No Place Like Home: Familiarity in Mutual Fund Manager Portfolio Choice

Transcription

1 No Place Like Home: Familiarity in Mutual Fund Manager Portfolio Choice Veronika Krepely Pool, Noah Stoffman, and Scott E. Yonker May 17, 2011 Preliminary and incomplete. Comments welcomed. Abstract We show that familiarity plays an important role in determining mutual fund managers portfolio decisions. Managers overweight companies headquartered in the states where they grew up by between 2.3% and 14.4%, compared to similar funds run by managers from other states. This effect is distinct from the fund location effect of Coval and Moskowitz (1999), and holds in specifications that control for time-invariant fund characteristics. This preference for home-state companies is not associated with higher risk-adjusted returns, indicating that the decisions are not driven by information. The overweighting is stronger in managers of smaller funds, managers who are early in their careers, and those who spent more time in their home state. As managers become more experienced, they decrease holdings of home-state companies in favor of companies located close to the mutual fund complex headquarters. We thank Utpal Bhattacharya, Tyler Shumway, and participants of the Indiana University finance brown bag for helpful comments. The authors are at Kelley School of Business, Indiana University. Contact information: Pool: Stoffman: Yonker:

2 I Introduction Investors invest in what they know (Merton, 1987). For example, individual investors invest more in nearby firms (Grinblatt and Keloharju, 2001; Ivković and Weisbenner, 2005; Seasholes and Zhu, 2010), and a home bias in equity holdings is pervasive around the world (French and Poterba, 1991; Tesar and Werner, 1995; Kang and Stulz, 1997). Whether the local equity preference of individual investors is driven by information or familiarity is a subject of debate in the literature. 1 Among professional investors, it would be surprising if a preference for local stocks were not driven by information, since these investors face higher costs from poor portfolio performance due, for example, to career concerns (Chevalier and Ellison, 1999). Consistent with this, Coval and Moskowitz (1999, 2001) find support for informed local investing by mutual funds. In this paper, we ask whether mutual fund managers also overweight stocks with which they are merely familiar. The challenge in testing whether familiarity plays a role in the portfolio choice of professional managers is to identify securities that are, ex ante, likely to be familiar to the fund manager, but for which the manager has no informational advantage. Companies that are geographically close to the fund do not fit this profile: although they are likely to be familiar to fund managers, the fund is also likely to be informed about these companies because its entire investment team (e.g. investment analysts) and employees share a connection to the location, and possibly to the nearby firms. In this paper, we therefore focus on firms headquartered in a mutual fund manager s home state. 2 For example, we hypothesize that a manager who grew up in Indiana and who works for a fund in New York would be likely to overweight holdings in Indiana-based firms, but the argument that this overweighting is due to information is a little more tenuous. For instance, since the home state represents the manager s past, whether the manager has an active connection to the area is more uncertain. Also, home-state securities are likely to be familiar to the manager of the fund, but not to the rest of the investment team. Since most of the fund s information gatherers do not have links to manager s home-state firms, but the manager does, we believe that manager home-state securities meet the criteria of familiar and unlikely to be informed. 1 Ivković and Weisbenner (2005) find support for the information hypothesis, while Grinblatt and Keloharju (2001), Huberman (2001), and Seasholes and Zhu (2010) find evidence that investors suffer from a familiarity bias. 2 Throughout the paper, we refer to the state where the manager grew up as their home state. We identify home states using Social Security data, as explained below. 1

3 Whether securities from managers home states fit these criteria ex post is an empirical question. Therefore, to investigate whether familiarity has a role in portfolio decisions, we proceed in two steps. First, we ask whether managers overweight companies from their home states. Evidence that they do is consistent with familiarity; but also with information. To distinguish between familiarity and information, we then ask whether funds earn higher returns on their home-state investments. Under our familiarity hypothesis, managers may simply be more aware of home-state companies, even if they have no real information about them. This would likely be true especially of small companies with few customers and low levels of advertising that operate primarily in the manager s home state. While professional fund managers are probably aware of most, if not all, stocks in their investment universe, if managers are more familiar with firms from their home states, then perhaps when choosing among stocks they choose the familiar one. 3 In contrast, under our information hypothesis, managers may have access to information about companies at home. For example, friends and family may still live in the manager s home state and work at the company. These contacts can provide information about general business conditions at their employer, even if they rarely provide any illegal inside information. The value of the information need not be especially high as long as the manager has a comparative advantage in gathering information from his home state, then he will be more likely to trade on it. The empirical predictions implied by these competing hypotheses differ starkly. If managers are informed about home companies, we would expect them either to overweight or underweight these companies depending on whether their information is positive or negative and generate positive abnormal returns from their information. If instead managers are simply more familiar with companies from home, they should overweight home-state companies without generating abnormal returns. In other words, the information hypothesis would suggest active over- or underweighting with positive abnormal returns, while the familiarity hypothesis suggests overweighting and no abnormal returns in a manager s home state holdings. 3 A related story is that managers may have an emotional connection to their home that leads them to feel sentimental about elements of life at home, including local companies. Hwang (2011) for example, finds a positive association between American investors demand for a country s securities and the popularity of that country among Americans. Our evidence indicates, however, that distance matters more than state boundaries, as discussed below. 2

4 We begin our empirical analysis by asking whether mutual fund managers invest more in the state where they grew up than do peers from other states. We find that they do. Fund managers invest more in companies that are headquartered in their home states than do managers of similar funds who grew up elsewhere. The overweighting in home-state companies is economically large: managers invest 14.4% more in their home state than would otherwise be expected. Even after including fund fixed effects, which identifies the effect using within-fund variation and presents a very high econometric hurdle, we find an overweighting of 2.3%. This translates to an overweighting of between $1.6 million and $10.1 million per fund-quarter. We then investigate performance to directly test our hypotheses and find strong empirical support for the familiarity hypothesis: The excess holdings in home-state firms by fund managers do not produce positive abnormal returns. Moreover, managers put more weight on home-state companies when they are early in their careers, suggesting they have not yet fully developed ways of gathering information. As they become more experienced and develop more connections to local firms, the overweighting shifts from home-state companies to companies located where the fund complex is headquartered. The former overweighting does not produce positive abnormal returns, while the latter does. Finally, managers who left their home state earlier in life and perhaps have less of a connection to home have less tendency to overweight home companies. Our paper contributes to an active literature on the relation between investors investment choices and their beliefs, values and experiences. Cohen (2009) finds that employees invest more of their retirement savings in the stock of their employer when they feel a sense of loyalty. Internationally, Morse and Shive (2011) show that investors in more patriotic countries invest more in domestic equity, while Bhattacharya and Groznik (2008) find that foreign direct investment from the U.S. is higher in countries with many emigrants living in the U.S. Grinblatt and Keloharju (2001) and Huberman (2001) find individual investors are more likely to invest in firms that are familiar to them. Hong, Kubik, and Stein (2003) and Shive (2010) show that social contact influences trading behavior among institutional investors and individuals, respectively. Ivković and Weisbenner (2005), find that individual investors earn more on their local investments than their non-local investments, suggesting that investors profit from an information advantage in local firms, although Seasholes and Zhu (2010) find contradictory results. Malmendier and Nagel (2010) show that investors who 3

5 have experienced periods of low market returns take less financial risk, while Korniotis and Kumar (2010) show that various life experiences, proxied by height, are significant predictors of risk-taking in the stock market. Among professional managers, Hong and Kostovetsky (2010) show that the political preferences of fund managers affect their portfolio choices, and Greenwood and Nagel (2009) find that younger fund managers were more heavily invested in technology stocks at their peak, and exhibit more trend-chasing behavior, than their older and more experienced peers. Coval and Moskowitz (1999, 2001) show that fund managers invest more in companies whose headquarters are located close to the fund, and earn higher returns on their local holdings. In other related work, Cohen, Frazzini, and Malloy (2008) identify networks of fund managers and board members based on college attendance and find that fund managers earn higher returns on firms to which they are connected. We show that fund managers invest in their home state for other reasons, and that college social networks do not drive the home state bias. A number of recent papers have also considered the relation between geography and information. Baik, Kang, and Kim (2010) find that trading by local institutions predicts future stock returns, while the predictive abilities are weaker for nonlocal institutions. Bernile, Kumar, and Sulaeman (2010) offer an alternative to institutional proximity to headquarters as a measure of information. They show that investors invest more in companies that are economically active in the state where the fund is based. Geographic effects also appear to have asset pricing implications: Hong, Kubik, and Stein (2008) show that geographic segmentation has stock price effects as local investors bid up the prices of the local stocks that are the only game in town. We offer another perspective on what it means for a stock to be local, and highlight one of the ways that fund managers choose stocks. The remainder of the paper is organized as follows. In section II we describe our data and explain how we construct our sample. We present our results in Section III and the robustness of the results in Section IV. Section V concludes. The Appendix provides additional details on sample construction, as well as definitions of all variables used in the paper. 4

6 II Data and Sample Construction We combine several data sources in this study. First, we draw information on fund managers from Morningstar. Morningstar reports the manager s name for each fund (including individuals on team-managed funds), their start and end dates with the fund, and information on the manager s educational background. We limit our sample to actively managed U.S. equity funds by filtering the observations using Morningstar style categories as well as manually screening the fund names. Our second data source is the Thomson Financial CDA/Spectrum Mutual Fund database. The database contains the quarter-end holdings reported by U.S.-based mutual funds in mandatory SEC filings. Thomson uses two date variables, RDATE and FDATE, which refer to the actual date for which the holdings are valid and the Thomson vintage date on which the data was cut, respectively. We follow standard practice and restrict the holdings to those observations where the FDATE is equal to the RDATE to avoid the use of stale data in our analysis. From this starting point, there are 4,731,878 quarterly fund-holding observations from the first quarter of 1996 to the fourth quarter of Since we are interested in the domestic portion of the funds portfolios, we next remove holdings in firms headquartered outside of the United States. This reduces the number of quarterly fund holding observations to 4,446,768. There are 48,077 quarterly fund observations with 2,552 distinct funds holding 13,908 different securities. 4 We obtain monthly return and total net asset value information for each of our sample funds from the CRSP Survivor-Bias-Free Mutual Fund database. Using the Morningstar data on fund categories, we restrict the sample to those funds with the following Morningstar categories: US OE Large Blend, US OE Large Growth, US OE Large Value, US OE Mid-Cap Blend, US OE Mid-Cap Growth, US OE Mid-Cap Value, US OE Small Blend, US OE Small Growth, and US OE Small Value. The rationale for this restriction is two-fold. First, by restricting the sample to these categories we are able to create benchmark portfolios that are both broad enough for the weights of these portfolios to be precisely measured and narrow enough to be relevant. Second, using only these broad categories mitigates the concern that any geographic overweighting in our sample may be driven by industry-specific funds, since these funds concentrate their holdings in certain industries and firms within industries tend to cluster geographically. This 4 Approximately, 97% of the value of all holdings remain after this filter. 5

7 filter further reduces the sample to 42,109 quarterly fund observations with 2,192 distinct funds. When constructing the weights in our benchmark portfolios this sample of 42,109 quarterly fund observations is used. Since the focus of our study is geographic location, we collect the following location information. First, we collect data on the managers home states from the Lexis Nexis Online Public Records Database, following the methodology proposed by Yonker (2010), who uses the first five digits of social security numbers to identify the state and year in which CEOs grew up. (See the Appendix for details.) We are able to determine the home state for 2,143 of the 4,236 unique managers in our restricted sample. 5 This reduces the sample to 27,914 quarterly fund observations with 1,810 unique funds. Additionally, using the college names listed in Morningstar, we manually determine the location of the managers colleges. We locate the management company s address from the CRSP Survivor-Bias-Free Mutual Fund database and use fund N-SAR filings from the SEC s website to find the address of each fund s advisers and sub-advisers during our sample period. (See the Appendix for more details.) Finally, we use the Compact Disclosure database to determine the location of the headquarters of each stock held in the mutual fund portfolios. Lastly, we remove mutual funds for which the location of the fund complex headquarters is missing. This final filter produces a sample of 27,416 quarterly fund observations (43,657 quarterly manager-fund observations) with 1,767 unique funds managed by 2,109 unique managers. Panel A of Table I shows the average quarterly composition of the sample by Morningstar fund category. There are 490 funds included in the sample on average each quarter. This represents 69.5% of the the funds covered by our benchmarks and 81% by total net assets (TNA). The largest Morningstar category represented, by both number of funds and TNA, is US OE Large Growth with on average 116 funds in the sample each quarter and with average aggregate TNA of $190 billion. The Morningstar category with the fewest funds is US OE Mid-Cap Value with an average of 18, while US OE Small Value is the smallest category, with an average quarterly aggregrate TNA of $8.3 billion. 5 The percentage of mutual fund managers identified is considerably lower than that in Yonker (2010) because unlike Execucomp, Morningstar does not provide precise data on an individual s age. Adding an age restriction greatly increases the probability of identifying a unique individual. 6

8 In Panel B of the table we report summary statistics for fund- and manager-specific variables, as well as information on the portfolio weights of managers and their benchmarks in stocks located within and outside of managers home states. The average (median) fund in the sample has TNA of $1.16 billion ($0.18 billion), 47% of the funds in the sample are growth funds, 56% are large-cap funds, and 32% employ a subadviser. The median sample fund is managed by two managers and 18% of funds employ more than 3 managers in a given quarter. The median manager is 45 years old, has worked for his current fund for 3.2 years, and lived in his home state for the first 31 years of his life. Approximately 4.5% of the manager-quarter observations are determined to be immigrants to the U.S., using the method explained in the Appendix. The summary statistics on portfolio weights show that mutual fund managers in the sample place greater weight on stocks in their home states than do their benchmarks. Managers allocate 6.89% to their home states, while the benchmark weight is only 6.05%. This overweighting of 83 basis points (bps) suggests that managers overweight their home states by 13.7%, although we formally test for this difference in the next section. The average portfolio weight that managers place in stocks in a state outside their home state is 1.86%. The fact that the benchmark portfolio s weight in the managers home states is larger than the benchmark weight in other states implies that portfolio managers tend to grow up in states with more companies. To see the geographic dispersion of managers and funds more clearly, we count the number of active funds located in a state each quarter during our sample, as well as the number of fund managers who come from each state. We report the time series average of these counts in Figure I. Both the location of funds and the home states of fund managers are correlated with population distribution. As would be expected, many funds are concentrated in New York, California, and Illinois, but funds are also strongly represented in Colorado, Connecticut, Florida, Massachusetts, Minnesota, Texas, and Wisconsin, among others. Fund managers are also likely to be born in these states, but home states are fairly dispersed and at least one manager comes from each state. (In the graph, averages of less than one mean that a state is not represented among active managers during some part of our sample period.) 7

9 III Empirical Analysis A Do fund managers overweight holdings in their home states? If mutual fund managers exhibit a familiariaty bias toward companies in their home states, then we should expect to find that mutual fund managers overweight these companies in their portfolios. Superior information could also lead managers to overweight the holdings in their home states. But superior information could also lead to underweighting if it is negative. We begin our empirical analysis by testing the hypothesis that mutual fund managers overweight their portfolio holdings in companies located in their home states. We do so by estimating various forms of the following regression equation using OLS estimation: w i,s,t = βmgrhmstate i,j,s,t + δmorningstarbmw t i,s,t + Γ Controls i,j,s,t + ɛ i,s,t, (1) where w i,s,t is the portfolio weight fund i allocates to all firms headquartered in state s during quarter t, MgrHmState i,j,s,t is a dummy variable that is one if state s is the home state of manager j of fund i during quarter t, MorningstarBMW t i,s,t is the average portfolio weight in state s of all funds within the same Morningstar category as fund i during quarter t, and Controls i,j,s,t is a vector of relevant control variables. If fund managers tilt their portfolios toward their home states, then we should find that the coefficient estimate on M grhmstate should be significantly greater than zero. In Table II, we report the coefficient estimates and White (1980) heteroscedasticity-consistent standard errors clustered at the fund-manager-level from the OLS estimation of various forms Equation (1). In column (1), we include only MorningstarBMW t to control for the average portfolio weight that funds in the same Morningstar category allocate to a given state during each quarter. 6 The coefficient estimate on MgrHmState is and is significant at greater than the 1-percent level. This estimate indicates that the mutual fund managers in the sample overweight their home states by 87 bps relative to other funds in the same nine-box Morningstar category. 6 We have also performed our tests with the restriction that the coefficient on MorningstarBMW t = 1 and find no qualitative difference in the results. 8

10 Given that the average portfolio weight allocated to a manager s home state is 6.05% in our sample, this implies that managers holdings in their home states are approximately 14.4% greater than similar funds whose managers do not call these states home. Coval and Moskowitz (1999) find that mutual funds holdings are geographically closer to funds advisers than are the holdings of the market portfolio. We test the magnitude of this result in our empirical framework in column (2) by estimating a model analogous to that estimated in column (1), but replacing MgrHmState with a dummy variable that takes a value of one if the mutual fund complex of fund i is headquartered in state s during quarter t (MF HQState). The coefficient estimate on MF HQState is 136 bps and is significant at better than the one percent level. This suggests that funds overweight their headquarter state by 136 bps relative to funds in the same nine-box Morningstar category. If labor markets for mutual fund managers are geographically segmented, then it is possible that funds hire managers close to their headquarters. If this is the case, then the result from column (1) that mutual funds overweight their managers home states could instead be due to the findings of Coval and Moskowitz (1999). To check this, in column (3) we decompose the effects of the fund location and the manager home state by estimating a model that includes both MgrHmState and MF HQState and an interaction between the two. The estimate on M grhmstate is interpreted as the fund s portfolio overweighting in its managers home states when the fund is not headquartered in those states. The coefficient estimate on M grhmstate drops by roughly half from the estimate in equation (1) to 44 bps, but is still significantly positive at better than the one percent level. The estimate also remains economically important, as mutual funds have 7.3 percent larger portfolio weights in their managers home states than do funds with similar strategies that do not have managers from those states. The decomposition in column (3) also provides some insight to the degree to which mutual funds local equity preferences are driven by portfolio managers or by other members of the fund complex. The coefficient on MF HQState falls by almost 20 percent from 136 bps in column (2) to 109 bps in column (6) once we control for the home states of the portfolio managers. Our measure of the mutual fund location, M F HQState, uses the location of the fund management company. The management company location and the adviser location are not always the same. Additionally, some mutual funds use subadvisers to manage the fund. In such funds it is possible 9

11 that the mutual fund management location and the adviser location are completely different. For example, the Vanguard Equity Income Fund belongs to the Vanguard fund complex, which is headquartered in Malvern, PA. This fund is subadvised by Wellington Management Company, headquartered in Boston, MA. In other cases, however, the subadviser is related to the fund complex. For example, the subadviser of many Fidelity funds is Fidelity Management & Research Company (FMR). In this case both the fund complex and the subadviser are based in Boston, MA. If the former case is prevalent, and the mutual fund manager labor market is geographically segmented, then M grhmstate may just proxy for adviser or subadviser locations. To mitigate these concerns, in columns (4) and (5) we control for the states in which the funds advisers are located 7 and in column (6) we estimate the model from column (3) using only those funds that are not subadvised. In each of the specifications the coefficient estimate on M grhmstate remains positive and statistically significant at the one percent level. To further mitigate the concern that the overweighting of funds in firms located in managers home states is due to a mutual fund location effect, in column (7) we confine the sample to quarterly fund-manager-state observations where state s is at least 500 miles from the state in which the fund s complex is headquartered. The coefficient estimate on M grhmstate is 58 bps and is significant at better than the one percent level. In untabulated results, we confirm that this general result holds using fund-manager-state observations that are at least 1000, 1500, or 2000 miles from the fund complex s headquarters. In column (8), we again test whether this result is due to managers home states proxying for the subadviser location by limiting the sample from column (7) to only those funds that are not subadvised. The results confirm all previous results: mutual funds overweight their managers home states. Finally, in columns (9) and (10) we include fund-state-level fixed effects to the model in column (1) in order to mitigate the concern that time-invariant fund-specific variables are driving the result. In these specifications the estimate on M grhmstate is identified from within-fund variation in managers home states generated by turnover in the management team, so this is clearly a very conservative test. In column (9) the estimate on MgrHmState is 14 bps and is significant at better than the one percent level. In column (10) we control for the adviser location since there is variation 7 We determine adviser locations from N-SAR filings, as explained in the Appendix. This reduces our sample by approximately one third. 10

12 in these locations over time. The estimate on M grhmstate is 16 bps and remains statistically significant. The estimates indicate that controlling for time-invariant fund-specific effects cause the estimate of overweighting in home states to drop from 14.4% to between 2.3% and 2.6%. B Fund characteristics One way to disentangle the information hypothesis from the familiarity hypothesis is to investigate which types of funds overweight the most. For instance, if information is driving the portfolio weights we would expect the local bias to be stronger in small-cap and growth funds, since the firms that these mutual funds invest in tend to be more opaque than larger firms or firms with greater asset tangibility. We investigate this hypothesis by testing for differences in the overweighting across the fund types by interacting M grhmstate with dummy variables that indicate the Morningstar styles of the funds: value (V alue), growth (Growth), small-cap (SmallCap), and large-cap (LargeCap). If there are differences in manager home state weightings across styles then we should find that these interaction terms are significantly different from zero. Table III shows the regression results for our tests. The baseline model used in columns (1) through (3) is from column (3) of Table II and controls for the location of the fund management company headquarters. In column (1) we test whether there are differences in manager home state weightings across value, growth, and blend funds. The coefficient estimates on the M grhmstate interactions with V alue and Growth are not statistically different from zero, indicating that there is no difference in the weightings that managers place on their home states across these funds. This finding is inconsistent with the information hypothesis. In column (2), we test for differences across size categories of funds. The estimates indicate that managers of mid-cap funds overweight their home states the most (101 bps), followed by small-cap funds (42 bps), and finally large-cap funds (23 bps). The coefficient estimates on the interactions between M grhmstate and SmallCap and M grhmstate and LargeCap are statistically negative at the five and one percent levels, respectively. The regression in column (2) suggests that large-cap funds do not display a significant overweighting in their managers home states, which is consistent with the information hypothesis. This finding is not inconsistent with familiarity. While all managers are likely to be equally aware 11

13 of the largest firms, it is likely that mutual fund managers are more familiar with the smaller firms from their home states than their peers. Smaller funds are likely to have fewer resources, and thus are more prone to rely on their managers ideas or biases. We find that this is indeed the case. In column (3) we test for differences in the local bias across fund sizes by TNA. The regression shows that mutual funds that are less than the median size by TNA in the sample place 40 bps more weight (significant at the 5 percent level) on their managers home states than do larger mutual funds in the sample. In column (4) of the table we investigate how the structure of the fund management team affects the manager home state overweighting. As we showed in Table I, many funds are run by more than one manager. We hypothesize that when there are many managers managing a fund, that the overweighting toward manager home states will be substantially reduced. In column (2) we include a dummy variable that is equal to one if a fund has more than three managers (ManyMgrs) and the interaction of this variable with M grhmstate. The estimates indicate that funds with more than three managers overweight those managers home states by 34 bps less than those with fewer than four managers (significant at better than the one percent level). These estimates imply that mutual fund managers in funds managed by three or fewer managers overweight their home states by 14.7%. In columns (5) through (8) we see how these test perform using the within fund estimator of M grhmstate. The only fund characteristic that seems to matter is fund size by TNA. Smaller funds in the sample overweight their manager s home states by 19 bps (statistically significant at the 5 percent level), while larger funds in the sample exhibit no manager home bias. C Fund manager characteristics Additional insights can be gained into what drives mutual fund managers to overweight their portfolios in their home states by testing which types of managers are more prone to overweighting. In Table IV we investigate whether differences in manager age, tenure, home state tenure, immigrant status, and college location are associated with the degree to which managers overweight their home states. We do so by interacting various dummy variables with M grhmstate using the 12

14 within-fund estimation specification from column (9) of Table II. If information is driving the portfolio overweighting, then we would expect to find that more experienced managers are driving the observed portfolio overweighting. If however, a bias is driving the result we should find that inexperienced managers overweight their home states more. We test this hypothesis in columns (1) and (2) of the table by interacting M grhmstate with dummy variables indicating managers who are older than the median age in the sample (AgeGT Med) and managers who have greater than the sample median tenure with their current fund (T enuregt M ed). If more experienced managers are driving the local bias in mutual fund portfolio holdings then the coefficient on these interaction terms should be positive. The regression estimates show that although age of the manager does not affect the manager s local portfolio bias, the tenure of the manager does. In fact, only managers who are early in their career exhibit the home bias in their portfolio holdings: the estimate on T enuregt Med is 26 bps and is significant at the one percent level. This finding is consistent with the results of Christoffersen and Sarkissian (2009), who find a positive relation between experience and returns among managers working in financial centers. The length of time a manager spends in his home state should also affect their propensity to overweight home-state companies. Both the information and familiarity hypotheses suggest that the longer a manager lives in his home state the more likely he is to overweight his state: he could have more contacts inside local firms, or could simply be more familiar with the companies of the state. To capture this we first estimate how long managers lived in their home states and then construct a dummy, HmT enuregt Med, that equals one when home tenure is greater than the median. The coefficient estimate on HmT enuregt M ed is significantly estimated at 36 bps, which indicates that managers who spent more time in their home states exhibit home-state equity preferences, while those that lived there for shorter periods of time do not overweight their home states. We next look at whether managers who immigrated to the U.S. have lower home-state biases than U.S.-born managers. We would expect that immigrants would have fewer connections to their home states since they didn t grow up in those states they are simply the first state the person lived after immigrating. Consistent with this explanation, the coefficient estimate on the interaction between M grhmstate and Immigrant is 20 bps, although the estimate is insignificant, which is not surprising as only 4.5% of managers are immigrants. 13

15 In columns (5) and (6) we test whether it is indeed the manager home state that matters for portfolio weighting, or whether it may be the state the manager went to college. In column (5) we include a dummy variable if the manager went to college in the state (M grcollstate), analogous to MgrHmState. We find that the coefficient estimate on this dummy is just 1 bp and statistically indistinguishable from zero. When we interact this dummy variable with M grhmstate in column (6) we find that managers who grew up and went to college in the same state exhibit the strongest home-state overweighting. The coefficient estimate on this interaction is 20 bps and is significant at the 5 percent level. The is consistent with our earlier findings on home state tenure: managers who spent more time in their home states tend to overweight those states more. In Table V we investigate the effect of manager characteristics on a manager s propensity to overweight companies close to the fund management company headquarters. If information drives this portfolio overweighting, as in Coval and Moskowitz (1999, 2001), then we should find that more experienced managers overweight companies in the state of the management company headquarters. We run tests analogous to those in Table IV, where we interact various manager experience measures with M F HQState. Consistent with previous research, we find that more experienced managers overweight companies close the fund complex headquarters more. Column (2) of the table shows that while managers early in their career overweight local companies by 95 bps, managers later in their careers overweight these local firms by an additional 29 bps (signficant at the 10 percent level). The coefficients on the interaction terms in these regressions are opposite to what we found when interacting the same dummies with M grhmstate, suggesting that while the fund complex state overweighting is driven by information, as shown by Coval and Moskowitz (2001), the manager home state overweighting is more likely driven by familiarity. We explore this finding more formally in the next section. We should also expect that managers who spend more time in the state of the fund complex headquarters are more likely to overweight firms geographically close to the fund. In column (3), we find that managers with greater than the sample median mutual fund management state tenure (MF HQStateT enuregt Med) overweight local companies by more than twice the amount that managers who have spent less time in the funds state of headquarters. 14

16 The results in this section show that when managers are early in their careers they overweight home-state companies, but that this overweighting declines as they become experienced, and instead they begin to overweight companies located close to the mutual fund headquarters. D Portfolio performance Our results thus far show that mutual fund managers invest significantly more in stocks that are headquartered in their home states than do managers who manage peer funds but grew up elsewhere. Moreover, the analyses in subsections B and C suggest that this overweighting may be due to familiarity. In this subsection, we formally test the information and familiarity hypotheses using direct, performance-based tests. The two hypotheses offer very different empirical predictions concerning fund performance. If managers overweight securities headquartered in their home states due to a comparative advantage they have in generating information about these companies, the home state portion of their portfolios should outperform. If instead the portfolio tilt is due to a familiarity bias, home company holdings should not deliver superior returns. To test these predictions, we begin by creating portfolios of manager home-state stocks. For each manager i, quarter t, and stock k, we have three important pieces of information: wk,t a (defined above as the actual portfolio weight of fund i in stock k during quarter t), wk,t b (the portfolio weight of fund i s Morningstar benchmark in stock k during the same quarter, as described above), and r k,t+1, a measure of the return on stock k in the quarter immediately following the portfolio reporting date. We use these quantities in a variety of tests to evaluate the performance of the manager s home-state investments. First, we follow Coval and Moskowitz (2001) and calculate home and distant portfolio returns for each fund manager-quarter as follows: P erformance i,h,t = k H ( wi,k,t a ) k H wa i,k,t r k,t+1 (2) 15

17 and P erformance i,d,t = k / H ( wi,k,t a ) k / H wa i,k,t r k,t+1 (3) where the weight on each stock s quarter t + 1 performance in the home and distant portfolios is based on the most recent holdings disclosure (filed at the end of quarter t). We then calculate the weighted average of the returns in (2) and (3) across funds at time t, weighting each fund s return by its total net asset value (size). 8 The aggregation generates 56 quarterly observations for each return series. Following Coval and Moskowitz (2001), we use two different excess return measures for r k : the average monthly raw return in excess of the one-month Treasury bill rate (obtained from Kenneth French s web site), and average monthly benchmark adjusted excess return as in Daniel et al. (1997, DGTW ). The DGTW benchmark adjustment procedure sorts stocks into size quintiles, and within each size quintile stocks are further sorted into book-to-market quintiles. Finally, each book-to-market quintile is divided into five momentum portfolios. The sorting process creates 125 stock characteristics groups, for which benchmark portfolios are formed by calculating the value-weighted average return of the stocks in each category. We calculate the DGTW benchmark returns in two ways: first, we include the CRSP universe of common stocks in the calculation; second we include only the stocks in our sample (i.e., stocks held by any mutual fund during the quarter). Finally, with each of the DGTW benchmark portfolio returns, we calculate the risk-adjusted excess returns by subtracting from each stock k s return the value-weighted average return of stocks with similar size, book-to-market, and momentum characteristics as defined by the triple sort benchmark portfolios. The average home (R H ) and distant (R D ) portfolio returns over the 56 quarters and the difference between these averages are presented in Panel A of Table VI. The home and the distant portfolio returns are not significantly different from each other. This result holds for both the raw and the risk-adjusted measures and implies that home-state portfolios are not informed investments. These findings are in contrast to those of Coval and Moskowitz (2001), who show that fund investments in stocks headquartered close to a fund s headquarters outperform distant portfolios. To gain 8 If a fund has multiple portfolio managers, we first average (2) and (3) across managers within the fund, and then compute the size-weighted average home and distant returns, respectively. 16

18 additional insight on how the two results are related, we perform additional analyses. First, since manager home states may coincide with the fund s location, in Panel B of the table, we exclude those managers from our sample for whom the home and fund headquarter states are the same. The difference between the local and distant portfolio returns is even weaker after this exclusion. In Panel C, we replicate the results of Coval and Moskowitz (2001) using our sample. 9 In this panel, the local and distant portfolios are based on the location of the fund, not on the home state of the manager. To highlight this distinction, we label the local portfolio as R L. The risk-adjusted returns on the portfolios that are local to the fund s headquarters are indeed significantly higher than those on the distant portfolios, consistent with the results reported in Coval and Moskowitz (2001). Finally, Panel D shows that the results of Coval and Moskowitz (2001) are even stronger when we confine the sample to funds that have local managers (managers who grew up in the state where the fund is located). This result supports our argument that the number of potential connection nodes between the fund and the company plays an important role in determining which trades are likely to be informed and which are driven by mere familiarity. A potential concern in our analysis is that some mangers in our sample do not tilt their portfolios toward home-state stocks. Under the information hypothesis, only managers better able to acquire information about home companies concentrate their holdings in their home state. Therefore, it is possible that our approach of averaging home and distant returns across all managers at time t masks the real performance effect of the portfolio tilt. To gain further insight, we sort funds into deciles based on the magnitude of their home state bias. We define home state bias as the fund s actual weight in the manager s home state normalized by the portfolio weight of the fund s Morningstar benchmark during the same time period: where w a i,h,t Bias i,h,t = w a i,h,t MorningstarBMW t i,h,t, (4) is the portfolio weight fund i allocates to all firms headquartered in the manager s home state H during quarter t and MorningstarBMW t i,h,t is the average portfolio weight in state H of all funds within the same Morningstar category as fund i during quarter t. We then determine the quarterly local and distant returns for each decile by averaging the performance measures in (2) 9 We define firms as local if they are headquartered in the state where the fund is located, while in Coval and Moskowitz (2001), a firm is local if it is within 100 kilometers of the funds headquarters. 17

19 and (3) across the decile funds at time t, weighting each fund s return by its total net asset value. The results are reported in Table VII. Panel A uses the full sample. In Panel B, we exclude funds for which the manager s home state and the location of the fund complex are the same. The table shows the average bias as defined in (4), the average weight in the home state, and the average weight that benchmark funds place on home state securities. Across deciles, we find that home and distant portfolio returns are not statistically significantly different from each other, except for the funds with the most extreme bias (decile 10), where home-state stocks underperform distant stocks. These results provide strong evidence against the information hypothesis and support the familiarity bias explanation. They also show that, in terms of economic significance, managers behavioral biases can be substantial. Among the most severely biased portfolios, managers forego up to 41 basis points monthly on the local portion of their portfolio, which comprises about 9 percent of their holdings. To provide additional robustness for our results, we also perform multivariate analyses. In these, we control for the location of the management company. Since in any given stock, a manager may under- or overweight her portfolio compared to her peer funds, our next set of performance metrics use (wi,k,t a wi,k,t)r b k,t+1, (5) where wi,k,t a is the actual portfolio weight of fund i in firm k in quarter t, wb i,k,t is the portfolio weight of fund i s Morningstar benchmark in firm k during the same time period, and r k,t+1 is a measure of the average monthly return of stock k in the quarter immediately following the portfolio reporting date. The measure above is negative, for instance, if relative to her peers, the manager places either a larger weight on a bad performer or a smaller weight on a good performer. As above, we use raw as well as the DGTW benchmark-adjusted returns for the average return in the subsequent quarter. Consistent with our empirical approach in subsection A above, for each fund i, we aggregate (5) in each quarter by state: P erformance i,s,t = k s(w a i,k,t w b i,k,t)r k,t+1. 18

20 Note that our measure sums the weighted returns across all stocks in state s that are held by at least one fund in fund i s Morningstar benchmark category in the given quarter even if the stock is not held by fund i. That is, our measure penalizes funds for choosing not to invest in stocks that exhibit good future performance and rewards funds that avoid stocks that underperform in the next quarter. We then estimate various forms of the regression P erformance i,s,t = βmgrhmstate i,j,s,t +δmorningstarbmw t i,s,t +Γ Controls i,j,s,t +ɛ i,s,t. (6) Under the information hypothesis, we expect the coefficient estimate on the MgrHmState i,j,s,t dummy to be positive and significant; if it is negative or flat, it would provide evidence in favor of the familiarity alternative. The results are reported in Table VIII. Columns (1) and (2) are based on raw and risk adjusted returns, respectively. In addition to the indicator variable for manager home state, the additional explanatory variables reported in Table VIII include an indicator variable for the management company state, an interaction variable between the manager home state and the management company state, and the average portfolio weight in state s of all funds within fund i s Morningstar category during quarter t. We include the later to avoid the possibility of giving a larger role to larger states or more popular stocks for which the weight allocated by funds in our sample and by the benchmark group is generally higher. Our results indicate that managers do not possess information related to their home state. While in column (1), our coefficient estimate is significant and positive, after adjusting for risk, the outperformance disappears. Moreover, the raw return results are not robust to alternative specifications as indicated in column (3) of Table VIII. While an investment in the manager s home state does not deliver abnormal performance, consistent with Coval and Moskowitz (2001), management company location matters. In particular, the subset of a fund s portfolio that are headquartered where the fund s management company is located exhibits higher performance. The management company effect is significant and positive in each specification reported in Table VIII. 19

21 IV Robustness A Subsample analysis In Table IX we perform a number of robustness checks. We showed in Figure I that many funds are headquartered in New York. In column (1) of the table we estimate our baseline within fund regression for only observations where the fund is headquartered outside of the state of New York. The coefficient estimate on M grhmstate is 12 bps and remains statistically significant. We estimate a similar regression in column (2) omitting all observations where the fund is headquartered in the financial centers of California, Illinois, New York, Massachusetts, or Pennsylvania. Again the estimate on M grhmstate is positive and significant. In columns (3) through (5) we omit observations of subadvised funds, states that are less than 500 miles from the mutual fund complex headquarters, and observations for which we were unable to find the home states of the complete management team. In all cases the coefficient estimate on M grhmstate remains positive and significant at the one percent level. Finally, in column (6) we restrict the coefficient estimate on MorningstarBMStateW t to be equal to one. The coefficient estimate on MgrHmState remains positive and significantly estimated at 14 bps. B Portfolio distance An alternative to investigating whether portfolio managers overweight their holdings of firms headquartered in their home states is to test whether the the stocks they hold are geographically closer to their home states than those of a benchmark portfolio. Since we know the state where the manager grew up, we need to choose a location in that state that represents the middle of the state. Rather than using the geographic centroid of each state, we use a population-weighted measure, which takes into account where people live. In other words, we choose the point in each state that minimizes the expected distance to a randomly selected person who lives in that state. These population-weighted centers are calculated by the U.S. Census Bureau using data 10 from the 2000 Census, and are shown in Figure II. 10 Available at 20

22 Figure III plots the excess weight (relative to the equally-weighted benchmark portfolio) for stocks as a function of stock headquarter distance from fund manager home state. To calculate this, we estimate a regression of excess weight on seven dummy variables for distances 0 50, , , , , , and >1000. We calculate clustered standard errors to allow for correlation within each fund-stock. The average excess weight in stocks located within 50 miles is 0.75 bps, and then declines for stocks located further from the manager s home state. For comparison, the average stock weight is 5.5 bps, 11 which implies an overweighting of 13.6%, and is comparable to the estimates reported above. In order to conduct our tests, we compute the manager bias (MB) test statistic, which is analogous to the local bias (LB) test statistic developed by Coval and Moskowitz (1999). Specifically, we define MB = 1 T M T M N (wi,j,t b wi,j,t) a d i,j,t d t=1 i=1 j=1 b, (7) i,t where wi,j,t b is the portfolio weight of fund i s Morningstar benchmark in firm j, wa i,j,t is the actual portfolio weight of fund i in firm j, d i,j,t is the distance (in miles) from the population-weighted center of fund i s managers home states to the centroid of the zip code of the headquarters of firm j, and d b i,t = n j=1 wi,j,t b d i,j,t (value-weighted distance from fund i s benchmark portfolio to its managers home states). In cases where there are more than one manager of the fund, d i,j,t is the average distance to each of the managers home states. To alleviate the concern that our distance measure is skewed by outliers, we remove all funds with managers who grew up in Hawaii or Alaska, as well as any funds that are headquartered in either of these states. This reduces our sample from 27,914 fund-quarter observations to 27,319. The benchmark weights are calculated as an equal-weighted average of the portfolio weights of all funds in the same Morningstar category as fund i, so our MB statistics is equivalent to the Equal-Value version of the LB statistic calculated by Coval and Moskowitz (1999) in Table II of their paper. The results of our portfolio distance tests are displayed in Panel A of Table X. The table shows that the average distance from mutual funds managers home states to the headquarters of the stocks 11 The regression includes all stocks within a fund s investment universe at each quarter, regardless of whether the fund holds the stock, so many of the quarter-fund-stock observations have zero weight. We include all stocks held by any fund in a Morningstar category in each fund s investment universe. 21

23 they own is 1,038 miles, while the average distance from their home states to their benchmark s holdings is 1,529 miles. The MB statistic is percent, which is significant at greater than the one percent level, indicating that the mutual funds in the sample hold firms that are percent closer to the funds managers home states than the holdings of a benchmark portfolio. This bias is larger in magnitude to that documented by Coval and Moskowitz (1999), who find that the Equal-Value LB in their sample of 2,183 funds is percent (Table II of their paper). For over 20 percent of the funds, the manager s home and the fund s headquarters are in the same state. It is therefore possible that the local bias uncovered by Coval and Moskowitz (1999) is driving the MB. To rule this out, in the second row of Panel A we remove all observations for which the manager home state is the same as the state in which the fund is headquartered. When we estimate MB for this sample, we find that MB increases from percent to percent. It does not appear that the mutual fund headquarters location is driving the manager bias. In the third line of Panel A we partition the sample even further by removing all funds that employ subadvisers. We do this to control for the concern that managers homes proxy for subadviser locations. The MB in this sample is and remains highly significant. We conclude that mutual funds hold stocks that are closer to their managers homes than that of benchmark portfolios. In Panel B of the table we conduct the same exercise substituting the mutual fund complex headquarters location for managers home states. Following Coval and Moskowitz (1999), we label the local bias measure LB. In order for our results to be comparable to those in Panel A, we use the population-weighted center of the state in which the fund complex is headquartered as the location for the fund. The average distance from fund holdings to the fund complex headquarters is 1,054 miles, which is approximately fifteen miles further away than the distance from the holdings to the funds managers home states. The average distance to the funds benchmark holdings is 1,081 miles, which is much shorter than the distance to the benchmark portfolio from funds managers home states. This is most likely due to fund complexes locating nearer to companies than the population in general. Of most interest is the LB statistic. We compute this statistic for the same three subsamples as we did for MB and although highly signficant, the magnitude of the LB statistic is much lower than that of the MB statistics. Specifically, the LB statistic for the three samples is 2.41, 1.68, and 2.42 percent, respectively. 22

24 C Alternative geographic classifications In Table XI we estimate a model without controls and the models from columns (1), (3), (6), and (9) of Table II using two alternative geographic classifications. Specifically, in columns (1) through (5) we use U.S. Census Divisions and in columns (6) through (10) we use U.S. Census Regions. 12 Observations are quarterly manager-fund-geographic classification observations. The variables of interest are M grhmdivision and M grhmregion, which are dummy variables defined analogously to M anagerhmstate using Census Divisions and Regions, respectively. The empirical estimates corroborate our finding that mutual fund managers overweight their holdings in firms located close to their home states. Using Census Divisions to classify geographic locations, the estimates in columns (1) and (2) suggest that the average portfolio weight in divisions that managers call home is 15.70% and that funds tend to overweight holdings in these divisions by 134 bps or 8.7% relative to funds that do not have managers from these divisions. When we control for the mutual fund complex headquarter location in column (3), the overweighting falls to 72 bps or 4.6% relative to other mutual funds which do not have managers from these divisions. The results in column (4), which are estimated excluding subadvised mutual funds, show that the main results are not driven by fund managers home divisions proxying for adviser locations. The within-fund estimator on M grhmdivision is substantially lower at 16 bps, but remains significant at the one percent level. In columns (6) through (10) we classify managers homes by Census Regions. The estimates in columns (6) and (7) suggest that the average portfolio weight in regions where managers call home is 26.77% and that funds tend to overweight holdings in these regions by 209 bps or 7.8% relative to other mutual funds which do not have managers from these regions. When we control for the mutual fund complex headquarter location in column (8), the overweighting falls to 117 bps. Once again, the results in column (9), which are estimated excluding subadvised mutual funds, show that the result that mutual funds overweight firms that are located close to their managers are not driven by fund managers home regions proxying for adviser locations. The estimate on M grhmregion in 12 The U.S. Census Bureau partitions states into four regions (Northeast, Midwest, South, and West) and nine divisions (NE New England, NE Mid Atlantic, MW East North Central, MW West North Central, S Atlantic, S East South Central, S West South Central, W Mountain, and W Pacific). 23

25 column (10) shows that their is even enough variation in M grhmregion within funds to conclude that funds overweight their managers home regions (coefficient estimate of 30 bps and significant at the one percent level). V Conclusion Whether familiarity plays a role in investment decisions has been a recurring question since at least Merton (1987). In this paper we investigate the influence of familiarity on the portfolio choices of mutual fund managers. These professional investors are a less likely candidate for behavioral biases; therefore finding evidence of a familiarity bias among them has implications for all other investor groups. To study familiarity in the context of mutual funds, we identify a group of securities with which the managers are familiar but on which they are not likely to have private information. Severing the tie between familiarity and information is difficult however. For instance, managers are likely to be familiar with local companies. However, due to the proximity of these companies to the managers, their private circle, and all other employees of the fund, the opportunity to acquire information about these firms is also very likely. We attempt to separate information and familiarity by identifying the states in which the managers grew up; we then ask whether managers over-weight fund holdings in these states. The advantage of focusing on stocks from manager home states is that these states represent the managers pasts. This reduces the probability that managers have valuable connections to these securities. Moreover manager home states are linked to the manager but not to other employees of the fund, which allows us to close several potential information channels in our analyses. We find that compared to managers of similar funds, but who grew up in different states, managers over-weight companies located in the states in which they grew up by as much as 14.4%. (The most conservative of our tests yields an estimate of 2.3%.) These results hold after controlling for a number of potentially important factors, including where the manager attended college and are robust to several alternative distance definitions. 24

26 Our findings suggest that the overweighting is not the result of information, but rather results from familiarity. Specifically, managers who overweight their home state firms do not deliver superior performance on these holdings. Moreover, we find significant underperformance in home state stocks among the funds with the highest home state bias. In terms of economic significance, managers behavioral biases can be substantial. In the upper extreme, among the most severely biased portfolios, managers forego up to 41 bps per month on the home state portion of their portfolio. These performance results are important for establishing familiarity because, while our identification strategy is designed to distinguish familiarity from information, managers may retain personal connections to their home states through family or friends who still live in the area. Finally, we find that the bias is stronger in managers who are earlier in their careers. As they become more experienced and develop more connections to local firms, the overweighting shifts from home-state companies to companies located where the fund complex is headquartered. The former overweighting does not produce positive abnormal returns, while the latter does. Managers who left their home state earlier in life and perhaps have less of a connection to home have less tendency to overweight home companies. Taken together, our findings suggests that familiarity biases are not limited to individual investors. Although smaller in magnitude, we show that even professional investors are prone to this bias. In addition, our results hint at the channel through which information flows between mutual funds and local firms leading to the superior performance in local holdings documented by Coval and Moskowitz (1999, 2001). Only mutual funds run by local managers make informed trades in local firms, suggesting that strong ties through the community can be valuable for investors. Further research will investigate this information flow more fully. 25

27 References Baik, B., Kang, J.-K., Kim, J.-M., Local institutional investors, information asymmetries, and equity returns. Journal of Financial Economics 97 (1), Bernile, G., Kumar, A., Sulaeman, J., Home away from home: Economic relevance and local investors. Working paper. Bhattacharya, U., Groznik, P., Melting pot or salad bowl: Some evidence from u.s. investments abroad. Journal of Financial Markets 11 (3), Chevalier, J., Ellison, G., Career concerns of mutual fund managers. Quarterly Journal of Economics 114 (2), Christoffersen, S. E. K., Sarkissian, S., City size and fund performance. Journal of Financial Economics 92 (2), Cohen, L., Loyalty-based portfolio choice. Review of Financial Studies 22 (3), Cohen, L., Frazzini, A., Malloy, C., The small world of investing: Board connections and mutual fund returns. Journal of Political Economy 116 (5), Coval, J. D., Moskowitz, T. J., Home bias at home: Local equity preference in domestic portfolios. Journal of Finance 54 (6), Coval, J. D., Moskowitz, T. J., The geography of investment: Informed trading and asset prices. Journal of Political Economy 109 (4), Daniel, K., Grinblatt, M., Titman, S., Wermers, R., Measuring mutual fund performance with characteristic-based benchmarks. Journal of Finance 52 (3), French, K. R., Poterba, J. M., Investor diversification and international equity markets. American Economic Review 81 (2), Greenwood, R., Nagel, S., Inexperienced investors and bubbles. Journal of Financial Economics 93 (2), Grinblatt, M., Keloharju, M., How distance, language, and culture influence stockholdings and trades. Journal of Finance 56 (3), Hong, H., Kostovetsky, L., Red and blue investing: Values and finance. Forthcoming Journal of Financial Economics. Hong, H., Kubik, J. D., Stein, J. C., Thy neighbor s portfolio: Word-of-mouth effects in the holdings and trades of money managers. Hong, H., Kubik, J. D., Stein, J. C., The only game in town: Stock-price consequences of local bias. Journal of Financial Economics 90 (1), Huberman, G., Familiarity breeds investment. Review of Financial Studies 14 (3),

28 Hwang, B., Country-specific sentiment and security prices. Journal of Financial Economics 100 (2), Ivković, Z., Weisbenner, S., Local does as local is: Information content of the geography of individual investors common stock investments. Journal of Finance 60 (1), Kang, J.-K., Stulz, R. A. M., Why is there a home bias? an analysis of foreign portfolio equity ownership in japan. Journal of Financial Economics 46, Korniotis, G. M., Kumar, A., Tall versus short: Height, lifelong experiences, and portfolio choice. Forthcoming, Journal of Finance. Malmendier, U., Nagel, S., Depression babies: Do macroeconomic experiences affect risktaking? Forthcoming Quarterly Journal of Economics. Merton, R. C., A simple model of capital market equilibrium with incomplete information. Journal of Finance 42 (3), Morse, A., Shive, S., Patriotism in your portfolio. Journal of Financial Markets 14 (2), Seasholes, M., Zhu, N., Individual investors and local bias. Journal of Finance 65 (5), Shive, S., An epidemic model of investor behavior. Journal of Financial and Quantitative Analysis 45 (01), Tesar, L. L., Werner, I. M., Home bias and high turnover. Journal of International Money and Finance 14, White, H., A heteroskedasticity-consistent covariance matrix estimator and a direct test for heteroskedasticity. Econometrica 48 (4), Yonker, S. E., Geography and the market for CEOs. Working Paper. 27

29 Appendix Identifying managers home states We identify the home states and ages of fund managers in a series of steps: 1. We begin with a search in the Lexis Nexis Public Records Database for the full name of each fund manager in our sample. The database provides the state in which the Social Security Number (SSN) was issued, the year of issue, and the recipient s month and year of birth. Prior to the Tax Reform Act of 1986, SSNs were typically assigned when a person got their first job or their driver s license usually at the age of 15 or so. We are therefore able to identify the place the person lived at this age, and we assume that this is where the manager grew up. This step uniquely identifies 1,138 fund managers. If this search does not provide a unique result, we continue to step In some cases, the Morningstar data includes the year in which the fund manager graduated from college. We use this to estimate the age of the manager by assuming the manager was between 18 and 24 years old when graduating. (We use this conservative window to avoid misidentifying managers.) This additional age information narrows the matches, allowing us to identify an additional 243 managers in this step. 3. For managers not uniquely identified in steps 1 or 2, we include the additional filter that the manager must have lived in the state where the fund complex (their employer) is based. An additional 547 managers are identified in this step. 4. Finally, it is sometimes possible to determine the home state of a manager even if there is no unique match to a record in the Nexis database. This occurs when all possible matches have the same home state. In this case, we can identify which state the manager is from, but we cannot determine the manager s age. This allows us to identify the home states of an additional 181 managers. Together, these steps allow us to uniquely identify 2,109 of the 4,236 managers in our original sample. (Since SSNs are assigned to immigrants when they immigrate for work, we assume a person who obtained their SSN at an age of 22 or older is an immigrant. This is the case for 109 (5.2%) of our identified managers.) None of the main results in the paper is sensitive to restricting our sample to those matches identified only in Step 1. Identifying fund adviser locations We identify the location of fund advisers from Form N-SAR, which is a required semi-annual SEC filing. The form records the mutual fund complex ( Registrant ) and all advisers and sub-advisers for each fund ( Series ) within the fund family. A fund many have many advisers, and different funds within a family may have different advisers. The N-SAR captures all of this information, as well as the business address of the each adviser. We begin by using mutual fund ticker symbols, available in the CRSP mutual fund data and in certain SEC filings, to create a link to the SEC s Central Index Key (CIK) identifier. We download all N-SAR filings for each fund and extract the fund ( series ) name and adviser zip code for each fund. We then use a name matching algorithm to match funds reported in the N-SARs to our data. When using adviser data, we restrict our sample to those cases where there is no ambiguity in the name matching. This reduces our sample by about 34 percent. Variable definitions MgrHmState i,j,s,t is a dummy variable that is one if state s is the home state of manager j of fund i during quarter t. MgrHmDivision i,j,s,t is a dummy variable that is one if division s is the home division of manager j of fund i during quarter t. 28

30 MgrHmRegion i,j,s,t is a dummy variable that is one if region s is the home region of manager j of fund i during quarter t. MorningstarBMW t i,s,t is the average portfolio weight in geographic classification s (state, division, or region) of all funds within the same nine-box Morningstar category (US OE Large Blend, US OE Large Growth, US OE Large Value, US OE Mid-Cap Blend, US OE Mid-Cap Growth, US OE Mid-Cap Value, US OE Small Blend, US OE Small Growth, or US OE Small Value) as fund i during quarter t. SeniorMgrHmState i,j,s,t is a dummy variable that is one if the home state of senior manager j of fund i during quarter t is state s and is zero otherwise. The manager with the longest tenure at fund i is considered the senior manager. ManyMgrs i,s,t is a dummy variable that is equal to one if fund i has more than three managers during quarter t and is zero otherwise. MF HQState i,s,t is a dummy variable that takes a value of one if the mutual fund management company of fund i is headquartered in state s during quarter t and is zero otherwise. MF HQDivision i,s,t is a dummy variable that takes a value of one if the mutual fund management company of fund i is headquartered in division s during quarter t and is zero otherwise. MF HQRegion i,s,t is a dummy variable that takes a value of one if the mutual fund management company of fund i is headquartered in region s during quarter t and is zero otherwise. MF AdvState i,s,t is a dummy variable that takes a value of one if the mutual fund adviser of fund i is headquartered in state s during quarter t and is zero otherwise. V alue is a dummy variable that is one if the mutual fund is categorized by Morningstar as a value fund. Growth is a dummy variable that is one if the mutual fund is categorized by Morningtar as a growth fund. SmallCap is a dummy variable that is one if the mutual fund is categorized by Morningtar as a small-cap fund. LargeCap is a dummy variable that is one if the mutual fund is categorized by Morningtar as a large-cap fund. MF SizeGT Med is a dummy variable that is one if the assets under management of the mutual fund are less than the median in the sample. Assets under management are calculated from the Thomson s holding data. AgeGT Med is a dummy variable that is one if the age of the manager is greater than the median in the sample. Data on age is hand-collected from the Lexis Nexis online public records database. T enuregt Med is a dummy variable that is one if the tenure of the manager is greater than the median in the sample. Tenure data is calculated from Morningstar. HmT enuregt Med is a dummy variable that is one if the number of years that the manager lived in his home state is greater than the median in the sample. A particular manager s home tenure is equal to his age if the manager s home state matches the state in which the mutual fund is headquartered. If the two states do not match, then if the manager attended college in the same state as his home state, the age at which the manager graduated from his degree program is considered the manager s home tenure. If the manager did not attend college in his home state and does not work for a fund headquarted in his state, then the manager is assumed to have left the state four years prior to obtaining a degree at an institution outside his home state. 29

31 MF HQStateT enuregt Med is a dummy variable that is one if the number of years that the manager lived in his home state is greater than the median in the sample. A particular manager s MF headquarter state tenure is equal to his age if the manager s home state matches the state in which the mutual fund is headquartered (unless the manager is foreign, in which case it is his current age minus the age that he obtained his social security number). If the two states do not match, then if the manager attended college in the same state as his home state, the age at which the manager graduated from his degree program is considered the manager s home tenure. If the manager did not attend college in his home state and does not work for a fund headquarted in his state, then the manager is assumed to have left the state four years prior to obtaining a degree at an institution outside his home state. Immigrant is a dummy variable that is one if the manager is determined to be a U.S. immigrant. A fund manager is considered an immigrant if he obtains his social security number at the age of 22 or older. MgrCollState is a dummy variable that is one if the manager of the fund went to college in the state. Data on the names of managers colleges are from Morningstar, and we manually match college names to states. 30

32 Table I: Summary Statistics Panel A of the table reports the average aggregate total net assets (TNA), the average fund s TNA, the average number of funds in the sample, the average percentage of aggregate TNA of the benchmark covered, and average percentage of benchmark funds covered per quarter by Morningstar category for the sample of 27,416 quarterly fund observations from the first quarter of 1996 through the fourth quarter of Panel B reports summary statistics for fund and manager characteristics for the sample. For fund-specific variables the unit of observation is fund-quarter, for manager-specific variables the unit of observation is fund-manager-quarter, and for portfolio weights the unit of observation is fund-manager-state-quarter. The sample includes 27,416 quarterly fund observations, 43,657 quarterly manager observations, or 2,226,507 fund-manager-quarter observations. Panel A Avg. Pct. Of Avg. Pct. Of Sample avg. Sample avg. Aggregate Benchmark Benchmark aggregate TNA fund TNA Sample avg. TNA Covered Funds Covered per quarter per quarter Funds per per quarter per quarter Morninstar Category ($ s millions) ($ s millions) quarter (%) (%) US OE Large Blend 126,757 1, US OE Large Growth 189,972 1, US OE Large Value 135,820 1, US OE Mid-Cap Blend 13, US OE Mid-Cap Growth 36, US OE Mid-Cap Value 15, US OE Small Blend 19, US OE Small Growth 20, US OE Small Value 8, Total 566,549 1, Panel B Variable Mean Median St. Dev N Total net assets ($ s millions) 1, , ,416 Growth fund dummy (Growth) ,416 Value fund dummy (V alue) ,416 Small-cap fund dummy (SmallCap) ,416 Large-cap fund dummy (LargeCap) ,416 Subadvised dummy (Subadvised) ,416 Number of managers ,416 More than 3 managers (ManyMgrs) ,416 Manager age (Age) ,777 Manager tenure (T enure) ,632 Manager home tenure (HmT enure) ,671 Immigrant manager dummy (Immigrant) ,777 Manager portfolio weight in home state ,632 Benchmark weight in manager home state ,632 Difference in weights in manager home state ,632 Manager weight in states outside manager home ,182,875 Benchmark weight in states outside manager home ,182,875 Difference in weights in states outside manager home ,182,875 31

33 Table II: Do Mutual Fund Managers Overweight Holdings in Their Home States? The table reports the coefficient estimates and White (1980) heteroscedasticity-consistent standard errors clustered at the fund-manager-level from the OLS estimation of various forms of the regression equation w i,s,t = βmgrhmstate i,j,s,t + δmorningstarbmw t i,s,t + Γ Controls i,j,s,t + ɛ i,s,t, where w i,s,t is the portfolio weight fund i allocates to firms headquartered in state s during quarter t, MgrHmState i,j,s,t is a dummy variable that is one if state s is the home state of manager j of fund i during quarter t, MorningstarBMW t i,s,t is the average portfolio weight in state s of all funds within the same Morningstar category as fund i during quarter t, and Controls i,j,s,t is a vector of relevant control variables. If mutual fund managers overweight their portfolios toward their home states, then we should find that the coefficient estimate on MgrHmState i,j,s,t should be significantly greater than zero. The sample includes 2,226,507 quarterly manager-fund-state observations from the first quarter of 1996 to the fourth quarter of 2009 and includes 1,767 unique funds managed by 2,109 unique managers. MF HQState i,s,t is a dummy variable that takes a value of one if the mutual fund complex of fund i is headquartered in state s during quarter t and is zero otherwise. MF AdvState i,s,t is a dummy variable that takes a value of one if the adviser of fund i is headquartered in state s during quarter t and is zero otherwise. Column (6) includes only observations where fund i during quarter t does not employ a subadviser. In column (7), the sample is limited to observations where state s is at least 500 miles from the state of fund i s fund complex headquarters. In column (8), the sample is limited to the intersection of the samples used in the estimation in columns (6) and (7). Columns (9) and (10) include fund-state-level fixed effects. Significance levels are denoted by *, **, ***, which correspond to 10, 5, and 1 percent levels, respectively. MF HQ MF HQ 500 mi. & Sample: Full Full Full Full Full No Subadv 500 mi. No Subadv Full Full (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) 32 MgrHmState *** *** *** *** *** *** *** *** *** (0.0010) (0.0008) (0.0009) (0.0009) (0.0010) (0.0012) (0.0016) (0.0003) (0.0005) MF HQState *** *** *** *** (0.0012) (0.0012) (0.0012) (0.0018) MgrHmState MFHQState * (0.0029) (0.0035) (0.0037) MF AdvState *** *** (0.0013) (0.0012) (0.0010) MgrHmState MFAdvState ** (0.0028) (0.0043) (0.0012) MgrHmState MFHQState MFAdvState (0.0053) MorningstarBMStateW t *** *** *** *** *** *** *** *** *** *** (0.0045) (0.0045) (0.0046) (0.0057) (0.0057) (0.0061) (0.0055) (0.0073) (0.0162) (0.0199) Intercept *** *** (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) (0.0003) (0.0004) AdjR N 2,226,507 2,226,507 2,226,507 1,461,456 1,461,456 1,440,954 1,678,557 1,091,969 2,226,507 1,461,456 Fixed Effects No No No No No No No No fund-state fund-state

34 Table III: Fund Characteristics The table reports the coefficient estimates and White (1980) heteroscedasticity-consistent standard errors clustered at the fund-manager-level from the OLS estimation of the regression equation estimated in Table II including interaction terms with various fund characteristics. Observations are quarterly fund-manager-state observations. V alue is a dummy variable that is one if the mutual fund is categorized by Morningstar as a value fund. Growth is a dummy variable that is one if the mutual fund is categorized by Morningtar as a growth fund. SmallCap is a dummy variable that is one if the mutual fund is categorized by Morningtar as a small-cap fund. LargeCap is a dummy variable that is one if the mutual fund is categorized by Morningtar as a large-cap fund. MF SizeLT Med is a dummy variable that is one if the assets under management of the mutual fund are less than the median in the sample. ManyMgrs is a dummy variable that is equal to one if the fund has more than three managers during the quarter and is zero otherwise. All specifications include a constant, and the levels for the interaction variable being tested when appropriate. The specifications fund-state-level fixed effects, Coefficient estimates and standard errors for these independent variables are not reported. Significance levels are denoted by *, **, ***, which correspond to 10, 5, and 1 percent levels, respectively. (1) (2) (3) (4) (5) (6) (7) (8) MgrHmState * *** ** *** ** *** *** (0.0021) (0.0023) (0.0010) (0.0010) (0.0006) (0.0007) (0.0004) (0.0006) MgrHmState V alue (0.0029) (0.0008) 33 MgrHmState Growth (0.0028) (0.0008) MgrHmState SmallCap ** (0.0027) (0.0009) MgrHmState LargeCap *** (0.0028) (0.0009) MgrHmState MF SizeLTMed ** ** (0.0016) (0.0008) MgrHmState ManyMgrs * (0.0017) (0.0007) MF HQState *** *** *** *** (0.0012) (0.0012) (0.0012) (0.0012) MgrHmState MF HQState * * * * (0.0029) (0.0029) (0.0029) (0.0029) MorningstarBMStateW t *** *** *** *** *** *** *** *** (0.0046) (0.0046) (0.0046) (0.0046) (0.0162) (0.0162) (0.0162) (0.0162) AdjR N 2,226,507 2,226,507 2,226,507 2,226,507 2,226,507 2,226,507 2,226,507 2,226,507 Fixed Effects No No No No fund-state fund-state fund-state fund-state

35 Table IV: Fund Manager Characteristics The table reports the coefficient estimates and White (1980) heteroscedasticity-consistent standard errors clustered at the fund-managerlevel from the OLS estimation of the regression equation estimated in Table II including interaction terms with various fund manager characteristics. Observations are quarterly fund-manager-state observations. AgeGT Med is a dummy variable that is one if the age of the manager is greater than the median in the sample. T enuregt Med is a dummy variable that is one if the tenure of the manager is greater than the median in the sample. HmT enuregt Med is a dummy variable that is one if the number of years that the manager lived in his home state is greater than the median in the sample (see the appendix for a description of how this is calculated). Immigrant is a dummy variabe that is one if the manager is determined to be a U.S. immigrant (see the appendix for a description of this calculation). MgrCollState is a dummy variable that is one if the manager went to college in the state. All specifications include fund-state-level fixed effects, a constant, and the levels for the interaction variable being tested. Coefficient estimates and standard errors for these independent variables are not reported. Significance levels are denoted by *, **, ***, which correspond to 10, 5, and 1 percent levels, respectively. (1) (2) (3) (4) (5) (6) MgrHmState *** *** *** (0.0005) (0.0005) (0.0008) (0.0004) (0.0005) MgrHmState AgeGT Med (0.0007) MgrHmState T enuregt Med *** (0.0007) MgrHmState HmT enuregt Med * (0.0019) MgrHmState Immigrant (0.0019) MgrCollState (0.0004) (0.0005) MgrHmState MgrCollState ** (0.0009) MorningstarBMW t *** *** *** *** *** *** (0.0169) (0.0162) (0.0255) (0.0169) (0.0211) (0.0211) AdjR N 2,079,627 2,226,507 1,054,221 2,079,627 1,433,457 1,433,457 Fixed Effects fund-state fund-state fund-state fund-state fund-state fund-state 34

36 Table V: Manager Age, Tenure, and Local Company Overweighting The table reports results from the regression equation estimated in Table II including interaction terms for various fund manager characteristics with M F HQState. Observations are quarterly fund-manager-state observations. AgeGT Med is a dummy variable that is one if the age of the manager is greater than the median in the sample. T enuregt Med is a dummy variable that is one if the tenure of the manager is greater than the median in the sample. MF HQStateT enuregt Med is a dummy variable that is one if the number of years that the manager lived in the state of the fund s complex headquarters is greater than the median in the sample (see the Appendix for a description of how this is calculated). All specifications include a constant and the main effect for the interaction variable being tested, but these are suppressed for brevity. We report White (1980) heteroscedasticity-consistent standard errors clustered at the fund-manager-level. Significance levels are denoted by *, **, ***, which correspond to 10, 5, and 1 percent levels, respectively. (1) (2) (3) MF HQState *** *** *** (0.0015) (0.0013) (0.0013) MF HQState AgeGT Med (0.0022) MF HQState T enuregt Med * (0.0017) MF HQState MF HQStateT enuregt Med ** (0.0023) MgrHmState *** *** *** (0.0008) (0.0008) (0.0008) M grhm State M F HQState * (0.0032) (0.0029) (0.0038) M orningstarbm StateW t *** *** *** (0.0048) (0.0046) (0.0049) AdjR N 2,079,627 2,226,507 1,972,119 Fixed Effects No No No 35

37 Table VI: Home State Performance The table reports returns for home state (H) and distant (D) holdings of fund managers. We use the performance measures ( ) and P erformance i,h,t = k H P erformance i,d,t = k / H ( wi,k,t a k H wa i,k,t wi,k,t a k / H wa i,k,t ) r k,t+1 r k,t+1 to calculate the home and distant returns each quarter by averaging across funds in quarter t, weighting each fund s return by its total net asset value. Panel A uses the full sample of data while Panel B reports returns using only the sample of funds where the manager s home state is not equal to the state in which the mutual fund complex is headquartered. In Panels C and D we replace the home-state return with a local return (R L ), using the location of the fund management company, to replicate the results of Coval and Moskowitz (2001). Panel C uses the full sample of data while Panel D reports returns using only the sample of funds where the manager s home state matches the state in which the fund complex is headquartered. Panel A R H R D R H R D p-value Raw DGTW (all stocks) DGTW (sample stocks) Panel B R H R D R H R D p-value Raw DGTW (all stocks) DGTW (sample stocks) Panel C R L R D R L R D p-value Raw DGTW (all stocks) DGTW (sample stocks) Panel D R L R D R L R D p-value raw DGTW (all stocks) DGTW (sample stocks)

38 Table VII: Manager Home State Performance by Home State Bias Deciles The table reports returns for local (R H ) and distant (R D ) holdings for deciles of the home state bias. We define home state bias as the fund s actual weight in the manager s home state s normalized by the portfolio weight of fund i s Morningstar benchmark in state s during the same time period. We then determine the quarterly local and distant returns for each decile by averaging the performance measures P erformance i,l,t and P erformance i,d,t defined in Table VI, across the decile funds at time t, weighting each fund s return by its total net asset value. In Panel A we report statistics using the full sample of funds. In Panel B, we report statistics by decile for the sample of funds where the manager s home state is not equal to the state in which the mutual fund complex is headquartered. Panel A Decile w a,l w b,l Avg. % bias R H R D R H R D p-value Panel B Decile w a,l w b,l Avg. % bias R H R D R H R D p-value

39 Table VIII: Portfolio Performance The table reports the coefficient estimates and White (1980) heteroscedasticity-consistent standard errors clustered at the fund-manager-level from the OLS estimation of the regression equation P erformance i,s,t = βmgrhmstate i,j,s,t + δmorningstarbmw t i,s,t + Γ Controls i,j,s,t + ɛ i,s,t, where MgrHmState i,j,s,t is a dummy variable that is one if state s is the home state of manager j of fund i during quarter t, MorningstarBMW t i,s,t is the average portfolio weight in state s of all funds within the same Morningstar category as fund i during quarter t, and Controls i,j,s,t is a vector of relevant control variables. Significance levels are denoted by *, **, ***, which correspond to 10, 5, and 1 percent levels, respectively. Raw DTGW Return Adj. Return (1) (2) M grhmstate ** (0.0018) (0.0013) MF HQState *** *** (0.0020) (0.0013) M grhmstate M F HQState (0.0046) (0.0030) M orningstarbm StateW t *** ** (0.0146) (0.0098) Intercept (0.0002) (0.0001) AdjR N 2,225,997 2,225,997 38

40 Table IX: Subsample Analysis The table reports the coefficient estimates and White (1980) heteroscedasticity-consistent standard errors clustered at the fund-level from the OLS estimation of the regression equation estimated in Table II for various subsamples. Observations are quarterly fund-state observations. The estimation in column (1) uses only observations where the mutual fund is headquartered outside of the state of New York. The estimation in column (2) uses only observations where the mutual fund is headquartered outside of the states of California, Illinois, New York, Massachusetts, and Pennsylvania. In column (3) the estimation uses only observations where the mutual fund does not employ a subadviser. In column (4) only observations where the state is at least 500 miles from the mutual fund s headquarters are used in the estimation. In column (5) only observations where MgrHmState was found for the complete management team were used in the estimatation. In column (6) the coefficient on MorningstarBMStateW t is constrained to be equal to one. All specifications include fund-state-level fixed effects. Significance levels are denoted by *, **, ***, which correspond to 10, 5, and 1 percent levels, respectively. No CA, IL, No NY, MA, or MF HQ Complete Sample: NY Funds PA Funds No Subadv 500 mi. MgrHmState Full (1) (2) (3) (4) (5) (6) MgrHmState *** *** *** *** *** *** (0.0004) (0.0005) (0.0004) (0.0005) (0.0007) (0.0003) MorningstarBMStateW t *** *** *** *** *** (0.0173) (0.0223) (0.0195) (0.0180) (0.0245) Intercept *** *** *** *** *** *** (0.0003) (0.0004) (0.0004) (0.0003) (0.0005) (0.0000) AdjR N 1,893,732 1,064,523 1,440,954 1,678,557 1,000,977 2,226,507 Fixed Effects fund-state fund-state fund-state fund-state fund-state fund-state 39

41 Table X: Portfolio Distance This table reports test statistics and significance levels for a manager bias (MB in Panel A) and a local bias (LB in Panel B) in mutual fund holdings for the sample of 27,319 (21,080 when excluding funds where the manager home state matches the state in which the fund complex is headquartered, and 13,293 when removing subadvised funds from the excluded match sample) fund-quarter observations from the first quarter of 1996 to the fourth quarter of 2009 (Column 4 of the table). The bias calculations follow Coval and Moskowitz (1999) as the average difference in distance between the benchmark holdings from the manager home (fund) and the actual holdings from the manager home (fund) as a percentage of the distance of the benchmark holdings from the manager home (fund). Additionally the average components of these test statistics are reported in columns (1) through (3), which include the average value-weighted distances (in miles) of mutual funds holdings and benchmark holdings from funds managers home states (Panel A) and funds fund complexes headquarters (Panel B) and the difference between the two. In the calculation of distance, the firm location of each firm is the population-weighted center of the zip code of the firm s headquarters. For both mutual fund complex headquarters locations and mutual fund managers home states, the population-weighted center of the state is used in the calculation of distance. Population centers are calculated using population data from the 2000 U.S. Census. Benchmark portfolio weights are calculated using the sample of 42,109 quarterly fund observations detailed in Section III of the paper (excluding those fund headquartered in Hawaii or Alaska). The portfolio weights for each zip code for the benchmark portfolio are calculated as the equal-weighted average value-weight of all funds within the same nine-box Morningstar category as the fund being benchmarked. Standard errors used in constructing t-statistics are clustered at the fund-level. Significance levels are denoted by *, **, ***, which correspond to 10, 5, and 1 percent levels, respectively. Panel A Avg. Distance From Managers Homes Pct. Bias Holdings Benchmark Difference MB N Full Sample 1, , *** 27,319 Exclude Manger Home = MF HQ 1, , *** 21,080 Exclude Manger Home = MF HQ & Subadvised 1, , *** 13,293 Panel B Avg. Distance From Fund Complex HQ Pct. Bias Holdings Benchmark Difference LB N Full Sample 1, , *** 27,319 Exclude Manger Home = MF HQ 1, , *** 21,080 Exclude Manger Home = MF HQ & Subadvised 1, , *** 13,293 40

42 Table XI: Alternative Geographic Classifications The table reports the coefficient estimates and White (1980) heteroscedasticity-consistent standard errors clustered at the fund-manager-level from the OLS estimation of the regression equation estimated in Table II using alternative geographic classifications. Observations are quarterly fund-geographic classification observations. In columns (1) through (4) the geographic classification is U.S. Census Division and in columns (5) through (8) we use U.S. Census Regions. The U.S. Census partitions states into 4 different regions (Northeast, Midwest, South, and West) and nine different divisions (NE New England, NE Mid Atlantic, MW East North Central, MW West North Central, S Atlantic, S East South Central, S West South Central, W Mountain, and W Pacific). MgrHmDivision i,j,s,t (MgrHmRegion i,j,s,t) is a dummy variable that is one if division (region) s is the home division (region) manager j of fund i during quarter t. MorningstarBMW t i,s,t is the average portfolio weight in division (region) s of all funds within the same Morningstar category as fund i during quarter t. MF HQDivision i,s,t (MF HQRegion i,s,t) is a dummy variable that takes a value of one if the mutual fund complex of fund i is headquartered in division (region) s during quarter t and is zero otherwise. Columns (4) and (9) includes only observations where fund i during quarter t does not employ a subadviser. Columns (5) and (10) includes fund-state-level fixed effects. Significance levels are denoted by *, **, ***, which correspond to 10, 5, and 1 percent levels, respectively. Census Divisions Census Regions No No Sample: Full Full Full Subadv Full Full Full Full Subadv Full (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) MgrHmDivision *** *** *** *** *** (0.0022) (0.0015) (0.0014) (0.0019) (0.0006) 41 MF HQDivision *** *** (0.0017) (0.0026) MgrHmDivision MFHQDivision (0.0030) (0.0041) MgrHmRegion *** *** *** *** *** (0.0028) (0.0021) (0.0022) (0.0029) (0.0010) MF HQRegion *** *** (0.0023) (0.0036) MgrHmReg MF HQRegion (0.0031) (0.0044) MorningstarBMW t *** *** *** *** *** *** *** *** (0.0061) (0.0062) (0.0084) (0.0167) (0.0126) (0.0125) (0.0168) (0.0198) Intercept *** *** *** *** (0.0002) (0.0006) (0.0006) (0.0009) (0.0019) (0.0007) (0.0032) (0.0031) (0.0043) (0.0050) AdjR N 392, , , , , , , , , ,528 Fixed Effects No No No No fund-state No No No No fund-state

43 Figure I: Fund Locations and Manager Home States This figure shows the average number of funds (per quarter) located in each state, and the average number of fund managers who grew up in each state. Data are calculated each quarter for each state, and then averaged across time. Alaska and Hawaii are displayed below the lower 48 states. An average of less than one means that a state is not represented among active managers or active funds during some part of our sample period. 42

44 Figure II: Population-Weighted State Centers Stars denote the population-weighted center of each state, as calculated by the U.S. Census Bureau using data from the 2000 Census. Gray circles denote cities with a population of 50,000 or more. 43