The People in Your Neighborhood: Social Interactions and Mutual Fund Portfolios

• DOI: http://doi.org/10.1111/jofi.12208
• Published date: 01 December 2015
• Date: 01 December 2015

THE JOURNAL OF FINANCE •VOL. LXX, NO. 6 •DECEMBER 2015

The People in Your Neighborhood: Social

Interactions and Mutual Fund Portfolios

VERONIKA K. POOL, NOAH STOFFMAN, and SCOTT E. YONKER∗

ABSTRACT

Wefind that socially connected fund managers have more similar holdings and trades.

The overlap of funds whose managers reside in the same neighborhood is considerably

higher than that of funds whose managers live in the same city but in different neigh-

borhoods. These effects are larger when managers share a similar ethnic background,

and are not explained by preferences. Valuable information is transmitted through

these peer networks: a long-short strategy composed of stocks purchased minus sold

by neighboring managers delivers positive risk-adjusted returns. Unlike prior em-

pirical work, our tests disentangle the effects of social interactions from community

effects.

DESPITE THE IMPORTANT ROLE professional money managers play in financial

markets, and decades of academic study, relatively little is known about how

they generate investment ideas. Research shows that managers invest in com-

panies headquartered nearby (Coval and Moskowitz (1999,2001)), and in com-

panies to which they are linked through school networks (Cohen, Frazzini, and

Malloy (2009)). They also choose stocks based on their political ideology (Hong

and Kostovetsky (2012)) and stocks with which they are merely familiar (Pool,

Stoffman, and Yonker (2012)).

But, as Aristotle famously noted, humans are social animals, so perhaps fund

managers also trade stocks that they learn about from other managers. While

numerous papers examine the effects of social interaction on choices in other

domains,1there is little empirical evidence on how word-of-mouth communica-

∗Pool and Stoffman are at the Kelley School of Business, Indiana University. Yonker is at

the charles Ho Dyson School of Applied Economic and Management, Cornell Univesity. We thank

Kenneth Ahern, Matt Billett, Martijn Cremers, Jessie Ellis, Ryan Israelsen, Ken Weakley, Kenneth

Singleton (the Editor), two anonymous referees, and an Associate Editor, as well as seminar

participants at Arizona State University,the College of William and Mary, Federal Reserve Board,

Miami University, University of Alabama, University of Massachusetts–Amherst, University of

Miami, University of Toledo, University of WesternOntario, the Early Career Women in Finance

Conference, the Financial Intermediation Research Society annual meeting, the IU-Notre Dame-

Purdue Summer Symposium, the Second Michigan State Federal Credit Union Conference, and

brown bags at The Ohio State University and Indiana University for helpful comments. Research

support was provided by the Indiana University Kelley School of Business. We have read the

Journal of Finance’s disclosure policy and have no conflicts of interest to disclose.

1For example, Grinblatt, Keloharju, and Ik¨

aheimo (2008) document a substantial influence of

near-neighbors on automobile purchases. Bayer, Ross, and Topa (2008) show the importance of

DOI: 10.1111/jofi.12208

2679

2680 The Journal of Finance R

tions influence professional investors’ decision to trade a stock. Hong, Kubik,

and Stein (2005) take an important first step in answering this question by

studying a broad sample of mutual funds. They show that the holdings and

trades of fund managers who work in the same city are correlated.2

Although these results are consistent with the hypothesis that professional

money managers transmit investment ideas socially,3the authors point to sev-

eral alternative hypotheses that are difficult to rule out with their data. Specif-

ically, the correlation in portfolios could be due to fund managers in the same

city being exposed to the same local media outlets, being visited by the same

corporate executives during investor-relations road shows, or herding with

local managers, which could be induced by geographic segmentation of the

job market combined with career concerns (Scharfstein and Stein (1990)and

Chevalier and Ellison (1999)). These alternative “community effects” would

imply that news travels through formal information channels, whereas the

social hypothesis implies that information travels through informal person-to-

person relationships. Of course, both channels can operate simultaneously. In

this paper we implement a test that allows us to disentangle the two effects.

If we could observe whether any two managers know and communicate with

each other, constructing an empirical test would be straightforward. In the

absence of such data, however, we rely on a unique identification strategy to

uncover person-to-person relationships. We argue that managers who live near

one another (“neighbors”) have a better chance of meeting—and subsequently

becoming acquaintances or friends—than managers who live further apart.4

For example, managers might meet at a neighborhood park or school, or while

taking the train to work. The longer they live near each other, the more such

opportunities to become friends will arise. Further, having become friends,

neighbors may have more ongoing interactions due to a higher probability of

random encounters or shared social connections through local schools or places

of worship. Importantly, we are not suggesting that these random encounters

are the main method for the transmission of ideas, but rather that they increase

the probability of planned interactions.

social interaction in labor markets, while Sacerdote (2001) finds strong peer effects on educational

outcomes among randomly assigned college roommates. Bertrand, Luttmer, and Mullainathan

(2000) find similar effects on welfare participation rates, as do Glaeser,Sacerdote, and Scheinkman

(1996) on crime rates.

2Ivkovi´

c and Weisbenner (2007) find similar results for individual investors who live within

50 miles of each other. Feng and Seasholes (2004) show correlated trading among proximate

individual investors in China by exploiting brokerage rules that investors must trade at their

branch office. In earlier survey research, Shiller and Pound (1989) find that both individual and

institutional investors report that their portfolio choices are driven in part by interpersonal com-

munication.

3Throughout the paper, we use the phrase “social interactions” in the traditional sense of

the social sciences literature. That is, any relation between two people, regardless of where the

interaction takes place.

4In this respect, our approach is similar in spirit to that of Cohen, Frazzini, and Malloy (2008),

who argue that board members and executives who attended the same college at the same time

are likelier to know each other.

Social Interactions and Mutual Fund Portfolios 2681

Todetermine which managers are neighbors, we collect the complete residen-

tial address history of fund managers in our sample from public records data,

and calculate the pairwise distance between the homes of all managers. We

classify managers as neighbors only if they live truly close to each other—for

example, just a fraction of a mile in densely populated areas in Manhattan

or Boston. (The distance cutoff varies by population density, as we explain

later.) Using our distance measure to proxy for social interaction creates vari-

ation within a city that is independent of sharing a media market, road shows,

career concerns–induced herding, or any other community effects.

Thus, while previous papers document correlated trading among professional

and individual investors using far coarser definitions of neighbors, we are able

to identify the effects of social contact by zeroing in on fund managers who are

likely to know each other, rather than treating all fund managers based in, say,

New YorkCity as neighbors. Prior studies rely on coarse definitions of neighbors

for two reasons. First, they do not have residential addresses for the investors

in their samples. Second, their empirical design tests whether the trades and

holdings of investors are more sensitive to those of nearby investors than to

those of a distant cohort. To perform such portfolio-based tests it is necessary

to have a sufficient number of nearby investors to form the nearby cohort for

each investor in the sample. As the distance between investors constituting

“nearby” gets smaller, fewer and fewer investors meet this criterion, making

such a test impossible to implement.

We circumvent this problem by designing an empirical test that does not

require every investor in our sample to have a neighbor. We construct measures

of pairwise overlap in holdings and trades for all funds, and test whether the

overlap is greater when fund pairs are managed by neighbors. This design

allows us not only to shrink the distance in the definition of neighbor, but also

to control for other common community effects that are difficult to separate

from the effects of social interactions in other empirical setups.

Remarkably, the portfolio overlap of funds managed by neighbors is 12%

higher in our baseline model than that of funds whose managers live in the

same city but are not neighbors—even after controlling for investment styles

and fund family memberships. This increases to 28% when we implement a

cleaner test by focusing on funds with just one manager. We find similarly

strong results for trades.

These results are economically large. The increase in portfolio overlap that

comes with being neighbors is 2.5 times that of funds whose managers are in

the same 50-mile radius media market, and five times as large as the effect

of being in the same city. Moreover, abnormal overlap for neighbor funds is

about half as large as it is for two funds whose managers comanage another

fund together, and it is about one quarter as large as the overlap between

funds that belong to the same fund family, which shares analysts and other

stock-selection resources. Despite the size of the estimates, however,our results

are likely to understate the true magnitude of the effects of social contact for

two reasons. First, our neighbor proxy is clearly a noisy measure of whether

managers socially interact. If, for example, only half of the managers whom we

classify as neighbors actually know each other, the true effect would be twice