioeconomic factors that induce higher expenditures in state
lotteries are also associated with greater investments in lottery-type stocks.
Poor, young, less educated single men who live in urban areas, undertake non-
professional jobs, and belong to specific minority groups (African-American and
1
I assume that a state that adopted state lotteries earlier and has a higher per capita lottery
expenditure has a favorable lottery environment.
1892 The Journal of FinanceR

Hispanic) invest more in lottery-type stocks. In addition, investors who live in

regions with a higher concentration of Catholics (Protestants) have a stronger
(weaker) preference for lottery-type stocks.
The results from cross-sectional analysis also indicate that local economic
conditions and regional lottery environments inf luence the demand for lottery-
type stocks. Investors who earn less than their neighbors (i.e., have lower “rel-
ative” income) and live in counties with higher unemployment rates invest
relatively more in lottery-type stocks. In addition, the proportional investment
in lottery-type stocks is higher in states that were early lottery adopters and
have higher per capita lottery expenditures. Collectively, the cross-sectional re-
sults indicate that state lotteries and lottery-type stocks act as complements
and attract very similar socioeconomic clienteles.
Turning to the portfolio performance of lottery investors, I find that investors
who invest disproportionately more in lottery-type stocks experience greater
underperformance. The average, annual, risk-adjusted underperformance that
can be attributed to investments in lottery-type stocks is 1.10% and the level of
underperformance is over 2.50% for investors who allocate at least one-third of
their portfolios to lottery-type stocks. A typical investor would have improved
performance by 2.84% if she had simply replaced the lottery component of her
portfolio with the nonlottery component. As a proportion of income, the degree
of portfolio underperformance has a striking resemblance to the evidence from
lottery studies. In both instances, the proportional level of underperformance
is greater among low-income investors.
Taken together, the empirical results provide evidence of strong similarities
between the behavior of state lottery players and individual investors who in-
vest disproportionately more in stocks with lottery features. The findings are
consistent with my main conjecture and indicate that a set of common personal
attributes determines people’s gambling preferences. Alternative explanations
for these results based on local bias, investor overconfidence, media coverage,
or microstructure effects have little empirical support.
The balance of the paper is organized as follows. In the next section, I use
the salient findings from the literature on state lotteries to develop the key
testable hypotheses. In Section II, I describe the data sources. In Section III,
I formally define lottery-type stocks and using the definition of lottery-type
stocks, in Sections IV to VII, I present the main empirical results. I conclude in
Section VIII with a brief summary.

I. Testable Hypotheses Motivated by Lottery Studies

In this section, I examine the empirical evidence from previous studies on
state lotteries and develop this paper’s main testable hypotheses.

A. Profile of Lottery Players

Extant evidence from the state lottery literature indicates that both lottery
participation rates and lottery expenditures are strongly inf luenced by people’s
 Who Gambles in the Stock Market? 1893

socioeconomic characteristics (e.g., Kallick et al. (1979)). For instance, relatively

poor individuals tend to spend a greater proportion of their income on lottery
purchases (e.g., Clotfelter and Cook (1989), Clotfelter (2000), Rubinstein and
Scafidi (2002)). Beyond income and wealth, age, education, gender, and marital
status inf luence lottery purchases. In particular, younger and less educated in-
dividuals find lotteries more attractive (e.g., Brenner and Brenne (1990)), and
relative to women, men are more likely to participate and spend disproportion-
ately more in lotteries. Further, single or divorced individuals are more active
lottery players than people who are married (e.g., Clotfelter et al. (1999)).
Lottery studies also document that race, ethnicity, and religious affiliation
inf luence people’s attitudes toward lottery-playing and gambling. Specifically,
both lottery participation rates and purchase levels are higher among African-
American and Hispanic minority groups (e.g., Herring and Bledsoe (1994), Price
and Novak (1999)). Among religious groups, Catholics and Jews are more active
participants in lotteries compared to Protestants and Mormons (e.g., Tec (1964),
Grichting (1986)).2
Geographically, lottery studies find that urban residents are more likely to
buy lottery tickets and spend more on their lottery purchases than individuals
in rural areas (e.g., Kallick et al. (1979)). Lottery participation rates and ex-
penditures also vary significantly across the United States, where the degree
of popularity of lotteries ref lects the overall social acceptability of gambling in
the state (e.g., Clotfelter and Cook (1989)).
Examining the effects of broad macroeconomic indicators (e.g., the unemploy-
ment rate), lottery studies demonstrate that people find the tiny probability of
a large gain more attractive when economic opportunities are not very bright.
As a result, during economic downturns, people are attracted more toward vari-
ous forms of gambling, including state lotteries (Mikesell (1994)). For instance,
during the Great Depression of the 1930s, the popularity of lottery-playing
and gambling had increased dramatically in the United States. (Brenner and
Brenner (1990)). Sweden experienced a similar phenomenon, where during the
Great Depression, gambling became extremely popular and gambling activities
such as soccer pools were made legal (Tec (1964)).

B. Main Testable Hypotheses

Overall, the empirical evidence from lottery studies indicates that demo-
graphic characteristics and economic factors jointly determine the propensity
to play lotteries. If lottery purchases and investments in lottery-type stocks are
both inf luenced by a set of common personality attributes that determines gam-
bling preferences, and if people’s gambling demands are not saturated, then the

2
Other forms of gambling such as casino gambling do not have stable and well-defined de-
mographic characteristics. See Section A of the Internet Appendix for a brief discussion. An In-
ternet Appendix for this article is online in the “Supplements and Datasets” section at http://
www.afajof.org/supplements.asp.
1894 The Journal of FinanceR

behavior of state lottery players and lottery investors would exhibit similarities
along multiple dimensions.
First, the socioeconomic characteristics of people who find state lotteries at-
tractive should be similar to those of investors who exhibit a greater propensity
to invest in stocks with lottery features. In particular, relatively poor, less ed-
ucated, young, single men who undertake nonprofessional jobs, live in urban
areas, and belong to specific minority (African-American and Hispanic) and
religious (Catholic) groups are expected to invest disproportionately more in
lottery-type stocks.
Second, local socioeconomic factors would inf luence investors’ holdings of
lottery-type stocks. In particular, if investors perceive stocks with lottery fea-
tures as gambling devices, investors located in regions with more favorable
lottery environments (states that adopted lotteries earlier and have higher per
capita lottery expenditures) can be expected to tilt their portfolios more to-
ward lottery-type stocks. In contrast, the demand for nonlottery-type stocks
in those regions should be relatively weaker. This conjecture is partially moti-
vated by the observation that the demand levels for various gambling devices
are positively correlated. For instance, lottery studies indicate that many types
of gambling devices were legal in states that were early lottery adopters, while
states without lotteries also had lower acceptability of other forms of gambling
(Clotfelter and Cook (1989)). In addition, survey evidence indicates that geo-
graphical regions with greater levels of lottery demand also exhibit stronger
levels of demand for other forms of gambling (Kallick et al. (1979)).
Additionally, because status-seeking individuals exhibit a stronger propen-
sity to gamble to improve their upward social mobility (e.g., Friedman and
Savage (1948), Brunk (1981), Brenner (1983), Becker, Murphy, and Werning
(2000)), the level of investments in lottery-type stocks should be greater among
investors who have a lower social status relative to their respective neighbors.
Specifically, investors with lower income relative to their neighbors are ex-
pected to invest more in lottery-type stocks because relative income is a good
proxy for relative social status and a feeling of overall well-being (e.g., Luttmer
(2005)).
Third, if economic conditions inf luence an individual’s gambling preference,
then like state lotteries, the aggregate demand for lottery-type stocks should be
higher in regions with relatively poor economic conditions (e.g., higher unem-
ployment). Over time, as economic conditions change, the aggregate levels of
demand for state lotteries and lottery-type stocks should be correlated. In par-
ticular, like state lotteries, investors are likely to exhibit a stronger preference
for lottery-type stocks during bad economic times.
Overall, there are four distinct testable implications of my main conjecture:

H1: Aggregate preference hypothesis: Relative to institutions, individual in-

vestors exhibit stronger aggregate preference for lottery-type stocks.
H2: Similar clienteles hypothesis: The socioeconomic characteristics of lot-
tery players and lottery investors are similar. Thus, state lotteries and
lottery-type stocks act as complements.
 Who Gambles in the Stock Market? 1895

H3: Location and social mobility hypothesis: Investors who live in regions
with higher unemployment rates and favorable lottery environments,
and who have lower social status relative to their neighbors, allocate
larger portfolio weights to lottery-type stocks.
H4: Time-series hypothesis: Similar to the demand for state lotteries, the
aggregate demand for lottery-type stocks is higher during economic
downturns.
In addition to testing these gambling-motivated hypotheses, I examine
whether the propensity to gamble with lottery-type stocks adversely inf luences
portfolio performance.

II. Data Sources

To test the gambling hypotheses outlined above, I primarily use data from
a major U.S. discount brokerage house. This data set contains all trades and
end-of-month portfolio positions of a set of individual investors during the 1991
to 1996 period. There are a total of 77,995 investors in the database, of which
62,387 trade common stocks. An average investor holds a four-stock portfolio
(median is three) with an average size of $35,629 (median is $13,869). For a
subset of households, demographic measures, including age, income, location
(zip code), total net worth, occupation, marital status, family size, gender, etc.,
are available. The demographic measures were compiled by Infobase Inc., in
June 1997.3
I enrich the individual investor database using data from several additional
sources. First, to identify sample investors’ racial and ethnic characteristics,
education level, and immigrant status, I obtain the racial and ethnic compo-
sitions of each zip code using data from the 1990 U.S. Census. I assign each
investor the appropriate zip code-level racial and ethnic characteristics. I also
assume that investors who live in more educated zip codes are likely to be more
educated and investors who live in zip codes with a greater proportion of for-
eign born people are more likely to be immigrants. Second, to characterize the
lottery environment of the state, I obtain the annual per capita lottery sales
data for the 37 U.S. states in which lotteries were legal during the sample pe-
riod.4 For a handful of states, I am also able to obtain zip code-level lottery sales
data directly from the state lottery agencies. Third, I obtain the religious profile
of all U.S. counties in 1990 using data from the Association of Religion Data
Archives.5 For each county, I compute the proportion of Catholics and the pro-
portion of Protestants. Using each investor’s zip code, I assign the appropriate
county-level religious characteristic to the investor.
3
Additional details on the individual investor database are available in Barber and Odean (2000)
and Barber and Odean (2001).
4
I thank Garrick Blalock for providing the lottery sales data. See Blalock et al. (2007) for addi-
tional details about the data.
5
The 1990 U.S. Census data are available at http://www.census.gov/main/www/cen1990.html.
The 1990 county-level religion data are available at http://www.thearda.com/.
1896 The Journal of FinanceR

In addition to detailed data on individual investors, I obtain quarterly institu-

tional holdings from Thomson Financial. These data contain the end-of-quarter
stock holdings of all institutions that file form 13f with the Securities and Ex-
change Commission. I obtain trading data from the Trade and Quote (TAQ)
and the Institute for the Study of Security Markets (ISSM) databases, where
small-sized trades (trade size below $5,000) are used to proxy for retail trades.6
I also use data from a few other standard sources. I obtain analysts’ quar-
terly earnings estimates from Thomson Financial’s Institutional Brokers Esti-
mate System (I/B/E/S) summary files and monthly macroeconomic data from
Datastream. For each stock in the sample, I obtain monthly price, return, vol-
ume turnover, and market capitalization data from the Center for Research
on Security Prices (CRSP), and quarterly book value of common equity from
COMPUSTAT. The monthly time series of the three Fama-French factors
and the momentum factor are from Kenneth French’s data library while the
characteristic-based performance benchmarks are from Russell Wermers’ web
site.7 Table I presents definitions and sources of the variables used in the em-
pirical analysis.

III. Lottery-Type Stocks

A. An Empirical Definition
Motivated by the salient features of state lotteries, I consider three stock
characteristics to identify stocks that might be perceived as lotteries: (i) stock-
specific or idiosyncratic volatility, (ii) idiosyncratic skewness, and (iii) stock
price. At the end of month t, I compute both the idiosyncratic volatility and
the idiosyncratic skewness measures using the previous 6 months (i.e., months
t − 6 to t − 1) of daily returns data. The idiosyncratic volatility measure is the
variance of the residual obtained by fitting a four-factor model to the daily stock
returns time-series. To measure idiosyncratic skewness, I adopt the Harvey and
Siddique (2000) method and decompose total skewness into idiosyncratic and
systematic components. Specifically, idiosyncratic skewness is a scaled measure
of the third moment of the residual obtained by fitting a two-factor model to
the daily stock returns time series, where the two factors are the excess market
returns and the squared excess market returns. The stock price refers to the
price at the end of month t − 1.
I consider all CRSP stocks and assume that stocks in the lowest k th stock
price percentile, the highest k th idiosyncratic volatility percentile, and the high-
est k th idiosyncratic skewness percentile are likely to be perceived as lottery-
type stocks. All three sorts are carried out independently. I choose k = 50 to

6
Additional details on the TAQ small-trades data set, including the detailed procedure for iden-
tifying small trades, are available in Barber et al. (2009).
7
The risk factors are obtained from http://mba.tuck.dartmouth.edu/pages/faculty/ken.french/.
The Daniel et al. (1997) characteristics-based performance benchmarks are available at
http://www.smith.umd.edu/faculty/rwermers/ftpsite/Dgtw/coverpage.htm.
 Who Gambles in the Stock Market? 1897

Table I
Brief Definitions and Sources of Main Variables
This table brief ly defines the main variables used in the empirical analysis. All volatility and
skewness estimates are obtained using 6 months of daily stock returns. The factors employed in the
multifactor models are RMRF (excess market return), SMB (size factor), HML (value factor), and
UMD (momentum factor). The data sources are as follows: (i) ARDA: Association of Religion Data
Archives, (ii) Brokerage: Large U.S. discount brokerage, (iii) BLS: Bureau of Labor Statistics, (iv)
Census: 1990 U.S. Census, (v) CRSP: Center for Research on Security Prices, (vi) DS: Datastream,
(vii) IBES: Institutional Brokers Estimate System from Thomson Financial, (viii) KFDL: Kenneth
French’s data library, (ix) LOTAG: State lottery agencies, and (x) 13f: 13f institutional portfolio
holdings data from Thomson Financial.

Variable Name Description Source

Panel A: Stock Characteristics Reported in Table II

Percentage of market Weight of a stock category in the aggregate market portfolio CRSP
constructed using all common stocks (share codes 10 and 11)
in CRSP.
Total volatility Standard deviation of daily stock returns. CRSP
Idiosyncratic Standard deviation of the residual from a four-factor model. CRSP
volatility
Total skewness Scaled measure of the third moment of daily stock returns. CRSP
Idiosyncratic Scaled measure of the third moment of the residual obtained by CRSP
skewness fitting a two-factor (RMRF and RMRF 2 ) model.
Systematic skewness Coefficient of the squared market factor in the skewness CRSP
regression.
Stock price End-of-month stock price. CRSP
Firm size End-of-month market capitalization (price × shares CRSP
outstanding).
Book-to-market ratio Ratio of the book-value and the market capitalization of the firm. CRSP
Past 12-month return Total monthly stock return during the past 12 months. CRSP
RMRF, SMB, and The loadings on RMRF, SMB, and HML factors in a three-factor KFDL
HML betas model, respectively.
Amihud illiquidity Absolute daily returns per unit of trading volume (Amihud CRSP
(2002)).
Monthly volume Shares traded divided by the number of shares outstanding. CRSP
turnover
Firm age Number of years since the stock first appears in CRSP. CRSP
Percentage dividend Proportion of firm in a stock category that paid a dividend at CRSP
paying least once during the previous 1 year.
% Without analyst Proportion of firm in a stock category without analyst coverage. IBES
coverage
Mean number of Mean number of analysts per stock. IBES
analysts
% Institutional Percentage of total shares outstanding owned by 13F 13f
ownership institutions.

Panel B: Additional Variables Used in Stock-Level Regressions (Table III)

Dividend paying Set to one if the stock paid dividends during the past 1 year. CRSP
dummy
S&P500 dummy Set to one if the stock belongs to the S&P500 index. CRSP
Nasdaq dummy Set to one if the stock belongs to the Nasdaq index. CRSP

(continued)
1898 The Journal of FinanceR

Table I—Continued

Variable Name Description Source

Panel C: Variables Used in Investor-Level Regressions (Table IV)

Demographic characteristics
Wealth Total net worth of the investor. Brokerage
Income Annual household income. Brokerage
Age Age of the head of the household. Brokerage
Education Proportion of residents in investor’s zip code with a Census
Bachelor’s or higher educational degree.
Professional dummy Set to one if the investor belongs to one of the Brokerage
professional (managerial or technical) job categories.
Retired dummy Set to one if the head of the household is retired. Brokerage
Male dummy Set to one if the head of the household is male. Brokerage
Married dummy Set to one if the head of the household is married. Brokerage
Investment experience Number of days since the brokerage account opening Brokerage
date.
Taxable account dummy Set to one if the investor holds only taxable accounts. Brokerage
Tax deferred account Set to one if the investor holds only tax-deferred (IRA or Brokerage
dummy Keogh) retirement accounts.
Location-based measures
Catholic (Protestant) Set to one if the proportion of Catholics (Protestants) in ARDA
dummy the county of investor’s residence is greater than the
mean proportion of Catholics (Protestants) across the
U.S. counties.
African Ratio of African-Americans and Whites in the investor’s Census
American-White zip code.
ratio
Hispanic-White ratio Ratio of Hispanics and Whites in the investor’s zip code. Census
Proportion foreign born Proportion of foreign born residents in the investor’s zip Census
code.
Income relative to Difference between the investor’s annual income and the Brokerage
neighbors mean income of sample investors located within 25
miles of her zip code.
Urban dummy Set to one if the investor resides within 100 miles of one Brokerage
of the largest 20 U.S. metropolitan areas.
County unemployment Unemployment rate in the investor’s county of residence. BLS
rate
State lottery Mean annual per capita expenditure in state lotteries in LOTAG
expenditure the investor’s state of residence.
State lottery age Number of years since the lottery adoption date in the LOTAG
investor’s state of residence.
Portfolio characteristics
Initial portfolio size Size of investor’s stock portfolio when she enters the Brokerage
sample.
Monthly portfolio Average of buy and sell turnover rates. Brokerage
turnover
Portfolio diversification Portfolio variance divided by the average variance of all Brokerage
stocks in the portfolio.
Portfolio dividend yield Sample period average dividend yield of the investor’s Brokerage
portfolio.

(continued)
 Who Gambles in the Stock Market? 1899

Table I—Continued

Variable Name Description Source

Panel C: Variables Used in Investor-Level Regressions (Table IV)

Portfolio local bias Proportion of the portfolio that is invested in stocks within Brokerage
100 miles of the investor’s zip code.
Industry Largest weight allocated to one of the 48 Fama-French Brokerage
concentration industries.
Portfolio factor RMRF, SMB, HML, and UMD betas of the investor portfolio. Brokerage
exposures

Panel D: Variables Used in State-Level Regressions (Table V)

Annual per capita Annual per capita expenditure on state lotteries in the state. LOTAG
state lottery
expenditure
State unemployment Monthly unemployment rate in the state. BLS
Catholic (Protestant) Set to one if the proportion of Catholics (Protestants) in the ARDA
dummy state of investor’s residence is greater than the mean
proportion of Catholics (Protestants) across all U.S. states.

Panel E: Time-Series Regression Variables (Table VI)

UNEMP U.S. monthly unemployment rate. DS

UEI Unexpected inf lation (current inf lation minus the average of DS
the past 12 realizations).
MP Monthly growth in industrial production. DS
RP Monthly default risk premium (difference between Moody’s DS
Baa-rated and Aaa-rated corporate bond yields).
TS Term spread (difference between the yields of constant DS
maturity 10-year Treasury bond and 3-month Treasury
bill).
EFC Mean monthly change in analysts’ earnings forecasts of IBES
lottery-type stocks.
LOTRET Mean monthly return of a portfolio of lottery-type stocks. CRSP
MKTRET Monthly market return. CRSP

Panel F: Additional Variables Used in Performance Regressions (Table VIII)

Lottery-type stock Set to one if an investor buys at least one lottery-type stock Brokerage
participation during the sample period.
dummy
Strong lottery-type Set to one if the lottery-type stock preference measure is in Brokerage
stock preference the highest decile.
dummy

have a considerable number of lottery-type stocks in the sample, but the main
results are very similar when I choose k = 33.
I use stock price as one of the defining characteristics of lottery-type stocks
because, like lotteries, if investors are searching for cheap bets, they should nat-
urally gravitate toward low-priced stocks. Thus, stock price is likely to be an
important characteristic of stocks that might be perceived as lotteries. Within
1900 The Journal of FinanceR

the set of low-priced stocks, investors are likely to be attracted more toward
stocks that occasionally generate extreme positive returns that cannot be justi-
fied by the movements in the market. In other words, investors are likely to find
stocks with high stock-specific or idiosyncratic skewness attractive. Therefore, I
use idiosyncratic skewness as the second defining characteristic of lottery-type
stocks.8
Finally, within the set of stocks that have low prices and high idiosyncratic
skewness, stocks with higher stock-specific volatility are more likely to be per-
ceived as lotteries. When idiosyncratic volatility is high, investors might believe
that the extreme return events observed in the past are more likely to be re-
peated. In particular, if investors adopt an asymmetric weighting scheme and
assign a larger weight to upside volatility and ignore or assign lower weight to
downside volatility, high idiosyncratic volatility could amplify the perception of
skewness. In contrast, if a low price–high idiosyncratic skewness stock has low
idiosyncratic volatility, the extreme return events observed in the past might be
perceived as outliers and the re-occurrence of an extreme return event might be
assigned a low probability. Consequently, higher idiosyncratic volatility could
amplify the estimates of the level of idiosyncratic skewness and the likelihood
of realizing extreme positive return in the future.9
Strictly speaking, the three stock characteristics identify stocks that appear
to be like lotteries, rather than stocks that are truly lotteries. Ideally, one would
classify stocks with higher probability of large positive returns (i.e., positive
skewness) in the future as lottery-type stocks. While it is conceivable that so-
phisticated institutional investors are able to predict future skewness, it is un-
likely that less sophisticated individual investors would be successful in identi-
fying those predictors. Rather, they are more likely to “naı̈vely” extrapolate past
moments into the future and pick stocks that appear like lotteries. Because my
study focuses on the investment choices of individual investors, I characterize
lottery-type stocks using measures that are more likely to be used by individual
investors to naı̈vely identify stocks with lottery features.

B. Main Characteristics
Table II presents the sample period averages of several important character-
istics of lottery-type stocks. For comparison, I also report the characteristics of
nonlottery-type stocks and the other remaining stocks in the CRSP universe.
The nonlottery-type stock category consists of stocks that are in the highest k th

8
Other mechanisms can generate a preference for skewness. For instance, over-weighting of very
low probability events (e.g., the probability of winning a lottery jackpot) can induce a preference
for skewness (Tversky and Kahneman (1992), Polkovnichenko (2005), Barberis and Huang (2008)).
Brunnermeier and Parker (2005) show that anticipatory utility (e.g., dream utility) can generate
a preference for skewness in portfolio decisions.
9
Of course, high volatility is not a characteristic that is unique to state lotteries. Other forms of
gambling such as casinos also share this feature. For robustness, I examine the sensitivity of my
main results by defining lottery-type stocks without the volatility characteristic. See Section C of
the Internet Appendix.
 Who Gambles in the Stock Market? 1901

Table II
Basic Characteristics of Lottery-Type Stocks
This table reports the mean monthly characteristics of lottery-type stocks, measured during the
1991 to 1996 sample period. For comparison, the characteristics of nonlottery-type stocks and stocks
that do not belong to either of the two categories (i.e., other stocks) are also reported. The stocks
in all three categories are defined at the end of each month using all stocks in the CRSP universe.
The stocks in the lowest k th price percentile, highest k th idiosyncratic volatility percentile, and
highest k th idiosyncratic skewness percentile are identified as lottery-type stocks. Similarly, stocks
in the highest k th price percentile, lowest k th idiosyncratic volatility percentile, and lowest k th
idiosyncratic skewness percentile are identified as nonlottery-type stocks. For the results reported
in the table, k = 50. Additional details on the definition of lottery-type stocks are available in
Section III.A and all reported measures are defined in Table I, Panel A.

Measure Lottery-Type Nonlottery-Type Others

Number of stocks 1,553 1,533 8,945

Percentage of the market 1.25% 50.87% 47.88%
Total volatility 78.57 3.29 22.14
Idiosyncratic volatility 75.56 2.96 20.36
Total skewness 0.330 0.175 0.237
Systematic skewness −0.202 −0.061 −0.110
Idiosyncratic skewness 0.731 −0.041 0.332
Stock price $3.83 $31.68 $17.51
Market beta 1.090 0.906 0.897
Firm size (in million $) 31.41 1650.87 539.66
SMB beta 0.804 0.378 0.617
Book-to-market ratio 0.681 0.314 0.348
HML beta 0.272 0.186 0.151
Past 12-month return 16.52% 20.22% 18.14%
Amihud illiquidity 70.16 0.465 15.13
Monthly volume turnover 84.72% 64.16% 57.90%
Firm age (in years) 5.78 12.10 11.87
Percentage dividend paying 3.37% 44.59% 57.03%
Percentage without analyst coverage 71.30% 21.19% 36.87%
Mean number of analysts 3.93 12.40 6.49
Percentage institutional ownership 7.35% 49.34% 30.09%

stock price percentile, the lowest kth idiosyncratic volatility percentile, and the
lowest kth idiosyncratic skewness percentile. The remaining stocks are classi-
fied into the “Other Stocks” category.
The summary statistics in Table II indicate that lottery-type stocks have
very low average market capitalization ($31 million), low institutional owner-
ship (7.35%), a relatively high book-to-market ratio (0.681), and lower liquid-
ity. These stocks are also younger (mean age is about 6 years), have low ana-
lyst coverage (about 71% of stocks have no analyst coverage), and are mostly
nondividend-paying stocks (only 3.37% pay dividends). Given the definition
of lottery-type stocks, not surprisingly, they have significantly higher volatil-
ity, higher skewness, and lower prices. Similarly, by definition, nonlottery-type
stocks have diametrically opposite features, and “other stocks” have character-
istics in between these two extremes.
1902 The Journal of FinanceR

I also find that lottery-type stocks are concentrated heavily in the energy,
mining, financial services, bio-technology, and technology sectors. The indus-
tries with the lowest concentration of lottery stocks include utilities, consumer
goods, and restaurants. As a group, lottery-type stocks represent 1.25% of the
total stock market capitalization, but in terms of their total number, they rep-
resent about 13% of the market.

C. How Might Investors Identify Lottery-Type Stocks?

The volatility and skewness measures are difficult to compute using the stan-
dard formulas and individual investors are unlikely to compute those measures
to identify lottery-type stocks. Given the clear differences between the stock
characteristics of lottery-type stocks and other stocks, it is conceivable that
relatively less sophisticated individual investors would use one or more of the
salient stock characteristics of lottery-type stocks to identify them. Some in-
vestors might even be attracted toward certain industries, which might have
strong lottery characteristics.
To formally examine whether a collection of common stock characteristics
could serve as a substitute for the three lottery features, I estimate cross-
sectional regressions in which one of the lottery features is the dependent vari-
able. The set of stock characteristics reported in Table II are the independent
variables. In untabulated results, I find that although the univariate regres-
sion estimates are strong, only a handful of stock characteristics are strongly
significant in a multivariate specification. When idiosyncratic skewness is the
dependent variable, only 2.85% of the cross-sectional variation in skewness can
be explained by these stock characteristics. Even when I include idiosyncratic
volatility and stock price in the set of independent variables, the explanatory
power increases to only 3.65%.10
I also examine whether the set of stock characteristics listed in Table II can
serve as a substitute for idiosyncratic volatility. When idiosyncratic volatil-
ity is the dependent variable, the explanatory power is higher (=17.33%), but
a large part of the cross-sectional variation in idiosyncratic volatility is still
unexplained. When I use the lottery-stock dummy as the dependent variable
and estimate a logit regression, the combined explanatory power of all stock
characteristics is higher (=21.24%), but the increase is driven primarily by the
presence of stock price in the dependent variable.
These regression results indicate that even a comprehensive set of stock char-
acteristics is unlikely to serve as an effective substitute for the three lottery
features. Although certain salient stock characteristics could steer investors
toward lottery-type stocks, realizations of extreme returns are necessary to
generate a perception of “lottery.” Investors who are looking for cheap ways of
buying a tiny probability of a very high return would most likely extrapolate

10
The inability of stock characteristics to explain the cross-sectional heterogeneity in skewness
is consistent with the evidence in previous studies that attempt to predict skewness (e.g., Chen,
Hong, and Stein (2001)).
 Who Gambles in the Stock Market? 1903

past extreme return events into the future, especially if the associated stocks
have low prices and high volatility. Even if investors do not compute skew-
ness and volatility according to the standard formulas, they would be able to
discriminate between high and low volatility stocks or high and low skewness
stocks. If both volatility and skewness levels are high, investors might be able
to identify those stocks with even greater ease.11

IV. Aggregate Preferences for Lottery-Type Stocks

In the first set of tests, I gather support for the first hypothesis (H1). Specifi-
cally, I characterize the aggregate stock preferences of individual investors and
compare them with the aggregate preferences of institutional investors.

A. Lottery-Type Stocks in Aggregate Investor Portfolios

To begin, I examine how the aggregate individual and institutional pref-
erences for lottery-type stocks vary over time. Figure 1 shows the monthly
weights allocated to lottery-type stocks in the aggregate individual and insti-
tutional portfolios, respectively. For comparison, I also show the total weight
of lottery-type stocks in the aggregate market portfolio. To construct the ag-
gregate individual investor portfolio, I combine the portfolios of all individual
investors in the brokerage sample. I construct the aggregate institutional port-
folio in a similar manner using the 13f institutional portfolio holdings data. I
define the aggregate market portfolio by combining all stocks within the CRSP
universe.
The figure indicates that, relative to the market portfolio, individual investors
significantly overweight lottery-type stocks. The average weights allocated to
lottery-type stocks in the aggregate retail and market portfolios are 3.74% and
1.25%, respectively. In contrast, the average weight allocated to lottery-type
stocks in the aggregate institutional portfolio is only 0.76%. The aggregate time-
series results indicate that individual investors exhibit a strong preference for
stocks with lottery features, while institutions exhibit a weak aversion for those
stocks.

B. Aggregate Stock Preference Measure

To characterize investors’ aggregate preferences for lottery-type stocks more
accurately, I estimate stock-level pooled and cross-sectional regressions and
compare the aggregate stock preferences of individual and institutional in-
vestors. In these regressions, I employ a set of stock characteristics as indepen-
dent variables. The set includes measures that investors might use to identify
lottery-type stocks. The stock preference in the aggregate investor portfolio is
the dependent variable.

11
Another channel through which investors might be driven toward lottery-type stocks is the
news media. I examine this conjecture in Section V.D.
1904 The Journal of FinanceR

6
Expected Weight
Retail Weight
Institutional Weight

4
-

0
Dec 91 Dec 92 Dec 93 Dec 94 Dec 95 Dec 96

Calendar Time (January 1991 to November 1996)

Figure 1. Aggregate weight in lottery-type stocks over time. This figure shows the time
series of the actual weights allocated to lottery-type stocks in the aggregate individual and in-
stitutional investor portfolios. The expected lottery weight time series, which ref lects the weight
allocated to lottery-type stocks in the aggregate market portfolio, is also shown. The aggregate
individual investor portfolio is formed by combining the portfolios of all individual investors in
the brokerage sample. The aggregate institutional portfolio is constructed in an analogous manner
using the 13f institutional portfolio holdings data. The aggregate market portfolio is obtained by
combining all stocks in the CRSP universe. The stocks in the lowest k th price percentile, highest k th
idiosyncratic volatility percentile, and highest k th idiosyncratic skewness percentile are identified
as lottery-type stocks. For the plot, k = 50. Additional details on the definition of lottery-type stocks
are available in Section III.A. The individual investor data are from a large U.S. discount broker-
age house for the period 1991 to 1996, while the institutional holdings data are from Thomson
Financial.

The aggregate investor preference for stock i in month t is the unexpected

(or excess) portfolio weight allocated to that stock. Specifically, this measure is
defined as
wipt − wimt
E W ipt = × 100. (1)
wimt
Here, wipt is the actual weight assigned to stock i in the aggregate investor
portfolio p in month t, and wimt is the weight of stock i in the aggregate market
portfolio in month t. The institutional preference for a stock is identified in an
identical manner using the aggregate institutional portfolio.
 Who Gambles in the Stock Market? 1905

If the sample investors were to randomly select stocks such that the proba-
bility of selecting a stock is proportional to its market capitalization, the weight
of each stock in the aggregate investor portfolio would be equal to the weight of
the stock in the aggregate market portfolio. Thus, for a given stock, a positive
(negative) deviation from the expected weight in the market portfolio captures
the aggregate individual investor preference (aversion) for the stock. While
other benchmarks exist for measuring the expected weight of a stock in a given
portfolio, I use the market capitalization-based benchmark because it is simple
and based on few assumptions.12

C. Stock-Level Fama–MacBeth and Panel Regression Estimates

In the first regression specification, the independent variables are the three
measures that ref lect the lottery characteristics of stocks: idiosyncratic volatil-
ity, idiosyncratic skewness, and stock price. I estimate the regression specifi-
cation at the end of each month using the Fama and MacBeth (1973) cross-
sectional regression method, and I use the Pontiff (1996) method to correct the
Fama–MacBeth standard errors for potential higher-order serial correlation.13
To ensure that extreme values are not affecting my results, I winsorize all vari-
ables at their 0.5 and 99.5 percentile levels. I standardize both the dependent
and the independent variables (the mean is set to zero and the standard devi-
ation is one) so that the coefficient estimates can be directly compared within
and across regression specifications.
The Fama–MacBeth regression estimates are presented in Table III. Col-
umn (1) reports the estimates for idiosyncratic volatility and skewness mea-
sures; for robustness, column (2) reports the estimates for total volatility and
skewness measures. The results indicate that individual investors assign a
relatively larger weight to stocks with higher idiosyncratic volatility, higher
idiosyncratic skewness, and lower prices. Thus, individual investors prefer to
hold stocks that might be perceived as lotteries. I find that the estimates in col-
umn (2) with the total volatility and skewness measures are very similar to the
estimates in column (1) where I consider idiosyncratic volatility and skewness
measures.
To examine which lottery characteristics have stronger inf luence on in-
vestors’ aggregate preferences, I compare the coefficient estimates of the three
characteristics that are used to define lottery-type stocks. The results indicate
that idiosyncratic volatility and idiosyncratic skewness similarly inf luence in-
vestors’ aggregate preferences as their coefficient estimates are comparable

12
For instance, one might conjecture that all stocks, irrespective of their size, would have an
equal probability of being chosen. Thus, all stocks would have an expected weight of 1/N, where N
is the number of stocks available in the market.
13
For each independent variable, I estimate an autoregressive model using the time series of
its coefficient estimates. The standard error of the intercept in this model is the autocorrelation
corrected standard error of the coefficient estimate. The order of the autoregressive model is chosen
such that its Durbin-Watson statistic is close to two. I find that three lags are usually sufficient to
eliminate the serial correlation in errors (DW ≈ 2).
1906 The Journal of FinanceR

Table III
Aggregate Stock Preferences of Individual and Institutional
Investors: Stock-Level Regression Estimates
This table reports the Fama and MacBeth (1973) cross-sectional regression estimates (columns (1),
(2), (3), and (6)) and the panel regression estimates with time fixed effects (columns (4), (5), (7),
and (8)) for the aggregate individual and institutional portfolios. Panel B reports panel regression
estimates from an extended specification that includes the independent variables from Panel A
along with the variables shown in Panel B. The dependent variable in these regressions is the
excess weight assigned to a stock in the aggregate individual or institutional portfolio (see equation
(1) in Section IV.B). All independent variables are measured at the end of month t − 1 and are
defined in Table I, Panels A and B. Total volatility and skewness measures are used in column
(2) of Panel A and columns (2) and (4) of Panel B. In the Fama–MacBeth regression estimation, I
use the Pontiff (1996) method to correct the Fama–MacBeth standard errors for potential higher-
order serial correlation in the coefficient estimates. In the panel regression estimation, to account
for potential serial and cross-correlations, I compute firm- and month-clustered standard errors.
The t-statistics, obtained using corrected standard errors, are reported in parentheses below the
estimates. I winsorize all variables at their 0.5 and 99.5 percentile levels. Both the dependent
variable and the independent variables have been standardized (the mean is set to zero and the
standard deviation is one).

Panel A: Baseline Estimates

Individuals Institutions
Variable (1) (2) (3) (4) (5) (6) (7) (8)

Intercept 0.003 0.003 0.007 −0.041

(10.67) (9.09) (3.23) (−6.31)
Idiosyncratic or 0.056 0.055 0.049 0.059 0.046 −0.044 −0.048 −0.051
total volatility (7.45) (7.24) (5.13) (8.85) (5.66) (−4.37) (−5.32) (−5.78)
Idiosyncratic or 0.047 0.049 0.038 0.052 0.049 −0.071 −0.070 −0.066
total skewness (5.06) (5.27) (5.01) (9.28) (6.14) (−5.55) (−4.04) (−3.69)
Stock price −0.191 −0.190 −0.137 −0.108 −0.124 0.061 0.062 0.059
(−8.99) (−9.93) (−9.77) (−7.97) (−8.73) (5.64) (8.26) (8.51)
Market beta 0.111 0.155 0.100 −0.006 −0.008 0.002
(6.79) (8.69) (8.03) (−2.47) (−2.18) (0.37)
Log(firm size) −0.189 −0.200 −0.183 0.185 0.220 0.156
(−10.41) (−10.79) (−8.81) (4.95) (10.57) (3.16)
Book-to-market ratio −0.071 −0.086 −0.064 0.052 0.058 0.065
(−7.63) (−10.92) (−6.32) (4.12) (5.58) (3.28)
Past 12-month stock return −0.015 −0.013 −0.021 0.024 0.030 0.012
(−2.51) (−1.95) (−2.53) (3.40) (6.96) (1.87)
Systematic skewness −0.012 −0.017 −0.011 0.020 0.014 0.001
(−3.36) (−3.15) (−2.34) (2.03) (2.26) (0.06)
Monthly volume turnover 0.125 0.133 0.150 −0.032 −0.038 −0.037
(8.72) (6.77) (9.51) (−6.51) (−5.07) (−3.70)
Dividend paying dummy −0.069 −0.100 −0.074 0.014 0.016 0.012
(−7.62) (−11.53) (−7.37) (4.37) (2.31) (3.14)
Firm age −0.038 −0.069 −0.047 0.014 0.016 0.011
(−6.63) (−6.65) (−7.17) (3.20) (2.75) (2.02)
S&P500 dummy −0.004 −0.005 −0.008 0.012 0.012 0.014
(−2.51) (−1.90) (−1.96) (3.25) (4.83) (4.22)

(continued)
 Who Gambles in the Stock Market? 1907

Table III—Continued

Panel A: Baseline Estimates

Individuals Institutions
Variable (1) (2) (3) (4) (5) (6) (7) (8)

Nasdaq dummy 0.033 0.024 0.030 −0.016 −0.023 −0.004

(2.96) (4.47) (3.50) (−3.44) (−5.81) (−1.12)
(Mean) Number of 5,979 5,979 5,310 377,010 256,813 4,238 101,761 78,028
observations
(Mean) Adjusted R2 0.049 0.050 0.116 0.103 0.114 0.109 0.132 0.141

Panel B: Robustness Test Results (Panel Regression Estimates)

Individuals Institutions
Variable (1) (2) (3) (4)

High volatility dummy 0.051 0.053 −0.013 −0.019

(5.22) (6.90) (−2.32) (−2.17)
High skewness dummy 0.046 0.042 −0.032 −0.042
(3.25) (3.39) (−3.03) (−2.16)
Low price dummy 0.092 0.107 −0.034 −0.031
(6.51) (6.09) (−3.17) (−3.00)
High volatility × high skewness 0.074 0.079 −0.010 −0.011
(6.32) (5.37) (−1.87) (−1.96)
High volatility × low price 0.028 0.025 −0.005 −0.004
(8.71) (7.01) (−1.16) (−1.13)
High skewness × low price 0.017 0.016 −0.007 −0.007
(3.27) (3.70) (−1.03) (−0.92)
High skewness × high volatility × low price 0.046 0.048 −0.037 −0.033
(4.26) (4.62) (−2.08) (−1.88)
(Other coefficient estimates have been suppressed.)

(0.056 and 0.047, respectively). I find that the stock price measure has the
strongest inf luence on aggregate stock preferences. Specifically, the magnitude
of the coefficient on stock price (= −0.191) is more than three times stronger
than the estimates of the idiosyncratic volatility and idiosyncratic skewness
measures.
To ensure that the stock-level regression results are not simply restating
individual investors’ known preferences for small-cap stocks, value stocks, div-
idend paying stocks, or “attention grabbing” stocks (e.g., Barber and Odean
(2000, 2001, 2008), Graham and Kumar (2006)), I estimate regression spec-
ifications with several control variables. This set includes market beta, firm
size, book-to-market, the past 12-month stock return, systematic skewness
(or coskewness), monthly volume turnover, a dividend-paying dummy, firm
age, an S&P500 dummy, and a Nasdaq dummy. Similar to the three main
independent variables, I measure the control variables at the end of month
t − 1.
1908 The Journal of FinanceR

The full specification results reported in column (3) indicate that the coef-
ficient estimates of all three lottery indicators remain significant in the pres-
ence of control variables. The coefficient estimates of control variables also
have the expected signs. For instance, the coefficient on Firm Size is strongly
negative, which indicates that individual investors exhibit a preference for rel-
atively smaller stocks. The positive coefficients on S&P500 dummy and Volume
Turnover indicate that investors exhibit a preference for relatively more visible
and liquid firms. The positive coefficient on the turnover measure is also con-
sistent with individual investors’ preferences for attention-grabbing stocks as
stocks with high monthly turnover are more likely to be in the news and thus
are more likely to catch the attention of individual investors. Interestingly, in-
dividual investors exhibit an aversion for stocks that have high coskewness and
increase the skewness of the overall portfolio.
Although I correct the Fama–MacBeth standard errors for potential higher-
order autocorrelations, to further ensure that the standard error estimates are
not downward biased I estimate a panel regression specification and compute
month- and firm-clustered standard errors (Petersen (2009)). The estimates
are reported in Table III, column (4). I find that the panel regression estimates
are qualitatively similar to the Fama–MacBeth regression estimates. In abso-
lute terms, the coefficient estimates of volatility and skewness increase, while
the coefficient estimate of stock price decreases. Nevertheless, the price coef-
ficient estimate is almost two times the estimates of volatility and skewness,
and it is still the strongest determinant of individual investors’ aggregate stock
preferences.
Since both dependent and independent variables have been standardized,
the stock-level regression estimates are easy to interpret in economic terms.
The variable EW has a mean of 1.01% and a standard deviation of 3.45%, and
in column (4) Idiosyncratic Volatility has a coefficient estimate of 0.059. This
estimate implies that, all else equal, a one-standard deviation increase in the
idiosyncratic volatility level of a stock would induce a 0.059 × 3.45 = 0.204%
increase in the EW measure for that stock. In percentage terms, relative to
the mean of EW, this corresponds to about a 20% increase in EW, which is
economically significant.
Among the other two lottery characteristics, the coefficient estimate for Id-
iosyncratic Skewness is slightly lower (= 0.052) than the volatility estimate,
while Price has a higher estimate (= −0.108). The mean stock price during
the sample period is $15.51 and its standard deviation is $15.31. Thus, all else
equal, two stocks with prices of $5 and $20 would have a 0.108 × 3.45 = 0.373%
difference in their EW measures. In percentage terms, relative to the mean of
EW, this corresponds to about a 37% difference in EW.
These rough calculations indicate that the statistically significant coefficient
estimates for the three lottery characteristics in stock-level regressions are also
economically significant. Overall, the stock-level regression estimates indicate
that individual investors exhibit a strong aggregate preference for stocks with
lottery features, even after I account for the known determinants of their stock
preferences.
 Who Gambles in the Stock Market? 1909

D. Aggregate Institutional Stock Preferences

Due to the aggregate summing-up constraints, the aggregate individual and
institutional preferences for stocks with lottery features should be roughly op-
posite.14 To investigate whether these fundamental constraints hold, I estimate
stock-level cross-sectional regressions to examine aggregate institutional pref-
erences. These regression estimates are also presented in Table III. Columns (6)
and (7) report the Fama–MacBeth and panel regression estimates, respectively.
Consistent with the summing-up constraints, I find that the individual and
institutional investor groups exhibit roughly opposite preferences. Most impor-
tantly, unlike individual investors, institutions exhibit a relative aversion for
stocks with lottery features, and they overweight stocks with higher coskew-
ness. The other coefficient estimates in the institutional regression are broadly
consistent with previous evidence on aggregate institutional preferences (e.g.,
Bennett et al. (2003), Frieder and Subrahmanyam (2005)).

E. Robustness Checks for Stock-Level Regression Estimates

I conduct additional tests to ensure that the stock-level regression estimates
are robust. In the first test, I ensure that my results are not strongly inf luenced
by microstructure issues or institutional constraints. The concern might be that
the results are induced mechanically by the constraints faced by individual and
institutional investors. For instance, individual investors might be constrained
to hold lower-priced stocks due to the small size of their portfolios. Similarly,
institutional constraints such as prudent man rules might prevent them from
holding lower-priced stocks (Badrinath et al. (1989), Del Guercio (1996)).
When I re-estimate the stock-level regressions after excluding stocks that
are priced below $5, the subsample coefficient estimates for both individual
and institutional portfolios are very similar to the reported full-sample results
(see columns (5) and (8)). Thus, the stock-level regression results do not appear
to be mechanically induced by potential microstructure effects or investors’
constraints.
In the next set of robustness tests, I introduce several interaction terms in
the regression specification to capture investors’ preferences for lottery-type
stocks more accurately. The interaction terms ref lect the definition of lottery-
type stocks more precisely. For these tests, I first define high volatility, high
skewness, and low price dummy variables. The high volatility dummy is set
to one for stocks that are in the highest three volatility deciles. The other two
dummy variables are defined in an analogous manner. Using the three dummy
variables, I define four interaction terms and include them in the regression
specification. For robustness, I consider specifications for both total and id-
iosyncratic volatility and skewness measures. The regression estimates are
presented in Table III, Panel B.
14
Very small institutions and very large and wealthy individual investors are not appropri-
ately represented in the sample. Therefore, the summing-up constraints are not expected to hold
perfectly.
1910 The Journal of FinanceR

The results from the extended regression specifications indicate that the indi-
vidual investors assign larger weights to stocks with higher volatility and skew-
ness levels and lower prices. The dummy variables as well as the interaction
terms have positive and statistically significant estimates for the aggregate in-
dividual investor portfolio. Moreover, the idiosyncratic and total measures yield
very similar results (see columns (1) and (2)). In contrast, when I re-estimate
the extended regression specification for the aggregate institutional portfolio,
the dummy variables and the interaction terms have negative and statistically
weaker coefficient estimates.15
Taken together, the stock-level regression estimates indicate that individual
investors overweight stocks that are more likely to be perceived as lotteries,
while institutions underweight those stocks. Thus, like state lotteries, stocks
with lottery characteristics attract a relatively less sophisticated individual
investor clientele. This evidence provides strong empirical support for the first
hypothesis (H1).

V. Socioeconomic Profile of Lottery Investors

In this section, to gather support for the second and third hypotheses (H2
and H3), I examine how the preference for lottery-type stocks varies cross-
sectionally within the individual investor category.

A. Measuring Individual Preference for Lottery-Type Stocks

I use five distinct but related measures to capture an investor’s preference for
lottery-type stocks. I compute the lottery-stock preference measures for each in-
vestor at the end of each month and use the sample period averages to quantify
an investor’s overall preference for lottery-type stocks. The preference mea-
sures in month t employ the set of lottery-type stocks identified using the stock
price, idiosyncratic volatility, and idiosyncratic skewness measures obtained at
the end of month t − 1.
The first measure of lottery-stock preference (LP) of investor i in month t is
the raw portfolio weight allocated to lottery-type stocks,

nijt P j t
j ∈Lt−1
LP(1)
it = × 100, (2)

Nit
nijt P j t
j =1

where Lt−1 is the set of lottery-type stocks defined at the end of month t − 1, Nit
is the number of stocks in the portfolio of investor i at the end of month t, nijt is
15
The stock-level regression results do not merely reflect the regional preferences of investors
from California (27% of the sample) or the hedging preferences of mutual fund investors. When I re-
estimate the stock-level regressions after excluding investors who reside in California or investors
who hold mutual funds, I find that the subsample coefficient estimates are very similar to the
full-sample estimates.
 Who Gambles in the Stock Market? 1911

the number of shares of stock j in the portfolio of investor i at the end of month
t, and Pjt is the price of stock j in month t.
The second lottery preference measure is the portfolio size adjusted weight in
lottery-type stocks. I define this alternative measure because, even merely due
to chance, an investor with a larger portfolio could allocate a larger weight to
lottery-type stocks.16 To ensure that a large weight in lottery-type stocks is not
mechanically generated by a large portfolio size, I compare the weight investor
i allocates to lottery-type stocks (LP(1)
it ) with an expected weight of lottery-type
stocks in her portfolio that is determined by the size of her portfolio. For ease of
interpretation, I normalize both the actual and the expected portfolio weights
such that they lie between zero and one. The second lottery preference measure
is defined as the percentage difference between the actual and the expected
normalized weight measures:
NW it − ENW it
LP(2)
it = × 100. (3)
ENW it

In equation (3), the actual and the expected normalized weights in lottery-
type stocks for investor i in month t are given by
 (1) 
LP(1)
it − min LPit
N W it =     (4)
max LP(1) it − min LP(1)
it

and
P Sizeit − min(P Sizeit )
ENW it = , (5)
max(P Sizeit ) − min(P Sizeit )
respectively. Here, PSizeit is the total size of the stock portfolio of investor i in
month t, min(PSizeit ) is the minimum portfolio size of the sample investors in
month t, and max(PSizeit ) is the maximum portfolio size of the sample investors
in month t. The min(LP(1) (1)
it ) and max(LPit ) measures are defined in an analogous
manner using the lottery weights of sample investors in month t.
The third lottery preference measure is the market portfolio adjusted weight
in lottery-type stocks. I compare the raw lottery preference measure (LP(1) it )
to the expected weight of lottery-type stocks determined on the basis of total
market capitalization of lottery-type stocks, and obtain the excess percentage
weight allocated to lottery-type stocks. Specifically, the third lottery preference
measure is defined as
LP(1) mkt
it − LPt
LP(3)
it = × 100, (6)
LPmkt
t
16
An investor holding a larger portfolio would hold a greater number of stocks and, thus, she
is more likely to select stocks from the subset of lottery-type stocks. This choice need not reflect
a preference for lottery-type stocks. I find that the correlation between the LP(1) measure and
portfolio size is significantly positive. However, the portfolio size–based adjustment used to define
the LP(2) measure eliminates this mechanically induced correlation between portfolio size and
portfolio weight allocated to lottery-type stocks.
1912 The Journal of FinanceR

where LPmktt is the weight allocated to lottery-type stocks in the aggregate

market portfolio in month t.
In the fourth lottery-type stock preference measure, I compare an investor’s
preference for lottery-type stocks with her preference for nonlottery-type stocks
and obtain a relative lottery preference measure. Specifically, the LP(4)
it measure
is defined as the difference between the excess percentage weight in lottery-type
stocks and the excess percentage weight in nonlottery-type stocks:

LP(2) (2)
it − NLPit
LP(4)
it = × 100. (7)
NLP(2)
it

Since the market capitalization of the nonlottery-type stock category is signifi-

cantly higher (about 40 times) than the capitalization of lottery-type stocks, it is
necessary to examine the excess weight differential. The raw weight differential
does not have a very meaningful interpretation.
Finally, I define a lottery-type stock preference measure using investors’
trades. Because the portfolio of lottery-type stocks changes monthly, under the
position-based measures of lottery-type stocks, a component of the total weight
in lottery-type stocks ref lects an investor’s “passive” preference for lottery-type
stocks. This is the weight allocated to those lottery-type stocks that did not have
lottery features at the time of purchase.
To identify whether investors actively seek lottery-type stocks, each month,
for each investor i, I compute the buy volume for lottery-type stocks (VBLit )
and the total buy volume for all stocks in the portfolio (VBit ). The trade-based
lottery preference measure is defined as the ratio between these two trading
volume measures:
VBLit
LP(5)
it = × 100. (8)
VBit

This measure ref lects the active preference of investor i for lottery-type stocks
in month t.
Given the similarities in their definitions, it is not surprising that the five
lottery preference measures are positively correlated. The average correla-
tion between the position-based lottery preference measures (LP(1) – LP(4) ) is
0.646, while the trade-based measure (LP(5) ) has a weaker correlation with the
position-based measures (average correlation = 0.521).

B. Choice of Independent Variables in Investor-Level Regressions

To characterize the heterogeneity in individual investors’ preferences for
lottery-type stocks, I estimate investor-level cross-sectional regressions, where
the dependent variable is one of the five lottery preference measures defined in
equations (2) to (8).17 A set of variables that capture investors’ socioeconomic

17
Examining the lottery-type stock participation rates, I find that the overall participation rate
is about 35% and it does not vary significantly across the income and wealth categories. See Section
B of the Internet Appendix for additional details.
 Who Gambles in the Stock Market? 1913

characteristics, local economic conditions, and portfolio characteristics are em-

ployed as independent variables. The focus of this analysis is on the coefficient
estimates of socioeconomic variables, which could provide empirical support for
the second and third hypotheses (H2 and H3).
For ease of interpretation, I group the independent variables into three broad
categories. The first set contains the key demographic variables that are known
to explain people’s preferences for state lotteries. The second set of independent
variables contains location-based demographic measures. The last set contains
a number of portfolio characteristics that serve as control variables. To ensure
that investors’ demographic characteristics are not just a nonlinear function of
income, I also include squared income as an additional control variable.

C. Investor-Level Cross-sectional Regression Estimates

The investor-level cross-sectional regression estimates are presented in
Table IV. In specifications (1) to (5), I use one of the five lottery preference
measures (LP(1) to LP(5) ) as the dependent variable.18 For brevity, the coeffi-
cient estimates of all control variables are suppressed.
I find that younger, less wealthy, less educated, nonprofessional single men
invest disproportionately more in lottery-type stocks. The propensity to gamble
with lottery-type stocks is lower among retired investors and among those who
only hold tax-deferred accounts.19 Thus, the demographic attributes that induce
greater lottery participation and expenditures are also associated with greater
investments in lottery-type stocks.
Examining the coefficient estimates of the religion, race, and ethnicity vari-
ables, I find that Catholic dummy has the strongest inf luence on an investor’s
propensity to invest in lottery-type stocks. Specifically, investors who live in
counties with a relatively greater concentration of Catholics (Protestants) in-
vest more (less) in lottery-type stocks. Investment in lottery-type stocks is
also higher in zip codes with a greater concentration of minorities (African-
Americans or Hispanics) and foreign born individuals.20 This evidence indi-
cates that, like state lotteries, investment in lottery-type stocks is correlated
with the religious, racial, and ethnic characteristics of individual investors.
Examining the effects of other geographical factors, I find that investors
who earn less than their “neighbors” (other investors who are located within a

18
As before, I winsorize all variables at their 0.5 and 99.5 percentile levels and standardize
both the dependent and the independent variables. I use clustered standard errors to account for
cross-sectional dependence within zip codes. The estimates are very similar when I assume that
data are clustered by counties or states.
19
I also experiment with an interaction dummy variable in the regression specification that
is set to one for investors who are retired and hold only tax-deferred accounts. I find that this
interaction dummy has a marginally negative coefficient estimate (estimate = −0.014, t-statistic =
−1.56). The evidence indicates that retired investors with only tax-deferred accounts are extra
cautious and allocate lower weights to lottery-type stocks.
20
The Catholic and Hispanic measures are positively correlated (correlation = 0.137) but they
are not substitutes for each other.
1914 The Journal of FinanceR

Table IV
Investor Characteristics and Preference for Lottery-Type Stocks:
Cross-sectional Regression Estimates
This table reports the estimates of investor-level cross-sectional regressions, where the dependent
variable is a measure of the investor’s preference for lottery-type stocks. The lottery-type stock
preference measures are defined in Section V.A and all explanatory variables are defined in Table I,
Panel C. In specifications (1) to (5), one of the lottery-type stock preference measures (LP(1) − LP(5) ,
respectively) is used as the dependent variable. In specification (6), the dependent variable is the
LP(2) measure, but stocks with price below $5 are excluded from the analysis. In specification (7), I
use the LP(1) preference measure for local lottery-type stocks only. In specification (8), I also use the
LP(1) preference measure, but I exclude active traders (investors with portfolio turnover in the top
quintile) from the sample. In all specifications, the set of control variables includes portfolio size,
monthly portfolio turnover, portfolio diversification, local bias, portfolio dividend yield, portfolio
industry concentration, the four factor exposures of the portfolio, and squared income. For brevity,
the coefficient estimates of these control variables are suppressed. The t-statistics for the coefficient
estimates are reported in parentheses below the estimates. Both the dependent and independent
variables have been standardized such that each variable has a mean of zero and a standard
deviation of one.

Variable (1) (2) (3) (4) (5) (6) (7) (8)

Intercept −0.019 0.008 −0.006 −0.007 −0.009 0.020 −0.010 −0.013

(−1.30) (0.82) (−0.68) (−0.48) (−0.91) (1.48) (−0.97) (−0.90)
Wealth −0.052 −0.081 −0.051 −0.068 −0.059 −0.110 −0.057 −0.040
(−3.64) (−5.30) (−4.40) (−4.43) (−4.95) (−4.51) (−3.97) (−3.33)
Age −0.044 −0.063 −0.051 −0.064 −0.038 −0.114 −0.048 −0.039
(−4.43) (−5.77) (−3.54) (−5.76) (−3.42) (−7.15) (−4.38) (−3.82)
Zip code education −0.052 −0.061 −0.046 −0.060 −0.036 −0.126 −0.051 −0.073
(−5.29) (−6.23) (−4.68) (−5.99) (−3.65) (−3.71) (−4.11) (−5.27)
Professional dummy −0.041 −0.023 −0.035 −0.027 −0.011 −0.059 −0.021 −0.036
(−3.70) (−2.06) (−3.17) (−2.48) (−1.97) (−2.75) (−1.98) (−2.24)
Retired dummy −0.037 −0.041 −0.027 −0.040 −0.011 −0.048 −0.036 −0.039
(−2.99) (−3.33) (−2.20) (−3.23) (−1.85) (−2.71) (−3.03) (−2.17)
Male dummy 0.033 0.024 0.023 0.036 0.034 0.046 0.027 0.029
(3.12) (2.16) (2.10) (3.45) (3.03) (2.73) (2.51) (2.59)
Married dummy −0.016 −0.010 −0.023 −0.011 −0.021 −0.030 −0.023 −0.011
(−1.55) (−1.10) (−2.19) (−1.19) (−1.99) (−2.06) (−2.16) (−1.72)
Investment experience 0.067 0.064 0.081 0.045 0.004 0.019 0.063 0.079
(7.23) (6.96) (8.79) (4.79) (0.41) (2.76) (6.43) (5.49)
Taxable account only 0.034 0.027 0.027 0.018 0.045 0.030 0.029 0.023
dummy (3.58) (2.24) (2.20) (1.79) (4.76) (2.48) (2.88) (1.89)
Tax deferred acc. only −0.022 −0.053 −0.015 −0.025 −0.029 −0.067 −0.048 −0.033
dummy (−2.39) (−4.67) (−1.31) (−2.54) (−3.24) (−5.24) (−4.05) (−1.95)
Catholic county 0.053 0.049 0.043 0.035 0.032 0.058 0.078 0.044
dummy (5.13) (3.69) (3.68) (3.85) (2.77) (3.98) (7.39) (3.79)
Protestant county −0.046 −0.035 −0.027 −0.024 −0.030 −0.051 −0.040 −0.051
dummy (−3.19) (−3.38) (−3.69) (−2.12) (−3.38) (−4.19) (−4.54) (−4.24)
Zip code Afr. 0.028 0.025 0.028 0.023 0.019 0.021 0.023 0.030
Am.–White ratio (3.51) (3.67) (2.64) (3.28) (2.14) (2.11) (3.19) (2.96)
Zip code Hispanic– 0.034 0.031 0.040 0.028 0.020 0.046 0.036 0.034
White ratio (2.97) (3.07) (3.56) (3.19) (2.20) (3.00) (3.28) (3.38)
Zip code prop foreign 0.020 0.017 0.011 0.021 0.011 0.007 0.016 0.020
born (2.28) (2.01) (1.64) (2.18) (1.45) (0.74) (1.89) (2.18)

(continued)
 Who Gambles in the Stock Market? 1915

Table IV—Continued

Variable (1) (2) (3) (4) (5) (6) (7) (8)

Income relative to −0.040 −0.042 −0.031 −0.044 −0.094 −0.103 −0.045 −0.037
neighbors (−3.61) (−4.97) (−3.07) (−5.14) (−8.65) (−8.34) (−4.92) (−3.52)
Urban dummy 0.030 0.032 0.025 0.018 0.017 0.004 0.029 0.015
(2.32) (3.15) (2.81) (2.75) (2.31) (1.28) (2.66) (1.85)
County unemployment 0.035 0.026 0.030 0.027 0.017 0.027 0.031 0.034
rate (3.94) (2.91) (3.14) (2.99) (2.50) (2.30) (4.14) (2.64)
State lottery 0.031 0.026 0.027 0.030 0.023 0.025 0.027 0.024
expenditure (3.17) (2.66) (2.97) (3.11) (1.96) (2.15) (2.46) (2.77)
State lottery age 0.044 0.046 0.037 0.036 0.031 0.039 0.051 0.030
(4.60) (4.75) (3.75) (3.70) (2.51) (2.69) (3.97) (3.12)
Number of investors 21,194 21,194 21,194 21,194 18,650 21,194 21,194 16,955
Adjusted R2 (0.043) (0.058) (0.035) (0.052) (0.031) (0.061) (0.055) (0.040)

25-mile radius) and live in urban regions exhibit stronger preference for lottery-
type stocks. This evidence indicates that to some extent gambling-motivated
investments are likely to be inf luenced by a desire to maintain or increase
upward social mobility. Local economic conditions, as captured by a county’s
unemployment rate are also associated with investors’ decisions to hold lottery-
type stocks. In particular, consistent with the evidence from lottery studies, the
propensity to gamble is greater in regions with higher unemployment rates.
Another intriguing piece of evidence that emerges from the coefficient esti-
mates of geographical factors is that investors who live in states with favorable
lottery environments invest more in lottery-type stocks. The average invest-
ment in lottery-type stocks is higher in states that adopted lotteries early and
that have higher per capita consumption of lotteries. Thus, greater acceptabil-
ity of gambling in a state is associated with greater investment in lottery-type
stocks. This direct link between lottery expenditures and investments in lottery-
type stocks indicates that they act as complements.
The coefficient estimates using the trade-based lottery-type stock preference
measure as the dependent variable are reported as specification (5) in Table IV.
I find that these coefficient estimates are qualitatively similar to the esti-
mates obtained using the position-based lottery preference measures reported
in columns (1) to (4). With the trade-based measure, education level, urban lo-
cation, and state lottery expenditure measures are the strongest correlates of
investors’ propensity to invest in lottery-type stocks.
The coefficient estimates of unreported control variables are also as expected.
For instance, investors who hold better diversified portfolios and exhibit a pref-
erence for high dividend yield stocks invest less in lottery-type stocks. In con-
trast, investors who hold portfolios with greater industry concentration exhibit
stronger preference for lottery-type stocks.
Since all variables in investor-level regressions are standardized, the coeffi-
cient estimates are easy to interpret in economic terms. For instance, Age has
1916 The Journal of FinanceR

a coefficient estimate of −0.044 in the first specification, which implies that,

all else equal, a one-standard deviation increase in the age of an investor is
associated with a 0.044 × 16.62 = 0.73% reduction in the weight allocated to
lottery-type stocks.21 Thus, a 65-year-old investor would allocate 2.19% lower
weight to lottery-type stocks than he would have allocated at the age of 30 (a
three-standard deviation change in age).
Another interpretation of the Age coefficient estimate is that the differential
in the lottery weights is 2.19% if two investors are similar on all dimensions
but their age differential is 35. If the older investor is also a Protestant, she
would further reduce the weight allocated to lottery-type stocks by 0.76%, and
the total weight differential would be 2.95%. In percentage terms, relative to
the mean of LP(1) , there is an economically significant (=28.48%) reduction in
lottery weight.

D. Robustness Checks for Investor-Level Regression Estimates

To examine the robustness of the investor-level regression estimates, I con-
duct five sets of additional tests. First, I examine whether the cross-sectional
regression estimates are sensitive to microstructure issues (e.g., large bid-ask
spread) that might make the identification of lottery-type stocks noisy. I re-
define the lottery-type stock preference measure such that I only consider
stocks with a price above $5. These estimates are also presented in Table IV
(column (6)). I find that these results are qualitatively similar to the baseline
estimates reported in column (2). Thus, investors’ gambling preferences rather
than microstructure effects are the primary drivers of the investor-level cross-
sectional regression results.
In the second robustness test, I examine whether the preference for lottery-
type stocks ref lects an informational advantage rather than a preference for
gambling. Ivković and Weisbenner (2005) find that individual investors exhibit
a preference for stocks in their vicinity, perhaps because they have better in-
formation about those stocks. Motivated by their evidence, I examine whether
investment in local lottery-type stocks ref lects an informational advantage.
Specifically, I compute the portfolio weight allocated to lottery-type stocks
using only investors’ local stocks (stocks that are within 100 miles of the in-
vestor’s location) and re-estimate the investor-level cross-sectional regression.
The results indicate that investors who prefer lottery-type stocks do not dif-
ferentiate between local and nonlocal lottery-type stocks (see column (7)). The
coefficient estimates with local lottery weights are very similar to those ob-
tained using total lottery weights (see columns (1) to (5)). The similarities in
these results indicate that gambling preferences rather than local bias-induced
informational advantage inf luence the cross-sectional relation between lottery
preferences and socioeconomic characteristics.22

21
The LP(1) measure has a mean of 10.36% and a standard deviation of 16.62%.
22
I also explicitly examine whether investors have superior information about local
lottery-type stocks. If investors are informed, the local lottery-type stocks they buy should
 Who Gambles in the Stock Market? 1917

In the third set of robustness tests, I entertain the possibility that a large
portfolio weight in lottery-type stocks is a ref lection of investor overconfidence
rather than an indicator of strong lottery preference. Each of the three lot-
tery characteristics used to define lottery-type stocks could potentially induce
greater overconfidence. In particular, stocks with high idiosyncratic volatility
are harder to value, provide noisier feedback, and could amplify investors’ be-
havioral biases such as overconfidence. Volatility and skewness are positively
correlated and, thus, skewness could have a similar effect on investor overcon-
fidence. Further, higher levels of valuation uncertainty (e.g., higher levels of in-
tangible assets) associated with low-priced stocks could induce greater overcon-
fidence (e.g., Daniel, Hirshleifer, and Subrahmanyam (1998, 2001), Hirshleifer
(2001), Kumar (2009)).
To distinguish between overconfidence- and gambling-based explanations, I
first examine whether investors who allocate a larger weight to lottery-type
stocks also trade actively. Since active trading is one of the defining features
of overconfidence, a positive lottery weight–turnover relation would be consis-
tent with the conjecture that investors over-weight lottery-type stocks due to
their higher levels of overconfidence. I find that investors who invest in lottery-
type stocks at least once during the sample period (lottery participants) trade
less frequently. The average monthly portfolio turnover of nonparticipants and
participants is 7.05% and 6.23%, respectively. Within the group of investors
who hold lottery-type stocks, portfolio turnover declines monotonically with
lottery weight. For the five lottery-weight (LP(1) ) sorted quintiles, the average
turnover rates are 11.34%, 7.58%, 5.37%, 4.23%, and 2.91%, respectively. If
portfolio turnover is a reasonable proxy for overconfidence, this evidence indi-
cates that larger investment in lottery-type stocks is unlikely to be induced by
overconfidence.
In the second overconfidence test, I exclude investors whose portfolio turnover
is in the highest quintile (active traders) and re-estimate the investor-level re-
gression for a subsample of investors who trade moderately and are unlikely to
exhibit the overconfidence bias. If overconfidence induces a strong relation be-
tween socioeconomic characteristics and lottery preferences, this relation would
be considerably weaker for the subsample of investors who are unlikely to ex-
hibit overconfidence. The subsample results are presented in column (8) of
Table IV. I find that the subsample estimates are very similar to the full-sample
estimates reported in column (2), which indicates that the relation between
investors’ socioeconomic characteristics and lottery preferences is unlikely to
ref lect overconfidence.
In the third overconfidence test, I examine whether overconfidence has an
incremental ability to explain investors’ decision to hold lottery-type stocks.

outperform the local lottery-type stocks they sell. However, I find that the average k-day re-
turns following purchases is lower than the average k-day returns following sales. For k =
5, 10, 21, 42, 63, 84, 105, 126, and 252, the average post-trade buy–sell return differentials are
−0.25%, −0.34%, −0.26%, −0.99%, −1.32%, −1.47%, −1.75%, −2.95%, and −5.64%, respectively.
This evidence is inconsistent with the local bias–induced information asymmetry hypothesis.
1918 The Journal of FinanceR

For this test, I define an Overconfidence dummy, which is set to one for in-
vestors who belong to the highest portfolio turnover quintile and the lowest
risk-adjusted performance quintile. The measure is defined under the assump-
tion that overconfident investors would trade most actively and those trades
would hurt their portfolio performance the most. When I include Overconfi-
dence dummy in investor-level regression specifications, I find that it has a sig-
nificantly positive estimate in all instances. For instance, in specification (2),
Overconfidence dummy has a significantly positive estimate (estimate = 0.079,
t-statistic = 7.85) and the other coefficient estimates reported in Table IV re-
main very similar.23 This evidence indicates that overconfidence has an incre-
mental ability to explain investors’ preference for stocks with lottery features.24
In addition to these new results, the main investor-level regression results
presented in Table IV do not have a meaningful economic interpretation un-
der the overconfidence-based explanation. For example, high levels of overcon-
fidence in high unemployment regions or a greater degree of overconfidence
among Catholics is not predicted by any overconfidence theory, but these re-
sults are strongly consistent with the evidence from lottery studies and have a
natural interpretation under the gambling hypothesis.
In the fourth robustness test, I examine whether investors over-weight lot-
tery stocks not because of their gambling preferences but merely because those
stocks are in the news more often. Specifically, I re-estimate the investor-level
regression, where the dependent variable is the first lottery preference mea-
sure, but the set of lottery-type stocks excludes stocks that have turnover in the
highest quintile and are more likely to be in the news. In untabulated results,
I find that the investor-level regression estimates for this subsample are qual-
itatively very similar to the full-sample estimates. For instance, the coefficient
on Wealth is −0.043 (t-statistic = −3.27), Education has an estimate of −0.084
(t-statistic = −4.49), and Catholic dummy has a strong positive estimate (co-
efficient = 0.056, t-statistic = 4.56). This evidence indicates that news is not
the primary channel through which investors identify lottery-type stocks. In-
vestors with socioeconomic characteristics of lottery players over-weight even
those lottery-type stocks that are less likely to be in the news.
In the last set of robustness tests, I investigate whether one of the lottery
characteristics or some combination of those characteristics are more important
for explaining investors’ gambling preferences. The results are discussed in
Section C of the Internet Appendix. The evidence indicates that stock price is
the most important lottery characteristic, followed by idiosyncratic skewness.
The least important lottery characteristic appears to be idiosyncratic volatility.

23
This result is not mechanically induced. See Section D of the Internet Appendix for further
details.
24
I also conduct two additional tests to entertain the overconfidence hypothesis. In the first test,
I define an alternative overconfidence proxy (the difference between the average k-day returns
following stock sales and purchases). Next, I consider a subsample of lottery stocks that have
moderate levels of intangible assets and are less likely to be associated with overconfidence. The
relation between socioeconomic characteristics and lottery weight remains strong in both cases
and I do not find evidence consistent with the overconfidence hypothesis.
 Who Gambles in the Stock Market? 1919

E. Regional Gambling Preferences and Investments in Lottery-Type Stocks

Although the evidence from investor-level regressions indicates that the lo-
cal lottery environment inf luences the propensity to gamble with lottery-type
stocks, the relation is identified with some noise because the variables used
in the regression model are measured at different levels of aggregation. For
greater accuracy, I re-examine the inf luence of the local lottery environment
on lottery investments using variables that are defined at the same (either zip
code or state) level of aggregation.
Focusing on the relation between the aggregate measures of lottery expendi-
ture and investment in lottery-type stocks, I find that the correlation between
per capita state-level lottery expenditure and mean state-level portfolio weight
in lottery-type stocks is significantly positive (correlation = 0.303, p-value =
0.035). The correlation between lottery age and the mean state-level portfolio
weight in lottery-type stocks is even stronger (correlation = 0.417, p-value =
0.014). Surprisingly, the correlations between the mean state-level portfolio
weight in nonlottery-type stocks and the lottery environment measures (per
capita lottery sales and lottery age) are significantly negative (correlations are
−0.172 and −0.284, and the p-values are 0.054 and 0.033, respectively).25
Because state-level lottery sales data might be a crude proxy for regional lot-
tery environment, I obtain zip code–level lottery sales data for several states.
In the empirical exercise, I focus on the zip code–level lottery sales data for
California, which has the largest (about 27%) proportion of sample investors.
Unfortunately, zip code–level data are available only for more recent years
(2005 and 2006). In spite of the nonoverlapping time periods for the lottery
sales and brokerage data sets, I find that the zip code–level per capita lottery
sales and the zip code–level investment in lottery-type stocks are positively cor-
related (correlation = 0.106, p-value = 0.035). Moreover, when I sort zip codes
using the per capita lottery sales measure, the LP(2) lottery preference mea-
sure for the lowest and the highest lottery sales deciles is 45.19% and 84.94%,
respectively.
These correlation estimates indicate that the mean investment levels in
lottery-type stocks are higher in regions with favorable lottery environments.
In light of the extant evidence from lottery studies, this evidence indicates that
individual investors are likely to perceive stocks with lottery features as valid
gambling devices.26

25
Given the positive correlation with lottery-type stocks, this negative correlation is not me-
chanically induced because lottery-type stocks represent only a small segment of the aggregate
portfolio and there is a large “other stocks” category between the lottery-type and nonlottery-type
stock categories.
26
While I find a strong correlation between per capita lottery sales and investment in lottery-type
stocks within a region, I am unable to establish a causal link. To establish causality, one could use
lottery advertisement expenses in a region as an instrument. The regional advertisement expense
is likely to be an effective instrument because it would be correlated with regional lottery sales but
there is no obvious link between the advertising measure and investment in lottery-type stocks
within a region. Unfortunately, lottery advertising data are confidential and are not available
from state lottery agencies. I thank an anonymous referee for suggesting this instrument and the
associated test.
1920 The Journal of FinanceR

Table V
State-Level Preference for Lottery-Type Stocks: Panel
Regression Estimates
This table reports the estimates from state-level panel regressions with month fixed effects. The
dependent variable is the average weight allocated to lottery-type stocks by brokerage investors
in state i in month t. In specifications (1) and (2), the first lottery preference measure is used.
Specifications (3)–(6) use lottery preference measures LP(2) –LP(6) , respectively. The lottery-type
stock preference measures are defined in Section V.A. The set of control variables includes mean
investor age, mean household income, squared income, mean education level, proportion of male
population in the state, proportion married, proportion African American, proportion Hispanic,
proportion foreign born, and mean local bias of investors in the state. For brevity, the coefficient
estimates of the control variables have been suppressed. Additional details on main independent
variables are provided in Table I, Panel D. I use state- and month-clustered standard errors to
compute the t-statistics. The t-statistics for the coefficient estimates are reported in parentheses
below the estimates. Both the dependent and independent variables have been standardized such
that each variable has a mean of zero and a standard deviation of one.

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

Annual per capita state lottery 0.057 0.050 0.107 0.037 0.109 0.121
expenditure (4.99) (3.66) (3.36) (2.04) (3.69) (5.49)
State lottery age 0.136 0.107 0.228 0.109 0.192 0.174
(2.66) (2.440 (8.98) (7.63) (6.45) (6.92)
Monthly state unemployment 0.068 0.062 0.026 0.047 0.126 0.129
rate (6.85) (2.56) (2.17) (2.18) (5.98) (2.02)
Catholic state dummy 0.231 0.192 0.096 0.232 0.259
(6.44) (7.91) (3.65) (9.85) (2.92)
Protestant state dummy −0.094 −0.114 −0.116 −0.176 −0.098
(−2.65) (−3.65) (−3.45) (−6.04) (−4.10)
(Estimates of control variables have been suppressed.)
Number of observations 2,236 2,236 2,236 2,236 2,236 2,236
Adjusted R2 0.024 0.047 0.097 0.067 0.092 0.057

To further quantify the relation between regional lottery environments

and investments in lottery-type stocks, I estimate state-level panel regres-
sions, where the dependent variable is the mean state-level preference for
lottery-type stocks. The independent variables capture the lottery environ-
ment and the socioeconomic characteristics of the state. I estimate one
regression specification for each of the five lottery preference measures. The
panel regression estimates are reported in Table V, where following Petersen
(2009), I use state- and month-clustered standard errors to compute the
t-statistics.
Consistent with the investor-level regression estimates and the correlation
estimates, I find that regional lottery environment inf luences investors’ pref-
erence for lottery-type stocks (see column (1)). The proportional investment
in lottery-type stocks is higher in states with favorable lottery environments
and higher unemployment rates. Adding the religion variables to the regres-
sion specification (see column (2)) does not change those estimates consider-
ably. More importantly, I find that the effect of religious affiliation on lottery
 Who Gambles in the Stock Market? 1921

investment is evident even in the aggregate state-level regressions. The mean

investment in lottery-type stocks is higher (lower) in states with a stronger
concentration of Catholics (Protestants). Even when I consider other lottery
preference measures (specifications (3)–(6)), the coefficient estimates are re-
markably similar to the baseline estimates reported in column (2).
The correlation estimates and state-level regression estimates indicate that
the demand for state lotteries and the mean state-level investment in lottery-
type stocks are associated with a common set of socioeconomic characteris-
tics. The state-level results also indicate that state lotteries do not saturate
the aggregate gambling demand of state investors. Overall, the results from
state-level regressions, in conjunction with the evidence from investor-level re-
gressions, provide strong support for the second and third hypotheses (H2 and
H3).

VI. Time Variation in Lottery Preferences

In this section, I test the fourth hypothesis (H4), which posits that lottery
demand and aggregate demand for lottery-type stocks is correlated over time
because they are induced by common economic factors. In particular, like aggre-
gate lottery demand, individual investors’ aggregate demand for lottery-type
stocks should increase during economic downturns.

A. Time-Series Regression Model

I examine the time variation in the aggregate demand for lottery-type stocks
by estimating the following time-series regression model

EBSIt = b0 + b1 UNEMPt−1 + b2 UEIt−1 + b3 MPt−1 + b4 RPt−1 + b5 T St−1

+ b6 EFCt−1 + b7 EFCt + b8 MKTRET t−1 + b9 MKTRET t
+ b10 LOTRET t−1 + b11 LOTRET t + b12 E BS It−1 + εt . (9)

The dependent variable in the model is the excess buy–sell imbalance

(EBSI) for lottery-type stocks in a given month. This measure captures the
change in investors’ bullishness toward lottery-type stocks relative to the
change in their bullishness toward other remaining stocks. It is defined
as EBSIt = LBSIt − OBSIt , where LBSIt is the month-t buy–sell imbalance
of a portfolio of lottery stocks, and OBSIt is the month-t buy–sell imbal-
ance of a portfolio that contains the other remaining stocks.27 The portfolios

27
In my analysis, the composition of portfolios of lottery-type stocks and other stocks changes
every month. However, the time-series regression estimates are very similar if those two portfolios
are defined at the beginning of each year or the beginning of the sample period and held fixed
during the entire year or the entire sample period, respectively.
1922 The Journal of FinanceR

of lottery-type stocks and other stocks are defined at the end of month
t − 1.28
The independent variables in the regression specification include the fol-
lowing five macroeconomic variables that vary significantly over the business
cycle (Chen, Roll, and Ross (1986), Ferson and Schadt (1996)): monthly U.S.
unemployment rate (UNEMP), unexpected inf lation (UEI), monthly growth in
industrial production (MP), monthly default risk premium (RP), and the term
spread (TS). To examine whether investors’ trading behavior is inf luenced by
changes in the expected future cash f lows of lottery-type stocks, I use revi-
sions in analysts’ forecasts of future earnings (EFC) as a proxy for changes in
investors’ expectations about future cash f lows.29
Additionally, investors are known to be sensitive to past returns. They might
trade in response to recent market returns or returns from lottery-type stocks
(e.g., Odean (1999), Barber and Odean (2008)). To capture the effects of returns
on investors’ trading activities, I include the market (MKTRET) and the lottery
portfolio returns (LOTRET) as additional independent variables. Last, I use
the 1-month lagged EBSI variable as an explanatory variable to control for
potential serial correlation in that measure.

B. Time-Series Regression Estimates Using the Brokerage Data

First, I use the brokerage sample to estimate the time-series regression
model. Although the January 1991 to November 1996 sample period is short,
the macroeconomic variables exhibit considerable time variation during this
period. Thus, if investors’ propensity to invest in lottery-type stocks is inf lu-
enced by changes in macroeconomic conditions, the trading intensity should
vary over time and the relation between macroeconomic indicators and trading
intensity may be identified.
The time-series regression estimates are presented in Table VI. The re-
sults indicate that higher unemployment rates are associated with greater
relative demand shifts for lottery-type stocks (coefficient estimate = 0.189,
t-statistic = 2.54). Furthermore, EBSI is higher when the default risk pre-
mium is higher, to compensate for the relatively poor state of the econ-
omy (coefficient estimate = 0.124, t-statistic = 2.75). The remaining three
macroeconomic variables have statistically insignificant coefficient estimates.

 N pt
28
The buy–sell imbalance (BSI) of portfolio p in month t is defined as BS I pt = 100
N pt i=1 BSIit ,
 Dt
(VBijt −VSijt )
where the BSI for stock i in month t is defined as BSIit = j =1
 Dt . Here, Dt is the number of
j =1
(VBijt +VSijt )
days in month t, VBijt is the buy volume (measured in dollars) for stock i on day j in month t, VSijt
is the sell volume (measured in dollars) for stock i on day j in month t, and Npt is the number of
stocks in portfolio p formed in month t. See Kumar and Lee (2006) for further details of the BSI
measure, including a discussion about why an equal-weighted BSI measure is more appropriate
for capturing shifts in investor sentiment.
29
If trading in lottery-type stocks is motivated mainly by investors’ gambling preferences, in-
vestors would not pay much attention to the fundamentals. Nevertheless, to choose a subset of
stocks from the larger set of lottery-type stocks, they might consider the fundamentals.
 Who Gambles in the Stock Market? 1923

Table VI
Macroeconomic Conditions and Demand Shifts: Time-Series
Regression Estimates
This table reports the estimation results for the time-series regression model defined in equation (9).
The dependent variable is the excess buy–sell imbalance (EBSI) in month t. Among the independent
variables, UNEMPt is the U.S. unemployment rate in month t, UEIt is the unexpected inf lation in
month t, MPt is the monthly growth in industrial production, RPt is the monthly risk premium, TSt
is the term spread, EFCt is the mean change in analysts’ earnings forecasts of lottery-type stocks
in month t, MKTRETt is the monthly market return, and LOTRETt is the mean monthly return
on lottery-type stocks. Table I, Panel E provides additional details on the independent variables.
In specifications (1) to (5), EBSI is computed using the individual investor data from a large U.S.
discount brokerage house for the 1991 to 1996 period. In specification (6), I use a proxy for retail
trading obtained from the ISSM and TAQ databases for the 1983 to 2000 period. Additional details
on the regression specification are available in Section VI.A. Both the dependent variable and the
independent variables have been standardized. Newey and West (1987) adjusted t-statistics for the
coefficient estimates are reported in parentheses below the estimates.

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

Intercept 0.001 0.972 1.399 0.972 0.023 −0.010

(0.20) (1.29) (1.55) (1.10) (0.25) (−0.16)
Lagged UNEMP 0.202 0.189 0.135
(2.74) (2.54) (3.14)
Lagged UEI −0.091 −0.072 −0.053
(−1.36) (−0.77) (−1.05)
Lagged MP −0.013 0.016 −0.022
(−0.69) (0.20) (−0.45)
Lagged RP 0.507 0.124 0.112
(4.59) (2.75) (3.16)
Lagged TS −0.114 −0.012 −0.066
(−0.44) (−0.50) (−0.91)
Lagged EFC 0.034 0.071 0.014
(0.78) (1.59) (1.44)
EFC 0.003 −0.014 0.005
(0.40) (−0.17) (0.32)
Lagged LOTRET 0.102 −0.037 −0.029
(1.02) (−0.12) (−0.32)
LOTRET 0.498 0.427 0.816
(3.54) (2.07) (11.02)
Lagged MKTRET −0.047 −0.017 −0.054
(−1.33) (−0.51) (−0.83)
MKTRET −0.127 −0.111 −0.219
(−1.56) (−1.45) (−3.30)
Lagged EBSI 0.412 0.215
(3.99) (3.00)
Number of Months 71 71 71 71 70 215
Adjusted R2 0.085 0.204 0.005 0.181 0.396 0.493

The time-series regression estimates also indicate that EFC, which proxies
for investors’ changing expectations about future cash f lows, has insignificant
coefficient estimates. This evidence indicates that changes in investors’ trad-
ing activities in lottery-type stocks are unlikely to be driven by changing ex-
pectations about stock fundamentals. The coefficient estimates of the control
1924 The Journal of FinanceR

variables are also as expected. For instance, lagged EBSI has a positive coeffi-
cient estimate, which indicates that there is persistence in investors’ differen-
tial demand shifts.30
In economic terms, the time-series regression estimates are significant. Dur-
ing the sample period, the U.S. unemployment rate has a mean of 6.40%
and a standard deviation of 0.78%, while EBSI has a mean of −0.75% and
a standard deviation of 8.06%. Because all variables have been standard-
ized, the unemployment rate increases of one percentage point (say, from 5%
to 6%) corresponds to a 1.28-standard deviation increase in unemployment.
Thus, a one percentage point change in unemployment rate corresponds to a
1.28 × 0.189 × 8.06 = 1.95% increase in EBSI. In absolute terms, this increase
is more than 2.5 times the mean of EBSI and represents an economically sig-
nificant shift.

C. Time-Series Regression Estimates Using “Small-Trades” Data

To examine the robustness of the time-series regression estimates, I construct
a proxy for retail trading using the ISSM and TAQ databases and re-estimate
the time-series regression. One of the main advantages of the ISSM/TAQ data is
that they are available from 1983 to 2000, which is considerably longer than the
6-year brokerage sample. The macroeconomic variables exhibit greater varia-
tion during the 18-year period and, therefore, their potential inf luence on the
aggregate demand for lottery-type stocks may be identified more accurately.
The small trades data capture retail trading quite well because the BSI time
series computed using the small-trades data is positively correlated with the
BSI time series obtained using the brokerage data. The correlations between
the two BSI time series for lottery-type stocks and other stocks are 0.504 and
0.533, respectively. Even the EBSI time series obtained using the two samples
have a strong, positive correlation of 0.526. These correlation estimates indicate
that the small-trades data from ISSM/TAQ capture retail trading reasonably
well.
The time-series regression estimates indicate that the coefficient estimates
obtained using the small-trades data are qualitatively very similar to those
obtained using the brokerage sample (see Table VI, column (6)). For exam-
ple, with the small-trades data, the coefficient estimates of lagged UNEMP,
lagged RP, and LOTRET are 0.135 (t-statistic = 3.14), 0.112 (t-statistic = 3.16),
and 0.816 (t-statistic = 11.02), respectively. In comparison, using the broker-
age data, the corresponding coefficient estimates are 0.189 (t-statistic = 2.54),
0.124 (t-statistic = 2.75), and 0.427 (t-statistic = 2.07), respectively. These com-
parisons indicate that the time-series relation between lottery demand and
30
For robustness, I consider additional lags of the EBSI variable in the regression specification.
In untabulated results, I find that those lagged variables have statistically insignificant coefficient
estimates. I also experiment with other regression specifications that include contemporaneous
values of macroeconomic variables, lagged unemployment rates measured over a quarter, and
innovations in unemployment rates. These estimates are qualitatively similar to the reported
results.
 Who Gambles in the Stock Market? 1925

economic indicators identified using the relatively short brokerage sample is

quite robust.
Collectively, the time-series regression estimates using brokerage and
ISSM/TAQ data indicate that, like lottery demand, investors’ propensity to
buy lottery-type stocks is higher during economic downturns. This evidence is
consistent with the fourth hypothesis (H4).

VII. Lottery Preferences and Portfolio Performance

In the last set of tests, I investigate whether investment in lottery-type stocks
has a positive or an adverse inf luence on portfolio performance. I also exam-
ine whether investment in lottery-type stocks is regressive, where portfolio
underperformance related to investment in lottery-type stocks decreases with
income.
On the one hand, if people with strong gambling preferences find the
small possibility of a very large return attractive, then like lottery play-
ers, lottery investors should be willing to invest in lottery-type stocks, even
when they are expected to underperform. Furthermore, the magnitude of
the underperformance induced by investment in lottery-type stocks might be
greater among investors with stronger lottery preferences (e.g., low-income
investors).
On the other hand, although lottery-type stocks earn lower average perfor-
mance, there is significant heterogeneity in their performance. It is therefore
possible that investors with strong gambling preferences are able to identify
lottery-type stocks with superior performance, assign larger weight to those
lottery-type stocks, and generate higher overall returns from their lottery in-
vestments. In this scenario, greater allocation in lottery-type stocks would be
induced by an informational advantage rather than investors’ pure gambling
preferences.

A. Performance of Lottery-Type Stocks

Prior to estimating the potential economic costs associated with invest-
ments in lottery-type stocks, I examine the performance of lottery-type stocks.
For comparison, I also report the performance of portfolios of “nonlottery
stocks” and “other stocks.” All three stock portfolios are defined in Section III.
Table VII reports the characteristics and performance of the three value-
weighted portfolios.
The performance estimates indicate that lottery-type stocks earn signifi-
cantly lower average returns, relative to both nonlottery and other stock cat-
egories. Specifically, relative to the nonlottery stock portfolio, the annualized
raw, characteristic-adjusted, and risk-adjusted performance differentials are
−7.96%, −4.98%, and −7.10%, respectively. Relative to the “other stocks” portfo-
lio, the annualized raw, characteristic-adjusted, and risk-adjusted performance
differentials are −6.74%, −4.19%, and −6.23%, respectively. Thus, irrespective
of the benchmark used and irrespective of the performance measure used,
1926 The Journal of FinanceR

Table VII
Performance of Lottery-Type Stocks: Time-Series
Regression Estimates
This table reports the characteristics and performance of three value-weighted portfolios for the
1980 to 2005 period: lottery-type stocks, nonlottery stocks, and other stocks. The construction of
these three stock portfolios is described in Section III. The following performance measures are
reported: mean monthly portfolio return (MeanRet), standard deviation of monthly portfolio re-
turns (SD), characteristic-adjusted mean monthly portfolio return (CharAdjRet), and the intercept
(Alpha) as well as the factor exposures (RMRF, SMB, HML, and UMD are the exposures to the mar-
ket, size, value, and momentum factors, respectively) from a four-factor model. The characteristic-
adjusted returns are computed using the Daniel et al. (1997) method. Only stocks with CRSP share
code 10 and 11 are included in the analysis. The t-statistics for the coefficient estimates are reported
in parentheses below the estimates.

Portfolio MeanRet SD CharAdjRet Alpha RMRF SMB HML UMD Adj. R2

Lottery (L) 0.472 7.934 −0.375 −0.552 1.204 1.130 −0.049 −0.442 0.880
(−2.95) (−3.22) (18.90) (15.15) (−0.78) (−8.05)
Nonlottery 1.135 4.025 0.040 0.041 0.920 −0.123 0.102 −0.008 0.963
(NL) (0.47) (0.84) (28.39) (−8.16) (5.77) (−0.82)
Others (O) 1.033 4.644 −0.026 −0.033 0.959 0.099 −0.103 −0.010 0.981
(−1.12) (−0.83) (18.39) (7.90) (−6.82) (−1.19)
L–NL −0.663 5.882 −0.415 −0.592 0.284 1.253 −0.151 −0.433 0.728
(−2.95) (−3.14) (−3.12) (6.15) (12.13) (−2.17) (−6.65)
L–O −0.562 4.846 −0.349 −0.519 0.244 1.031 0.051 −0.431 0.685
(−2.57) (−2.93) (−3.08) (5.96) (10.61) (0.83) (−5.96)

I find that lottery-type stocks earn at least 4% lower average annual

returns.31

B. Investment in Lottery-Type Stocks and Portfolio Performance

The lower average returns of lottery-type stocks suggest that greater invest-
ment in lottery-type stocks is likely to be associated with greater average un-
derperformance. The exact magnitude of portfolio underperformance, however,
depends upon the subset of lottery stocks chosen by the investor, the weights
allocated to those lottery-type stocks, and the holding periods of those stocks.
To isolate the level of underperformance that is associated with investment in
lottery-type stocks, I estimate the degree of underperformance that is generated
in a well-diversified market portfolio when a part of that portfolio is replaced by
the lottery-stock component of an investor’s portfolio. This method is equivalent
to replacing the nonlottery portfolio component of an investor’s portfolio by

31
For robustness, I also estimate Fama–MacBeth cross-sectional regressions to examine the per-
formance of lottery-type stocks and find that lottery-type stocks earn lower average risk-adjusted
returns. The results are reported in Table IA.I of the Internet Appendix. In these tests, I use a
longer time period (1980 to 2005) to obtain more accurate estimates of the characteristics and per-
formance of the three portfolios, but I also report the estimates for the 1991 to 1996 sample period.
See Section E of the Internet Appendix for an additional discussion.
 Who Gambles in the Stock Market? 1927

the market portfolio.32 I conduct this exercise for every investor who holds
lottery-type stocks and obtain the risk-adjusted performance of investor-specific
hypothetical portfolios.
I find that the average annualized risk-adjusted underperformance (the four-
factor alpha) of hypothetical portfolios is 1.10% and it increases almost mono-
tonically with lottery weight. For instance, investors who allocate one-third of
their portfolios to lottery-type stocks underperform by about 2.50% annually
on a risk-adjusted basis.
To better examine the relation between investors’ propensity to invest in
lottery-type stocks and portfolio performance, I estimate cross-sectional re-
gressions, where the dependent variable is the performance of an investor’s
hypothetical portfolio. The main independent variables in these performance
regressions are a lottery-stock participation dummy, one of the five lottery-stock
preference measures (LP(1) –LP(5) ), and a strong lottery preference dummy to
capture potential nonlinearity in the lottery-type stock preference and perfor-
mance relation. The strong lottery preference dummy is set to one for investors
whose portfolio weights in lottery-type stocks are in the highest decile.
The performance regression specification also includes the known determi-
nants of portfolio performance as control variables. This set contains demo-
graphic variables, including the investor’s age, investment experience, annual
household income, plus zip code education level, a male dummy, a retired
dummy, and race/ethnicity identifiers. I also consider the following four portfo-
lio characteristics as control variables: initial portfolio size, monthly portfolio
turnover, dividend yield of the portfolio, and portfolio diversification.
The performance cross-sectional regression estimates are reported in
Table VIII, where I use clustered standard errors to account for cross-sectional
dependence within zip codes. In specifications (1) to (4), I consider the first
lottery preference. To ensure the robustness of the performance regression es-
timates, I consider lottery preference measures LP(2) –LP(5) in specifications (5)
to (8), respectively.33
The coefficient estimates from specification (1) indicate that the average an-
nual risk-adjusted underperformance is 3.00% (0.250 × 12) for an investor who
trades lottery-type stocks at least once during the sample period.34 The level of
underperformance is significant (0.189 × 12 = 2.27%), even after I account for
other known determinants of portfolio performance (see specification (2)).
The estimates from specifications (3) and (4) indicate that the degree of un-
derperformance is greater for investors who allocate a larger portfolio weight to
lottery-type stocks. The incremental annual risk-adjusted underperformance is
3.16% (0.263 × 12) for an investor who increases the investment in lottery-type

32
I thank an anonymous referee for suggesting this test.
33
As before, to allow for direct comparisons among the coefficient estimates, I standardize all
independent variables, and to keep the discussion focused on the incremental effects of investors’
preference for lottery-type stocks, I suppress the coefficient estimates of control variables.
34
The low adjusted R2 s in the cross-sectional regressions are consistent with the evidence in
Barber and Odean (2001, p. 280).
1928 The Journal of FinanceR

Table VIII
Preference for Lottery-Type Stocks and Portfolio Performance:
Cross-sectional Regression Estimates
This table reports the estimates for performance cross-sectional regressions. In specifications (1)
to (8), the dependent variable is the risk-adjusted performance measure (four-factor alpha) of a
hypothetical portfolio that is formed by replacing the nonlottery component of an investor portfolio
by the market portfolio. Lottery-type stocks are defined in Section III.A. In specification (9), the
dependent variable is the performance differential between the actual and a hypothetical portfolio
that is defined by replacing the lottery component of an investor portfolio by the nonlottery compo-
nent of her portfolio. The set of independent variables includes a participation dummy, lottery-type
stock preference measure, and strong lottery-type stock preference dummy. In specifications (3) and
(4), the main independent variable is the LP(1) lottery-stock preference measure. Specifications (5)
to (8) use one of the lottery-type stock preference measures (LP(2) – LP(5) , respectively) as the main
independent variable. The set of control variables includes the investor’s age, investment experi-
ence, annual household income, plus zip code education level, a male dummy, a retired dummy,
two race/ethnicity dummies, initial portfolio size, monthly portfolio turnover, dividend yield of the
portfolio, and portfolio diversification. For brevity, the coefficient estimates of the control variables
have been suppressed. The lottery-type stock preference measures are defined in Section V.A and
other variables have been defined in Table I, especially Panel F. Clustered standard errors are used
to account for potential cross-sectional dependence within zip codes. All independent variables have
been standardized. The t-statistics for the coefficient estimates are reported in parentheses below
the estimates.

Variable (1) (2) (3) (4) (5) (6) (7) (8) (9)

Intercept −0.368 −0.273 −0.317 −0.309 −0.291 −0.292 −0.292 −0.294 −0.157
(−6.78) (−3.09) (−6.19) (−3.89) (−3.74) (−2.76) (−2.73) (−2.76) (−2.43)
Lott-stock −0.250 −0.189
part. dummy (−4.59) (−3.32)
Lott-type −0.402 −0.263 −0.299 −0.226 −0.237 −0.193 −0.173
stock pref (−8.51) (−4.86) (−3.88) (−3.13) (−3.29) (−2.88) (−2.73)
Strong lott. −0.069 −0.048 −0.079 −0.078 −0.075 −0.070 −0.008
pref. dummy (−2.36) (−2.17) (−1.75) (−1.73) (−1.71) (−2.16) (−0.36)
Control variables No Yes No Yes Yes Yes Yes Yes Yes
(Estimates of control variables have been suppressed.)
Number of 40,476 27,565 34,588 26,204 25,229 25,229 25,229 25,229 25,229
investors
Adjusted R2 0.009 0.039 0.017 0.041 0.039 0.042 0.040 0.036 0.045

stocks by one standard deviation. Furthermore, if the investor is in the high-

est lottery weight decile, the portfolio underperforms by an additional 0.83%
(0.069 × 12) annually and the total annual, risk-adjusted underperformance is
3.99%.
When I consider other lottery preference measures (specifications (5) to (8)),
the coefficient estimates are qualitatively similar. Even when I consider a trade-
based lottery preference measure (specification (8)), the degree of underperfor-
mance that can be attributed to investment in lottery-type stocks is econom-
ically significant. The incremental annual risk-adjusted underperformance is
3.04% for an investor who increases the buying intensity in lottery-type stocks
by one standard deviation and belongs to the highest portfolio weight decile.
 Who Gambles in the Stock Market? 1929

Collectively, the performance regression estimates indicate that portfolios of

investors who invest more in lottery-type stocks experience greater under-
performance, even when I account for other known determinants of portfolio
performance.
Because low-income and less wealthy investors invest disproportionately
more in lottery-type stocks (see Table IV), their lottery-type stock investments
generate larger underperformance, both when measured in dollar terms and
when measured as a proportion of the total annual income. As a proportion of
income, the degree of portfolio underperformance due to investments in lottery-
type stocks has a striking resemblance to the evidence from state lottery stud-
ies. In both instances, the proportional level of underperformance decreases
with income. Therefore, like state lotteries, lottery-type stocks appear to be
regressive.

C. Performance of Lottery and Nonlottery Portfolio Components

To better identify the mechanisms that generate portfolio underperformance,
I compare the performance of lottery and nonlottery components of investor
portfolios. If investors who prefer lottery-type stocks are “bad” investors in gen-
eral, both the lottery and the nonlottery portfolio components of their portfolios
would perform poorly and the two performance measures should be positively
correlated. In contrast, if the underperformance of the lottery component of the
portfolio is induced by certain specific behavioral biases that are induced ex-
clusively or get amplified by lottery-type stocks, then the correlation between
the lottery and the nonlottery portfolio components should be zero. The third
possibility is that investors hold a layered portfolio that contains a large well-
diversified component along with a small component containing lottery-type
stocks (e.g., Shefrin and Statman (2000)). In this scenario, the nonlottery port-
folio component should perform relatively better than the lottery component
and the two performance measures should be negatively correlated.
In the first test, I compute the average correlation between the performance
of lottery and nonlottery components of investors’ portfolios. Every month, I de-
compose the total portfolio performance of each investor into the performance
of lottery-type stocks and nonlottery stocks and compute the time-series corre-
lation between the two performance measures.35 I find that the average lottery-
nonlottery performance correlation is weak and mildly positive. The mean cor-
relation estimate is 0.004 (t-statistic = 3.83) and the median is 0.001. This
evidence is weakly consistent with the conjecture that investors earn low re-
turns from their investments in lottery-type stocks due to their overall lack of
financial sophistication.
In the second test, I compare the performance levels of lottery and nonlottery
portfolio components. Specifically, I estimate the additional return a lottery in-
vestor would have earned if she had simply replaced the lottery component of
her portfolio with the nonlottery component. I compute the performance of this

35
The correlation is only defined for investors who hold both lottery- and nonlottery-type stocks.
1930 The Journal of FinanceR

hypothetical portfolio and examine the difference between the performance lev-
els of the actual and the hypothetical portfolios. Such performance differential
would ref lect both the relative underperformance of lottery-type stocks and the
additional biases that an investor might exhibit when she holds lottery-type
stocks.36 I find that a typical lottery investor would have been able to earn
0.237 × 12 = 2.84% higher annual returns on average if she simply replaced
her lottery investments with her nonlottery investments.
To examine whether the potential for performance improvement is greater
among investors with stronger gambling preferences, I estimate a cross-
sectional regression where the performance differential between the actual and
the hypothetical portfolios is the dependent variable. The set of independent
variables is identical to the performance regression estimated earlier. I find
that the level of relative underperformance is greater among investors who
allocate a larger portfolio weight to lottery-type stocks (see Table VIII, speci-
fication (9)). The incremental annual risk-adjusted relative underperformance
is 2.08% (0.173 × 12) for an investor who increases the weight in lottery-type
stocks by one standard deviation.
Taken together, these performance comparisons indicate that individual in-
vestors earn lower returns from their lottery investments. This underperfor-
mance of the lottery-type stock component of investor portfolios ref lects both
the underperformance of lottery-type stocks and investors’ behavioral biases.

VIII. Summary and Conclusion

This paper shows that the gambling preferences of individual investors are
ref lected in their stock investment decisions. Using monthly portfolio holdings
and trading data from a large U.S. brokerage house, I find that individual in-
vestors invest disproportionately more in stocks that have the qualitative fea-
tures of state lotteries. Within the individual investor category, socioeconomic
factors that induce greater expenditure in state lotteries are also associated
with greater investments in lottery-type stocks. And similar to lottery demand,
individual investors’ demand for lottery-type stocks increases when economic
conditions worsen.
Investors who invest disproportionately more in lottery-type stocks experi-
ence greater underperformance and the degree of portfolio underperformance
resembles the evidence from lottery studies. In both instances, the level of
underperformance as a proportion of income is greater among low-income in-
vestors.
Overall, these empirical findings indicate that state lotteries and lottery-type
stocks act as complements and attract very similar socioeconomic clienteles.
There are striking similarities between the behavior of state lottery players
and individual investors who invest disproportionately more in stocks with
lottery features.

36
When investors invest in lottery-type stocks, they might exhibit new types of biases or the
biases that they exhibit with nonlottery stocks get amplified due to stocks’ lottery characteristics.
 Who Gambles in the Stock Market? 1931

The finding that socioeconomic characteristics of individual investors inf lu-

ence their stock preferences is not entirely surprising because the psychological,
social, economic, political, and religious identities of an individual supersede
her identity as an investor. Portfolio choice models that recognize this potential
link could better explain the portfolio decisions of individual investors. Further,
if socioeconomic characteristics inf luence portfolio choice, those characteristics
could also be ref lected in stock prices. For instance, the return generating pro-
cess of a stock with an older investor clientele might be inf luenced by the pref-
erences and biases that are unique to older investors. Similarly, it is easy to
imagine a Catholic stock and a Protestant stock, an African-American stock
and a White stock, or a Democrat stock and a Republican stock.
In broader terms, the evidence in the paper suggests that the link between
changes in socioeconomic environment and stock market behavior might be
stronger than currently believed. For example, on the one hand, as the U.S.
population becomes older, the aggregate level of gambling-motivated trading
in financial markets could decline, which in turn could affect the equilibrium
returns, volume, and volatility of lottery-type stocks. On the other hand, as
gambling attains wider acceptability in society and the level of gambling activ-
ities increases, the level of speculative trading in financial markets could rise.
These social shifts could be associated with higher levels of trading, higher
volatility, and lower average returns. Future research could examine how the
interactions among different social processes inf luence stock market behavior.

REFERENCES
Amihud, Yakov, 2002, Illiquidity and stock returns: Cross-section and time-series effects, Journal
of Financial Markets 5, 31–56.
Badrinath, S. G., Gerald D. Gay, and Jayant R. Kale, 1989, Patterns of institutional investment,
prudence, and the managerial “safety-net” hypothesis, Journal of Risk and Insurance 56, 605–
629.
Barber, Brad M., and Terrance Odean, 2000, Trading is hazardous to your wealth: The common
stock investment performance of individual investors, Journal of Finance 55, 773–806.
Barber, Brad M., and Terrance Odean, 2001, Boys will be boys: Gender, overconfidence, and common
stock investment, Quarterly Journal of Economics 116, 261–292.
Barber, Brad M., and Terrance Odean, 2008, All that glitters: The effect of attention and news on
the buying behavior of individual and institutional investors, Review of Financial Studies 21,
785–818.
Barber, Brad M., Terrance Odean, and Ning Zhu, 2009, Do noise traders move markets? Review of
Financial Studies 22, 151–186.
Barberis, Nicholas, and Ming Huang, 2008, Stocks as lotteries: The implications of probability
weighting for security prices, American Economic Review 98, 2066–2100.
Barsky, Robert B., F. Thomas Juster, Miles S. Kimball, and Matthew D. Shapiro, 1997, Prefer-
ence parameters and behavioral heterogeneity: An experimental approach in the health and
retirement study, Quarterly Journal of Economics 112, 537–579.
Becker, Gary S., Kevin M. Murphy, and Ivan Werning, 2000, Status, lotteries and inequality,
Working paper No. 160, George J. Stigler Center for the Study of the Economy and the State,
University of Chicago.
Bennett, James A., Richard W. Sias, and Laura T. Starks, 2003, Greener pastures and the impact
of dynamic institutional preferences, Review of Financial Studies 16, 1203–1238.
1932 The Journal of FinanceR

Blalock, Garrick, David R. Just, and Daniel H. Simon, 2007, Hitting the jackpot or hitting the skids:
Entertainment, poverty, and the demand for state lotteries, American Journal of Economics
and Sociology 66, 545–570.
Brenner, Reuven, 1983, History—The Human Gamble (University of Chicago Press, Chicago, IL).
Brenner, Reuven, and Gabrielle A. Brenner, 1990, Gambling and Speculation (Cambridge Univer-
sity Press, Cambridge, UK).
Brunk, Gregory G., 1981, A test of the Friedman-Savage gambling model, Quarterly Journal of
Economics 96, 341–348.
Brunnermeier, Markus K., and Jonathan A. Parker, 2005, Optimal expectations, American Eco-
nomic Review 95, 1092–1118.
Chen, Joseph, Harrison Hong, and Jeremy C. Stein, 2001, Forecasting crashes: Trading volume,
past returns, and conditional skewness in stock prices, Journal of Financial Economics 61,
345–381.
Chen, Nai-Fu, Richard Roll, and Stephen A. Ross, 1986, Economic forces and the stock market,
Journal of Business 59, 383–403.
Clotfelter, Charles T., 2000, Do lotteries hurt the poor? Well, yes and no, Working paper, Terry
Sanford Institute of Public Policy, Duke University.
Clotfelter, Charles T., and Philip J. Cook, 1989, Selling Hope: State Lotteries in America (Harvard
University Press, Cambridge, MA).
Clotfelter, Charles T., Philip J. Cook, Julie A. Edell, and Marian Moore, 1999, State lotteries at
the turn of the century: Report to the national gambling impact study commission, Working
paper, Duke University.
Daniel, Kent D., Mark Grinblatt, Sheridan Titman, and Russell Wermers, 1997, Measuring mutual
fund performance with characteristic-based benchmarks, Journal of Finance 52, 1035–1058.
Daniel, Kent D., David Hirshleifer, and Avanidhar Subrahmanyam, 1998, A theory of overconfi-
dence, self-attribution, and security market under- and over-reactions, Journal of Finance 53,
1835–1885.
Daniel, Kent D., David Hirshleifer, and Avanidhar Subrahmanyam, 2001, Overconfidence, arbi-
trage, and equilibrium asset pricing, Journal of Finance 56, 921–965.
Del Guercio, Diane, 1996, The distorting effect of the prudent man law on institutional equity
investments, Journal of Financial Economics 40, 31–62.
Fama, Eugene F., and James D. MacBeth, 1973, Risk, return, and equilibrium: Empirical tests,
Journal of Political Economy 81, 607–636.
Ferson, Wayne E., and Rudi W. Schadt, 1996, Measuring fund strategy and performance in changing
economic conditions, Journal of Finance 51, 425–461.
France, Clemens J., 1902, The gambling impulse, American Journal of Psychology 13, 364–407.
Frieder, Laura, and Avanidhar Subrahmanyam, 2005, Brand perceptions and the market for com-
mon stock, Journal of Financial and Quantitative Analysis 40, 57–85.
Friedman, Milton, and Leonard J. Savage, 1948, The utility analysis of choices involving risk,
Journal of Political Economy 56, 279–304.
Graham, John R., and Alok Kumar, 2006, Do dividend clienteles exist? Evidence on dividend pref-
erences of retail investors, Journal of Finance 61, 1305–1336.
Grichting, W. L., 1986, The impact of religion on gambling in Australia, Australian Journal of
Psychology 38, 45–58.
Harvey, Campbell R., and Akhtar Siddique, 2000, Conditional skewness in asset pricing tests,
Journal of Finance 55, 1263–1295.
Herring, Mary, and Timothy Bledsoe, 1994, A model of lottery participation: Demographics, context,
and attitudes, Policy Studies Journal 22, 245–257.
Hirshleifer, David A., 2001, Investor psychology and asset prices, Journal of Finance 56, 1533–1597.
Ivković, Zoran, and Scott Weisbenner, 2005, Local does as local is: Information content of the
geography of individual investors’ common stock investments, Journal of Finance 60, 267–
306.
Kallick, Maureen, Daniel Smits, Ted Dielman, and Judith Hybels, 1979, A survey of American
gambling attitudes and behavior, Survey Research Center Research Report, Institute for Social
Research, University of Michigan.
 Who Gambles in the Stock Market? 1933

Kumar, Alok, 2009, Hard-to-value stocks, behavioral biases, and informed trading, Journal of
Financial and Quantitative Analysis (forthcoming).
Kumar, Alok, and Charles M. C. Lee, 2006, Retail investor sentiment and return comovements,
Journal of Finance 61, 2451–2486.
Luttmer, Erzo F. P., 2005, Neighbors as negatives: Relative earnings and well-being, Quarterly
Journal of Economics 120, 963–1002.
Markowitz, Harry, 1952, The utility of wealth, Journal of Political Economy 60, 151–158.
Mikesell, John L., 1994, State lottery sales and economic activity, National Tax Journal 47, 165–
171.
Newey, Whitney K., and Kenneth D. West, 1987, A simple, positive semi-definite heteroskedasticity
and auto-correlation consistent variance–covariance matrix, Econometrica 55, 703–708.
Odean, Terrance, 1999, Do investors trade too much? American Economic Review 89, 1279–1298.
Petersen, Mitchell A., 2009, Estimating standard errors in finance panel data sets: Comparing
approaches, Review of Financial Studies 22, 435–480.
Polkovnichenko, Valery, 2005, Household portfolio diversification: A case for rank-dependent pref-
erences, Review of Financial Studies 18, 1467–1502.
Pontiff, Jeffrey, 1996, Costly arbitrage: Evidence from closed-end funds, Quarterly Journal of Eco-
nomics 111, 1135–1151.
Price, Donald I., and E. Shawn Novak, 1999, The tax incidence of three Texas lottery games:
Regressivity, race, and education, National Tax Journal 52, 741–751.
Rubinstein, Ross, and Benjamin Scafidi, 2002, Who pays and who benefits? Examining the distri-
butional consequences of the Georgia lottery for education, National Tax Journal 55, 223–238.
Shefrin, Hersh M., and Meir Statman, 2000, Behavioral portfolio theory, Journal of Financial and
Quantitative Analysis 35, 127–151.
Shiller, Robert J., 1989, Market Volatility (MIT Press, Cambridge, MA).
Shiller, Robert J., 2000, Irrational Exuberance (Princeton University Press, Princeton, NJ).
Statman, Meir, 2002, Lottery players/Stock traders, Financial Analysts Journal 58, 14–21.
Tec, Nechama, 1964, Gambling in Sweden (Bedminster Press, Totowa, NJ).
Tversky, Amos, and Daniel Kahneman, 1992, Advances in prospect theory: Cumulative represen-
tation of uncertainty, Journal of Risk and Uncertainty 5, 297–323.
Walker, Michael B., 1992, The Psychology of Gambling (Pergamon Press, Oxford, UK).