The Predictability of Managerial Heterogeneities in Mutual Funds Jun Huang School of Accountancy Shanghai University of Finance and Economics No.777 Guoding Road, Shanghai, China Yan (Albert) Wang 1 Department of Finance Chinese University of Hong Kong Shatin, N.T., Hong Kong This Version: January 14, 2013 1 Corresponding author: Phone: 852-39431914, Email: albertwang@cuhk.edu.hk. We appreciate helpful comments from Jeffrey Pontiff, Jennifer Huang and Zheng Lu. All errors are our own.
The Predictability of Managerial Heterogeneities in Mutual Funds Abstract We empirically assess the role of managerial heterogeneities in mutual fund performance. Using a sample of Chinese mutual funds with relatively high managerial turnover rates, we follow the method in Abowd, Kramarz and Margolis (1999) to explicitly estimate the heterogeneities among fund managers. We find that funds with higher manager fixed effects outperform those with lower manager fixed effects by 2% per year. We also find that fund performance improves after managers with larger fixed effects are hired. The results are consistent with the notion that the fixed effects are associated with managerial innate ability or human capital, which in turn leads to better performance. Finally, we find that investors pay special attentions to these managerial attributes above and beyond the traditional performance measures, which provides some valid empirical evidence to the rational model of active portfolio management as in Berk and Green (2004). JEL Classification: G20, J24 Keywords: Manager Fixed Effect, Fund Performance, Fund Flows, Managerial Turnover 1
1. Introduction A large body of research has studied the characteristics of fund managers to see whether they can predict future fund performance. Not surprisingly, many studies have found that certain types of fund managers reliably outperform their counterparties. For example, Ding and Wermers (2009) find that experienced managers of large funds (i.e., with assets under management above the median) outperform less experienced managers by 92 bps a year. Other manager characteristics that can help predict outperformance include social connections, academic background, and co-investment. 2 While the documented manager characteristics are mostly based on the observables (i.e. tenure, degree, etc.), some of the characteristics are not easy to observe (i.e. risk preference, private information source, innate ability, etc.). Unfortunately, little is known about the importance of time invariant managerial attributes (i.e. the unobservable ones) in determining fund performance. Furthermore, if unobservable managerial heterogeneities are correlated with the observable characteristics, empirical methods that do not explicitly account for the unobservable heterogeneities could result in omitted variable bias. 3 Another difficulty when assessing the role of manager characteristics on fund performance is that all previous papers essentially do not consider the contributions of managers and funds separately and, hence, treat them as a manager-fund combination. However, it seems reasonable to entertain the notion that part of the performance of a mutual fund resides in the manager, who is responsible for the investment decisions, and part resides in the fund (or fund family), 4 which can influence performance 2 Chevalier and Ellison (1999) document that fund managers who graduate from colleges whose students have higher average SAT scores tend to outperform other managers. Similarly, Gottesman and Morey (2006) find that the quality of a manager s MBA program is positively correlated with future fund performance. Cohen, Frazzini and Malloy (2008) show that fund managers take larger positions in companies in which they have social connections (i.e., the officers or board members attended the same colleges as the fund managers), and these holdings outperform non-connected holdings on average. Using a large sample of hedge fund managers, Li, Zhang and Zhao (2011) find that managers from higher-sat undergraduate institutes are associated with higher risk-adjusted performance, more inflows and less risk. 3 For example, Hsiao (2002) concludes that ignoring the individual effects that exist among cross-sectional units, but are not captured by the included explanatory variables can lead to inconsistent or meaningless estimates of interesting parameters. 4 We treat fund and fund family as interchangeable in this paper, because to separate the role of each is beyond the scope of this paper. 2
through economy of scale in trading, administrative procedures, quality of research, marketing and distribution, etc. In reality, however, one never observes performance outcomes of managers and funds separately, let alone assesses the importance of unobservable managerial heterogeneities. The primary objective of this paper is to examine the role of time invariant managerial heterogeneities (or manager fixed effects) in determining mutual fund performance. The simplest way to examine the role of manager heterogeneities is to include a dummy variable for each unique fund-manager combination (which can be called a spell ). However, the spell approach does not separately identify manager and fund heterogeneities and, thus, does not permit estimation of their relative importance. Thus, this approach is not sufficient to help us quantify how much of the variation in fund performance is attributable to manager and fund heterogeneities, respectively. The alternative method is based on Abowd, Kramarz and Margolis (1999; AKM henceforth). The AKM method identifies manager and fund fixed effects through group connection. The AKM approach has two advantages. First, it allows us to separate manager and fund heterogeneities even when they are simultaneously observed in the same fund performance. Second, the method can lever the potentially small number of mover observations (i.e., managers who move across funds) to derive information about non-movers who work in the same funds that have hired at least one mover. This allows us to separate the manager and fund fixed effects for not only movers, but also non-movers, increasing the number of observations under study. We apply AKM method to estimate the effect of managerial heterogeneities on mutual fund performance in China. Chinese mutual funds offer a unique setting to implement such an experiment. First, turnovers among Chinese mutual fund managers are almost three times as high as those among their peers in the U.S. The annual turnover rates for mutual fund managers in China range from 31.2% to 49.8% between 2003 and 2010. 5 5 According to Hu, Hall and Harvey (2000), the annual turnover rate among U.S. mutual fund managers is 12.5% in 1990s. 3
Large numbers of movers add power to the AKM method, lending a hand to a precise estimation of managerial heterogeneities. Second, given the complexity and sophistication of how the Chinese name is recorded, 6 we can effectively track down almost every manager s career path, along with their personal information such as age, education and work experience since the profile information of fund managers is publicly disclosed in China. Last, but not the least, Chinese mutual fund industry has been experiencing an unprecedented growth compared to other developed countries. The fact that China overtook Japan as the world s second largest economy in 2011 makes the study of Chinese capital market in general and Chinese mutual fund industry in particular intriguing by its own means, which also has ample implication for markets in other places. We first demonstrate that ignoring manager fixed effects in estimating a fund performance regression (as is typically done) could yield biased coefficient estimates. For example, we find that the magnitude of coefficient on fund size decreases by approximately 60% when manager fixed effects are added to the specification. We also show that manager fixed effects are significantly correlated with observable time-variant firm and manager characteristics in a meaningful way. For example, managers with higher fixed effects tend to work for bigger funds, receive higher management fees and trade more. If managerial heterogeneities are proxies for valuable human capital related to investment skills or strategies, then we should observe a positive correlation between managerial heterogeneities and future fund performance. This is indeed what we find. Using out-of-sample testing, we find that funds with higher manager fixed effects tend to outperform funds with lower manager fixed effects. One standard deviation increase in manager fixed effects can increase the fund net return in the following year by 2% after controlling for past fund performance and other relevant fund and manager 6 Chinese use characters instead of a Latin alphabetic spelling using 26 letters. There are 7,000 commonly used characters and much more less-than-common characters in the modern Chinese language. This helps us to identify fund managers by their names. Furthermore, none of the 722 fund managers in our sample have identical names. 4
characteristics. Having identified an important role played by managerial heterogeneity, we next investigate whether or not employing a fund manager with a high fixed effect increases fund performance. Using the difference-in-differences (DiD) approach can allow us to mitigate the potential endogeneity concern that superior managers are likely to be attracted to superior mutual funds. In particular, we examine the change of fund performance during the quarters surrounding managerial turnover and find that both fund net return and benchmark-adjusted return improve significantly when incoming managers have higher fixed effects than out-going managers. The result that managers with higher fixed effects increase fund performance around the turnover is once again consistent with our notion that manager fixed effects capture some innate ability or elements of human capital. This interpretation is also consistent with the finding that managers with higher fixed effects are associated with higher management fees such that the marginal productivity is matched with extra rents they receive. In addition to the practical value of identifying superior managers, understanding the effect of managerial attributes on mutual fund performances also provides a way of testing the rational model of active portfolio management as in Berk and Green (2004). For example, one important assumption of Berk and Green is that investors can learn about the ability of fund managers based on past performance and allocate money to funds based on their Bayesian updates. However, if the theory of Berk and Green is true, then it could be difficult to identify cross-sectional differences in fund managers based on returns, because the competitive supply and redeployment of capital drives the abnormal performance of all funds to zero. Moreover, if fund performance is partly determined by managers, who change funds frequently, then fund return would not exhibit any long-term persistence even if managers have relevant skills. Thus, by taking managerial turnover into consideration and estimating the departing and incoming manager attributes respectively, one may be able to better understand whether performance persistence exists, and even if it does whether it can be attributed 5
to managerial skill. In addition, if manager fixed effects contain information about managerial skills that may not be entirely captured in the traditional performance measures, investors would rationally chase managers with higher fixed effects after controlling for past performance. Consistent with the model in Berk and Green (2004), we find that, controlling for past fund performance and other fund-specific characteristics, funds with higher manager fixed effects can attract higher inflows, indicating that investors pay special attention to managerial attributes that are important determinants of fund performance. We also find that after the incumbent manager leaves the fund, mutual fund investors respond more strongly to the fund performance if the incoming manager has a higher fixed effect than the outgoing manager. The discontinuity of flow performance sensitivity provides some supporting evidence to the premise that investors are Bayesian updating managerial skills when investing in mutual funds, a key assumption in Berk and Green (2004). Finally, we examine the role of managerial heterogeneities in determining the likelihood of managerial turnover. Consistent with Khorana (1996), our baseline probit regression confirms that past performance of mutual funds can predict managerial turnover. Our result further suggests that manager fixed effects contribute additionally to explain the likelihood of managerial replacement. By classifying managerial turnover into separate cases of promotion and demotion, we show that the probability that a manager is likely to be fired or demoted is negatively associated with manager fixed effects, whereas the promotion probability is positively related with manager fixed effects after we control for fund past performance. Our paper marks several contributions to the literature. First, our paper adds to the literature of mutual fund performance valuation by providing the first empirical study on the role of unobservable manager heterogeneities in determining fund performance. The existing literature on fund performance has studied how fund return is affected by 6
observable manager characteristics. Given the importance of latent factors (such as innate ability, preference, risk aversion, personality, etc.) in shaping investment decision, we provide supporting evidence that it is crucial to incorporate the role of unobservable managerial heterogeneities into the determinants of fund performance. Second, the prospects for detecting abnormal fund performance or managerial skill by using traditional asset pricing models are poor due to model misspecification. Thus, even if there are significant differences among fund managers, it is possible that one would never produce evidence that they have superior stock-picking or market timing abilities. Furthermore, to the extent that unobservable managerial heterogeneities are correlated with the observable characteristics, empirical methods that do not explicitly account for the unobservable heterogeneities could produce biased estimates for managerial skill. Using a sample of Chinese mutual funds with relatively high managerial turnover rates and applying AKM method, we precisely estimate the managerial heterogeneities of fund managers. Third, this paper contributes to the growing literature on how managerial attributes affect corporate outcomes. Bertrand and Schoar (2003) find significant manager fixed effects in corporate activities such as return on assets, investment, leverage, and cash holdings. Frank and Goyal (2007) provide some evidence that adding manager fixed effects to a regression analysis of the determinants of firm leverage significantly increases the model fit. Graham, Li and Qiu (2012) find that manager fixed effects explain a majority of variation in executive compensation. We extend this line of research to mutual fund performance evaluation. Last, we find that managerial turnover leads to a significant and structural change for flow performance sensitivity. Moreover, we show that the change in sensitivity depends on the characteristics of incoming managers. This provides supporting empirical evidence to a key assumption in Berk and Green (2004) that investors update their posteriors about managerial skills. Despite the importance of this assumption in 7
theoretical models, few empirical studies directly validate this assumption. The rest of the paper proceeds as follows. Section 2 reviews prior research and gives a brief introduction of China mutual fund industry. Section 3 discusses the estimation methodology. Section 4 describes our sample and data. Section 5 presents empirical results. Section 6 concludes. 2. Literature and Institution 2.1. The role of managers in mutual fund performance There are two views in the literature that offer insights into the issue of managerial role in fund performance. According to the first view, managers are considered as having no skill or talent and, thus, are regarded as homogeneous. Under this view, we would not expect individual managers to matter in terms of fund performance. In other words, two funds sharing similar technology, research support and organization structure will make similar investment decisions, whether or not they have the same manager. Consistent with the first view, numerous studies of fund performance, starting with Jensen (1968), have concluded that fund managers are unable to beat the market. Among recent studies, Wermers (2000) finds that the average U.S domestic equity funds underperformed its benchmark of overall market, size, book-to-market, and momentum by 1.2% per year between 1975 and 1994. Similarly, Fama and French (2010) document that only a few funds produce benchmark-adjusted expected returns sufficient to cover transaction costs. Researchers have also studied persistence in mutual fund performance. Using fund return as proxy for managerial skill, a number of papers have documented that performance is largely unpredictable (Gruber, 1996; Carhart, 1997; Bollen and Busse, 2001). Furthermore, the evidence that it does persist (Bollen and Busse, 2005; Mamaysky, Spiegel and Zhang, 2008; Berk and Tonks, 2007)), is generally found only in low-liquidity sectors or at shorter horizons, and is modest in magnitude. The lack of 8
long-term persistence is widely regarded as implying that superior performance is due to luck, rather than to variation in abilities across managers. The alternative view posits that fund decisions can be affected by managerial heterogeneity. Managers may differ in their unobserved personal characteristics, such as innate abilities, social network, personalities, effort, etc. Therefore, unobservable manager characteristics could account for a significant part of the variation in fund performance. By applying some parametric estimation and assuming some degree of investors beliefs about managerial skill, Baks (2003) finds that manager and fund strategy are equally important when determining the cross-sectional variations in fund performance. Kosowski et al. (2006) employ a bootstrap methodology and provide further evidence that a sizable minority of fund managers (about 10%) have skill. Berk and van Binsbergen (2012) show that average fund managers add the value about $2 million per year in mutual funds, suggesting the existence of fund managers abilities. Furthermore, recent studies that focus on fund manager attributes suggest a strong relation between managers characteristics and their performance. Golec (1996) finds that younger managers with MBA degree are associated with better performance. Chevalier and Ellison (1999) show that mutual fund managers who attended higher Scholastic Assessment Test (SAT) undergraduate institutions have systematically higher returns. Such education premium also presides for hedge funds (Li, Zhang and Zhao, 2011). Gottesman and Morey (2006) provide some evidence that the quality of a manager s MBA program is positively correlated with future fund performance. Cohen, Frazzini and Malloy (2008) documents that fund managers place larger bets on firms that they are connected to through their education network, and perform significantly better on these holdings relative to their non-connected holdings. Theoretical models in the mutual fund literature typically assume a significant degree of investor sophistication and learning ability. Berk and Green (2004) posit that investors can learn about the ability of fund managers through past performance and 9
allocate money to funds based on their Bayesian updates. Lynch and Musto (2003) and Huang, Wei and Yan (2007) provide alternative explanations for the convex flow-performance relation. In both papers, the conclusions are based on the premise that investors rationally learn about the managerial ability from past performance. However, according to Berk and Green (2004), it would be difficult to identify cross-sectional differences in fund managers using returns, because the competitive supply and redeployment of capital, combined with decreasing return to scales in assets management, drive the abnormal performance of all funds to zero. Therefore, managerial attributes are particularly important if investors can learn about manager abilities from various sources other than past fund performance. For example, investors may gather additional information about fund managers from media. 7 If managerial heterogeneity is related to innate ability or human capital and if investors, or at least some of them, are sophisticated Bayesians, we would expect that fund flows respond to manager fixed effects above and beyond the traditional fund performance measures. 2.2. Mutual fund industry in China In order to help stabilize the stock market and strengthen corporate governance, China government made a strategic decision in year 2000 to develop the mutual fund sector in the capital market. After China s first launch of an open-end fund, Hua An Chuang Xin, in September 2001, the mutual fund industry has become one of the fastest growing industries in the Chinese capital market. Figure 1 shows the total net asset value and the number of open-end mutual funds from 2001 to 2010. The Chinese mutual fund industry experienced dramatic growth from 2001 to 2007 with a slight decrease during the credit crisis after 2007. The total net assets under management by the end of 2010 are approximately US$ 374 billion, with an average annual growth rate of 116%. On the other hand, the number of open-end Chinese mutual funds has been increasing steadily since 2001, adding 70 new funds on average each year. By the end 7 Some Chinese fund managers are reported to have lots passion for luxurious products, great interest in gambling in Las Vegas and invest heavily in real estate market in China. For more details, see links below (http://www.chinaacc.com/new/207_434_201203/13ch383316588.shtml, http://finance.sina.com.cn/world/gjjj/20080711/18375083696.shtml, http://finance.qq.com/a/20100807/000864.htm). 10
of 2010, there are 637 open-end Chinese mutual funds. With the rapid growth, mutual funds play an important role in Chinese capital market. According to a recent report issued by Securities Association of China in 2010, the mutual fund market represents 11.06% of GDP, 7% of household deposits and 7.91% of the total market capitalization of A-shares in China. A survey conducted by China Securities Journal at the end of 2007 shows that 83% out of 14,800 respondents would pick mutual funds as the first choice for their wealth management. There are 34 million fund investment accounts at the end of 2010, of which 99% are held by individual investors. In China, all mutual funds are contractual-form funds whereby investors enter into a contract with a fund management firm (the sponsor and trustee) and unit holders are the beneficial owners of the fund (Lu, 2006). In contrast to the situation in the U.S., a mutual fund in China does not have a board of directors to act as a fund governance body and a buffer between the mutual fund investors and fund managers. Additionally, compared with the corporate-form mutual funds that prevail in the U.S., contractual-form mutual funds provide fewer voting rights for fund investors (i.e., unit holders), and are therefore more prone to agency problems, which are particularly acute in China, because of the weak and capricious legal system and high information asymmetry. The government also has a considerable influence on the mutual fund industry in China. First, the CSRC is the regulatory body that oversees mutual funds and their powers (vested by the Securities Investment Fund Law of 2003) are much greater than its U.S. counterpart (the SEC) in matters such as authorizing the setting up of fund management firms (Articles 13 and 14) and the launching of new funds (Article 40), approving the appointment of senior managers of fund management firms (Article 17), and revoking the license of a fund management firm when it deems necessary (Article 21). Second, the government directly or indirectly controls almost all mutual fund 11
management firms (i.e. they are often sponsored by state owned entities). While the CSRC has a lot of influence on the set-up and management of fund companies, it has less leverage on the actions of individual fund managers (Firth, Lin and Zou, 2010). Like their peers in the U.S., Chinese mutual fund managers (particularly, managers of open-end mutual funds) are under constant pressure to deliver good returns because (a) their management fees depend on investment performance and fund size and (b) the investors of open-end funds can choose to withdraw at any time and thereby create a market disciplinary mechanism for fund managers. Besides, because fund management firms are often in contract with owners, they have strong incentives to monitor their portfolio performance and to take action in order to protect their investments against erosion in value. The frequent turnover in fund managers is actually, to a large extent, an outcome of labor market competition and fund company s incentive structure. 3. Methodology Consider the following three-way fixed effects of mutual fund performance regression: R j, t xi ( j, t) t wj, t i( j, t) j t j, t (1) Funds are indexed j=1, 2,, J. Fund managers are indexed i=1, 2,, N. Managers can change funds over time, and the function i ( j, t) maps fund j to manager i at time t. The dependent variable is R j, t, which is fund j s net return in time t. x i j, t) t ( and w jt are vectors of observable manager-level and fund-level covariate. i( j, t) and j are vectors of unobservable manager-level and fund-level covariate (fixed effects). The third component t represents the unobserved time effect. j, t is the error term reflecting the abnormal fund performance that is not captured by manager or fund observed and unobserved contributions. We assume the error term is independent of manager s mobility decision. We can easily obtain consistent estimates of and by taking differences or by 12
time-demeaning within each unique combination of manager and fund (i.e. so called spell). The spell method uses the full sample and addresses the possible omitted variable bias, but it cannot separate the fund and manager heterogeneities. We can also include a dummy variable for each fund, and sweep out the manager heterogeneity by time-demeaning over i. Fund dummies are not different from any multi-category dummy so long as managers can move from one fund to another fund in the sample over time (i.e. movers). Bertrand and Schoar (2003) use this approach to examine whether unobserved managerial heterogeneity has power to explain return on assets, investment, leverage and cash holdings. One potential difficulty with the mover dummy variable approach is that movers may be significantly different from non-movers, resulting in selection bias and limiting the generalizability of the result. Furthermore, the sample that can be studied is usually quite small due to infrequent managerial turnovers. Or, in case of a sample with large number of funds, this method may be computationally infeasible as it requires inverting a covariate matrix with too many dummy variables. The alternative method we use to separate managerial heterogeneity from fund heterogeneity is based on Abowd, Kramarz and Margolis (1999). The AKM method allows us to separate manager fixed effects ( ) and fund fixed effects ( ) for departing managers, as well as retained managers who work in the same funds with departing managers. The identification is implemented through group connection. When a group of managers and funds are connected, the group contains all the managers who ever worked for any of the funds in the group and all the funds at which any of the managers were ever employed. Abowd, Kramarz and Margolis (1999) formally prove that the connectedness is necessary and sufficient for the separate identification of manager and firm heterogeneities in a group of connected S&P 1500 firms. From a statistical perspective, connected groups of managers and funds block-diagonalize the estimation equation or matrix (the columns associated with non-connected groups are formalized to zero) and permit the precise statement of identification restrictions on the 13
manager and fund fixed effects. 4. Data Our mutual fund dataset covers all open-end mutual funds from year 2003 to 2010. We draw data from two sources. One is China Stock Market Accounting Research (CSMAR) database, which provides fund characteristics and fund manager data. Another is Wind Financial Terminal database, which contains monthly fund returns and net assets. In addition, we use the web search engine to fill some missing data such as fund manager age and identify whether fund managers get promotion or demotion after their turnover. In Table 1, we present the summary statistics of fund type and manager mobility in our sample. Our full sample includes 722 unique managers who have worked for 637 unique mutual funds during the period from 2003 to 2010. Panel A shows the distribution of funds across assets class. There are 326 equity funds, 169 balanced funds and 142 bond funds. At the annual horizon, the percentage of funds experiencing manager turnover ranges from 31.2% to 49.8%, with an average of 37.5%. Whereas, 106 funds do not have any managers moving across funds, the remaining 83.4% (531 funds) have managers switching at least once during the sample period. Using AKM estimation, we are able to identify the fixed effects for all managers who are or were in these 531 funds, no matter whether these managers move or not. Out of 722 managers, half of them (50.7%) have switched funds during the sample period. The remaining 49.3% of fund managers work in only one fund. For movers, about 31% of managers work for two different funds, and the remaining 19.6% of managers work in more than two different funds. By connecting the managers to 531 funds experiencing manager turnover, our connected sample (or the sample on which we are able to perform the AKM approach) includes 488 (or 67.6% of the population) unique managers who have worked for those 531 funds. 14
In Table 2, we examine the basic characteristics of funds and managers in our sample. We first investigate whether the connected sample that includes only funds with movers is representative of the original fund sample. Panel A summarizes the variables at the fund level, whereas Panel B includes all observable manager characteristics. TNA denotes total net assets of fund. Fund Age is the number of years since the fund inception. Expense refers to total annual management and administrative expenses divided by TNA, but Management Fee only refers to management expenses. Purchase and Redemption represent front-end and back-end load charges, respectively. Turnover is the minimum of annual aggregate purchase and sale of securities, divided by TNA. Volatility is measured as the standard deviation of daily fund return in one month. We use a dummy variable to indicate whether the fund is team managed (Team=1) or otherwise (Team=0). Top10 is also a dummy variable, equal to one if the fund family is among the top ten fund management firms based on assets under management. We employ two fund performance measures: raw return and benchmark-adjusted return. Benchmark is defined as the average performance of all funds in the same Morningstar Style Box. We construct two fund flow measures: net flows and benchmark-adjusted flows by subtracting the average flows of all funds in the same Morningstar Style Box. Net flow is defined as [ TNAt TNAt Flow TNA 1 (1 Rt )] where TNA t denotes total net assets at the end of each month t and R t denotes the fund return in month t. t 1 (2) At the manager level, tenure is defined as the number of years that a fund manager stays in the current fund. We also collect manager age, although about one-third of the sample has missing data. Other observable manager characteristics include manager gender and education degree, which will be in one of the following four types: community college, undergraduate, master and Ph.D. As politics and network are much more important in emerging markets, we also classify a fund manager as connected (Connection=1) if he or she has working experience in the government. The detailed 15
definitions of variables are presented in Appendix. First, let us focus on the full sample in Panel A. The average fund size is about 625 million U.S. dollars, with a little over a three-year history. The expense ratio is 3.06%, of which 1.28% is the management fee. The sales charge, including both front-end load and back-end load, is 1.72%. About one third of funds in our sample are team-managed. The top 10 fund management companies account for about one-third market share. Monthly raw return is 0.86% after fee and -0.2% by subtracting the benchmark return, which is the average fund performance in the same Morningstar Style Box. Monthly net flow is 6.35%, and -2.76% after subtracting the average flow of all funds in the same Morningstar Style Box. The average manager tenure is about 1.56 years, indicating the high manager turnover in Chinese mutual funds. The average fund manager age is 36 years old. Nine out of ten fund managers are male. The majority (92%) of fund managers have a master s degree or higher. It is not common for fund managers to have political connection. Only 4% of them are connected to government, indicating that they worked in the government before they took the position of fund manager. By comparing the connected sample to the full sample, it seems that the connected sample is fairly representative of the full sample in all dimensions except that the connected sample funds are somewhat smaller than the full sample funds. On average, connected funds have US$ 40 million less assets than those in full sample. As it is likely that the difference in size may also be related to the difference in fund performance, we control for fund size in all our subsequent regression analyses. 5. Results 5.1. Estimating manager fixed effects by AKM approach With detailed fund manager s career path, combined with monthly fund performance data and other fund characteristics, we are able to estimate the three-way fixed effect model in Equation (1) through the AKM method. We use benchmark-adjusted return on 16
the left-hand side of the regression to address the distinct nature of underlying assets across fund types. The benchmark is defined as the average performance for all funds in the same Morningstar stylebox classification. In short, we estimate for each of 488 fund managers in our connected sample. Panel A of Table 3 presents the summary statistics on manager fixed effects. Prior researchers have argued that mutual funds with different underlying assets might require different skill or ability (Deli, 2002). Accordingly, we examine manager fixed effects in three different groups, namely equity funds, balanced funds and bond funds. The estimate of fixed effect for equity fund managers is 0.003, with a standard deviation of 0.028. The null hypothesis that the fixed effect is not different from zero cannot be rejected at the 10% level. However, there exists a substantial dispersion in heterogeneities among equity managers as indicated by the large number of standard deviation. Since an index fund is mainly holding a portfolio of passive benchmark (i.e. Shanghai Composite Index), one might suspect that index fund managers should not possess any valuable heterogeneities. Accordingly, we further divide equity funds into index funds and non-index funds. There are 68 index funds and 189 non-index funds in our sample. The estimate of manager fixed effects for non-index funds manages is 0.003 and significant at the 10% level, but the fixed effect for index funds managers is close to zero. The estimate of manager fixed effects for balanced funds is -0.001, which is not significant from zero. On the other hand, bond fund managers seem to have a much lower fixed effect. The mean is -0.025, which is significantly with a t-statistic of -8.91. Finally, the average estimate of manager fixed effects for all fund managers is -0.003 and significant at the 5% level. This is mainly caused by the low manager fixed effects for bond fund managers. To directly test whether the fixed effects for fund managers are different from zero might be problematic, because the mean and the location of the estimated fixed effects may change when different benchmarks are used to measure fund performance. Alternatively, we compare the estimates of fixed effects across different assets classes 17
and show the results in Panel B of Table 3. Among equity funds, we find that the estimate of fixed effects of index fund managers is not different from that of non-index fund managers (p-value=0.33). Hence, the difference in managerial heterogeneity between active-managed fund managers and index fund managers is economically small and insignificant. This is consistent with existing literature that active management on average does not provide investors better returns than passive benchmark. On the other hand, bond fund managers seem to possess much lower fixed effects than either equity or balanced fund managers. This could be due to the fact that bond investment is so different from equity investment, and therefore the required human capital may be distinct in nature. Figure 2 presents the distribution of manager fixed effects for the connected sample. The graph suggests that the estimated manager fixed effects are very close to a normal distribution. To the extent that manager fixed effects can be interpreted as a indicator of time invariant manager attributes, the magnitude of standard deviation suggests that there is a fair amount of variations in heterogeneity among fund managers (Jones and Wermers, 2011; Wang, 2011). Identifying managers with superior performance through an undocumented channel could have profound welfare implication for investors. 5.2. Managerial heterogeneities and fund performance In this section, we analyze how managerial heterogeneities are related to fund performance by comparing the coefficient estimates in the ordinary least squares (OLS) model with those in the three-way fixed effects model. We use two measures of fund performance, raw return and benchmark-adjusted return. We follow prior research to control for other fund and manager characteristics that determine the fund performance. Panel A of Table 4 shows the regression results of raw return. Column (1) is pooled OLS regression without controlling for either fund or manager fixed effects. We include manager fixed effects in column (2) and fund fixed effects in column (3). 18
Column (4) adopts the three-way fixed effect method by including both firm and manager fixed effects. An inspection of column (1) and (4) in Panel A reveals that the signs of the coefficients are similar for both models with some notable exceptions. For instance, fund size is negatively related to fund performance with the coefficient of -0.133 and -2.523 in the OLS and the three-way fixed effect model, respectively. The negative impact of fund size on performance is consistent with the notion of diseconomy of scale in assets management suggested by Chen et al. (2001) and Berk and Green (2004). However, the magnitude of such impact is much larger in three-way fixed effect regression. This result is particularly interesting in light of recent theory developed by Gabaix and Landier (2008), who argue that better skilled managers match with larger firms. To the extent that managerial skill contributes positively to the fund performance followed by new investors (Sirri and Tufano, 1998), additional capitals can lead to a subsequent larger fund size. In other words, fund size has implicitly incorporate two competing factors: diseconomy of scale and managerial skill. The latter is manager specific and will be absorbed at least partially into manager fixed effects. As a result, after controlling for manager fixed effects in the regression, the diseconomy of sale becomes more significant. For a similar reason, fund expense has a negative rather than positive impact on fund performance after we control for manager fixed effects. Presumably, fund expenses could capture both managerial skills (i.e. good managers charge higher fees) and management entrenchment. We also find that the coefficient of fund age is significantly negative in the OLS regression, but insignificantly in the three-way fixed effect regression. It is consistent with the phenomenon that newly set-up funds opt to employ star managers to attract flows and these managers usually have higher fixed effects. After controlling for this, newly set-up funds do not present better performance. The above discussion suggests that including manager fixed effects changes both the magnitude and interpretation of the estimated coefficients on the observable fund and manager characteristics that might be attributed to fund performance. The findings are robust to alternative fund performance measures. The dependent variable in Panel B of Table 4 is benchmark-adjusted return. The results are qualitatively similar to those in Panel A. 19
We next explore the relative economic importance of managerial and fund heterogeneities. We employ the AKM method on the connected sample to separate out the following components: observable time-variant manager characteristics ( x ˆ ), i( j, t) t observable time-variant fund characteristics ( w jt ˆ ), manager fixed effects ( ˆ i( j, t) ), fund fixed effects ( ˆ ), year effects ( ˆ ), and residuals ( j t ˆ j, t ). We examine how much each of these components contributes to the total variation in fund performance, using the covariance between fund return and each of the components, normalized by the variance of fund return. 8 After applying the AKM method, we find that the fractions of the model sum of squares attributed to observables fund and manager characteristics, manager fixed effects, fund fixed effects, year fixed effects are 72%, 8%, 4%, 6%, respectively. It is not surprising that the observable fund and manager characteristics capture most of the variations in fund performance, because a lot of those attributes are implicitly related to unobservable fund and manager heterogeneities. However, after we isolate manager fixed effects separately, we still find that a significant amount of variation in fund performance beyond what is explained by existing determinants can be contributed to unobservable human capital. While the entire mutual fund industry has become more competitive (Wahal and Wang, 2011), picking mutual funds with higher manager fixed effects might be of particularly important to individual investors (Jones and Wermers, 8 Note that model R-squared is calculated as cov( Rˆ, ) 2 jt R jt R var( R jt ) cov( x ˆ ˆ ˆ ˆ i( j, t) t w jt ˆ i( j, t) j t, R jt ) var( R jt ) cov( x ˆ, ) cov( ˆ, ) cov( ˆ, ) cov( ˆ, ) cov( ˆ i( j, t) t R jt w jt R jt i( j, t) R jt j R jt t, R jt ) var( R jt ) var( R jt ) var( R jt ) var( R jt ) var( R jt ) where R jt is the dependent variable fund performance. Therefore, the normalized covariance may be interpreted as a decomposition of model R-squared. 20
2011). 5.3. Predictability of managerial heterogeneities If managerial heterogeneity is related to human capital and innate ability, we would expect managers with higher fixed effects to outperform those with lower fixed effects. To test that, we perform the out-of-sample test for the predictability of manager fixed effects. To be more concrete, in each month between year 2008 and 2010 (i.e. the last three years in our sample period), we estimate the manager fixed effects by AKM method, using a rolling estimation window including all the fund and manager data up to the previous month (i.e., to predict the fund return in June of 2008, we use all the data from January of 2003 to May of 2008 to estimate the manager fixed effects by the AKM method). This allows us to use as many data observations as possible to estimate manager fixed effects with maximum precision. We then run a Fama-BacMeth cross-sectional regression in each month to explore the extent to which the variations of subsequent annual performance can be determined by the estimated managerial heterogeneities. 9 More specifically, we will run the following performance regression in the testing period. R j, t ˆ i( j, t 1) xi( j, t), t wj, t j, t 1 t j, t (3) The time-serious averages of the coefficients are presented in Table 5. The standard errors are controlled for serial correlation and heteroscedasticity. In column (1), we only include fund and manager time-variant characteristics. 10 The results are similar to those found in the previous literature. For example, funds with higher expense ratios underperform their counterparts. Funds that trade more frequently underperform their counterparts. Funds with single manager perform better than team-managed funds. 9 This is in a similar spirit of sorting managers based on their fixed effects and examining the subsequent fund performance. However, in a multivariate regression, we are able to control for other observable manager and fund characteristics that are related to fund performance. 10 We do not include time-invariant fund and manager variables (for example, top10, gender, education, etc.) in the model, because it facilitates the comparison with other models in the same table that have controlled for manager and fund fixed effects. 21
While the performance of old funds is worse than that of young funds, fund manager tenure is positively related to fund returns. More interestingly, past fund performance seems to have some predictory power on subsequent performance. Given that the coefficient on past performance is 0.128 and significant at the one percent level, it suggests that previous winners are more likely to become future winners. To test the predictability of managerial heterogeneities for future performance, we include manager fixed effects in column (2). The coefficient on manager fixed effects is 3.289 and significant at one percent level, suggesting that higher manager fixed effects lead to better fund performance in the future. One may argue that since past fund performance should also reflect managerial skill, it might be enough to just control for past return. To address this concern, we include both manager fixed effects and past fund returns in column (3). The coefficient on manager fixed effects remains positive and significant; indicating that even after controlling for past return, manager fixed effects can still predict future fund performance. Interestingly, we find that the coefficient on past return is no longer positive, suggesting that historical performance no longer has predictory power for future fund return after controlling for manager fixed effects. Finally, we include other manager and fund attributes that might be related to fund performance in column (4). After we control for all the relevant determinants for future fund performance, manager fixed effects remain positive and significant. In terms of economic magnitude, a standard deviation increase in manager fixed effects leads to an increase of annual return by 2%, which is quite substantial compared to the average annual fund return of 10.2% for our sample. Other control variables present similar results to column (1) except fund flow. Negative coefficient reflects the diseconomy of scale caused by large flows emerges after controlling for manager fixed effects, which might be captured by fund flow. To summarize, the evidence here suggests that managerial heterogeneities are related to 22
manager s human capital, or innate ability, and hence it is persistent and has some predictory power for future fund performance beyond the usual suspicious elements. 5.4. The correlation between observable and unobservable managerial attributes If the estimate of manager fixed effects is proxy for human capital or innate ability, we would expect the estimate to be correlated with the observable characteristics of funds and managers in a meaningful way. Table 6 presents the Pearson correlation coefficients. On average, managers with higher fixed effects tend to work in bigger funds. This is consistent with the notion that fund size can be an indirect measure of managerial skill (Berk and Green, 2004). There is a positive relationship between management fees and manager fixed effects. Recall that manager fixed effects are positively associated with future performance; it suggests that managers with higher marginal productivity are able to extract higher rents from investors. Moreover, funds employing managers with higher fixed effects charge higher fees for front-end purchase and back-end redemption. Another interesting observation is that managers with higher fixed effects tend to trade more (higher turnover) and their fund returns are more volatile (higher volatility). Again, assuming manager fixed effects are related to skill, the turnover pattern suggests that those managers are more likely to exploit mispricing by trading more (Pastor and Stambaugh, 2002). We also find that manager fixed effects are negatively correlated with fund age. This might be due to the fact that young funds often hire some star managers to attract flows. For example, ICBCCS Core Value Equity Fund employed the famous fund manager Hui Jiang when it was set up in 2005, and Bank of Communications Schroder Fund employed the famous fund manager Xuli Li when it was set up in 2006. 11 Finally, managers with higher fixed effects are more likely to be 11 For the detail, see the link http://business.sohu.com/20100305/n270608807.shtml. 23