Fund Manager Educational Networks and Portfolio Performance. Botong Shang. September Abstract

Transcription

1 Fund Manager Educational Networks and Portfolio Performance Botong Shang September 2017 Abstract In this study, I investigate the relation between social connections among fund managers and portfolio performance. The social connection in this study is built between mutual fund managers through shared educational background, and I examine whether this connection aects portfolio performance. Specically, I use ve measures of portfolio performance, portfolio net return, one-factor alpha, four-factor alpha, Shape ratio and appraisal ratio, and I nd that the most closely connected teams outperform the others. In the dynamic setting, the portfolio performance drops after breaking the most connected team, and forming such a team with most connected managers improves the portfolio performance. Furthermore, the relation between social connection and portfolio performance is robust to the inclusion of controls for other performance-related variables and social connection-related variables. I am grateful for helpful comments and suggestions from Lauren Cohen, Ron Kaniel, Robert Novy-Marx, Robert Ready, Bill Schwert, Erin Smith and Jerold Warner. I thank Lauren Cohen for generously sharing his data on manager educational connections. All remaining errors are mine. Simon Business School, University of Rochester. botong.shang@simon.rochester.edu.

2 1 Introduction Agents exchange information and use the information for decisions. How does social connection aect this exchange process in nancial markets and what are the associated economic consequences? In this paper, I study the relation between mutual fund managers' social connection and the performance of the portfolio they jointly manage. In particular, I conjecture that the connection between managers improves the quality of information exchange within the team. That is, when managers know each other better or they are more connected, it is possible that they can communicate more eciently and more eectively. Regarding the impact of information, the quality of information exchange would have a predictable impact on the outcome of managers' actions. Hence, one can investigate the importance of the quality of information exchange through social connections in a team by testing the outcome of their actions. One possible channel of the connectedness is the connection formed by shared educational backgrounds. The shared educational backgrounds, albeit sometimes independent of the information exchanged later in work, could have formed connectedness between the managers, and this connectedness could have a positive eect on the quality of information exchange within the team. In turn, this quality of information exchange, or this quality of communication, within a team could have a positive eect on the performance of the team managed mutual fund. The objective of this paper is to investigate whether the quality of information exchange among managers is higher when the team members are more connected through their educational background by examining whether this social connection links to a better portfolio performance. 1

3 The relatively scarce existing literature explores the dierence in performance between team managed funds and single managed funds (Prather and Middleton, 2002; Bliss, Potter, and Schwarz, 2008; Bär, Kempf, and Ruenzi, 2010; Patel and Sarkissian, 2015), and even fewer studies investigate the characteristics of team managers (Sen and Tan, 2016). The main dierence between this study and the previous literature on team managed mutual fund performance is the link between the managers' level of social connectedness and portfolio performance. Instead of information exchange, previous literature applies various theories to explain the dierent performances of teams from single managers. For example, the returns of team managed mutual funds appear to have lower standard deviation, and previous literature suggests that the team managed mutual funds bear lower risk (Prather and Middleton, 2002; Bliss, Potter, and Schwarz, 2008; Bär, Kempf, and Ruenzi, 2010). Particularly, in Prather and Middleton (2002), the authors show that a higher risk-adjusted returns of team managed mutual funds can be explained by behavioral decision making theory. Namely, as single decision maker possesses the absolute decision making power, a team achieves decisions by consensus which leads to returns with lower standard deviation. Another theory to explain the outperformanance of team managed mutual fund is the diversication of opinions theory. In Bär, Kempf, and Ruenzi (2010), the authors nd that returns of team managed mutual fund are more stable, because team's investment style is less extreme and their portfolios are less industry-concentrated. More closely related to this study, (Sen and Tan, 2016) argue that as the fraction of the team who graduated from top MBA programs increases, the risk-adjusted return of the team managed mutual fund is higher with lower expense ratio. Similarly, teams which are more educational diverse, in terms of nance versus non-nance degree or quantitative versus non-quantitative majors, also have higher risk-adjusted returns. 2

4 What previous literature has overlooked is the dierence between social conncetion and diversity. While managers with dierent degrees or majors may diversify the knowledge base of a team, the social connection between managers is a form of social capital or team synergy, which should be categorized as another type of team capital apart from the diversity of knowledge. How connected the team members feel towards each other, therefore, should also be considered a team characteristic. As a particular form of social network, connectedness has attracted researchers' attention in its eect on agents' behavior in nancial markets. In Hong, Kubik, and Stein (2005), the authors explore the eect of word-of-mouth on portfolio choices between mutual fund managers in the same city. In Kuhnen (2005), the author suggests that past business connections between mutual fund directors and advisory rms aect future contracting decisions. Hochberg, Ljungqvist, and Lu (2007) demonstrates that the venture capital companies who are more connected with other venture capital companies via syndicating invest in rms with better outcomes in initial public oerings, in acquisitions, and with higher probability of surviving the subsequent nancing. As a particular basis for social network, educational institutions are also a key place where people connect. Besides education, people tend to build social connections through interactions with other students while attending school. The interactions potentially inuence people's decisions and behavior by forming a common information set. Massa and Simonov (2004) argue that the past educational backgrounds of individual investors have an impact on portfolio choices. Shue (2013) states that the shared educational background in executive peers inuences managerial decisions from various perspectives, such as executive compensation and acquisitions. Similarly, in Fracassi (2016), the author shows that if executives and directors are connected through employment, education or other activities, they have more 3

5 similar capital investments. Moreover, when the managers of a bank and a rm are more connected through education or previous work, the lending interest rate for these rms is signicantly lower, and subsequent credit ratings and stock return of these rms are much higher (Engelberg, Gao, and Parsons, 2012). In particular, there are two studies that are closely related to this paper:cohen, Frazzini, and Malloy (2008) and Cohen, Frazzini, and Malloy (2010). In Cohen, Frazzini, and Malloy (2008), the authors analyze the eect of shared educational background between mutual fund managers and corporate board members on mutual fund returns. In Cohen, Frazzini, and Malloy (2010), the authors also identify the impact of shared educational background between sell-side analysts and senior corporate members on stock return premia. Building upon the previous literature on educational background as the basis of social connections, this study will focus on the potential eect of the connectedness between managers in a team on fund performance. In this paper, I nd a strong and robust relation between the connectedness in teams and the performance of team managed portfolios. In particular, this relation only shows in teams where the managers share the closest connections. That is, when two managers are likely to know each other or have built a friendship during school, their portfolio performs better. This result remains signicant with various controls and xed eects. Furthermore, comparing to the team with the closest connection, the portfolio performance of managers who are less connected through schools, such as alumni, is not statistically dierent from the performance of the non-connected teams or single managers. In addition to the net return, I also use four other risk-adjusted measures of portfolio performance: one-factor alpha (CAPM alpha), four-factor alpha, Sharpe ratio and appraisal ratio, and I nd positive relations between the strongest connection and all the risk-adjusted performance measures. 4

6 To further analyze the portfolio performance of the most closely connected team, I examine the change of the portfolio performance when the team composition is changed. In other words, I examine how forming a team or dismissing a team with most connected managers aects the portfolio performance. Using a dierence-in-dierence approach, I nd that after assigning two managers who were classmates during school to one team, the performance of the portfolio they manage improves signicantly. After dismissing such a closely connected team, the performance of the portfolio they managed drops signicantly. This dynamic analysis based on team composition lends additional support for the idea that the relationship between fund managers through their educational background is positively associated with portfolio performance through its impact on the quality of information exchange within a team. In addition, I also explore the dierence in investment behavior between the most connected teams the others. In particular, I nd the managers who connected the most during school have remarkably dierent patterns in risk exposure and asset allocations, and it is possible that this dierence drives the higher portfolio performance of these teams. The main results are robust, taking into the account the potential team eect and other team characteristics. I conduct several other robustness tests addressing the concerns of educational institution, degree diversity, age similarity and sample selection. Notably, the estimates for the variable that captures the strongest connection remain signicant regardless of the specications. I conjecture that this strong relation can be explained by the eect of social connection on the quality of information exchange, or less friction in communication, between the managers in a team. With less friction in communication, managers are more 5

7 likely to be able to work in a more ecient and eective manner, which leads to better performance aftermath. The reminder of the paper proceeds as follows. Section II of the paper describes the data on mutual funds and managers connections. Section III provides the preliminary results of the empirical analysis of portfolio performance and team managers' connectedness, the dynamic analysis and the investment behavior. Section VI consists of four robustness checks. Section V concludes. 2 Data 2.1 Sample I collect my data from serveral data sources. The main data on mutual funds are obtained from the Survivor-Bias-Free U.S. Mutual Fund data from Center for Research on Security Prices (CRSP). The data include a description of each mutual fund, asset allocations, holdings and fee structures. In particular, I use the monthly and daily total returns, daily total net assets, and daily net asset value for actively managed equity funds from 1990 to the rst quarter of I exclude index funds following Berk and Van Binsbergen (2015), and the nal mutual fund sample consists of 1,697 U.S. actively managed equity funds. The biographical information of managers is from Cohen, Frazzini, and Malloy (2008, 2010). The original biographical data is from Morningstar, which include the educational background of mutual fund managers: undergraduate and graduate degrees received as well as the year and institution where the degrees were granted. The data shared by Professor 6

8 Lauren Cohen include 1,776 unique portfolio managers from 1990 to rst quarter of 2005 and 323 academic institutions. The top ve most connected academic institutions in the sample are Harvard University, University of Pennsylvania, University of Wisconsin (Madison), Columbia University and New York University, which in total comprise approximately 20.3% of the sample. As in Cohen, Frazzini, and Malloy (2008, 2010), there are six groups of degrees: (1) business school (MBA), (2) medical school, (3) general graduate school (Master of Arts or Master of Science), (4) Doctor of Philosophy, (5) law school, and (6) general undergraduate degree. I use quarter as the unit for time, and the nal sample consists of 44,703 unique portfolio-quarter observations, of which approximately 22.1% are managed by teams. 2.2 Summary Statistics Table 1 reports the summary statistics for the portfolios in the sample from January 1990 to March Panel A shows the mean, median and standard deviation of portfolio's return, net asset value, total net asset and number of managers in each quarter. All four variables are winsorized at 1 percent level. The median of the quarterly return, total net asset (TNA) and the value per share of the portfolio (NAV) are 3.3 percent, 399 million and dollar, respectively. Also, since approximately 22.1% of the portfolios are managed by team managers, the median of the number of managers for each portfolio-quarter in the sample is 1, and the mean is Panel B provides information about the distribution of managers' degrees and the median of the graduation years for each degree. Almost all of the fund managers in the sample hold a bachelor degree. Nearly half of the sample holds a degree in Business Administration 7

9 (47.86%), while only % hold a master degree (non-mba). The fractions of the managers who received a doctoral degree or Law degree are 2.03 % and 1.18 %, respectively. The median of the graduation year for all the undergraduate in the sample is 1977, and the median of the graduation year for all managers with MBA degrees is Panel C shows the distribution of graduation years including all types of degrees % of the portfolio managers graduated in the 80s (1980 to 1989), 27.27% of them graduated in the 70s (1970 to 1979), and 25.97% graduated in the 60s (1960 to 1969). In addition, some managers may appear more than once in this panel. For example, one manager can receive an undergraduate degree at 1979 and a MBA degree at 1981, thus, he would appear twice (once in the range and once in the range) in the sample. [Table 1 about here.] 3 Empirical Analysis 3.1 Social Connection and Portfolio Performance In this section I investigate the relation between the fund managers' connectedness and portfolio performance. If the level of connectedness between managers can potentially aect the quality of information exchange in the team, then we would expect a positive relation between the level of connectedness and team managed portfolio performance. Specically, I use various levels of connectedness to test how managers' social connection is associated with the portfolio net returns and risk-adjusted returns. In order to study the social connection among fund managers, I construct the CONNECT 8

10 variables to indicate the level of connectedness following Cohen, Frazzini, and Malloy (2008, 2010). Specically, for any two mutual fund managers, if they graduated from the same school, they are categorized as CONNECTED1. If they graduated from the same school and received the same degree, these managers are categorized as CONNECTED2. If they attended the same school at the same time, they are categorized as CONNECTED3. Lastly, if they attended the same school at the same time and received the same degree, they are categorized as CONNECTED4. A typical pair of entries on biographical data with CON- NECTED4 is as follows: In year 1998, Mr. Black and Mr. White are two managers for the ABC fund. Mr. Black attended Unicorn University in 1994 and received a MBA degree in Mr. White has one year overlap with Mr. Black. He attended Unicorn University one year later than Mr.Black in 1995, and received his MBA degree in It is important to recognize that the CONNECTED variables are nested in each other. For example, when two managers are categorized as CONNECTED4, i.e. they attended the same school at the same time for the same degree, they must also be categorized as CONNECTED1. Hence, to separate the dierence in connectedness between dierent levels of social connections, I adjust the CONNECTED variables so that none of them overlap with one another. Furthermore, regardless the ambiguous relationship between CONNECTED2 and CONNECTED3, CONECTED1 is treated as the lowest level of connectedness and CONNECTED4 is considered the highest level of connectedness. Since the likelihood of two CONNECTED4 managers knowing each other is quite high, CONNECTED4 potentially also capture the mutual trust and friendship between the managers, which the other levels of connectedness do not have. In addition, if the connection between managers aects the the quality of their information exchange, then the highest level of connectedness, CONNECTED4, should have the strongest 9

11 eect on how managers communicate with each other, and hence have the strongest positive relation with portfolio performance. In other words, I conjecture that a higher level of connectedness would lead to a higher quality of information exchange, or a higher quality of communication, within a team of fund managers, and this higher quality of information exchange will be associated with a better performance of the portfolio the team manages. As discussed earlier in the paper, instead of the connectedness between the fund managers, it is possible that team eect is the reason of portfolio performance in the sample (Prather and Middleton, 2002; Bliss, Potter, and Schwarz, 2008; Bär, Kempf, and Ruenzi, 2010; Patel and Sarkissian, 2015). I separately run the regressions with only the team dummy, and it appears that team managed portfolio does have higher return (not tabulated). Similar with portfolio returns, I also separately run a regression of these risk-adjusted measurements of portfolio performance on a team dummy, and it conrms that team managed portfolio outperforms the single manager portfolios (not tabulated). I therefore include Team dummy in all the specications to better isolate the social connection eect in the analysis. The baseline regression is as follows: 4 P ortfoliop erformance it = α t + η i + β j CONNECT ED i,t + j=1 (1) β 5 T eamdummy i,t + γ Controls i,t + ɛ it where P ortfoliop erformance it are the rolling net return of a quarter using the monthly mutual fund return data from CRSP and four risk-adjusted measurements of portfolio performance, which includes the alpha from CAPM model, the four-factor alpha, the Sharpe 10

12 ratio and appraisal ratio. To better measure the manager's performance, I use daily data of a short window to construct the quarterly risk-adjusted performances. That is, I take the daily data of the quarter when the managers are managing the portfolio to calculate the manager's alphas, sharpe ratios and appraisal ratios. The CAPM alpha, or one-factor alpha, is calculated only using the market excess return as a factor. The four-factor alpha is calculated using the market excess return, the size factor, the book-to-market factor as in Fama and French (1993) and the momentum factor as in Carhart (1997). 1 The Sharpe ratio and appraisal ratio are included in the analysis, taking into account the volatility of returns. The Sharpe ratio is calculated as the ratio of the average portfolio excess return over the standard deviation of the portfolio return, and the appraisal ratio is calculated as the ratio of four-factor alpha of the portfolio over the residual standard deviation of the portfolio. Due to data restriction, this sample is from the third quarter of 1998 to the rst quarter of 2005 with a total of 20,455 observations. The control variables include the Team dummy, the number of managers in a team, the portfolio's total net asset from the last quarter and the raw return from the last quarter. Fixed eects are included as indicated, and all standard errors are clustered by quarter and rm. Table 2 summarizes the results from ordinary least squares (OLS) pooled regression of fund performance on CONNECT variables, control variables and xed eects. The unit of observations is a portfolio-quarter. For each quarter, the CONNECTED4 groups signicantly outperform all the other groups, and this high performance holds for all performance measurements. When two managers attended the same academic institution at the same time 1 I also use portfolio alphas calculated for the one- and four-factor alphas using the monthly returns of the previous three years. The results remain consistent with the managers' alpha used in the main analysis. 11

13 and received the same degree, the portfolio they manage on average earns approximately 0.6% higher in net returns, 120bps in one-factor alpha and 90bps in four-factor alpha more than the other connected teams or single managers each quarter. In addition, comparing to the others, the sharpe ratio and appraisal ratio of the CONNECTED4 groups are also and higher, respectively. Interestingly, the other CONNECTED groups do not show any signicant higher performance for any of the measures. The results support the idea that the most connected teams are more likely to have interacted with each other in the past and hence more likely to have built or to build mutual trust and friendship that potentially facilitate the quality of information exchange within the team. Under such a scenario, the CONNECTED4 groups with the advantage of tightest connectedness between managers are the only groups with connected managers in a team who markedly outperform the others. [Table 2 about here.] 3.2 Dynamic Analysis In this subsection, I provide additional evidence on the relation between the CONNECTED4 team and their portfolio performance in a dynamic setting. How the fund is managed, by team or single managers, is changing over time. Hence, I can study the changes in portfolio performance before and after a CONNECTED4 team is formed or dismissed. Specically, I use a dierence-in-dierence approach and focus on the changes in team compositions, Initiation and T ermination. The Initiation is dened as when a fund that was managed by non-connectd4 managers transitions to a CONNECTED4 team. Analogously, the T ermination is dened as when a fund that was managed by a CONNECTED4 team no 12

14 longer has the CONNECTED4 level of connectedness. The regression is as follows: P ortfoliop erformance i,t = β 0 + β 1 T reated i,t + β 2 T reated i,t After + α t + η i + ɛ i,t, where the P ortfoliop erformance i,t are the quarterly net returns of the portfolio or a quarterly risk-adjusted measure of performance. T reated is the dummy that equals to one for the funds that have a team composition change, Initiation or T ermination, over time. Since the time of team composition change is not the same across the sample, the control groups in this analysis are all the other managers. After is a dummy variable which is equal to one after Initiation or T ermination. In addition, I also include the quarter and fund xedeect to control for the time trend and fund characteristics. The coecient of interest is β 2, which captures the change of portfolio performance when the CONNECTED4-related team composition changes. [Table 3 about here.] Table 3 summarizes the results from the dierence-in-dierence analysis. Panel A shows the results when a CONNECTED4 team is formed. The coecients on T reated Af t are all positive and signicant at 1% level for net returns and at 5% level for the risk-adjusted measures for performance. That is, after a team is formed with two or more managers who are the most connected by their educational background, the performance of the fund they manage improves signicantly. Comparatively, the results in Panel B suggest that when a CONNECTED4 team was dismissed, the performance of the fund they managed drops signicantly, as the coecients on T reated Af t are all negative and statistically signicant. It is interesting that the eect from CONNECTED4 group to portfolio performance is two- 13

15 sided, that we see both signicant changes in performance after Initiation and T ermination. 3.3 Investment Behavior In the previous sections, I have shown that there is a positive and signicant relation between the connectedness of fund managers and portfolio performances. To further investigate the dierence in the portfolio performance, in this section I examine the dierences in investment behaviors between dierent groups. By comparing the beta loadings and the portfolio allocations, I found a clear pattern in managers' investment behavior of the CONNECTED4 group Beta Loading Given the dierences in portfolio performances, it is intriguing to examine the dierences in factor exposures and portfolio allocations between the CONNECTED team and the others. Panel A of Table 4 shows the comparison of market premium exposures between the CON- NECTED4 group and the other groups based on CAPM model. The CONNECTED4 team has signicantly higher market exposure than the others at the 1% level. Panel B shows the comparison of beta loading as in the four-factor model. Interestingly, comparing to all the other managers, the CONNECTED4 groups tend to have signicantly less exposure on the momentum but markedly more exposure on all three factors. The distinct dierence in beta loading between the CONNECTED4 teams and the others clearly shows a pattern in the exposure to systematic risks, which may lead to the higher manager's alphas of the CONNECTED4 groups. [Table 4 about here.] 14

16 3.3.2 Allocation In addition to beta loading, I also explore the dierences in the allocations of portfolios, management fees, turnover ratios and expense ratios. The Column 1 and Column 2 are the mean (µ) and standard deviation of the full sample. Then, I split the sample by team or single managers, and compare the dierence in allocations. Lastly, I compare the non- CONNECTED team, other level of CONNECTED team and CONNECTED4 teams with the sample mean. As shown in Table 5, the portfolios of other level CONNECTED managers allocate the most in common stocks, and the managers with the closest connections have the lowest allocation in common stock. Comparatively, the non-connected teams tend to allocate more in preferred stocks, but there is no signicant dierence between CONNECTED4 and the other CONNECTED teams. From Row 3 to Row 7, Table 9 shows the percentages of portfolio allocations in dierent types of bonds. Comparing to the single managed portfolios, the portfolios of teams on average allocate more in convertible bonds, corporate bonds and government bonds. For the portfolios managed by CONNECTED4 teams, they appear to allocate less in convertible bonds and more in government bonds. Moreover, the cash allocations also vary signicantly with groups. The managers in the CONNECTED4 group allocate the highest fraction of their portfolio in cash, while the single manager allocate the lowest fraction in cash. Furthermore, teams on average have a lower turnover ratio than the single managed portfolios, but CONNECTED4 teams have the highest turnover on average in the sample. It is interesting to see that most of these dierences in allocations are also statistically signicant at the 1% level, which again indicates a pattern in investment for each group. 15

17 Regarding to the fees, we can see that the management fee over average net asset (MGMT Fee/ANA) is on average higher for team managers than the one for single managers, but this ratio is not the highest for CONNECTED4 teams, which does not support the positive association between portfolio performances and fees. The expense ratio shows the opposite pattern of MGMT Fee/ANA, but again the CONNECTED4 teams do not have the highest expense ratio. [Table 5 about here.] In summary, I investigate how team-managed portfolio performs when the managers are socially connected through their educational background. The connectedness between fund managers is constructed following Cohen, Frazzini, and Malloy (2008, 2010), where CON- NECTED1 is the least connected teams and CONNECTED4 are the most closely connected teams. In addition, I use both the portfolio's quarterly net return and the manager's riskadjusted measures of performance in the analysis, which includes one-factor alpha, four-factor alpha, Sharpe ratio and appraisal ratio. For all the measurements for portfolio performance, the most closely connected teams signicantly outperform the others. The main dierence between the closest connection and the other levels of connections is that the managers who are categorized as CONNECTED4 are the ones who most likely knew each other the best during school or they are more likely to have a close relationship, and this relationship can potentially improve the quality of information exchange within the team. The results show that the connectedness between managers positively aects the portfolio performance and dierentiates their investment behavior from others. 16

18 4 Further Analysis There are various potential reasons why the connectedness between fund managers may reect other confounding factors. In this section, I present a series of robustness checks that address several of these concerns. These specications focus on managers' educational institutions, degree diversity, age similarities and sample selection. The results of these robustness checks are in favor of social connections between managers in explaining the higher performance of CONNECTED4 groups. 4.1 Educational Institutions It has been documented in the previous literature that fund managers' characteristics significantly aect mutual fund performance (Chevalier and Ellison (1999); Atkinson, Baird, and Frye (2003); Gottesman and Morey (2006)). Particularly, the education of fund managers has a positive eect on fund returns (Chevalier and Ellison (1999); Atkinson, Baird, and Frye (2003); Gottesman and Morey (2006)). In this paper, educational institutions can potentially aect the main analysis through two channels. Firstly, if the schools where managers graduated characterize the abilities of managers and the dierence in the abilities of managers is the leading factor to explain the dierence in fund performance, then the statistical significance of connectedness may disappear after controlling for specic schools. Alternatively, if a big fraction of the connected fund managers graduated from only several institutions, then the association of the connectedness may merely reect the value of attending these institutions. To examine if the managers' educational institutions inuence their portfolio performance, I follow Cohen, Frazzini, and Malloy (2008, 2010) and construct a sub-sample 17

19 by excluding observations where the managers attended one of the top ve academic institutions as indicated earlier in the paper. Using this method, the sample consists of 36,038 observations for the analysis of portfolio net returns and 15,578 observations for the analysis of risk-adjusted portfolio performances. I then re-estimate the baseline regression. The results are shown in Table 6. Comparing with Table 2, all the coecients for CON- NECTED4 are still positive and statistically signicant at the 5% or 1% level, which indicates that the association between the connectedness and portfolio returns is not driven by the managers' educational institutions. Hence, even without the potential inuence from the top 5 institutions in the sample, the positive relation between the CONNECTED4 teams and their portfolio performance still holds. That is, the results show that the higher performance of the most connected team is not driven by their educational institutions. [Table 6 about here.] 4.2 Degree Diversity Previous literature has documented a positive relation between the diversity of degrees in a team and performance. However, Sen and Tan (2016) also argue that the higher fraction of managers with MBA degrees, the better is the portfolio performance. Hence, the relation between diversity of degrees and portfolio performance is ambiguous, since the fraction of managers with MBA degrees is inversely related to the diversity of the team. I therefore investigate the robustness of the main results to this concern by including a diversity measure for managers' degrees and its interactions with CONNECTED variables in the regressions. Specically, I use the ratio of the fraction of fund managers with MBA degrees in a team 18

20 to the fraction of managers without a MBA degree in the team as the diversity measure (Diversity). Then, I include the variable Diversity and the corresponded interaction terms with the CONNECTED variables in all the specications. As shown in Table 7, the coecients of CONNECTED4 remain positive and statistically signicant in all performance measures. The teams where fund managers are the most closely connected outperform the others in both net returns and risk-adjusted measures, taking into the account of potential inuence on the diversity of managers' degrees. Interestingly, the diversity also has positive and signicant eect on the CAPM alpha and Sharpe ratios, which indicates the team with managers have more diverse degrees also has a better portfolio performance but not to all measures of performance. Hence, the main results appear to be a separate eect apart from the diversity of a team, and the estimated association between the connectedness and portfolio performances do not change when the degree diversity measure is included in the specications. [Table 7 about here.] 4.3 Age Similarity I now check the robustness of the main results to the potential eect from age similarity. If managers with similar age can communicate better and thus have less friction in information exchange, then age similarity between fund managers could be a confounding factor that leads to similar results on portfolio performance as the ex-ante connectedness between fund managers. Since the data on each manager's age are not available, I use the managers' graduation years to construct a new sub-sample. Specically, since almost all the managers 19

21 in my sample hold bachelor degrees and the graduation years are included in the data, I approximate the managers' ages by assuming that all managers received their undergraduate degree at the same age. Given any two fund managers, if the dierence in ages is no more than ve years, I treat these managers as having similar age. Then, I construct a sub-sample by only including the portfolio-quarter observations with at least two managers with similar age. The selection criteria of team managed funds and managers with similar age reduce the sample sizes to 5,447 portfolio-quarter for the analysis of portfolio returns and 1,231 for the analysis of risk-adjusted portfolio performances. Table 8 reports the regression results. The coecients on CONNECTED4 variable are again statistically positive in all measurements of portfolio performance, and both the magnitude and statistical signicance are close to the main results in Table 2. Therefore, even when team managers are of similar age, the association between ex-ante connectedness of managers through educational connection and the portfolio performance is still robust, which supports the idea that the closely connected managers might have built or be easier to build a friendship that facilitates their communication and hence improves the performance of the portfolio they manage together. [Table 8 about here.] 4.4 Sample Selection Lastly, I conduct a test to examine whether the positive relation between the most closely connected team and their portfolio performance is due to sample selection bias. To do so, I construct a sub-sample by only including the connected groups and the funds who would 20

22 be connected somewhere in time to capture the potential inuence from unobserved fund characteristics. That is, by comparing the dierent time periods of the same fund, I can investigate whether the higher portfolio performance is due to the social connection between the managers or the fund itself in spite of the managers or the composition of teams. In Table 9, I report the regression results for this robustness check. Given the same funds, the portfolio performance is much higher when the CONNECTED4 team jointly manages the portfolio than at other times. The coecients on CONNCETED4 groups for all portfolio performance measures are positive and signicant at the 1% level. In addition, the size of the coecients in all ve regressions are greater than the main results in Table 2. This indicates that even though there may be some unobserved fund-specic characteristics, such as specic trading strategy or location, which may be associated with high performance, the fund still has better performance when it's managed by the most closely connected team. Hence, the positive relation between the most closely connected team and portfolio performance is not driven by sample selections. [Table 9 about here.] Overall, the positive relation between managers' connectedness and portfolio performance is only found in the strongest tie, where the managers attended the same school at the same time and received the same degree. As discussed earlier, these managers are the ones who most likely knew each other the best during school and have built or easier to build a mutual trust and friendship. This social relationship between the managers potentially improves the quality of information exchange within the team, and, in turn, positively aects the portfolio performance. In this section, the results show that not only this relation is a separate eect 21

23 from team eect, but also robust to other performance-related concerns addressed in this study. 5 Conclusion In this paper, I examine the relation between fund managers' social connection and portfolio performance. To quantify the level of connectedness between fund managers, I follow Cohen, Frazzini, and Malloy (2008, 2010) and use the social connections built through educational institutions. The measures of portfolio performance used in this study include the portfolio quarterly net return, one-factor alpha, four-factor alpha, Sharpe ratio and appraisal ratio. By constructing dierent CONNECTED-variables that reect dierent levels of connectedness of managers through educational institutions, I nd the only positive and signicant relation is between the teams of the most closely connected managers and their portfolio performance. The main dierence between the most closely connected teams and the others is that these managers more likely knew each other and have built a friendship while in school. This familiarity potentially has a positive eect on the quality of information exchange, which in turn improves the performance of their portfolio. I then further investigate the relation between the most closely connected teams and the portfolio performance in a dynamic setting. Using a dierence-in-dierence approach, I nd that when a team is formed with managers who are the most closely connected, the performance of the portfolio improves remarkably, and when such a team is dismissed, the performance of the portfolio drops signicantly. Having established a robust relation between connectedness of the managers and portfolio performances, I study the dierences in investment behaviors. By comparing the beta loading and the portfolio allocations, I provide evidence that the most 22

24 connected groups have distinct investment patterns from the other. Furthermore, the results are not driven by the team eects, school selections, the diversity of managers' degrees, the age similarity of managers nor sample selection bias. Overall, even though the social connection between fund managers through educational institutions is often overlooked as a salient team characteristic and sometimes mistaken as a part of the team diversity, this connection relationship between managers truly has a pronounced and robust association with portfolio performance, which is the main contribution of this paper. 23

25 References Atkinson, S. M., S. B. Baird, M. B. Frye, Do female mutual fund managers manage dierently?. Journal of Financial Research 26(1), 118. Bär, M., A. Kempf, S. Ruenzi, Is a team dierent from the sum of its parts? Evidence from mutual fund managers. Review of Finance p. rfq014. Berk, J. B., J. H. Van Binsbergen, Measuring skill in the mutual fund industry. Journal of Financial Economics 118(1), 120. Bliss, R. T., M. E. Potter, C. Schwarz, Performance characteristics of individually-managed versus team-managed mutual funds. Journal of Portfolio Management 34(3), 110. Carhart, M. M., On Persistence in Mutual Fund Performance. Journal of Finance 52(1), Chevalier, J., G. Ellison, Are Some Mutual Fund Managers Better Than Others? Sectional Patterns in Behavior and Performance. Journal of Finance 54(3), Cross- Cohen, L., A. Frazzini, C. Malloy, The Small World of Investing: Board Connections and Mutual Fund Returns. Journal of Political Economy 116(5), , Sell-Side School Ties. Journal of Finance 65(4), Engelberg, J., P. Gao, C. A. Parsons, Friends with money. Journal of Financial Economics 103(1), Fama, E. F., K. R. French, Common risk factors in the returns on stocks and bonds. Journal of nancial economics 33(1), 356. Fracassi, C., Corporate nance policies and social networks. Management Science. Gottesman, A. A., M. R. Morey, Manager education and mutual fund performance. Journal of empirical nance 13(2), Hochberg, Y. V., A. Ljungqvist, Y. Lu, Whom you know matters: Venture capital networks and investment performance. The Journal of Finance 62(1), Hong, H., J. D. Kubik, J. C. Stein, Thy neighbor's portfolio: Word-of-mouth eects in the holdings and trades of money managers. The Journal of Finance 60(6), Kuhnen, C. M., Social networks, corporate governance and contracting in the mutual fund industry.. Massa, M., A. Simonov, History versus geography: The role of college interaction in portfolio 24

26 choice and stock market prices.. Patel, S., S. Sarkissian, To group or not to group? Evidence from mutual funds. Journal of Financial and Quantitative Analysis, forthcoming. Prather, L. J., K. L. Middleton, Are N+1 heads better than one?: The case of mutual fund managers. Journal of Economic Behavior & Organization 47(1), Sen, A., K. M. Tan, The Eect of Fund Managers' Educational Background on Fund Performance and Money Flows.. Shue, K., Executive networks and rm policies: Evidence from the random assignment of MBA peers. Review of Financial Studies 26(6),

27 Table 1: Summary statistics This table contains summary statistics for the nal sample, consisting of 44,703 portfolio-quarter observations based on CRSP Survivor-Bias-Free U.S. Mutual Fund data from the rst quarter of 1990 to the rst quarter of In addition, all four variables are winsorized at 1 percent level. The statistics shown are mean, median and standard deviation (Std. Dev.). Panel A contains portfolio information for the full sample. The data in the rest of the Table is from Cohen, Frazzini, and Malloy (2008, 2010). Panels B provides the distribution of the mutual fund managers' academic degrees and the mean and median year of graduation for each degree. Panel C provides the distribution of the graduation years. Variable Obs Mean Std. Dev. P25 P50 P75 Return Net Asset Value Total Net Asset Number of Managers Panel B: Degree Distribution Degree # Managers Percentage Median Grad. Year Business School Medical School Graduate School (non-mba) PhD Law School Undergraduate 1, Panel C: Grad. Year Distribution Year Range # Obs. Percentage < , Total

28 Table 2: OLS Regressions of Portfolio Performance on Managers' Connectedness This table shows estimated coecients from OLS pooled regressions of mutual funds' returns on the CONNECTED-variables. The sample period is the rst quarter of 1990 to the rst quarter of The unit of observation is portfolio-quarter. The dependent variable is the portfolio's rolling return for the quarter. The independent variables of interests are the CONNECTED-variables. Specically, for any two mutual fund managers, if they graduated from the same school, they are categorized as CONNECTED1. If they graduated the same school and received the same degree, they are categorized as CONNECTED2. If they attended the same school at the same time, they are categorized as CONNECTED3. Lastly, if they attended the same school at the same time and received the same degree, they are categorized as CONNECTED4. The control variables are the portfolio's total net asset (Lag TNA), raw return (Lag Return) of last quarter and if the portfolio is managed by team. Quarter xed eects are included in all the regressions, and portfolio xed eects are included as indicated. Standard errors are clustered by quarter. Signicance at the 10%, 5%, and 1% level is indicated by *, **, and ***, respectively. T-statistics are shown in parentheses. (1) (2) (3) (4) (5) Return CAPMAlpha 4FactorAlpha Sharpe Appraisal CONNECTED ** (-0.26) (-0.32) (-1.28) (-0.51) (-2.33) CONNECTED (-0.54) (0.64) (0.74) (0.05) (1.11) CONNECTED (0.48) (0.57) (0.16) (-0.91) (-0.45) CONNECTED *** 0.012** 0.009** 0.025*** 1.945** (3.59) (2.32) (2.28) (2.98) (2.28) NonConTeam *** (0.79) (0.52) (0.83) (-2.62) (0.37) N adj. R Fixed Eects Y N N N N Controls Y Y Y Y Y 27

29 Table 3: OLS Regressions of Portfolio Performance with Dierence in Dierence Analysis after Changes of Team Composition This table shows estimated coecients from OLS pooled regressions of mutual funds' returns on CONNECTED-variables. Similar with Table 2, the sample period is the rst quarter of 1990 to the rst quarter of The unit of observation is portfolio-quarter. The dependent variable is the portfolio's rolling return for the quarter. The independent variables of interests are the CONNECTED-variables. Specically, for any two mutual fund managers, if they graduated from the same school, they are categorized as CONNECTED1. If they graduated the same school and received the same degree, they are categorized as CONNECTED2. If they attended the same school at the same time, they are categorized as CONNECTED3. Lastly, if they attended the same school at the same time and received the same degree, they are categorized as CONNECTED4. The control variables are the portfolio's total net asset and raw return of last quarter. Quarter xed eects are included in all the regressions, and portfolio xed eects are included as indicated. Standard errors are clustered by quarter. Signicance at the 10%, 5%, and 1% level is indicated by *, **, and ***, respectively. T-statistics are shown in parentheses. (1) (2) (3) (4) (5) Return CAPMAlpha 4FactorAlpha Sharpe Appraisal Panel A: Initiations Intercept *** ** 0.219*** *** (-11.16) (-0.11) (-2.36) (48.42) (-3.88) Treated ** (0.91) (-0.39) (-2.04) (-1.53) (-1.36) Treated*Aft 0.541*** 0.014** 0.011** 0.013** 2.052** (3.04) (2.37) (2.55) (2.32) (2.13) N adj. R Fixed Eects Y Y Y Y Y Panel B: Terminations Intercept *** ** 0.219*** *** (-11.38) (-0.06) (-2.32) (48.44) (-3.86) Treated 0.433*** 0.012*** 0.006** 0.007** (4.73) (3.08) (2.15) (2.02) (1.55) Treated*Aft *** *** *** *** ** (-4.60) (-3.36) (-3.26) (-2.85) (-2.23) N adj. R Fixed Eects Y Y Y Y Y 28

30 Table 4: Beta Loading Dierences on Team, Connection and Connected Variables This table shows dierences between allocations based on whether mutual funds is managed by teams, or whether fund managers are linked by CONNECTED-variables. Specically, for any two mutual fund managers, if they graduated from the same school, they are categorized as CONNECTED1. If they graduated the same school and received the same degree, they are categorized as CONNECTED2. If they attended the same school at the same time, they are categorized as CONNECTED3. Lastly, if they attended the same school at the same time and received the same degree, they are categorized as CONNECTED4. Signicance at the 10%, 5%, and 1% level is indicated by *, **, and ***, respectively. T-statistics are shown in parentheses. Panel A: CAPM Model Market CON NonCON Dierence 0.040*** (4.59) Panel B: Four-Factor Model Market SMB HML MOM CON NonCON Dierence 0.007*** 0.010*** 0.043*** *** (2.64) (2.93) (8.83) (-4.60) N