Employment Networks and Mutual Fund Performance. A thesis presented. Michael Buckley. Applied Mathematics

Transcription

1 Employment Networks and Mutual Fund Performance A thesis presented by Michael Buckley To Applied Mathematics in partial fulfillment of the honors requirements for the degree of Bachelor of Arts Harvard College Cambridge, Massachusetts April 1, 2010

2 ABSTRACT This paper explores the role of social networks in the portfolio allocation decisions and performance of mutual fund managers. Using a novel dataset, I examine the past employment networks of both mutual fund managers and firm executives. I find that fund managers place larger bets and perform significantly better on firms when they are connected to a senior executive of that firm through overlapping employment networks. A portfolio of connected stocks outperforms a portfolio of unconnected stocks by over 10% per year. The results suggest that fund managers may be benefiting from improved information secured as a result of their networks. *I would like to thank my adviser Lauren Cohen, along with John Beshears, Richard Townsend, Jeremy Hoon, Zach Frankel, and an anonymous private wealth management adviser.

3 As social websites such as Facebook, LinkedIn, and Twitter have gained popularity over the last few years, the relevance of social networks in our lives has become increasingly apparent. It has also become clearer that these social networks are not passive, but can in fact influence our lives enormously. Indeed, anecdotal evidence for the effects of social connections is abundant. For instance, a recent article in the Washington Post references the role of the old-boy network in Washington. 1 In this paper, I attempt to empirically document social networks and their effects. In particular, I examine a specific network created through employment experiences. I define the network as follows: when an individual joins a firm, he or she is permanently connected to this firm and other individuals who worked for this firm. Over an individual s entire career, he or she may be connected to many different firms and thus many different groups of people. I refer to this entire network of connections as the individual s employment network. I examine employment networks in a specific setting: the mutual fund space. In making investment decisions, mutual fund managers rely on information and research. Superior performance can result from the ability to accurately interpret new information before it is fully disseminated into the market. 2 Therefore, advantages in acquiring information should give managers the opportunity to act before the information is fully reflected in prices. If a fund manager is connected to a firm executive through his or her employment network, we might expect the manager to have an advantage in gathering information about the firm. I test this hypothesis by developing a dataset on the employment backgrounds for a number of mutual fund managers and firm executives. 1 The Washington Post, March 28, 2010, Why did health-care reform pass? Nancy Pelosi was in charge. 2 Fama (1965) points out that investors who can interpret new information quickly are precisely those who make markets more efficient. 1

4 Using portfolio holdings data, I find that fund managers place larger bets and perform significantly better on firms when they are connected to a senior executive of that firm through employment networks. A portfolio formed by buying connected holdings and shorting unconnected holdings yields over 10% per year. Additionally, I find that a portfolio formed by buying connected holdings and shorting unheld connected stocks yields a similar return of over 10% per year. The paper proceeds as follows: Section I discusses relevant research related to the role of social networks in financial markets. Section II outlines the sources and uses of data. In Section III, I discuss the methodology and results for pooled OLS regressions in order to determine any differences in portfolio weighting of connected and unconnected holdings. In Section IV, I assess whether connected and unconnected holdings differ by performance. In Section V, I outline several additional tests for robustness. Section VI includes a small discussion on two additional types of connections: gender and ethnicity. Finally, Section VII concludes. I. Literature Review There has been a recent influx of literature dealing with the role of social networks in finance. In general, researchers have succeeded in identifying networks and documenting their effects. For instance, Kuhnen (2009) finds that connections between fund directors and advisory firms, developed through past business relations, lead to mutual preferential hiring. Similarly, Hwang and Kim (2009) analyze social connections based on shared education experiences and other observable characteristics. They find that connections between CEOs and directors impact executive compensation, CEO turnover, and earnings management, illustrating that social ties can interfere with a 2

5 director s role in a similar manner as financial and familial ties. In the private equity universe, Rider (2009) finds that prior education and employment affiliations between investors of different firms increases the likelihood that the firms will engage in coinvestments in the future. Although network sociology has clearly been applied to the financial markets in some form, relatively few studies have attempted to merge these concepts with portfolio decisions and investment performance. In a study of venture capital firms, Hochberg, Ljungqvist, and Lu (2007) find that well-connected firms enjoy significantly better returns than their peers, where network connections are defined by relationships formed through prior syndicate investments. More similar to this paper, Cohen, Frazzini, and Malloy (2008) find that mutual fund managers place larger bets and perform better on firms when they are connected to a board member through a common education experience. They also find similar results in the performance of sell-side equity analysts who were connected to executives through education networks (Cohen, Frazzini and Malloy 2009). In a similar study, Tang (2009) finds that fund managers who previously worked as sell-side analysts place larger bets and perform better on stocks which they covered as analysts. I attempt to add to this literature by identifying a new channel through which social connections between fund managers and firm executives may be formed: prior employment. II. Data This study required a significant amount of data collected from a number of sources. First, I constructed a sample of 206 mutual funds using Morningstar, beginning with the universe of Morningstar funds from 2003 to I restricted the universe of 3

6 funds to those that were actively managed and possessed at least 80% of their holdings in domestic equities. Additionally, I immediately excluded any funds in which there was no biographical information available for any managers over the time period. I included both single-manager and multi-manager funds, for a total of 429 unique managers. For the fund managers, I collected employment information from biographies provided on Morningstar s monthly discs. I supplemented the Morningstar biographies with data collected from Zoominfo.com, 3 specifically the untabulated data in its original source form. I matched the managers from Morningstar to Zoominfo.com by using the biographies and full names. When relevant, I also included biographical data from the official websites of mutual funds and from manager data circulated to a private wealth management adviser. The additional sources were necessary since manager biographies often did not list past employment or only listed recent employment in the financial services sector. Next, I collected a variety of information pertinent to firms. I gathered mutual fund holdings data from CRSP, which was available either quarterly or semi-annually. I matched the Morningstar funds to CRSP primarily by using the fund names and tickers. I then assembled a list of 5594 firms which were held at any point by any mutual fund in the sample. For these firms, I obtained monthly stock return data from CRSP and balance sheet items from Compustat. From the latter, I specifically collected market value of equity and the items necessary to calculate book equity. 4 Next, I obtained a list of firm executives from Execucomp. Because of this, my sample consists of the executives for all firms that were part of the S&P1500 at any point between 2003 and 2008 and which 3 Zoominfo.com is a search engine which provides biographical data that is aggregated from other web sources. 4 Book equity was calculated using the method outlined by Kenneth French here: 4

7 were also held by one of the mutual funds in the sample at any point during the time period. This step reduced my sample of firms from a total of 5,594 to Next, I executed a variety of string searches for the chief executive officer, chief financial officer, and chairman of the board for each firm, which ultimately left 7,124 executives. The string searches often yielded executives who possessed one of these titles for a subsidiary company, as opposed to the main firm traded on a stock exchange. Regardless, all of the executives in the sample are senior executives who possess the ability to convey important information to mutual fund managers. Additionally, I collected executive biographical information from Forbes.com. 5 Although not explicitly stated, it is likely that only executives of larger firms have biographies on the website, and so the resulting sample is probably slightly skewed towards larger firms. I matched executives between Forbes and Execucomp using the full name, the current employer, and the biography when there was any ambiguity. From an original sample of 7124 executives, I was left with 4689 which were available on Forbes.com. This step also reduced the number of firms in my sample from 1914 to From the biographical information, I constructed an aggregated list of all firms at which any individual ever worked. I did not include board memberships in this list, so that any findings would be completely orthogonal to the research which has been done on connections through board memberships. Using this aggregated list of all past employers, I employed Capital IQ and executed web searches to trace the corporate structures of each firm. This was necessary in order to determine a firm s parent corporation, but was often difficult since many firms in the list were private. To illustrate the process, consider if someone once worked for a subsidiary of IBM. The fact that the firm was a subsidiary 5 People.forbes.com contains a searchable database of executives by name or by ticker. 5

8 of IBM was not always included in the biography. Thus, I used Capital IQ and web searches to determine if a firm has a parent corporation, which in this case is IBM. I then considered this firm to be IBM in the aggregated list. Additionally, I had to account for historical corporate events such as mergers and acquisitions. For instance, two firms that once were completely different may have become the same firm through a merger or acquisition. In the aggregated list, I considered two firms to be the same firm if a merger or acquisition occurred before the beginning of my sample period, or If a mutual fund manager and firm executive are only connected through a past employer because of a merger immediately before 2003, at the very least they may be aware of this merger and thus the connection. On the other hand, this connection is likely not nearly as strong as one developed by two individuals who worked for the same firm at the same time, or even different firms that merged closer to the individuals employment dates at the firms. Thus, the methods I used to sort through corporate structures may increase the number of connections at the possible expense of the quality of some connections. In light of the relatively low number of connections, 6 these methods seem justified and unavoidable. In order to avoid the issue of corporate structures, one could possibly construct time series employment data on individuals and executives where the exact dates of employment were included. This is resource intensive and outside the scope of the analysis, however, but could be of further interest. Next, I determined each firm s fiscal year from Compustat and used this in conjunction with the executive employment data, which had unspecific employment dates, such as Whenever there were two executives with the same title and the same year, I assumed that the more recent executive took over the position midway through the fiscal year. 6 On average roughly 1.54% of fund assets are connected stocks. See Table III. 6

9 Ultimately, I restricted my sample to after January 2003 because it became more difficult to collect biographical data on managers farther in the past. Moreover, I did not include 2009 in the analysis because the executive data from Execucomp was not as complete at the time of compilation. Since the first mutual fund holdings report date in my sample is not until March 2003, the final sample of mutual funds and firms lies between March 2003 and December Table I gives the final sample sizes along with additional summary statistics, such as the past employers with the most connections. The data reveal that past employers of mutual fund managers are more concentrated than past employers of firm executives. For instance, PricewaterhouseCoopers is the most connected past employer of firm executives, but is connected to only 4.12% of the firms in my sample on average each year. In contrast, Bank of America, the most connected past employer of fund managers, is connected to nearly 11% of funds on average each year. This makes sense given that fund managers tend to originate in financial services. Furthermore, for the main analysis, I use a filtered sample of funds which includes only those in the top fifty percentile of total connections within fund holdings. This removes a number of funds which have an extremely low number of connections. In fact, 63 of the original 206 funds possessed zero connections within their holdings over the entire time period. A low number of connections can result from the fact that certain managers may simply have a limited potential to form connections through their employment networks. For instance, many Fidelity managers began their careers at Fidelity as analysts, and so they have few if any additional firms in their employment backgrounds. Since funds were chosen prior to viewing the employment background in biographies, this was an unavoidable consequence. On the other hand, using the filtered 7

10 sample helps attenuate a measurement issue with regard to the data collection. When a manager possesses a large number of connections in the sample, it indicates that the manager s employment network is likely more accurate and complete. For many fund managers, biographical data was difficult to find, and so it is likely that biographical data is often incomplete. Thus, when a manager possesses a very low number of connections, it could indicate that I have not been able to completely discern the manager s employment network, which could lead some connected stocks to be misclassified as unconnected stocks, attenuating any holdings or performance differences. Additionally, if a fund possesses more connections in holdings, then the managers may be more cognizant of the connections and thus may be more likely to utilize them. If a fund manager has only a few fund-firm-quarter connections in their holdings over the six year period, they may not even be aware of them, given that a fund holds almost one hundred stocks each quarter. By removing observations such as these, I remove observations that may be false positives from the point of view of the manager, a priori any results. It would have been possible to restrict the sample based on total connections, as opposed to connections only within holdings. However, simply because a fund has connections to firms in the entire firm universe does not mean this fund has the capability to invest in these firms. Therefore, I attempt to restrict the sample to funds that are connected to firms and also that have the ability to invest in them. For all of these reasons discussed, I believe this filtered sample to be the most appropriate for analysis. Nonetheless, I include brief results for a few filtered and unfiltered samples in Section V. 8

11 III. Results of Portfolio Holdings First, I explore whether or not portfolio managers are influenced in their holdings decisions by their employment networks. Specifically, I seek to compare the portfolio weights of connected stocks with the portfolio weights of unconnected stocks, in order to determine if either group is treated differently by managers. I define a binary variable CONNECTED for every fund-firm-quarter observation in fund holdings. In a given quarter, a fund is connected to a firm if any of the fund managers have overlapping employment networks with any of the senior executives at the firm. More simply, if any of the fund managers worked at a firm where one of the senior executives also worked, then I define this to be a connection. In a variety of pooled OLS regression specifications, I consistently find that managers place larger bets on firms to which they are connected; these results are both statistically and economically significant. With controls, I find that funds invest on average an additional 24 basis points in connected stocks compared to unconnected stocks. In each regression, the units of observation are fund-firm-quarter. The dependent variable is the percent of total net equity assets of the security within the fund portfolio in a given quarter. The independent variable is the binary variable CONNECTED. In different regression specifications, I include different control variables. Specifically, I experiment with combinations of the momentum factor (Carhart (1997)) and the three Fama - French factors (1993). Using these controls is motivated by the fact that connectedness may in fact be correlated with observable factors. Indeed, when size is considered, the magnitude of the effect drops significantly from roughly 35 basis points to 23 basis points. This implies that funds may overweight connected stocks simply because connected firms tend to be larger than the average firm. One explanation is that 9

12 there may be more connections at larger firms, since the employment numbers are so much larger. Additionally, this effect is likely an unintended consequence of the fact that employment data was easier to find for larger firms, given that it originated from Forbes.com. Furthermore, I also include quarter fixed effects in most specifications and experiment with additional fixed effects such as firm, firm-quarter, and industry code, the latter of which is based on the Fama-French (1997) 49-industry codes. In each regression I cluster standard errors by quarter. Table II illustrates that the additional control variables have little effect on the results of the regression. Indeed, all of the results tell a similar story: managers place larger bets on stocks within their employment network. Across my sample managers invest on average 121 basis points of their portfolio in each unconnected stock, and thus the 24 basis points difference I found is nearly a 20% difference. I also find that these results are robust within sub-periods of the time series. (Table IV) Furthermore, I cluster standard errors by fund-firm in unreported tests and find statistical significance at the 5% level for each specification. Although the average stock portfolio weight seems high, given the propensity for mutual funds to hold often more than 100 stocks, some of this difference can be attributed to the fact that I was more likely to find data on larger firms. In addition, I suspect the particular funds in the sample possessed fewer stocks than the average mutual fund. One might expect these results for a variety of reasons. First, managers may tend to overweight stocks within their employment networks simply because of familiarity bias. 7 Additionally, managers may invest in executives with overlapping employment networks because they believe this background to be indicative of positive future 7 Cohen, Frazzini, and Malloy (2008) discuss this explanation with respect to connections through education networks; they mention the related results of Huberman (2001) who finds that shareholders in Regional Bell Operating Company tend to live in its service areas. 10

13 performance. Due to his or her own self-image, a fund manager might simply have greater respect for executives with whom he or she shares employment experiences. Thus, the manager may tend to overweight these executives because of a belief about positive future performance. It would be difficult to disentangle the relative strength of past employers; therefore, I cannot completely discount this as at least a partial explanation. Next, managers may overweight connected stocks because these stocks correlate closely with another attribute that might lend itself to overweighting. In particular, connected stocks might be concentrated in an industry that performed especially well during this period. However, the momentum control helps address this concern, and the results on portfolio performance discussed in the next section tend to refute each of the reasons just mentioned. Indeed, managers may overweight connected stocks because they have better information on these stocks. This improved information could be the result of many factors related to overlapping employment networks. First, it might be easier to gather evaluative information on the executives, since the fund manager and firm executive are more likely to have mutual contacts. In this case, the fund manager may feel more confident on his assessment of a stock and its executives, and thus might be more likely to place larger bets in connected stocks. Larger bets on a firm could result in extreme overweighting and underweighting. In the latter case, some connected stocks could be weighted to zero and drop completely out of the sample due to a manager s negative conviction; this would further enlarge the average portfolio weight of connected stocks. Second, it might be easier for a fund manager to gather public information on a connected firm, due to improved access to the firm s executives. Alternatively, corporate executives 11

14 might provide better, more timely information to fund managers within their employment network. This could be the indirect result of being comfortable during a discussion or could be the direct result of intentionally favoring someone within their employment network. An executive might feel a strong allegiance to a past employer, and thus might be inclined to assist anyone who also worked for that employer. In addition, an employment connection between a manager and executive raises the possibility that the pair may actually know each other, which could facilitate communication and possibly favoritism. 8 The potential positive effects of these social connections are grounded in homophily, or the idea that people tend to associate with, or have affinity for, those who are most similar to them. (McPherson, Smith-Lovin, and Cook 2001) IV. Portfolio Performance Results I directly test the hypothesis that managers may have advantages in gathering information about their connected stocks. If improved information is truly flowing to the manager, then we would expect to see managers perform better on connected stocks than unconnected stocks. In order to test this hypothesis, I construct two monthly calendartime portfolios from the fund-firm-quarter observations: one based on connected stocks and one based on unconnected stocks. Since holdings data is only available quarterly or semiannually, I assume that holdings do not change between reports. 9 I then create connected and unconnected calendar-time sub-portfolios for each fund. For each subportfolio, I calculate value-weighted monthly returns, using the monthly returns of the stocks and weighting by fund s dollar holdings in each stock. I rebalance sub-portfolios each calendar quarter based on the most up to date holdings data. Next, I value-weight 8 See Tang (2009) for a discussion on direct, personal connections versus social connections. 9 See Cohen, Frazzini and Malloy (2008), Tang (2009), and Grinblatt and Titman (1989). 12

15 the sub-portfolio monthly returns across funds, weighting each by the fund s total net equity assets. In each quarter a large number of funds have zero connections, thus it is important to note that these funds are not included when weighting the connected portfolio within a given quarter. The result is two calendar-time portfolios, one based on connected stocks and one based on unconnected stocks. With the portfolios I compute risk-adjusted returns, using Carhart s (1997) four-factor model, which adds his momentum factor to the three Fama and French (1993) factors. I also compute raw returns, which are summarized in Table III along with the main results. The results are consistent with the hypothesis: the connected portfolio has a higher risk adjusted alpha and a higher annual raw return than the unconnected portfolio. The effect is both statistically (1%) and economically significant, with an annual riskadjusted alpha of approximately 12.16%. In addition, a portfolio constructed by buying connected holdings and shorting unconnected holdings yields a statistically significant (2%) risk-adjusted alpha of approximately 10.91%. Therefore, I find conclusive evidence that managers outperform the relevant benchmarks on connected holdings. Additionally, the effect of employment network connections appears larger than the effect of education network connections documented by Cohen, Frazzini and Malloy (2008). This suggests that connections forged through work environments may be more effective than connections forged through common educational backgrounds, at least in this setting. The annual average number of past employers (3530) is much greater than that of past academic institutions (341); 10 consequently, on average, there should be fewer connections to each past employer than connections to each educational institution. Thus, there may be more solidarity between managers and executives who are connected 10 See Table I in Cohen, Frazzini and Malloy (2008). 13

16 through employment networks. This is implied by in-group bias, or preferential treatment towards members of one s own group, which has been shown to increase with the distinctiveness of a group. (Gerard and Hoyt 1974) The greater solidarity could help explain the greater effect. Additionally, the outperformance on connected holdings does not come at the expense of greater risk; the Sharpe ratio on connected holdings is.79 versus a Sharpe ratio of.20 for the unconnected holdings. Finally, the fact that mutual fund managers do in fact earn abnormal returns on connected holdings indicates that managers are not necessarily acting irrationally by placing larger bets on connected stocks. I perform an additional test on each sample to rule out the possibility that all connected stocks, held and not held, earn abnormal returns. This could be a concern, for instance, if mutual fund managers happen to be connected to outperforming firms. For instance, if a mutual fund manager was previously employed by a very elite firm, then one might expect current executives who previously worked there to outperform. Additionally, mutual fund manager connections might simply be concentrated in industries that performed abnormally well during the sample period. Therefore, I construct a portfolio comprised of purchasing connected stocks in which funds hold and shorting connected stocks in which funds do not hold, where the latter stocks are value weighted within each fund sub-portfolio by their market weight. This long-short portfolio yields a risk-adjusted alpha of 10.66%, which directly refutes the two proposed explanations. Ultimately, this result strengthens the conclusion that fund managers are truly obtaining improved information on connected stocks, enabling the managers to selectively invest in them. Additionally, if mutual funds were capable, we might even 14

17 see them shorting a batch of these connected stocks that are not held, in which received information on the corporation may have indicated negative future performance. V. Robustness In this section I attempt to address some of the issues raised and discuss further considerations. First, I explore the use of the main filtered sample versus other possibilities. As Panel A of Table IV indicates, using several different filtered samples yields similar results. As expected, the magnitude and significance of both the portfolio weight and performance seem to increase as the sample becomes more filtered, which is likely the result of improved data quality and more manager awareness of connections. Next, Panel B outlines several additional tests on the main sample related to portfolio weights. The results clearly indicate that managers place larger bets on connected stocks, and this is consistent throughout different sub-periods. Interestingly, I find a much stronger effect in the earlier part of the period compared to the latter. During this period, roughly , the stock market experienced a major boom as the value of the S&P increased over 60%. 11 Indeed, executives may have been more willing to share positive information during the boom, which could have led managers to overweight these stocks more generously. Alternatively, the difference may be attributable to manager actions in 2007 and The stock market stopped climbing in 2007 and experienced a massive fall in 2008 as part of the financial crisis. 12 Thus, managers may have had advanced knowledge about their connected firms which led them to unwind those positions. However, the results on performance discussed next tend to refute the latter explanation. 11 On January , the S&P closed at , and on December 31, 2007, the S&P closed at On October 5, 2007, the S&P closed at , and on Dececmber 26, 2008, the S&P closed at

18 Panel C displays several additional tests related to portfolio returns for connected holdings. First, I split my sample of firms into the top half and bottom half by market capitalization; I recalculate connected and unconnected portfolio returns, restricting the portfolios to only large firms (top half) or only small firms (bottom half). I find similar connection premiums of 10.23% for large firms and 11.00% for small firms, where the connection premium is the annual four-factor alpha of a portfolio formed by buying connected holdings and shorting unconnected holdings. However, the small firms result is without statistical significance, which is consistent with the fact that information on executives was more easily obtained for larger firms, making the CONNECTED variable more reliable in this instance. In order to account for this, I perform a separate four-factor regression on the full sample of firms using a portfolio of large cap stocks (top 20% of NYSE/AMEX/NASDAQ) in place of the market portfolio, and I find a reliable connection premium of 10.81%. Additionally, the inclusion of the SMB factor helps account for any potential bias. Therefore, it does not appear the main results are driven by abnormal returns to all larger firms. Next, within different sub-periods of the sample, I find that connection premiums are all positive, although without statistical significance, which is not surprising given the smaller sample size. I find a much lower premium in the latter half of the period (3.37%), which is consistent with the lower portfolio weights of connected stocks during the period. This may indicate that the negative returns in 2008 extended even to connected holdings, which would be expected given the discovery by Cohen, Frazzini, and Malloy (2008) that abnormal information flow from executive to fund manager tends to only be positive in nature. Alternatively, during this massive fall in the stock market, it is possible 16

19 that stocks were trading on reasons other than fundamentals; thus it may be a time period in which information advantages are less likely to convert to abnormal returns. Nonetheless, I cannot reject the null hypothesis that the abnormal returns from this period are equal to zero. Additionally, Panel C includes tests meant to determine whether or not connected holdings display any performance persistence. Lags are introduced between the report dates for fund holdings and the actual stock returns from CRSP. The connection premium tends to persist notably with both a one month lag (8.01%) and a two month lag (5.84%), the first of which is significant at the 6% level. These results imply that an investor could profit by replicating a portfolio of connected stocks, since funds are required by the SEC to file holdings within 60 days after a fiscal period end. 13 Additionally, since the premium tends to persist but also reduce with more lag time, it seems to reaffirm that the abnormal returns on connected holdings are the result of information advantages, which would presumably manifest at precise points in time, such as earnings announcements. VI. Gender and Ethnicity In this section I briefly explore two other bases for social networks: gender and ethnicity. The consideration of both of these is motivated by the principle of in-group bias. In particular, women have been shown to display strong gender in-group bias (Rudman and Goodwin 2004). In the context of mutual funds, we might expect women to display even more bias, since they are undoubtedly the minority. 14 Of the entire sample, 13 See the SEC website: 14 Gerard and Hoyt (1974) find that in-group bias tends to increase with the distinctiveness of the group. Therefore, minority groups should tend to display relatively more in-group bias. 17

20 only 14% of fund managers are female and less than 6% of senior executives are female. (Table V) In general, I employ the same methodology used in the main study of this paper and in the same time-frame, I determined the genders of mutual fund managers from their first names, and when there was any ambiguity, I used newspaper article references. I collected the genders of 7124 executives from Execucomp. For the analysis, within each quarter I restrict the sample of funds to those with at least one female manager (female funds). I use a binary variable CONNECTED which equals 1 when at least two senior executives are female, and 0 otherwise. Thus, there is a connection when at least one fund manager and two executives are female. I find that female funds place an additional 50 basis points in connected holdings, compared to a weighting of about 127 basis points in unconnected holdings. This difference is both economically and statistically significant (10%). Interestingly, I find that male funds (zero female managers) placed slightly less weight in stocks with two female executives, significant at the 5% level. (Table V) Thus, women seem to place much larger bets on firms in which multiple women hold senior executive positions. This could be a signal of support, or could be the result of information advantages provided by female executives to female managers. I thus construct a long-short portfolio using the method outlined in Section IV, which consists of buying connected holdings and shorting unconnected holdings. The portfolio yields a four-factor risk-adjusted annual alpha of approximately 4.1%, but without statistical significance. This is possibly a consequence of the fact that the number of connections in the sample is very low: on average, only 29 funds each year are female 18

21 funds and female funds place roughly 1.2% of their assets in connected holdings. In a separate configuration, I assessed connections between female fund managers and firms with at least one female executive and failed to find any effects. This may have been a result of the fact that annually nearly 13% of firms in the sample have at least one female executive. Thus, these firms are relatively common, and are much less distinct than firms with at least two female executives. Last, I consider whether or not ethnic groups produce any of the characteristics observed with employment networks. For both executives and fund managers, I used Ancestry.com 15 to find the ethnic origins of last names to proxy for an individual s ethnic background. However, I failed to find significant results using several different definitions of connectedness. With a larger sample of funds and managers, this may have been a more fruitful exercise. VII. Conclusion The results from this paper indicate that mutual fund managers obtain information advantages through their employment networks. Mutual fund managers place larger bets on firms to which they are connected, and they also perform better on these holdings. A portfolio formed by buying connected holdings and shorting unconnected holdings yields over 10% per year, and these returns do not come at the expense of greater risk. To a certain extent, the results may reveal that fund managers face lower costs in obtaining some types of public information about firms. On the other hand, fund managers may have access to private information by virtue of their employment networks. Significant 15 Ancestry.com includes a searchable utility that allows someone to find the meaning and history of surnames, which includes the name s origin. 19

22 publicity has surrounded insider trading recently 16 and this paper arguably provides empirical evidence that insider trading is both meaningful and prevalent, unconfined to the isolated cases that become publicized. Additionally, this study is related to the literature on actively-managed mutual fund performance. Although my research does not speak to the broad performance of funds, it does inform regarding the process by which managers may outperform on individual stocks. Connected holdings comprise a relatively low percentage of a fund portfolio, so their effect on aggregate performance is modest. However, connections through education networks (Cohen, Malloy, and Frazzini 2008) and past business dealings (Tang 2009) were shown to provide similar benefits for mutual fund managers. Further research could include all three networks, along with other possible networks that remain to be explored. I briefly explored two of these, gender and ethnicity, both of which deserve further consideration. If networks are aggregated, the effect of connection premiums may no longer be so modest. Of the very few managers who may have true stock picking ability, 17 it remains to be seen to what extent their returns are driven by information advantages from social networks. 16 Many have recently been indicted in an insider-trading case involving New York based hedge fund, Galleon Group An ongoing debate persists on whether or not actively managed mutual funds outperform the market. For instance, Carhart (1997) finds managers do not outperform benchmarks and Wermers (2000) finds that managers can outperform passive indices before expenses. 20

23 REFERENCES Carhart, Mark, 1997, On Persistence in Mutual Fund Performance, Journal of Finance 52, Cohen, Lauren, Andrea Frazzini and Christopher Malloy, 2008, The Small World of Investing: Board Connections and Mutual Fund Returns, Journal of Political Economy 116, Cohen, Lauren, Andrea Frazzini and Christopher Malloy, 2009, Sell Side School Ties, Working Paper, Harvard University. Fama, Eugene F., 1965, Random Walks in Stock Market Prices, Financial Analysts Journal 21, Fama, Eugene F., and Kenneth R. French, 1993, Common Risk Factors in the Returns on Stocks and Bonds, Journal of Financial Economics 33, Fama, Eugene F., and Kenneth R. French, 1997, Industry Costs of Equity. Journal of Financial Economics 43, Gerard, Harold B. and Michael F. Hoyt, 1974, Distinctiveness of Social Categorization and Attitude toward Ingroup Members, Journal of Personality and Social Psychology 29, Grinblatt, Mark and Sheridan Titman, 1989, Mutual Fund Performance: An Analysis of Quarterly Portfolio Holdings, Journal of Business 62, Hochberg, Yael V., Alexander Ljungqvist, and Yang Lu, 2007, Whom You Know Matters: Venture Capital Networks and Investment Performance, Journal of Finance 62, Huberman, Gur, 2001, Familiarity Breeds Investment, Review of Financial Studies 14, Hwang, Byoung-Hyoun, and Seoyoung Kim, 2009, It Pays to Have Friends, Journal of Financial Economics, forthcoming. Kuhnen, Camelia M., 2009, Business Networks, Corporate Governance and Contracting in the Mutual Fund Industry, Journal of Finance 64, McPherson, Miller L., Lynn Smith-Lovin, and James M. Cook, 2001, Birds of a Feather: Homophily in Social Networks, Annual Review of Sociology 27,

24 Rider, Christopher I., 2009, Embedding Inter-Organizational Relations in Organizational Members Prior Education and Employment Networks, Working paper, Emory University. Rudman, Laurie A. and Sephanie A. Goodwin, 2004, Gender Differences in Automatic In-Group Bias: Why Do Women Like Women More Than Men Like Men?, Journal of Personality and Social Psychology 87, Tang, Yue, 2009, Business Connections and Informed Trading of Mutual Fund Managers, Working Paper, Emory University. Wermers, Russ, 2000, Mutual Fund Performance: An Empirical Decomposition into Stock-Picking Talent, Style, Transaction Costs, and Expenses, Journal of Finance 56,

25 Table I: Summary Statistics 23

26 Table II: Portfolio Weights of Connected vs. Unconnected Stocks 24

27 Table III: Returns of Connected Holdings 25

28 Table IV: Robustness 26

29 Table V: Gender Study 27