Friends Do Let Friends Buy AAPL, and F, and IPET...

Rawley Z. Heimer∗

Brandeis University, International Business School

August 9, 2012

Abstract

This research is the �rst to provide empirical evidence that social interaction is more

prevalent amongst active rather than passive investors. While previous empirical work,

spearheaded by Hong, Kubik, and Stein (2004), shows that proxies for sociability are

related to participation in asset markets, the literature is unable to distinguish between

the types of participants because of data limitations. I address this shortcoming by

using data from the Consumer Expenditure Quarterly Interview Survey on individual

holdings, and buying and selling of �nancial assets as well as expenditure variables which

imply variation in the level of social activity. My �ndings o�er a new explanation for

the over-trading puzzle and lends microfounded support to asset pricing models that

incorporate consumption externalities (i.e. "keeping up with the Joneses") in consumer

preferences.

∗Brandeis International Business School Mailstop 032, P.O. Box 549100, Waltham, MA 02454, USA.e-mail: rheimer@brandeis.edu.

I am grateful for support from Brandeis University and to faculty for advice. I thank David Simonfor much helpful conversation as well as comments, Daniel B. Bergstresser, Debarshi Nandy, Carol L. Osler,and Raphael Schoenle for comments, and Eric Zwick for helping title the paper. This version is preliminaryand incomplete. All errors are my own.

1

1 Introduction

This paper is the �rst to provide empirical evidence that social interaction is more prevalent

amongst active rather than passive investors. By using data from the Consumer Expenditure

Quarterly Interview Survey (CEQ) on individual holdings, and buying and selling of �nan-

cial assets as well expenditure variables which imply variation in the level of social activity,

I show that conditional on owning securities, the odds of being an active investor increase

by about twenty percent if the individual is social. While this �nding has previously gone

undocumented, it should not be unexpected; as suggested by Becker (1991), one's �nancial

wheeling and dealings may be a more robust topic of conversation than the strategy recom-

mended by economists, buying and holding the market portfolio.1 Otherwise, why would

investment clubs, such as those documented in Barber and Odean (2000) continue to exist

despite underperforming relative to a broad-based market index?

My �ndings o�er two key contributions to the understanding of �nancial markets and

its participants. First, it provides an explanation for the over-trading puzzle documented in

Barber and Odean (2000) whereby individual investors trade actively but lose on average.

Secondly, it o�ers micro-level support for asset pricing models that incorporate positive

consumption externalities (Gali, 1994) to explain the equity-premium puzzle documented

in Mehra and Prescott (1985). The representative agent in this strand of models places a

greater share of their wealth in risky assets in order to �keep up with the Joneses�.

The empirical work in this paper is a decedent of Hong, Kubik, and Stein (2004). Their

work, which uses Health and Retirement Survey (HRS) data, shows that social individuals

� those who claim to �know their neighbors�, �visit their neighbors�, or �attend church� � are

more likely to be stock market participants. This paper validates their �ndings in several

ways. First, it uses a di�erent dataset, the CEQ, to con�rm their results. The CEQ expands

the range of individuals surveyed since the HRS only samples a cohort born between 1931 and

1941 while the CEQ is representative in age. In response to the �ndings of Malmendier and

Nagel (2011) that the macroeconomic conditions experienced during an individual's lifetime

in�uence their risk tolerance and thus their willingness to participate in stock markets, it is

important to validate the �ndings in HKS using a broader set of respondents. It is possible

that certain cohorts are more susceptible to word-of-mouth while others � perhaps those

with fresh memory of the Mado� scandal or those who experienced Black Friday � view

conversation about investment activity with skepticism. This research also expands the set

1Becker (1991) presents the keen observation that prices will remain at levels that appear to be not bepro�t-maximizing when the good provided by the �rm is more enjoyable when more people consume thegood. This would explain why long lines will persist at some restaurants and the owners do not raise prices,a phenomena that does not appear to occur amongst �rms that do not provide �social� goods.

2

of proxies for sociability by including a list of variables on expenditures towards items related

to social activity: church donations, admission to sporting events, and participation sports.

While validating existing research is a worthwhile endeavor,2 the most novel contribution

of this paper is not to further examine whether social interaction leads to participation in

�nancial markets, but to see what type of participation social interaction is more closely

associated with, active or passive investing. Unfortunately, since it asks respondents to

identify the total value of stock purchases and sales made over the previous two years,

the HRS does not provide any way to infer frequent turnover or active management of an

individual's portfolio. On the other hand, the CEQ asks respondents to report how much they

have spent on assets over the previous year and to separately identify how much they have

bought and how much they have sold. By using these variables I can infer that households

who report buying and/or selling stock are more likely to be active participants in �nancial

markets.

My empirical strategy is as follows. I �rst con�rm that my proxies for active investing are

consistent with known characteristics of active investors, namely that they are male, urban,

and educated, they skew younger, and are technologically savvy and risk seeking. After

con�rming that I have correctly identi�ed respondents in the CEQ who are more likely to be

active investors I incorporate the proxies for social interaction into the empirical analysis. I

begin by replicating the Hong, Kubik, and Stein (2004) results showing that these variables

are related to participation in asset markets. I then ask the question, �conditional on being

an investor, are social individuals more likely to be active or passive investors?�. I provide an

answer by restricting the regressions to the set of respondents who own securities and seeing

if the social variables are related to active management. I �nd strong evidence throughout

that engaging in social activities increases the odds of being an active investor by anywhere

between ten and twenty-�ve percent depending on which proxy is used.

I rule out several alternative explanations for the empirical relationship between social

interaction and active investing. First, active investing could simply arise because of shocks

to disposable income which would also increase expenditures regardless of whether or not

they are related to social activity. Secondly, I �nd that aggregate returns and volatility

as well as expectations of future returns are not the driving force behind the relationship.

Third, it may be that I am capturing response propensity and an interviewee's willingness

to document expenditures on social activities is associated with an increased likelihood of

responding to questions about their �nancial activity. I control for this potential complication

throughout the empirical analysis. I also provide an alternative speci�cation of active versus

passive investing, namely that social interactions are more strongly related to other forms of

2Georgarakos and Pasini (2011) con�rm the �ndings of HKS when examining stockholdings across Europe.

3

security ownership than to what could be considered an extremely passive form of investing,

ownership of U.S. savings bonds.

There are many reasons why it is important to distinguish between the type of partici-

pation associated with social interaction. First, there are welfare considerations since active

individual investors are unpro�table on average and at an information disadvantage in most

�nancial markets. Barber, Lee, Liu and Odean (2009) �nd that Taiwan's retail investors

under-perform the market by 3.8 percent and accumulate losses that amount to 2.2 per-

cent of Taiwan's GDP annually. This stresses the importance of having �nancial literacy

educational programs (van Rooij, Lusardi, and Alessie (2011)) rather than informal word-

of-mouth communication promote participation in �nancial markets. Also, Kaustia and

Knüpfer (2011) note the correlation between new stock market entrants and the presence

of asset price bubbles which can have a disastrous e�ect on the overall economy. If entry

is driven through informal communications between individuals and new entrants brought

about through such channels are tilted towards active investing then it further justi�es the

notion that social interaction plays a role in the formation of bubbles.

This research is useful for theoretical work that directly models the �ow of information

between investors, either through word-of-mouth (Ellison and Fudenberg (1993, 1995)) or

observation (Banerjee (1992) and Bikchandani, Hirshleifer, and Welch (1992)). It justi�es

an extension of these models in which certain �forceful� agents dominate the spread of infor-

mation (Acemoglu, Ozdaglar, and ParandehGheibi (2010)). Furthermore, models that are

concerned with the role of non-participation on asset prices (Allen and Gale (1994), among

others) may wish to consider shocks to the participation rate that are skewed towards certain

groups of investors.

Active investors are often considered analogous to rational speculators or noise traders.

In Stein (1987), the introduction of rational speculators into an asset market is shown to

lower the informativeness of prices to other traders and imposes an externality that reduces

total welfare. In DeLong et al. (1990), noise traders impact liquidity and the volatility of

assets, and their presence can explain how prices can persistently diverge from fundamental

values. The link between noise trading and individual investors has been justi�ed by a

number of empirical studies. Foucault, Sraer and Thesmar (2011) �nd that increasing the

cost associated with active retail trading on Euronext Paris reduces the volatility of daily

returns by about a quarter of its standard deviation, while Barber, Odean, and Zhu (2009),

Kumar and Lee (2006), and Hvidkjaer (2008) document that trades of individual investors

tend to be correlated and may a�ect asset prices.

The role of social networks in areas outside of �nance is well-known. Bayer, Ross, and

Topa (2008) use U.S. Census data to document the role of neighborhood referrals on labor

4

market outcomes.3 Christakis and Fowler (2007), in an epidemiological study conducted in

Massachusetts between 1971 to 2003, �nd that obesity appears to spread through social ties.

Social in�uences are also found to have an impact on educational outcomes. Calvò-Armengol

et al. (2009), using data on friendships from the National Longitudinal Survey of Adolescent

Health, �nd a relationship between the centrality of a student within a network and their

performance in school. Social interactions can explain the cross-city variance of crime rates

(Glaeser et al. (1996)) and a recent string of ��ash mob� robberies organized through social

media o�ers support for this theory (Jouvenal and Morse (2011)).

It is a natural extension to believe that social interaction can have an e�ect on invest-

ment behavior as well. In fact, there is substantial evidence that participation and investor

behavior in �nancial markets are in�uenced by social interaction (Shiller (1984, 1989) and,

Shiller and Pound (1989)). Brown et al. (2008) identi�es a causal e�ect of the level of stock

ownership in a community on the individual's decision to own stock. Among mutual fund

managers, Hong, Kubik, and Stein (2005) demonstrate that portfolios exhibit higher corre-

lation if they are from the same town while Cohen, Frazzini, and Malloy (2008) show that

they place greater bets on �rms whose board members are from their education network.

Shive (2010) uses an epidemic model and data on Finnish stock-holdings to study how social

contact can predict investor trading. Correlation across investments in retirement accounts

are also observed by Madrian and Shea (2000) and Du�o and Saez (2002, 2003). Researchers

document that investors are in�uenced by the investment decisions of others including fa-

mous investors like Warren Bu�ett (Sandler and Raghavan (1996)), insiders (Givoly and

Palmaon (1985)), and readers of the Wall Street Journal's Dartboard column (Barber and

Loe�er (1993)).

Several papers go further and show that communication about investments may be bi-

ased towards positive realized returns. Heimer and Simon (2011), in a novel dataset on

retail foreign exchange traders who participate in a social network, �nd that individuals are

more likely to issue communication about returns following short-term gains. Furthermore,

recipients of communications increase their trading activity in response to issuers of com-

munications with higher returns. Kaustia and Knüpfer (2011) show that social interactions

promote stock market participation when peer performance is positive. Hirshleifer (2010)

emphasizes the role of self-enhancing transmission bias (the tendency of individuals to tout

their successes while downplaying their failures), shows that it can explain active invest-

ing, and formalizes the consequences of its contribution towards asset markets. This paper

expands upon this aspect of the literature by showing that social interactions are biased

3Calvó-Armengol and Jackson (2004) provide the most reputable model of the role of social interactionon labor market outcomes.

5

towards active forms of investing.

The paper is organized as follows. Section 2 describes the data as well as con�rms that

my proxies for active investing are consistent with known facts. Section 3 presents base-

line results on the relationship between active investing and social interaction. Section 4

addresses several concerns about the observed empirical relationship and provides an alter-

native speci�cation of active versus passive investing. The �nal section concludes.

2 Data

2.1 The Consumer Expenditure Quarterly Interview Survey

The data comes from the Consumer Expenditure Quarterly Interview Survey (CEQ) a na-

tionwide household survey conducted by the U.S. Bureau of Labor Statistics (BLS) to esti-

mate household expenditures. The interview survey is a rotating panel survey and, according

to Chopova, et al. (2008), approximately 15,000 households are visited each quarter of the

year, with roughly half participating. Households, which in most cases represent a consumer

unit (CU), are contacted for an interview every three months for �ve consecutive quarters.

The �rst interview is not used in the public data nor in the published estimates of national

and regional aggregates; it is merely used for bounding purposes which counteracts the prob-

lem of �telescoping� or the tendency of survey respondents to report expenditures to have

been made more recently than they actually were. While the CEQ has been in existence

on a yearly basis since 1980, I append and use years 2000Q2 - 2010Q1 of the CEQ public

micro-use survey data. Only the �fth interview asks for household asset holdings and the

sample used herein is restricted to respondents who complete the �fth interview.

The interview is conducted by an interviewer and the respondent or �reference person�

provides information about the household or CU. The CEQ includes data on the age, years

of education, race/ethnicity, and marital status of both the respondent and his or her spouse.

It also includes information on household wealth, income, and household composition such

as the number of dependent individuals, as well as the place of residence, U.S. state and

an indicator for urban/rural. It may be that the data is biased by the possibility that

respondents are more likely to be non-income earning members of the household and that

these individuals are less aware of the family �nances. Accordingly, I refer to the demographic

characteristics of the respondent when the respondent is also one of the income earning

members of the household. When the respondent does not earn income, I refer to the

demographic characteristics of the spouse.4

4The CEQ contains the variable �earncomp� which asks for the composition of earners. Answers 1 through

6

2.1.1 Financial variables

The �fth interview of the CEQ asks respondents to provide information about their �nances

and asset holdings. In particular, interviewees are asked, �what was the estimated value of

securities, such as stocks, mutual funds, private bonds, government bonds or Treasury notes

owned by you (or any members of your CU) on the last day of (last month)?�. They are also

asked separately the value of securities purchased and the value of securities sold within the

last twelve months. Summary statistics on participation rates and the value of securities are

reported in Tables 1 and 2 respectively. Notably, of the roughly 75,000 respondents in my

sample, 12.8 percent of respondents report owning securities with just 1.8 percent declaring

both buying and selling of securities. Furthermore, 8.8 percent of respondents own securities

but neither bought or sold any in the preceding period and individuals purchase securities

more often than they sell them (3.8 versus 2.7 percent). It is worth mentioning that, as

expected, all of the aforementioned rates are increasing in household income and that there

is a downward trend in security ownership over the sample period � roughly 14 percent of

respondents owned securities in the early years of the 2000s, a rate that drops to around 11

percent in 2008 and 2009. The median value of securities owned (conditional on ownership)

is about 40,000 USD in 2010 prices, but the median value of that bought and sold is much

smaller � roughly 5,500 and 7,500 USD respectively.5

Unfortunately, the questions in the CEQ related to securities are not disaggregated and

there is likely no way to infer the composition of asset holdings per household.6 At best, by

using additional data from the Survey of Consumer Finances (Bucks et al. (2009)), I can

deduce that the majority of households who report owning securities hold stocks, but not

4 of this variable include the reference person as an earner. When the respondent answers �spouse only�(answer 5) or �spouse and others� (answer 6), I use the demographic characteristics of the respondent'sspouse.

5I normalize all reported dollar amounts into March 2010 dollars using the Consumer Price Index (CPI)for all urban consumers produced by the BLS. Nominal amounts X on item j in time t are adjusted asfollows:

Xj,t ∗CPI3/2010

CPIt= X̄j,t (1)

where X̄ indicates the real price of expenditures.6Vissing-Jørgensen (2002) attempts to distinguish between stock and bond owners. While her measure

of stock ownership is in line with other sources (21.75 percent of her sample), she classi�es 31.40 percentof her sample as bond owners. First, this is not consistent with external, representative surveys such asthe SCF. By restricting the sample to those who participate in all �ve interviews, she likely over-samplesthe higher income classes and homeowners (Heimer (2011) shows that complete respondents are found tobe wealthier and homeowners), categories which are positively correlated with bond ownership. Even theweights published by the BLS do not take into account longitudinal attrition and not adjusting them forattrition leads to over-sampling these strata. Secondly, the fact that her sample yields a 50 percent greateramount of households who own bonds versus stocks is strongly inconsistent with other sources and suggestsit may be inappropriate to replicate her methods.

7

bonds or treasury notes. As reported in Table 3, the percentage of households that owned

stocks was 20.7 and 17.9 in 2004 and 2007 respectively. Meanwhile, bonds were owned by

only 1.8 percent of families in 2004 and 1.6 percent in 2007. Pooled investment funds �

held by 11.4 percent of families in 2007 � are also found to typically contain stocks. 10.2

percent of families in 2007 owned a pooled investment fund with just stocks. Tax-free bond,

government bond, and other bond pooled investment funds are held by 2.1, 1.2, and 1.0

percent of households respectively.

I use the variables in the CEQ to create three indicator variables that establish the respon-

dent's status as an investor. First, I create a one/zero indicator variable for all respondents

that report having securities valued above zero which I call �investor �.7 The second variable,

�active�, is equal to one if the respondent reports having either bought or sold securities. The

third variable I create is called �very active� and is equal to one if the respondent reports both

buying and selling securities. Obviously, when conditioning on being an investor, the two

active variables imply that the respondent is either an active or passive investors. I create

two variables for active investing because buying and selling may simply represent liquidity

needs. My strategy for verifying that these variables are indicative of active investing is to

examine if known characteristics about active investors are associated with them, a task I

accomplish in Section 2.2.

2.1.2 Social expenditures

The primary purpose of the CEQ is to estimate U.S. aggregate and regional expenditures by

households. For instance, the survey is currently used to generate expenditure weights for use

in the U.S. Consumer Price Index and to make cost of living adjustments for military families.

The list of items for which respondents are asked to report expenditures on is exhaustive

and detailed, from necessities such as food and clothing to luxury items such as aircraft,

purchases required by law such as vehicle registrations and those related to entertainment

such as �dating services�. In particular, several of the items in which respondents are asked

to report on, namely �cash contributions to churches/religious organizations�, �admission to

sporting events (out-of-town)�, and �fees for participant sports (out-of-town)�,8 are related

to their propensity to be engaged in social activity.

�Cash contributions to churches/religious organizations� implies church attendance which

is related to the propensity to engage in social activities. Putnam (2000, pg. 66) suggests,

7Only 12 out of a possible 75,189 respondents report having securities valued zero. The rest of theobservations are recorded as missing observations.

8The CEQ asks respondents to report expenditures on admissions to sporting events and fees for par-ticipant sports both in the location of residence and while out-of-town. I combine in-town and out-of-townexpenditures per CU for each of these two variables.

8

�[r]eligious institutions directly support a wide range of social activities well beyond con-

ventional worship�, and provides many examples of those activities such as seminars titled,

�Women in the Marketplace� and �Career Builder's Workshop�. Church attendance is a key

explanatory variable in HKS. In their sample of households in the Health and Retirement

Study, HKS report that 76 percent report attending church at least once a year. In my

sample of households in the CEQ, 17 percent report having made donations to churches

or religious organizations (Table 4) (26 percent report making donations when the data is

restricted to respondents who consult records). The di�erence is not surprising considering

that those who donate to church are likely a subset of those who attend and the CEQ asks

for expenditures made within the last three months, while the Health and Retirement Study

assesses attendance over the preceding year.9 Regardless of these di�erences, this variable

has the potential to provide robustness to HKS's results as well as answer the questions

raised by this paper.

The relationship between participation sports and social activity is less clear as it may

represent individual athletic activities such as ski passes. In fact, popular literature such as

�Bowling Alone: The Collapse and Revival of American Community� (Putnam, 2000) chron-

icles how participation sports have become increasingly less of a social activity. However,

the proliferation of leagues for playground sports such as kickball in urban locales which are

clearly designed to conglomerate young professionals speaks to the ability of sporting activ-

ities to function as a social lubricant.10 While individuals may attend sporting events alone

and refrain from communication with others in attendance, the experience � be it collegiate

or professional athletics, or an international amateur competition � is clearly designed to

inspire a shared sense of fandom and comradery. Internet forums for athletic teams are fre-

quently used to organize tailgating and meet-ups to watch the games. The main confounding

factor common to these two variables is that they may re�ect an individual's own athletic

ability which could be indicative of over-con�dence. There is extensive literature attributing

active investing to over-con�dence and there is no way to disentangle over-con�dence from

social interaction in these variables.11

Table 4 o�ers an overview of the social variables used in this study. Between 12 and

17 percent of respondents claim to have had expenditures on admissions to sporting events,

9There are clearly exceptions to the idea that to donate to church an individual must also attend. Forexample, televangelists accept donations (as well as sell products such as recordings and books) over thephone.

10A casual scan over the webpages of local kickball leagues reveals frequent sponsorship provided bybars and pubs as well as advertisements by alcoholic beverage purveyors. The websites often contain linksto �pub-crawls� and other organized parties. One such example can be found here: http://www.kickball-rochester.com/

11See Oberlechner and Osler (2011) for one example.

9

participation fees for sports, or made donations to church or religious organizations. Not

surprisingly, all variables are increasing in household income which provides clear indication

that expenditures on all items are at least partly explained by the presence of disposable

income.

The correlations between these variables are presented in Table 5. The strongest correla-

tion between pairs of items occur between the two variables related to sports with a Pearson's

correlation coe�cient of 31.48 percent. The correlation between the other variables tends to

be between around 10 percent which suggests that there is relatively independent informa-

tion on the propensity to be social across most variables. For the most part, the correlation

between variables falls when the relationship between them is estimated conditional on being

a respondent who consults their records at least half the time. This suggests that at least

part of the correlation is driven by individual response propensity a feature I control for in all

of the empirical analysis. Likewise, the correlation between the variables falls when analyzed

within income classes, which again hints at the need to control for disposable income.

2.2 Facts about active investing

The purpose of this section is to ensure that I am correctly identifying proxies for active and

passive investing, a task I accomplish by verifying that my model speci�cation is consistent

with known facts about active investing.12 I run three logistic regressions with three di�erent

dependent variables: (1) whether or not the respondent owns securities (investor), (2) buys

or sells securities (active), and (3) buys and sells securities (very active). The latter two

regressions are conditional on the individual being an investor while the third regression

also excludes those who only buy or sell securities but not both. In all speci�cations the

independent variables are age and household income before taxes, and indicator variables

for gender, marital status, whether or not a member of the household is college educated,

and if they live in a rural setting. I also include categorical variables for respondent race

and geographic region of the country, year �xed e�ects, as well as two variables that capture

response propensity, whether an individual consulted records while answering the survey

and the number of items the respondent reports expenditures on.13 Lastly, I create two

additional variables intended to capture individual risk tolerance and their willingness to use

technology. To summarize my �ndings (presented in Table 6) which are in accordance with

known facts about active investing, active investors are more likely to be male, younger,

urban, educated, technologically savvy, and risk seeking.

12This method is similar to that of Kumar (2009), in which the demographic characteristics of those whopurchase lottery tickets are compared with those who buy �lottery-type� stocks.

13The number of items the respondent reports expenditures on may also capture disposable income.

10

One of the established facts about active investing is its relationship with overcon�dence

and that since men are more overcon�dent than women, they are more likely to be active

traders. Barber and Odean (2001) document that men trade 45 percent more than women.

This �nding could easily be explained if men were more pro�table traders, but trading is

found to reduce the performance of male investors by about a full percentage point more

than that of females. The speci�cation I use is consistent with these �ndings. First, men

are more likely to be investors; the odds are about 16 percent higher for males. Secondly,

conditional on being an investor, men are more likely to be active investors. The odds of

reporting buying or selling are about 28 percent higher for a male than female and about

40 percent higher for reporting both buying and selling. One concern is that this result

may simply re�ect a tendency for males to have more control over household �nances and

thus a better knowledge of the ins-and-outs of their family's portfolio. First, as mentioned

previously, the demographic characteristics belong to those of the wage-earner. Secondly, I

address this concern by restricting the regressions to include only unmarried respondents. I

�nd that this does not change any of the results.14

Linnainmaa (2003) �nds that active investors are more likely to be younger and urban

individuals.15 My results are consistent with these facts. The probability of being an investor

is increasing in age, a result that is in accordance with the accumulation of assets over the

course of an individual's life. However, similar to Linnainmaa (2003), the probability of

being an active investor is decreasing in age. Furthermore, investors are no more or less

likely to be urban individuals, but urban individuals are about one-third more likely to

be active investors. One other relationship between security ownership and demographic

characteristics is that Caucasians are more likely than all other races/ethnicities to own

securities, a fact also documented in HKS. There is no research linking active investing to

race/ethnicity and I �nd no evidence of a relationship in the CEQ data.

To the best of my knowledge, there are no existing studies linking education and active

investing. The closest substitute for this missing strand of literature is that which attempts

to link SAT scores (Gottesman and Morey (2006)) and IQ (Grinblatt et al. (2010, 2011))

to participation and performance as an investor. These studies �nd a positive relationship

between intelligence and both participation and performance. Considering the likely cor-

relation between intelligence and the educational attainment of respondents to the CEQ,

I expect to see and do indeed �nd a positive, statistically signi�cant relationship between

whether or not a member of the household is college educated and ownership of �nancial

14I do not report these regression results.15Linnainmaa (2003) is speci�cally concerned with documenting facts about day-traders. I assume that

day-traders and active investors are interchangeable.

11

assets as well as my proxies for active investing.

Another established relationship is between investing and the use of technology. HKS �nd

that individuals who report not being able to use a computer or word processor are less likely

to be investors. I �nd that respondents to the CEQ are more likely to be investors if they

have expenditures on personal computer hardware or software, results that are consistent

with HKS. Furthermore, Barber and Odean (2002) �nd that investors trade more frequently,

speculatively, and less pro�tably upon switching from phone-based to online trading which

leads me to expect to �nd a positive relationship between computer expenditures and active

investing. The coe�cients in Table 6 take on the expected sign and the relationship is statis-

tically signi�cant amongst very active investors. The failure to �nd a statistically signi�cant

relationship in both speci�cations could be due in part to survivorship bias amongst active

investors. At any given point in time I am observing only those who are current investors

and I am missing the fraction that quit trading due to performance. If internet investors are

less pro�table than otherwise then the bias is stronger amongst the technologically adept

cohort of traders.

I also attempt to capture individual tolerance for risk by including a variable for whether

or not the respondent has expenditures on either renter's or life insurance.16 Presumably

respondents who purchase insurance are more risk averse than those who do not. HKS

�nd that individuals who are risk tolerant are more likely to be stock market participants.

Their �nding makes sense amongst an older cohort of individuals such as those in the HRS.

However, the theory behind this relationship when extended to individuals of all age groups

is less clear since a standard intertemporal consumption/savings model predicts that a risk

averse agent will invest in the asset to bu�er against shocks to their earnings potential. This

is evident in the fact that I �nd that those who purchase insurance are more likely to be

investors. On the other hand, it is clear that active investors engage in risk seeking behavior.

The relationship between active investing and insurance is negative in both speci�cations and

statistically signi�cant at the one percent level when the dependent variable is an indicator

for having bought and sold assets. In the latter speci�cation, the odds of being an active

investor are about 22 percent lower for an individual who purchases insurance.17

16I also attempt to control for risk tolerance by including a variable for whether or not the respondent re-ports lottery or other parimutuel losses. The relationship between lotteries and active investing is statisticallyinsigni�cant which may suggest that lotteries and active investing are substitutes in consumption.

17It could also be argued that having insurance is actually a proxy for the �nancial sophistication of theindividual. If this is the case, the expected results would be the same. A �nancially sophisticated individualunderstands the value of holding securities and would also understand that active investing is an unpro�tableventure on average.

12

3 Empirical Analysis

3.1 Baseline regressions

In this section I analyze the relationship between my proxies for active investing (Section

2.1.1) and the proxies for sociability I discuss in Section 2.1.2. I run the same three regressions

as in Section 2.2, in which I verify that my proxies for active investing are consistent with

facts established in prior literature, but now include separately in each regression my proxies

for sociability. Recapping my procedure, I run three logistic regressions with three di�erent

dependent variables: (1) whether or not the respondent owns securities (investor), (2) buys

or sells securities (active), and (3) buys and sells securities (very active). The second and

third logistic regressions are conditional on the individual owning securities, while the third

excludes all those who buy or sell securities but not both. The control variables are the

same as those listed in Section 2.2. To preview my results, I �nd a positive relationship

between social interaction and ownership of securities (consistent with HKS), furthermore,

social interaction is more strongly related to active than to passive investing.

Baseline regression results are presented in Table 7. The �rst column presents the rela-

tionship between the candidate social variables and whether or not the respondent reports

ownership of securities. The relationship is positive and statistically signi�cant at the 99

percent level for all three social variables. The log-odds of owning securities are around 25

to 40 percent higher for those who report expenditures on social activities.

The second and third columns in Table 7, address the question, �conditional on owning

securities, do social interactions tilt individuals towards active or passive investing?�. This set

of logistic regressions includes the same set of control variables as before but the dependent

variables are now whether or not the respondent engages in active investing and are run

conditional on being an investor. The dependent variable in the second column is whether or

not the respondent is an active investor and in the third column, a very active investor. The

regression in the third column excludes those who are active investors, but not very active

so as to mitigate concern over the variables capturing household liquidity needs rather than

investing style. The results are largely consistent with the hypothesis that social interactions

are more prevalent amongst active investors. The relationship between expenditures on social

activities and active investing is positive in all speci�cations. Amongst the three social

variables that are statistically related to investing (all variables with the exception of �party

supplies�), there is a statistically signi�cant relationship between social expenditures and

active investing in �ve out of six possible speci�cations. The magnitude of the relationship

is economically signi�cant as well, however, the coe�cients are smaller than that of the pure

participation regressions. The odds of being an active investor are about nine percent larger

13

for those who purchase admission to sporting events, while the largest coe�cient, that on

church donations in the very active speci�cation, implies that the odds of being a very active

investor are 26 percent greater.

It should also be noted that there is a substantial decline in the �t of the model upon

going from the question of whether or not the investor owns securities to the regressions

that ask, �conditional on investing, are you an active or passive investor?�. The Pseudo

R-squared is around 20 percent in the logistic model whose dependent variable is whether

or not the individual is an investor. In other words, the model does a relatively good job

determining who owns securities. The Pseudo R-squared falls to �ve percent in the active

investing regression and seven percent in that with very active as the dependent variable.

The fact that it is larger in the very active versus the active regression suggests that removing

those who are active but not very active improves the model �t. This further implies that

individuals who buy or sell but not both may be more likely to buy or sell for liquidity

reasons rather than speculative purposes.

4 Concerns and Robustness

In this section, I rule out several alternative explanations for the empirical relationship

between social interaction and active investing. First, active investing could simply arise

because of shocks to disposable income which would also increase expenditures regardless of

whether or not they are related to social activity. Secondly, it may be that I am capturing

response propensity and an interviewee's willingness to document expenditures on social

activities is associated with an increased likelihood of responding to questions about their

�nancial activity. I control for this potential complication throughout the empirical analysis.

Third, I �nd that aggregate returns and volatility as well as expectations of future returns

are not the driving force behind the relationship. I also provide an alternative speci�cation

of active versus passive investing, namely that social interactions are more strongly related

to other forms of security ownership than to what could be considered an extremely passive

form of investing, ownership of U.S. savings bonds.

4.1 Is it disposable income?

A chief concern is that my regressions are simply picking up shocks to disposable income

that lead to increases in both investment activity and expenditures on non-essential items

regardless of whether or not they are social in nature. I control for this possibility in two ways

in the baseline regressions. First, all regressions include before tax income as an independent

14

variable. Presumably, those with more income have the opportunity to purchase more items

including securities. However, simply controlling for income may not be su�cient since

wages, Social Security income, alimony payments, and the other items included in this

variable, are typically slow-moving and do not account for shocks to disposable income such

as bequests and the repayment of debt. The second control variable, the number of items the

respondent reports expenditures on, may do a better job of capturing shocks to disposable

income since one would be expected to expand the set of items the consumer purchases.

The �rst method I employ to rule out the possibility that shocks to disposable income

are driving the relationship found in my results is to run the same set of regressions, but

instead of including social variables as regressors, I use other expenditure variables from the

CEQ that do not have a clear relationship to one's propensity to be engaged in social activity

(Table 8). This includes items such as online entertainment and television purchases as well

as photographic equipment and musical instruments. First, it is di�cult for me to entirely

rule out the disposable income story when it comes to the decision to own securities � �ve of

the eleven variables I present are statistically signi�cant and positively related to being an

investor. However, the expenditure variables fail to exhibit correlation with active investing

when I condition the regressions on being an investor. Only one of the twenty-two possible

speci�cations (very active regressed on stamps or coin collecting) yields a statistically sig-

ni�cant and positive relationship between the variable of interest and active investing. In

fact, three of the expenditure variables namely photographer fees, musical instruments, and

hunting and �shing, exhibit a negative and statistically signi�cant relationship with active

investing. This suggests that conditioning on whether or not an individual owns securities

adequately controls for the e�ect of disposable income.

Secondly, in future work I plan on instrumenting for disposable income by examining the

e�ect of the Bush and Obama stimulus packages on expenditures and seeing if the increase

in expenditures plays a role in determining investor activity. This method has considerable

potential since the Bush and Obama stimulus packages represent exogenous shocks to the

disposable income of liquidity constrained investors.

4.2 Is it response propensity?

Another possibility is that the relationship between investing and the proxies for sociability

is being driven by the propensity to respond to survey questions. In other words, having

provided information about their expenses on social activity makes the individual more likely

to respond to questions about their �nances. I refute this theory in several ways. First, in

Section 4.1, I show that many non-social expenditure variables are unrelated to investing

15

activity. These variables would display a positive relationship with investing throughout if

response propensity were behind my �ndings.

I also include two additional variables in all regressions that attempt to control for the

interveiwee's likelihood of responding to any given question. The �rst variable is a simple

count of the number of items the household reports expenditures on. It likely captures

the ability of the household to consume, but when controlling for income the variable also

represents the respondent's willingness to answer individual questions from the survey. The

second variable is recorded by the interviewer and asks, �[i]n answering questions about

expenses, did the respondent consult bills, receipts, check stubs, expense books, tax returns,

or other records, 90 percent, 50 to 90 percent, 10 to 40 percent, or less than 10 percent of the

time?�. This variable proxies for the respondent's enthusiasm towards participating in the

survey and its inclusion in the regressions controls for a spurious correlation between social

interaction and investing caused by individual response propensity. Furthermore, I run the

baseline regressions conditional on consulting records more than half and again for less than

half of the time, and �nd that it does not change the empirical relationship between investing

and socialability.

4.3 Are returns driving both investment activity and social expen-

ditures?

It may be that returns are driving both increased investment activity and expenditures on

items related to social activity. In the baseline speci�cations I include year �xed e�ects

to capture general macroeconomic conditions common to all individuals. In order to add

robustness to my results I drop the yearly �xed e�ects and include measures of excess returns

and realized volatility common to all individuals in a given month of a given year. Doing so

does not change my �ndings and the relationship between aggregate returns and volatility

and household investing is not strong.

Unfortunately, the year �xed e�ects and aggregate market statistics are common to all

individuals at a given point in time and it may be that I need to capture idiosyncratic

expectations of future returns. The CEQ does not provide any direct way to measure house-

hold expectations; however, I can proxy for the expectation of future returns by including

a variable on the current consumption of durable goods per household.18 I include in the

baseline speci�cations a variable that is the sum of durable goods consumption per household

normalized by household income.19 This variable is positively related to the ownership of

18I attribute Daniel L. Tortorice with suggesting this proxy.19The durable goods variable is the sum of household expenditures on: �mattress and springs�, �other

bedroom furniture�, �sofas�, �living room chairs�, �living room tables�, �kitchen/dining room furniture�,

16

securities and statistically signi�cant at one percent. Furthermore, it is positive and statisti-

cally signi�cant at the ten percent level when the dependent variable asks whether or not the

respondent is an active investor, but statistically insigni�cant when the dependent variable is

very active. However, the inclusion of durable goods consumption in the regression analysis

has no e�ect on the proxies for sociability, working against the theory that the expectations

of future returns are simultaneously driving the propensity to invest and engage in social

activity.

4.4 U.S. bonds versus other securities

In this section I attempt an additional test of the relationship between social interaction

and active investing. The CEQ includes another variable, �[o]n the last day of (last month),

what was the total amount your CU had in U.S. Savings bonds?�. While active investing

is typically considered to be over-management of one's portfolio it may also represent the

aggressive pursuit of high returns with little regard for the riskiness of the asset or strategy.

Therefore, other variables that capture this distinction, such as equities versus U.S. savings

bond ownership, may make for good proxies. If social interaction is tilted towards active

investing then I would expect to see a stronger relationship between the social variables and

my variable �investor � than with an indicator variable for U.S. savings bond ownership.

I test this theory by estimating two logistic regressions in which the dependent variables

are (1) whether or not the individual owns securities and (2) whether or not they own U.S.

savings bonds. The independent variables of interest are the proxies for social interaction

presented in Section 2.1.2 and the control variables are the same as those used throughout.

The results are presented in Table 9. Of course, theory predicts that social interactions

are related to participation in �nancial markets and as expected, there is a positive and

statistically signi�cant relationship between the proxies for social activity and U.S. bond

ownership. However, when I compute marginal e�ects evaluated at the means of all other

variables the discrete change from non-social to social has a statistically larger impact on

security ownership than on U.S. bond ownership in all speci�cations except participation

sports. This implies that social interaction has a smaller impact on passive forms of investing

such as U.S. bond ownership.

�purch/inst refrig/freezer rntr (ownd)�, �purch/inst clothes washer rntr (ownd)�, �purch/inst clothes dryerrntr (ownd)�, �stoves, ovens rntr (ownd)�, �lawn and garden equipment�, �power tools�, �o�ce furniture homeuse�, �watches�, �jewelry�, �new cars�, �new trucks�, and �new motorcycles�, divided through by before taxincome. All items are converted to March 2010 prices.

17

5 Conclusion

This research reinforces existing literature on the relationship between social interaction

and asset market participation, and goes one step further by showing that social interac-

tion is biased towards active investing. I do so by showing that those with expenditures on

items related to social activity are positively associated with active investing conditional on

participation in asset markets. I rule out several other potential explanations; the relation-

ship is not driven by the spurious presence of disposable income, aggregate returns or the

expectations of future returns, or response propensity. Furthermore, I o�er an alternative

speci�cation and show that the relationship between social interaction and ownership of U.S.

savings bonds while also positive, is smaller than that of other securities.

This �nding has many implications. First, much of the research that looks under the hood

at the participation rate has the normative implication that most households would bene-

�t from increased use of asset markets. If informal communications motivate participation

then my �ndings call into question whether or not they improve the welfare of individuals.

Secondly, my �ndings can be viewed in the context of many models that include specula-

tors or noise traders. Asset price bubbles may also be driven by word-of-mouth between

active investors. Lastly, this research o�ers an explanation for active investing by individual

investors, a puzzle that lacks explanations rooted in rational behavior.

While other research has identi�ed a causal relationship from sociability to participation

in asset markets, unfortunately, this research is unable to address the direction of causality

in the relationship between social interaction and active investing. Furthermore, theory fails

to point de�nitively in one direction or the other. For one, models such as Hirshleifer (2010)

show that behavioral biases may make individuals susceptible to conversation with active

investors. On the other hand, individual investors may seek information about asset returns

through informal communication with others (Ellison and Fudenberg (1995)). This would

lead active investors to court those in-the-know with dinners out at fancy restaurants and

box seats to big sporting events.

Regardless, identifying the presence of this relationship is valuable for several reasons.

It reinforces the notion that education about �nancial markets should occur through formal

means. Models that include informal communication between investors should be cognizant

of this empirical relationship. It also points to the need for future research that explores

the channels of communication between individual investors that pins down the direction of

causality.

18

References

[1] Daron Acemoglu, Asuman Ozdaglar, and Ali ParandehGheibi, Spread of

(mis)information in social networks, Games and Economic Behavior 70 (2010), no. 2,

194�227.

[2] Franklin Allen and Douglas Gale, Limited market participation and volatility of asset

prices, American Economic Review 84 (1994), no. 4, 933�55.

[3] Abhijit V. Banerjee, A simple model of herd behavior, The Quarterly Journal of Eco-

nomics 107 (1992), no. 3, 797�817.

[4] Brad M. Barber, Yi-Tsung Lee, Yu-Jane Liu, and Terrance Odean, Just how much

do individual investors lose by trading?, Review of Financial Studies 22 (2009), no. 2,

609�632.

[5] Brad M. Barber and Douglas Loe�er, The �dartboard� column: Second-hand infor-

mation and price pressure, Journal of Financial and Quantitative Analysis 28 (1993),

no. 02, 273�284.

[6] Brad M. Barber and Terrance Odean, Too many cooks spoil the pro�ts: The performance

of investment clubs, Financial Analyst Journal 56 (2000), no. 1, 17�25.

[7] , Boys will be boys: Gender, overcon�dence, and common stock investment, The

Quarterly Journal of Economics 116 (2001), no. 1, 261�292.

[8] , Online investors: Do the slow die �rst?, Review of Financial Studies 15 (2002),

no. 2, 455�488.

[9] Brad M. Barber, Terrance Odean, and Ning Zhu, Do retail trades move markets?, Re-

view of Financial Studies 22 (2009), no. 1, 151�186.

[10] Patrick Bayer, Stephen L. Ross, and Giorgio Topa, Place of work and place of residence:

Informal hiring networks and labor market outcomes, Journal of Political Economy 116

(2008), no. 6, 1150�1196.

[11] Gary S. Becker, A note on restaurant pricing and other examples of social in�uences on

price, Journal of Political Economy 99 (1991), no. 5, 1109�16.

[12] Sushil Bikhchandani, David Hirshleifer, and Ivo Welch, A theory of fads, fashion, cus-

tom, and cultural change in informational cascades, Journal of Political Economy 100

(1992), no. 5, 992�1026.

19

[13] Fischer Black, Noise, Journal of Finance 41 (1986), no. 3, 529�543.

[14] Je�rey R. Brown, Zoran Ivkovic, Paul A. Smith, and Scott Weisbenner, Neighbors

matter: Causal community e�ects and stock market participation, Journal of Finance

63 (2008), no. 3, 1509�1531.

[15] Brian K. Bucks, Arthur B. Kennickell, Traci L. Mach, and Kevin B. Moore, Changes in

u.s. family �nances from 2004 to 2007: Evidence from the survey of consumer �nances,

Federal Reserve Bulletin (2009).

[16] Antoni Calvó-Armengol and Matthew O. Jackson, The e�ects of social networks on

employment and inequality, American Economic Review 94 (2004), no. 3, 426�454.

[17] Antoni Calvó-Armengol, Eleonora Patacchini, and Yves Zenou, Peer e�ects and social

networks in education, Review of Economic Studies 76 (2009), no. 4, 1239�1267.

[18] James J. Choi, David Laibson, and Andrew Metrick, How does the internet a�ect trad-

ing? evidence from investor behavior in 401(k) plans, Journal of Financial Economics

64 (2002), no. 3, 397�421.

[19] Boriana Chopova, Jennifer Edgar, Je�rey Gonzalez, Susan King, Dave McGrath, and

Lucilla Tan, Assessing nonresponse bias in the ce interview survey: a summary of four

studies, Paper Presented at the Joint Statistical Meetings of the American Statistical

Association, Washington DC (2009).

[20] Nicholas A. Christakis and James H. Fowler, The spread of obesity in a large social

network over 32 years, New England Journal of Medicine 357 (2007), no. 4, 370�379.

[21] Lauren Cohen, Andrea Frazzini, and Christopher Malloy, The small world of investing:

Board connections and mutual fund returns, Journal of Political Economy 116 (2008),

no. 5, 951�979.

[22] Stefano DellaVigna and Ulrike Malmendier, Paying not to go to the gym, American

Economic Review 96 (2006), no. 3, 694�719.

[23] Bradford J. Delong, Andrei Shleifer, Lawrence H. Summers, and Robert J. Waldmann,

Noise trader risk in �nancial markets, Journal of Political Economy 98 (1990), no. 4,

703�38.

[24] Esther Du�o and Emmanuel Saez, Participation and investment decisions in a retire-

ment plan: the in�uence of colleagues' choices, Journal of Public Economics 85 (2002),

no. 1, 121�148.

20

[25] , The role of information and social interactions in retirement plan decisions: Ev-

idence from a randomized experiment, The Quarterly Journal of Economics 118 (2003),

no. 3, 815�842.

[26] Glenn Ellison and Drew Fudenberg, Rules of thumb for social learning, Journal of Po-

litical Economy 101 (1993), no. 4, 612�43.

[27] , Word-of-mouth communication and social learning, The Quarterly Journal of

Economics 110 (1995), no. 1, 93�125.

[28] Thierry Foucault, David Sraer, and David J. Thesmar, Individual investors and volatil-

ity, The Journal of Finance 66 (2011), no. 4, 1369�1406.

[29] Jordi Gali, Keeping up with the joneses: Consumption externalities, portfolio choice,

and asset prices, Journal of Money, Credit and Banking 26 (1994), no. 1, 1�8.

[30] Dimitris Georgarakos and Giacomo Pasini, Trust, sociability, and stock market partici-

pation, Review of Finance 15 (2011), no. 4, 693�725.

[31] Dan Givoly and Dan Palmon, Insider trading and the exploitation of inside information:

Some empirical evidence, Journal of Business 58 (1985), no. 1, 69�87.

[32] Edward L. Glaeser, Bruce Sacerdote, and Jose A. Scheinkman, Crime and social inter-

actions, The Quarterly Journal of Economics 111 (1996), no. 2, 507�48.

[33] Aron A. Gottesman and Matthew R. Morey, Manager education and mutual fund per-

formance, Journal of Empirical Finance 13 (2006), no. 2, 145�182.

[34] Mark Grinblatt, Matti Keloharju, and Juhani T. Linnainmaa, Iq, trading behavior, and

performance, Journal of Financial Economics forthcoming (2011).

[35] Rawley Z. Heimer, Panel attrition and unit nonresponse adjustment in the consumer

expenditure survey, Internal BLS Manuscript (2011).

[36] Rawley Z. Heimer and David Simon, Facebook �nance: How social interaction propagates

active investing, Working Paper (2012).

[37] David Hirshleifer, Self-enhancing transmission bias and active investing, Working Paper

(2010).

[38] Harrison Hong, Je�rey D. Kubik, and Jeremy C. Stein, Social interaction and stock-

market participation, Journal of Finance 59 (2004), no. 1, 137�163.

21

[39] , Thy neighbor's portfolio: Word-of-mouth e�ects in the holdings and trades of

money managers, Journal of Finance 60 (2005), no. 6, 2801�2824.

[40] Soeren Hvidkjaer, Small trades and the cross-section of stock returns, Review of Finan-

cial Studies 21 (2008), no. 3, 1123�1151.

[41] Justin Jouvenal and Dan Morse, Police probe germantown �ash-mob thefts, The Wash-

ington Post (2011).

[42] Markku Kaustia and Samuli Knüpfer, Peer performance and stock market entry, Journal

of Financial Economics forthcoming (2011).

[43] Alok Kumar, Who gambles in the stock market?, Journal of Finance 64 (2009), no. 4,

1889�1933.

[44] Alok Kumar and Charles M.C. Lee, Retail investor sentiment and return comovements,

Journal of Finance 61 (2006), no. 5, 2451�2486.

[45] Albert S. Kyle, Continuous auctions and insider trading, Econometrica 53 (1985), no. 6,

1315�35.

[46] Juhani T. Linnainmaa, The anatomy of day traders, mimeo (2003).

[47] Ulrike Malmendier and Stefan Nagel, Depression babies: Do macroeconomic experiences

a�ect risk taking?, The Quarterly Journal of Economics 126 (2011), no. 1, 373�416.

[48] Rajnish Mehra and Edward C. Prescott, The equity premium: A puzzle, Journal of

Monetary Economics 15 (1985), no. 2, 145�161.

[49] Thomas Oberlechner and Carol Osler, Survival of overcon�dence in currency markets,

Journal of Financia and Quantitative Analysis forthcoming (2011).

[50] Robert D. Putnam, Bowling alone: The collapse and revival of american community,

Simon & Schuster, New York, 2000.

[51] Linda Sandler, Salomon holders watch for a possible bu�et!-ing, Wall Street Journal

(1996), C1.

[52] Robert J. Shiller, Stock prices and social dynamics, Brookings Papers on Economic

Activity 15 (1984), no. 2, 457�510.

[53] , Market volatility, vol. 1, MIT Press Books, no. 0262691515, The MIT Press,

1989.

22

[54] Robert J. Shiller and John Pound, Survey evidence on di�usion of interest and informa-

tion among investors, Journal of Economic Behavior & Organization 12 (1989), no. 1,

47�66.

[55] Sophie Shive, An epidemic model of investor behavior, Journal of Financial and Quan-

titative Analysis 45 (2010), no. 01, 169�198.

[56] Jeremy C. Stein, Informational externalities and welfare-reducing speculation, Journal

of Political Economy 95 (1987), no. 6, 1123�45.

[57] Maarten van Rooij, Annamaria Lusardi, and Rob Alessie, Financial literacy and stock

market participation, Journal of Financial Economics 101 (2011), no. 2, 449�472.

[58] Annette Vissing-Jorgensen, Limited asset market participation and the elasticity of in-

tertemporal substitution, Journal of Political Economy 110 (2002), no. 4, 825�853.

23

Table 1: Financial Participationincome, X, (in March 2010 USD)

Variable all income X < 50,000 50,000 ≤ X < 100,000 X ≥ 100,000

% own securities 12.78% 6.29% 17.27% 30.25%

% buy securities 3.83% 1.35% 4.94% 11.50%

% sell securities 2.70% 1.31% 3.30% 7.02%

% buy or sell securities (active management) 5.36% 2.26% 6.86% 14.75%

% buy and sell securities (very active management) 1.08% 0.49% 1.38% 3.77%

% own securities, no buy or sell (passive management) 8.82% 4.85% 12.23% 18.44%

Observations 75,189 44,614 18,845 11,730

24

Table 2: Summary Statistics on FinancesNominal Expenditures (USD)

Variable Obs Mean Median Std. Dev. Min Max

amount owned in securities 9,619 183,435.5 35,000 498,140.6 0 3,846,362

amount bought securities 2,890 59,023.82 5,000 422,393.3 0 5,392,857

amount sold securities 2,044 77,083.07 6,600 606,236 0 8,192,727

income before taxes 75,189 49,209.67 34,000 56,194.57 -144,604 1,107,217

income before taxes (if own securities) 9,619 87,672.27 69,690 78,464.65 -144,604 1,107,217

Expenditures in March 2010 Prices (USD)

Variable Obs Mean Median Std. Dev. Min Max

amount owned in securities 9,619 204,683.9 40,232.45 548,683.2 0 4,576,459

amount bought securities 2,890 70,067.47 5,683.432 521,752.4 0 6,868,009

amount sold securities 2,044 91,713.89 7,413.461 749,493.7 0 1.00E+007

income before taxes 75,189 55,007.19 38,375.57 62,142.85 -145,967.1 1,140,043

income before taxes (if own securities) 9,619 98,522.49 78,520.85 86,249.72 -145,967.1 1,140,043

25

Table 3: Financial Participation, Survey of Consumer FinancesPercentage of families holding asset

Bonds Stocks Pooled Investment Funds

year2004 1.8 20.7 15.02007 1.6 17.9 11.4

Note: Data comes from the Survey of Consumer Finances, Bucks et al. (2009)

26

Table 4: Social Participationincome, X, (in March 2010 USD)

Variable all income X < 50,000 50,000 ≤ X < 100,000 X ≥ 100,000

% Church Donations 17.14% 11.87% 20.80% 31.28%

% Admission to Sports 15.79% 10.21% 20.19% 29.92%

% Participation in Sports 15.60% 9.68% 19.84% 31.28%

Observations 75,189 44,614 18,845 11,730

27

Table 5: Social Participation Correlation

Admission to Sports Participation in Sports Church Donations

Admission to Sports 1

Participation in Sports 0.3148 1

Church Donations 0.1188 0.124 1

28

Table 6: The Facts About Investingconditional on investing

investor active very activeage 1.018*** 0.991*** 0.997

(0.001) (0.002) (0.003)female 0.791*** 0.709*** 0.543***

(0.021) (0.035) (0.050)married 0.941* 0.930 0.863

(0.032) (0.058) (0.095)college educated 1.687*** 1.377*** 1.684***

(0.055) (0.085) (0.198)rural 0.943 0.885 0.648**

(0.047) (0.089) (0.142)log income 1.490*** 1.211*** 1.464***

(0.029) (0.045) (0.113)technology 1.163*** 1.096 1.219**

(0.043) (0.066) (0.122)insured 1.151*** 0.968 0.780***

(0.030) (0.048) (0.066)race yes yes yesregion �xed e�ects yes yes yesyear �xed e�ects yes yes yesN 67,501 9,441 7,260pseudo R2 0.19 0.05 0.07

Odds ratios; Robust standard errors in parentheses∗ p < 0.10 , ∗∗ p < 0.05 , ∗∗∗ p < 0.01

Description: The data comes from the Consumer Expenditure Quarterly Interview Survey(CEQ), 2000Q2 - 2010Q1. The dependent variables are de�ned as follows. �investor� is azero/one indicator, one if the consumer unit (CU) reports ownership of stocks, mutual funds,private bonds, government bonds or Treasury notes. �active� is if �investor� equals one andthey report either buying or selling one of the assets. �very active� is the same as �active�except the CU reports both buying and selling. The third column excludes those who are�active� but not �very active�. All logistic regressions are estimated with only one socialvariable at a time.

29

Table 7: Baseline Regressionsconditional on investing

investor active very activechurch donations 1.375∗∗∗ 1.181∗∗∗ 1.264∗∗∗

(0.0408) (0.0607) (0.113)admission sports 1.272∗∗∗ 1.091∗ 1.188∗∗

(0.0395) (0.0563) (0.105)participation sports 1.262∗∗∗ 1.164∗∗∗ 1.126

(0.0388) (0.0593) (0.0994)control variables (Table 6) yes yes yesrace yes yes yesregion �xed e�ects yes yes yesyear �xed e�ects yes yes yesN 67,501 9,441 7,260pseudo R2 0.19 0.05 0.07

Odds-ratios; Robust standard errors in parentheses∗ p < 0.10 , ∗∗ p < 0.05 , ∗∗∗ p < 0.01

Description: The data comes from the Consumer Expenditure Quarterly Interview Survey(CEQ), 2000Q2 - 2010Q1. The dependent variables are de�ned as follows. �investor� is azero/one indicator, one if the consumer unit (CU) reports ownership of stocks, mutual funds,private bonds, government bonds or Treasury notes. �active� is if �investor� equals one andthey report either buying or selling one of the assets. �very active� is the same as �active�except the CU reports both buying and selling. The third column excludes those who are�active� but not �very active�. All logistic regressions are estimated with only one socialvariable at a time. The reported Pseudo R2 come from the regressions that include �diningout� as an explanatory variable. The other variables related to social interaction do not havea signi�cant impact on the Pseudo R2.

30

Table 8: Is it Disposable Income?conditional on investing

investor active very activestamps or coins collecting 0.957 0.957 1.845∗

(0.135) (0.237) (0.634)photograph equipment 1.121∗ 1.070 0.964

(0.0693) (0.104) (0.163)photographer fees 1.045 0.900 0.542∗∗∗

(0.0626) (0.0864) (0.105)legal fees 1.216∗∗∗ 1.181 1.208

(0.0764) (0.123) (0.209)vehicle inspection 1.133∗∗∗ 0.967 0.973

(0.0490) (0.0742) (0.127)musical instruments 0.987 0.721∗∗ 0.920

(0.0806) (0.0962) (0.191)hunting and �shing 1.137∗ 0.803∗ 0.956

(0.0802) (0.0939) (0.190)winter sports 1.041 0.931 0.530

(0.146) (0.191) (0.214)water sports 1.407∗∗ 0.744 1.121

(0.189) (0.146) (0.331)online entertainment 0.997 0.970 1.199

(0.155) (0.273) (0.497)television 0.984 1.123 1.132

(0.078) (0.166) (0.52)control variables (Table 6) yes yes yesrace yes yes yesregion �xed e�ects yes yes yesyear �xed e�ects yes yes yesN 67,501 9,441 7,260pseudo R2 0.19 0.05 0.07

Odds ratios; Robust standard errors in parentheses∗ p < 0.10 , ∗∗ p < 0.05 , ∗∗∗ p < 0.01

Description: The data comes from the Consumer Expenditure Quarterly Interview Survey(CEQ), 2000Q2 - 2010Q1. The dependent variables are de�ned as follows. �investor� is azero/one indicator, one if the consumer unit (CU) reports ownership of stocks, mutual funds,private bonds, government bonds or Treasury notes. �active� is if �investor� equals one andthey report either buying or selling one of the assets. �very active� is the same as �active�except the CU reports both buying and selling. The third column excludes those who are�active� but not �very active�. All logistic regressions are estimated with only one of thedisposable income variables at a time. The reported Pseudo R2 come from the regressionsthat include �stamps or coin collecting� as an explanatory variable. The other variablesrelated to social interaction do not have a signi�cant impact on the Pseudo R2.

31

Table 9: Securities versus US bond ownershipindep var depend var dy/dx se 95% con�dence interval

church donationsinvestor 0.0280∗∗∗ 0.00286 0.0224 0.0336US bonds 0.0134∗∗∗ 0.00224 0.00875 0.0175

admission sportsinvestor 0.208∗∗∗ 0.00288 0.0151 0.0264US bonds 0.00860∗∗∗ 0.00221 0.00427 0.0129

participation sportsinvestor 0.0201∗∗∗ 0.00284 0.0146 0.0257US bonds 0.0117∗∗∗ 0.00226 0.00732 0.0162

Marginal e�ects; Robust standard errors∗ p < 0.10 , ∗∗ p < 0.05 , ∗∗∗ p < 0.01

Description: The data comes from the Consumer Expenditure Quarterly Interview Survey(CEQ), 2000Q2 - 2010Q1. The dependent variables are de�ned as follows: �investor� is azero/one indicator, one if the consumer unit (CU) reports ownership of stocks, mutual funds,private bonds, government bonds or Treasury notes. �US bonds� is a zero/one indicator, oneif the respondent reports ownership of US bonds. All logistic regressions are estimated withonly one social variable at a time.

32