Mutual Fund Stars': The Performance and Behavior of U.S. Fund ...


Published on

Published in: Economy & Finance, Business
  • Be the first to comment

  • Be the first to like this

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide

Mutual Fund Stars': The Performance and Behavior of U.S. Fund ...

  1. 1. Mutual Fund ‘Stars’: The Performance and Behavior of U.S. Fund Managers Bill Ding Department of Finance Leeds School of Business University of Colorado at Boulder Boulder, CO 80309 Russ Wermers Department of Finance Robert H. Smith School of Business University of Maryland at College Park College Park, MD 20742-1815 Phone: (301) 405-0572 July 2002 Web address: Wermers gratefully acknowledges research support from INQUIRE-UK for this project. Also, we thank Morningstar and Thomson Wiesenberger for providing mutual fund manager data.
  2. 2. Mutual Fund ‘Stars’: The Performance and Behavior of U.S. Fund Managers Abstract Do mutual fund “star” managers exist? Past studies of mutual fund performance have ignored the role of managers in the performance of funds. Our study assembles the most complete database of U.S. fund managers to date, and merges this manager database with a comprehensive database of fund stockholdings, net returns, and other characteristics. This merged database allows us to investigate several issues related to whether talent resides at the manager level, including the role of managerial experience and stockpicking track-record in predicting the future performance of a manager—this unique database, which extends from 1985 to 2000, allows the creation of several new measures that provide insights into these issues. We Þnd that experience matters, but only for growth-oriented fund managers—the most expe- rienced growth managers have substantially better stockpicking skills than their less-experienced colleagues. We also show that the career stockpicking track-record of a fund manager holds the most signiÞcant predictive power for future fund manager performance—managers with the best career records choose portfolios that beat their style benchmarks by almost two percent per year, while managers with the worst records provide an insigniÞcant level of performance—thus, manage- rial talent strongly persists. Finally, we Þnd that the replacement of a fund manager, for whatever reason, has an impact on fund performance, but only because the new manager has a substantially better track record than the replaced manager. During the year of replacement, fund managers underperform their counterparts by about one percent per year—this underperformance vanishes af- ter the manager is replaced. Our results add several new insights to the mutual fund performance and performance persistence literature by highlighting the role of the manager in generating fund performance.
  3. 3. I Introduction A good deal of attention is focused on professionals who manage money, in the form of television interviews, best-selling books, and frequent articles in the popular press. The media often focuses on the investment results of a few “star” mutual fund managers, such as Bill Miller of the Legg- Mason Value Trust Fund or Scott Schoelzel of the Janus 20 Fund. The implication of the media spotlight on star managers is that experienced managers, or managers with a good track-record, outperform other managers in addition to passively managed funds on a consistent basis. However, do star fund managers really exist? Over the past few decades, several papers have analyzed the performance of mutual funds, ignoring (in general) the role of the fund manager. Overall, these studies Þnd that, adjusted for the return premia earned from loading either on the overall stock market (relative to Þxed- income investments), or on certain equity “style characteristics,” mutual funds have provided a slightly negative level of abnormal returns. Examples of papers that examine this issue using net returns data include Malkiel (1995) and Carhart (1997), and examples that examine the issue using stockholdings data include Grinblatt and Titman (1989, 1993) and Wermers (2000). These papers indicate that our view of the average mutual fund’s performance depends on whether style investing represents a systematic risk: if we do not deduct the return premia for loading on style characteristics, the average manager, rather than exhibiting a negative abnormal return, exhibits an abnormal return of about zero. However, if talent resides at the manager level rather than at the fund level, then all prior tests may lack power in detecting stockpicking ability. With a couple exceptions, prior work has not considered the role of the fund manager in generating fund performance. Our paper provides fresh evidence on the role of managers in both the characteristics and the performance of mutual funds. Chevalier and Ellison (1999), using a sample of mutual funds over a short time period, are the Þrst to analyze the impact of experience on fund performance. Baks (2001) examines managers in the CRSP Mutual Fund database over the 1992 to 1999 period to separate the impact of the fund manager from the impact of the non-manager characteristics of a fund on the fund’s performance. Our paper contributes to the literature in several ways. First, our manager database covers the 1985 to 2000 period, which is the longest time-series of manager data assembled to date. The manager data is compiled from several sources, and includes basic information about a manager, 1
  4. 4. such as the starting and ending dates of the manager with each fund managed over her career. Second, we merge the manager database with an updated version of the merged Thomson/CDA mutual fund stockholdings and CRSP mutual fund net returns and characteristics database that is Þrst examined in Wermers (2000).1 And, third, the nature of our merged database allows us to design several new measures of manager and fund characteristics, such as the career stockpicking record of a manager and the level of “style drift” experienced by a fund. These new measures provide us with the ability to investigate several determinants of manager and fund characteristics, which, in turn, allow us to measure the correlates of these characteristics with fund performance and performance persistence. We study three basic issues in this paper to determine whether mutual fund star managers exist. First, we examine whether the experience of a fund manager, over her entire career, has any impact on the performance of the fund. There are several reasons why we may believe that seasoned fund managers have superior talents—these reasons include the increasing ability of the fund manager to interpret the research provided by internal and external stock analysts as well as the increasing access that fund managers may gain to corporate managers as the fund managers’ careers progress.2 Second, we measure the past stockpicking record of a fund manager to investigate whether man- agers with past success have persistent stockpicking skills, independent of their level of experience at a certain date. And, third, we examine mutual fund performance during the time surrounding the replacement of a manager, which provides a sharp test of whether stockpicking talent resides at the manager level. Our results provide several interesting insights. First, we Þnd that managerial experience is an important predictor of future stockpicking success for growth-oriented fund managers, but not for income-oriented managers. This Þnding indicates that experience is important for success in picking growth stocks, perhaps because of the difficulty in accurately forecasting earnings growth for these stocks, relative to value stocks. Growth-oriented managers may either develop specialized 1 This merged database, along with our new manager database, provides several advantages over past work. For example, we are able to more precisely measure the stockpicking talents of managers by using portfolio holdings data— these data also allow us to provide a complete attribution analysis for each mutual fund, before and after trading costs and other fund expenses. 2 This increasing access to corporate managers may result from several inßuences, including an increase in the size of positions in stocks that may result as a result of seasoned managers taking on the responsibility for larger funds. In addition, a relationship with a corporate manager may develop over time, as the fund manager potentially becomes a “long-term” shareholder of the Þrm. Regulation FD, implemented by the SEC to prevent an information advantage for institutions and other shareholders, was not in effect during the majority of our sample period. 2
  5. 5. skills over time, or, alternatively, they develop valuable relationships with corporate managers that give them access to private information on future earnings. Second, we Þnd that the past stockpicking track record of a fund manager is the most important predictor of the future performance of the fund. Managers with the best past stockpicking records outperform those with the worst records by almost two percent per year, even though these “star” managers do not have appreciably greater experience levels than their counterparts. SpeciÞcally, the signiÞcance of the track record variable remains strong when experience is added in a multivariate regression setting. This Þnding indicates that managerial talent persists for multiple-year periods, which is consistent with the Þndings of Wermers (2002b). We also Þnd that the replacement of a manager has a substantial effect on fund performance, but only because the new manager has a substantially better track record than the replaced manager. While the pre-replacement benchmark-adjusted return of a fund (before expenses and trading costs) is reliably lower than that of other funds, this difference vanishes after the manager is replaced. Thus, our paper indicates that managerial talent does persist over long time periods, and that the labor market for fund managers appears to work efficiently by replacing managers when their stockpicking talents have Þnally faded. Our Þnal tests look at the role of managerial aversion to risk in explaining fund performance. SpeciÞcally, we add two proxies for managerial risk aversion to determine whether managers who trade more aggressively on their private information exhibit a level of performance different from other managers. Adding these two proxies to the multivariate setting above, we Þnd no evidence that risk aversion inßuences future performance. This Þnding indicates that, if risk-aversion matters in generating portfolio performance, it is highly correlated with the other variables in our regressions (i.e., experience or track-record). The remainder of this paper is organized in four sections. The construction of our database is discussed in Section II, while our measures of manager characteristics and fund performance and costs are outlined in Section III. Section IV presents empirical Þndings. We conclude the paper in Section V. II Data Our mutual fund characteristics data is drawn from the merged CDA—CRSP mutual fund database of Wermers (2000). For each U.S. equity fund portfolio that exists anytime between January 1975 3
  6. 6. and December 2000, CDA—CRSP contains data on various fund statistics, such as the monthly net return, total net assets, annual expense ratio, annual turnover ratio, and quarterly stock hold- ings of each fund. This database is the longest time-series having both stockholdings and net returns/characteristics information that has been assembled to date. See Wermers (2000) for more information on the construction and limitations of an earlier version of this database. We merge the CDA-CRSP database with a newly constructed database of mutual fund managers that covers the 1985 to 2000 (inclusive) time period. In constructing our database of managers, we focus on U.S. equity funds, that is, funds having a self-declared investment objective of aggressive growth (AG), growth (G), growth and income (GI), income or balanced (I or B) at the beginning of a given calendar quarter. The fund manager data is assembled from three separate sources of manager data: the 2001 Morningstar Principia Pro database, the CRSP Survivor-Bias Free Mutual Fund Database, and a database of fund managers that was purchased from Thomson/Wiesenberger in 1999. We combine the fund manager data from these three sources based on the manager’s name and the name of the managed fund to ensure that we create a manager database that is as complete as possible.3 SpeciÞcally, for each fund manager, we collect her name, the names of funds managed by her during her career, the start and end dates for that manager at each fund over her career, and other manager characteristics, including CFA designation, universities attended, prior analyst experience, and other items such as marital status and personal interests. The fund manager data are then matched with the CDA—CRSP database of portfolio holdings, net returns, and fund characteristics. In conducting our study, we focus our attention on the lead manager of each mutual fund, assuming that this manager has the greatest decision-making power for that fund. As a proxy to identify the lead manager, we choose the manager with the longest tenure at a given fund (if team managed) to decide on which manager is the lead manager.4 Counts of our sample of lead managers over the entire 1985 to 2000 period, as well as counts at the beginning of 1985, 1990, 1995, and 2000 are presented in Table I. There are a total of 2,272 CDA—CRSP funds and 2,229 lead managers in our sample. Growth funds account for the majority of the fund universe, and about 80% of the fund managers have experience in managing at least one 3 We note that in some (rare) cases there are inconsistencies in the manager’s Þrst name abbreviation (e.g. Robert and Bob) and name suffix (e.g. none vs. Jr.) among the three fund manager data sources. In these cases, we use other information, such as the historical fund manager name, managed fund name, and start and end dates to ensure the accuracy of matching. 4 If there is tie in the start date, we use the total career experience as the tie-breaker, i.e., we pick the currently active fund manager who becomes a fund manager (of any fund) at the earliest date. 4
  7. 7. growth fund during 1985 to 2000. Not surprisingly, the number of funds and fund managers grows rapidly with the expansion of the whole fund industry in our sample period. The average number of funds lead-managed by a given fund manager increases gradually from 1.27 at the beginning of 1985 to 1.57 at the beginning of 2000. To check the completeness of our matched manager/fund database, we further examine the CDA—CRSP funds that fail to be matched with any fund manager, and report the results in panels C and D of Table I. Overall, we are able to identify at the lead manager for almost 94 percent of funds in our CDA—CRSP database. In addition, more than 85 percent of all fund-months during 1985 to 2000 in the merged CDA—CRSP database contain information about the lead manager. A close look at the number of missing managers at four different points in time reveals more detailed information. Thirty-three percent of the funds that exist at the beginning of 1985 are unable to be matched with a manager during 1985, but this fraction steadily declines over our sample period to 6.1 percent and 4.8 percent during 1995 and 2000, respectively. One reason that post-1995 manager data is noticeably more complete than pre-1995 data is that our data sources, in general, begin to formally collect manager data in the Þrst half of the 1990s, and probably backÞlled previous manager data. In Panel D, a further comparison is provided between funds with complete manager data and funds that have missing manager data. This panel presents data on the total net assets under management and the net return, in excess of the S&P 500 index return, between funds having manager data and funds with missing manager data at the beginning of each Þve-year period, as well as for the entire sample period of 1985 to 2000. We Þnd that funds with missing manager data tend to be smaller and perform somewhat worse than those with complete manager data. III Methodology A Measures of Manager Characteristics Since the fund manager is the unit of analysis for our study, we construct measures that quantify various manager characteristics, such as experience, track record in picking stocks, attitude toward risk-taking, and aggressiveness in trading stocks. The richness of our fund characteristics and port- folio holdings data available from CDA—CRSP allow us to design several measures that accurately capture these proxies for various attributes that might be associated with superior stockpicking 5
  8. 8. skills. In this subsection, we describe these measures, and then present summary statistics on the measures over the sample period. The Þrst manager characteristic of interest is experience, which we is simply deÞned as the total number of months that an individual has served as a fund manager over her entire career. To capture the track record of a fund manager, we develop three measures. The Þrst track record variable is the time-series average of monthly net return in excess of the S&P500 index return, or t 1 X i T rackRecordi = 1,t S&P (Rτ − Rτ 500 ) (1) t − ti 0 i τ =t0 where ti is the month at which manager i Þrst becomes a lead manager for any fund. Ri and 0 τ S&P Rτ 500 are fund i’s return and the S&P500 index return for month τ , respectively. We choose the S&P 500 index as our Þrst benchmark, since this benchmark is the most common one used by the U.S. fund industry. The second measure that we use to proxy for the track record of a fund manager is the time- series average of monthly investment objective-adjusted returns, which is deÞned as the manager’s net return minus the average return of all funds with the same investment objective as the managed fund during the same time period. This measure for manager i at month t is t 1 X i T rackRecordi = 2,t (Rτ − RK ) τ (2) t − ti 0 i τ =t0 where K is the investment objective of fund i and RK is the average return across all funds with τ objective K at month τ . The rationale of using the average investment objective return as a second benchmark is that managers likely have an incentive to outperform their peer funds, regardless of their performance relative to the S&P 500 index. The third track record that we use is the stockpicking talent of the fund manager, as de- Þned by the Characteristic Selectivity measure of Daniel, Grinblatt, Titman, and Wermers (1997) (henceforth, DGTW), where mutual fund performance is evaluated against characteristic-based benchmarks. SpeciÞcally, we use the time-series average of a manager’s Characteristic Selectivity (CS) measure (henceforth, CS measure), over the entire career of the manager, to measure the manager’s track record in picking stocks. The CS track record measure (CST ) for manager i at 6
  9. 9. month t is calculated as t Jτ 1 XX b CSTti = wj,τ (Rj,τ − Rτj,τ ) (3) t − ti 0 i j=1 τ =t0 where wj,τ is manager i’s portfolio weight on stock j at the end of the calendar quarter just b preceding month τ ; Rj,τ is the month τ return of stock j; Rτj,τ is the month τ return of stock j’s characteristic-matched portfolio (matched on market capitalization, the ratio of book-equity to market-equity, and the prior one-year return on stocks); Jτ indicates the number of stocks held in the fund(s) managed by manager i at the end of the quarter preceding month τ . An advantage of the CS measure is that it uses portfolio holdings information, which DGTW argue provides a more precise measurement of performance relative to regression-based methods. Further information on the construction of this measure is given in the next section, when we further describe this measure. A manager’s risk attitude may determine her choice of stocks to hold in the managed fund portfolio, and, thus, may affect fund performance. In some cases, managers may take on, or avoid, risk in response to labor-market incentives (see, for example, Chevalier and Ellison (1997) or Brown, Harlow, and Starks (1996)). The measures we use to characterize a fund manager’s risk attitude are, respectively, the standard deviation of her monthly excess return and the standard deviation of her monthly investment objective-adjusted return, i.e.,  1 2 t X  1  RiskAttitudei 1,t = i (Rτ − RS&P 500 τ − T rackRecordi )2  1,t (4) t − ti 0 τ =ti 0  1 2 t X  1  RiskAttitudei 2,t = i (Rτ K − Rτ − T rackRecordi )2  2,t (5) t − ti 0 τ =ti 0 Some managers may be more aggressive in trading stocks than others, perhaps because they have better private information about stock values than others, because they believe they have superior stock-picking skills (perhaps due to overconÞdence), or because they are simply less risk- averse than other fund managers in using their private information. We would believe that such aggressiveness would lead to higher trading frequency and volume. As such, a manager’s aggressive- ness in managing her portfolio is measured as the time-series average turnover ratio of the fund(s) 7
  10. 10. managed by her.5 The expression for the aggressiveness of manager i through month t is t 1 X Aggressivenessi = t T U RN OV ERi . τ (6) t − ti 0 i τ =t0 Our Þnal manager characteristic measure captures the tendency of the manager to shift between different equity investing styles through time. For example, some managers may be less focused in their style of investing, and believe that they can Þnd underpriced stocks in several different style categories (“bottom-up” investing). Alternatively, Brown, Harlow, and Starks (1996) show that managers that underperform their peers during early periods may later move to investments that are more risky in order to attempt to “catch up” to their peers. This risk-taking behavior may involve shifting to a different style category, relative to the manager’s peers. In any case, we measure the tendency of the manager to actively shift between different styles with X¯ ¯ 3 ¯ i ¯ ActiveStyleDrif ti t = ¯dj,t ¯ , (7) j=1 where di equals the drift of manager i in style dimension j (j=size, value/growth, or momen- j,t tum/contrarian) during year t. To measure the active drift in a given style dimension, we measure the difference in the portfolio-weighted style number between the current portfolio, at the end of June of year t, and that of the portfolio that would have resulted, had the manager passively held the prior-year’s June 30th portfolio. Thus, the active style drift (ASD) measure captures move- ments in style that are solely due to active trades during the year ending on June 30th. Following Wermers (2002a) and DGTW, we use a non-parametric characterization of each stock in three dimensions: the market capitalization, the ratio of the industry-normalized book-equity to market- equity, and the prior-year return of the stock. Further details on the assignment of style dimension numbers to each stock during each year are provided in Section B.1 below, as well as in DGTW. Details on the computation of the style drift of funds is given in Wermers (2002a). The sum of the absolute values of the active drift in each style dimension, during a year t, is our measure of the active style drift that results from a manager’s actions during that year. 5 The annual turnover ratio of a fund is deÞned, by CRSP, as the lesser of securities purchased and sold, divided by average monthly total net assets during the year. 8
  11. 11. B Measures of Mutual Fund Performance and Costs In this study, we use several measures that quantify the ability of a mutual fund manager to choose stocks, as well as to generate superior performance at the net return level. These measures, in general, decompose the return of the stocks held by a mutual fund into several components in order to both benchmark the stock portfolio and to provide a performance attribution for the fund. The measures used to decompose fund returns include: 1. the portfolio-weighted return on stocks currently held by the fund, in excess of returns (during the same time period) on matched control portfolios having the same style characteristics (selectivity) 2. the execution costs incurred by the fund 3. the expense ratio charged by the fund 4. the net returns to investors in the fund 5. the benchmark-adjusted net returns of the fund. The Þrst component, which measures the style-adjusted return of a given mutual fund before any trading costs or expenses are considered, is brießy described next.6,7 We estimate the execution costs of each mutual fund during each quarter by applying recent research on institutional trading costs to our stockholdings data—we also describe this procedure below. Data on expense ratios and net returns are obtained directly from the merged CDA-CRSP mutual fund database. Finally, we describe the Carhart (1997) and Ferson and Schadt (1996) regression-based performance measures, which we use to benchmark-adjust net returns. B.1 The Characteristic Selectivity Measure The Þrst component of performance measures the stock-picking ability of the fund manager dur- ing a given month, controlling for the particular style used by that manager.8 This measure of 6 This measure is developed in Daniel, Grinblatt, Titman, and Wermers (1997), and is more fully described there. In that paper, the authors argue that decomposing performance with the use of benchmark portfolios matched to stocks on the basis of the size, book-to-market, and prior-year return characteristics of the stocks is a more precise method of controlling for style-based returns than the method of decomposing performance with factor-based regression techniques that is used by Carhart (1997). 7 Due to the limited frequency (usually quarterly) of our holdings database, this component of performance assumes that a fund manager holds a portfolio (a buy-and-hold strategy) from the date of the holdings data, until the next holdings data become available. 8 This study does not take a position on whether fund managers should be rewarded for holding stocks with certain characteristics (e.g., momentum stocks) during long periods of time when those stocks outperform the market. However, we provide an accurate decomposition of the returns of winners and losers into style-based returns and style- adjusted returns to allow the reader (and investors) to draw their own conclusions about which method to use to 9
  12. 12. stock-picking ability, which is called the “Characteristic-Selectivity” measure (CS) (which was also described earlier in our measure of career stockpicking talent), is developed in DGTW, and is computed during quarter t as N X CSt = ˜ ˜ ˜b wj,t−1 (Rj,t − Rt j,t−1 ), (8) j=1 ˜ where wj,t−1 is the portfolio weight on stock j at the end of quarter t − 1, Rj,t is the quarter t ˜ ˜b buy-and-hold return of stock j, and Rt j,t−1 is the quarter t buy-and-hold return of the characteristic- based benchmark portfolio that is matched to stock j at the end of quarter t − 1. To construct the characteristic-based benchmark portfolio for a given stock during a given quarter, we characterize that stock over three characteristics—the size, book-value of equity to market-value of equity ratio, and past returns of that stock. Benchmarking a stock proceeds as follows—this procedure is based on Daniel, Grinblatt, Titman, and Wermers (1997), and is described in more detail in that paper. First, all stocks (listed on NYSE, AMEX, or Nasdaq) having book value of equity information in Compustat, and stock return and market capitalization of equity data in CRSP, are ranked, at the end of each June, by their market capitalization. Quintile portfolios are formed (using NYSE size quintile breakpoints), and each quintile portfolio is further subdivided into book-to-market quintiles, based on their book-to-market data as of the end of the December immediately prior to the ranking year. Finally, each of the resulting 25 fractile portfolios are further subdivided into quintiles based on the 12-month past return of stocks through the end of May of the ranking year. This three-way ranking procedure results in 125 fractile portfolios, each having a distinct combination of size, book-to-market, and momentum characteristics.9 The three-way ranking procedure is repeated at the end of June of each year, and the 125 portfolios are reconstituted at that date. Value-weighted returns are computed for each of the 125 fractile portfolios, and the benchmark for each stock during a given quarter is the buy-and-hold return of the fractile portfolio of which that stock is a member during that quarter. Therefore, the benchmark-adjusted return (also called the “DGTW-adjusted return”) for a given stock is computed as the buy-and-hold stock return minus the buy-and-hold value-weighted benchmark return during the same quarter. Finally, the rank mutual funds. 9 Thus, a stock belonging to size portfolio one, book-to-market portfolio one, and prior return portfolio one is a small, low book-to-market stock having a low prior-year return. 10
  13. 13. Characteristic Selectivity measure of the stock portfolio of a given mutual fund during quarter t, CSt , is computed as the portfolio-weighted DGTW-adjusted return of the component stocks in the portfolio, where the stock portfolio is normalized so that the weights add to one. B.2 Execution Costs Wermers (2000) uses past literature on the trading costs of institutional investors to construct an equation that describes the total trading costs of a fund manager in a given stock. This method is based on the empirical results of Keim and Madhavan (1997) and Stoll (1995), and should be viewed as an approximation of the expected total trading cost faced by a fund manager from the time a trade decision is made by a fund manager to the time it is fully executed. Thus, this trading cost estimate includes both the price impact (pre-trade price drift and trade price concession) and the explicit brokerage commission paid by the fund. This equation captures the cross-sectional dependence of total institutional trading costs on the market in which a stock is traded (i.e., NYSE or AMEX vs. Nasdaq), the size of the trade, the market capitalization and price of the stock, and whether the trade was a “buy” or a “sell.” SpeciÞcally, the equation for estimating the total cost of executing a purchase of stock i during quarter t, as a percentage of the total value of the trade, B Ci,t , is: " Ã !# B Nasdaq 1 Ci,t = Ytk · 1.098 + 0.336Di,t + 0.092T rsizei,t − 0.084Logmcapi,t + 13.807 . Pi,t Nasdaq Di,t is a dummy variable that equals one if the trade occurs on Nasdaq, and zero otherwise, T rsizei,t is the ratio of the dollar value of the purchase to the market capitalization of the stock, Logmcapi,t is the natural log of the market capitalization of the stock (expressed in $thousands), and Pi,t is the stock price at the time of the trade. Finally, Ytk is the year t trading cost factor for market k (k=NYSE/AMEX or Nasdaq). This factor captures the year-to-year changes in average trading costs over our time period in the different markets—these factors are based on Stoll (1995). S Similarly, our equation for estimating the percentage cost of selling stock i during quarter t, Ci,t , is " Ã !# S Nasdaq 1 Ci,t = Ytk · 0.979 + 0.058Di,t + 0.214T rsizei,t − 0.059Logmcapi,t + 6.537 . Pi,t Further details on the development of these equations are provided in Wermers (2000). 11
  14. 14. B.3 The Carhart Measure Carhart (1997) develops a four-factor regression method for estimating mutual fund performance. This four-factor model is based on an extension of the Fama and French (1993) factor model, and is described as Rj,t − RF,t = αj + bj · RMRFt + sj · SM Bt + hj · HMLt + pj · P R1Y Rt + ej,t . (9) Here, Rj,t −RF,t equals the excess net return of fund j during month t (the fund net return minus T- bills); RMRFt equals the month t return on a value-weighted aggregate market proxy portfolio; and SMBt , HMLt , and P R1Y Rt equal the month t returns on value-weighted, zero-investment factor- mimicking portfolios for size, book-to-market equity, and one-year momentum in stock returns. We use the Carhart (1997) regression measure of performance, α, to estimate the performance of mutual funds from their net return time-series data. B.4 The Ferson-Schadt Measure Ferson and Schadt (FS, 1996) develop a returns-based performance measure that controls for return predictability using dynamically evolving public information on relevant economic variables. In essence, the measure identiÞes a fund manager as providing value to shareholders if the manager provides excess net returns that are signiÞcantly higher than the fund’s matched factor benchmarks, both unconditional and conditional. These conditional benchmarks control for any predictability of the factor return premia that is due to evolving public information. Managers, therefore, are only labeled as superior if they possess superior private information on stock prices. FS also Þnd that these conditional benchmarks help to control for the response of consumer cashßows to mutual funds. For example, when public information indicates that the market return will be unusually high, consumers invest unusually high amounts of cash into mutual funds, which reduces the performance measure, “alpha,” from an unconditional model (such as the Carhart model). This reduction in alpha occurs because of the unconditional model does not control for the “market timing” inherent in using the public information to decide when to invest cash into the market—it is well-known that unconditional models exhibit a downward-biased alpha for funds with market timing “abilities” (see, for example, Treynor and Mazuy (1966)). Since the FS measure controls for the effect of public information, it also provides a control for 12
  15. 15. the effect of consumer cashßows on fund performance. The version of the FS model used in this paper starts with the unconditional Carhart four-factor model and adds a market factor that is conditioned on the Þve FS economic variables. This model is described as, 5 X Rj,t −RF,t = αj +bj ·RMRFt +sj ·SMBt +hj ·HM Lt +pj ·P R1Y Rt + Bj,i [zi,t−1 ·RMRFt ]+ej,t , i=1 where zi,t−1 is the deviation of information variable i from its unconditional (time-series) mean at time t − 1, and Bj,i is the response of fund manager j’s loading on the market factor, RMRFt , in response to the observed realization of zi,t−1 .10,11 The intercept of the model, αj , is the FS performance measure for fund j. C Summary Statistics on Funds and Fund Managers Table II provides four “snapshots” (at the beginning of 1985, 1990, 1995, and 2000) and the full- sample (1985 to 2000) summary statistics of manager characteristics. These characteristics are presented in two ways: the characteristics of the manager over that manager’s career with the current fund (only), and the full-career characteristics of that manager. The average manager career experience is roughly consistent throughout our sample period—average career experience is 7.4 years at the beginning of 1985 and 7.6 years at the beginning of 2000. Consistent with the Þndings of Wermers (2000), the mean and median manager track records, measured as the return in excess of the S&P 500 index (“Excess Return”), the return in excess of the same investment-objective average fund return (“Objective-Adjusted Return”), or as the CST measure (“DGTW”), are mostly positive. This is also consistent with the Þnding in Khorana (1996) that underperforming managers are more likely to be replaced than the average manager. Interestingly, fund managers take on somewhat higher portfolio risk in the Þve-year post-1995 period than in the pre-1995 period. For example, at the beginning of the year 2000, the mean career risk-tolerance, measured as the time-series standard deviation of excess return relative to the S&P 500 index or the standard deviation of investment objective-adjusted return are 10.2 percent and 8.9 percent, respectively, which are higher than their levels at the beginning of 1995. This higher 10 The public information variables of FS include (1) the lagged level of the one-month T-bill yield, (2) the lagged dividend yield of the CRSP value-weighted NYSE and AMEX stock index, (3) a lagged measure of the slope of the term structure, (4) a lagged quality spread in the corporate bond market, and (5) a dummy variable for January. 11 Note that, to maintain model simplicity, we use only the market equity premium to construct conditional factors. However, it is likely that the majority of public information concerns the return on the broad market, versus the return premia due to various styles. 13
  16. 16. average risk level is likely due to the increasing style specialization of mutual funds over this time period, and the subsequent increased volatility corresponding to the decreased style diversiÞcation of funds. Finally, the mean and median career aggressiveness of fund managers has risen gradually during the 15-year period. As noted by Wermers (2000), this increased trading activity is likely due to the substantially lower trading costs at the end of our sample period, compared with earlier years. More aggressive trading over time may also reßect more frequent portfolio adjustments required because of the increased market volatility toward the end of our sample period. IV Results A Does Experience Matter? We begin with an analysis of the effect of manager experience on mutual fund characteristics and performance. The extant literature, in general, has not examined whether more seasoned managers have better skills in picking stocks. We might believe that a manager gains skills in picking stocks as her career progresses, from perhaps several sources. For example, it may take some time for the manager to assemble and train her stock analysts, or to learn how to best use the analysts already in place at a fund complex. Also, over time, managers may develop relationships with corporate managers that provide them with privileged information on the prospects of Þrms. Chevalier and Ellison (1999) study the impact of the experience on the managerial stock-picking behavior, approaching the issue from the perspective of career concerns of fund managers. They Þnd that young managers are more risk averse and more likely to herd in picking stocks; however, the short time-series contained in their database of managers prevents them from following individual fund managers over their entire careers.12 To test the effect of manager experience on the performance and characteristics of a mutual fund, we sort all funds, at the end of each calendar year, on the level of career experience of the “lead manager” of the fund. We then measure the characteristics and performance of each ranked fractile of funds during the following calendar year—the process is repeated at the end of each year, starting December 31, 1985 and ending December 31, 1999. For a mutual fund with only one 12 In addition, their database, which is obtained from Morningstar, has a large number of missing managers during the time period under study. By contrast, our manager database contains the vast majority of managers, especially during the last 10 years of our sample period. 14
  17. 17. manager, that manager, by construction, is our lead manager. For funds that are team-managed, the lead manager is deÞned as the manager with the earliest start date as manager of the fund. We base our proxy for experience on the lead manager’s career experience because we believe that this manager probably has the biggest role in the decision-making process of the fund. If, on the other hand, the non-lead managers play a huge part of the decision-making process of a mutual fund, this will simply add noise to our tests. For example, for the year ending December 31, 1985, we rank all funds having an investment objective (at that date) consistent with holding mainly U.S. equities on the number of months of career experience of their lead managers.13 Then, funds are placed in quintile, decile, or ventile portfolios. Various average characteristics and measures of performance are computed for these fractile portfolios during the following “test” year. In computing test-year measures for statistics that are available at least quarterly (such as net returns or performance measures), we compute, for each test-year calendar quarter, the equal-weighted measure across all funds in a given fractile. If a fund disappears during the test year, we include it in the appropriate fractile portfolio until the beginning of the quarter in which the fund disappears, then we rebalance the fractile portfolio for the next quarter. For return or performance measures, we compound these rebalanced equal- weighted measures over all four quarters in the test year. For non-return characteristics, such as managerial turnover, the quarterly measures are cumulated over the test year. In computing test-year measures for statistics that are available only annually (such as portfolio turnover), we compute the equal-weighted average measure across all funds having data for that measure during the test year. The reader should note that all tables that follow will use these procedures for computing test-year average measures. Table III shows the results of our ranking on career lead manager experience. Panel A of that table shows the characteristics of the fractile portfolios over the year following the sort of funds on career manager experience. SpeciÞcally, the table shows the number of funds in each fractile, the average total net assets of funds in each fractile, the coefficients from a regression of the EW-average excess net return on the four Carhart factors, and the EW-average (over all event years): career aggressiveness of the lead manager (the average portfolio turnover level over all funds managed over her career), career experience of the lead manager, lead manager turnover level (percentage of lead managers that are replaced), portfolio turnover level, and active style drift (the sum of the 13 That is, funds must have a self-reported investment objective of “aggressive growth,” “growth,” “growth and income,” “income,” or “balanced” at the end of a given ranking year. 15
  18. 18. absolute values of the active style movements of the fund over the test year). The third column of the panel shows the average level of career experience of the sorted fractiles. The most experienced managers (the Top 5% fractile) have 343 months of experience, while the least experienced managers (the Bottom 5% fractile) have only 18 months. The panel also shows that more experienced lead managers oversee much larger pools of mutual fund assets than their less-experienced counterparts. For example, the Þve percent of managers with the most experience manage, on average, funds that are about seven times the size of funds managed by the least experienced Þve percent of managers ($2.2 billion vs. $321 million, respectively). The coefficients for the Carhart regressions show that more experienced managers have slightly less exposure to the broad stock market, small-capitalization stocks, and value stocks. All fractiles of fund managers show similar exposures to momentum stocks. Overall, as previously shown by Carhart (1997) and Wermers (2000), mutual fund managers hold about 90 percent of their assets in the stock market (vs. Þxed income and other investments), hold more small stocks than the broad market, and have a slight value and momentum tilt. The career aggressiveness measure shows the average portfolio turnover of each lead-manager fractile over the managers’ entire careers. More experienced fund managers exhibit much lower levels of career aggressiveness than less experienced managers—this may either be due to these managers trading less frequently as their careers progress (to avoid trading and other costs), or to these managers holding much larger portfolios than their less-experienced counterparts during the latter parts of their careers. These managers may simply be avoiding high levels of turnover of their large positions in order to avoid large trading impacts by their actions. The test year portfolio turnover column conÞrms that these managers trade much less frequently than their counterparts during this stage of their careers. The Þnal two columns of panel A show the percentage of managers that are replaced, and the level of active style drift during the test year, respectively. Both relatively experienced and relatively inexperienced managers are replaced at a higher rate than their mid-career counterparts. We would expect that many of the most experienced managers leave a fund either to retire or to manage a larger fund, while many of the least experienced managers may either be Þred, or (if successful) may leave to manage a larger fund. These two groups of managers also have a higher tendency to exhibit “active style drift” (ASD) than their mid-career peers. Less-experienced managers, who manage smaller portfolios consisting of heavier holdings of small-capitalization stocks, relative to 16
  19. 19. their counterparts, may move around in the style dimensions in an attempt to outperform their peers. Or, these managers may need to move across style categories due to the limited number of liquid small-capitalization stocks in a particular style category. For example, perhaps a small- capitalization growth manager Þnds it necessary to invest in some small-capitalization value stocks in response to large cash inßows from fund shareholders. Panel B of Table III presents a performance attribution for each manager-experience fractile. Experienced managers hold portfolios of stocks with slightly lower returns, both before and after trading costs and fund expenses, as shown by the “Gross Return” and “Net Return” columns. However, manager talent is best measured by the CS measure of stockpicking talent—here, expe- rienced managers show a level of talent that is not statistically distinguishable from their rookie counterparts. SpeciÞcally, the most experienced managers, those in the Top 5 percent fractile, ex- hibit a CS measure of 1.9 percent per year, while those managers in the Bottom 5 percent fractile exhibit a CS measure of one percent per year. The difference between these two measures is not signiÞcant. Of interest is the level of expenses charged by experienced managers, as we might expect that experienced managers charge higher expenses for their presumed greater skills. However, the ex- penses of experienced managers, which average 1.2 percent per year for experienced managers, are actually slightly lower than the expenses of inexperienced managers. To some degree, this reduc- tion in expense ratios might be related to the economies-of-scale in running funds, and experienced managers may still be capturing a larger net fee than other managers. Inßows from consumers are signiÞcantly higher for the most experienced fractile of managers, relative to the least experienced. However, this appears to be mainly driven by the reluctance of consumers to invest in funds managed by the most inexperienced managers (the Bottom 5 percent fractile), as none of the other inßow differences are signiÞcant. Finally, the Carhart and Ferson- Schadt alphas indicate that all fractile groups exhibit negative performance, net of all costs and expenses (except load fees and taxes), but none of these alphas are signiÞcant. In unreported tests, we repeat the sorting procedure of this section, limited to funds having a growth-oriented investment objective (an investment objective, at the end of a given ranking year, of either “aggressive growth” or “growth”). The results are consistent with our baseline results for all funds above: experienced managers exhibit no higher level of stockpicking talent than inexperienced managers. 17
  20. 20. To summarize our results from this section, experienced managers tend to manage much larger funds, and exhibit lower levels of trading activity than inexperienced managers. However, our results provide no support for the hypothesis that the stockpicking skills of fund managers improve over their careers. This Þnding seems somewhat surprising, since we might reasonably believe that a manager without stockpicking talents would be forced to leave the industry before the latter part of her career, as investors become more certain from the longer time-series of manager returns available, that the manager does not have talent. Apparently, the labor market for fund managers does not function effectively in this dimension. B Does Past Performance Matter? While our last section rejects the notion that experience is correlated with talent, we are also interested in whether some managers, at any experience level, have persistent stockpicking skills. In this section, we investigate this issue by examining whether lead fund managers with the best career stockpicking records have skills that persist in the future. We measure career stockpicking talent using our characteristic selectivity track record (CST ) for each manager, as described by Equation (3). Analogous to the ranking procedure of the last section, we sort all fund managers, at the end of each calendar year starting December 31, 1985 and ending December 31, 1999, on their CST measure at the end of that year. Then, we measure the following-year characteristics and performance of each fractile that results from this sorting procedure. In Panel A, we present the characteristics of these manager career-record fractiles. The panel shows that managers with the best track records do not have substantially more experience (92 months) than the average fund manager (108 months), although managers with the worst track records do have substantially less experience than average (52 months). Thus, experience, by itself, does not appear to be associated with career stockpicking talent; consistent with the results of the prior section, the majority of experienced managers appear to have no stockpicking talents. The results also show that managers with extreme stockpicking track records (either good or poor) tend to be more aggressive traders than the average fund manager. This Þnding holds both for their entire careers (up to and including the test year—shown in the “Career Aggressiveness” column) and during the test year alone (shown in the “Portfolio Turnover” column). Managers with the best track records may know that their talents will persist, and, therefore, may trade frequently to capitalize on their talent. Alternatively, these managers may be exhibiting overconÞdence based 18
  21. 21. on their past success, which would result in unnecessary costly trading of stocks in the future. On the other hand, managers with poor track records may be trading frequently in order to try to reverse their fortunes, or, alternatively, to appear to have stockpicking skills. Finally, managers with extreme stockpicking track records (either good or poor) experience higher managerial replacement rates and higher levels of active style drift. For managers with the best records, we would expect that they depart from a fund to either retire or to manage a larger fund, based on their past success. For managers with the worst records, we would expect a large proportion of dismissals or transfers to smaller funds. Baks (2001) studies this issue and provides Þndings that are consistent with this. The high levels of active style drift that we observe among successful managers may occur because these managers have talents that span across more than one style category. In contrast, the high levels of active style drift that we observe among unsuccessful managers may be due either to the difficulty of maintaining a style focus with a fund that invests in smaller-capitalization stocks, or to the manager taking active “bets” in order to attempt to outperform her counterparts (by luck). In Panel B, we provide a performance attribution for each track-record fractile of fund managers. The evidence shows that fund managers with the best career records have persistent stockpicking skills—for example, the Top 5% fractile of managers—those with the very best career stockpicking records—hold stocks that outperform their characteristic benchmarks by two percent per year. Fund managers in the bottom 40% of fractiles, by contrast, have no ability to pick underpriced stocks. In addition, the difference in stockpicking talents between managers with the best and worst career records is large and statistically signiÞcant. For example, the top decile of managers hold stocks that outperform the stocks held by the bottom decile of managers, adjusted for their characteristics, by a statistically signiÞcant 1.7 percent, averaged over all test years. Consistent with the higher portfolio turnover levels found earlier for the extreme fractiles (Panel A), execution costs are somewhat higher for these fractiles (Panel B) than for the average fund. In addition, the management companies of these extreme fractile funds charge higher average expense ratios—to some extent, this is due to the smaller portfolios managed by the top and bottom fractile managers. As shown by Collins and Mack (1999), strong economies-of-scale exist in the mutual fund industry, resulting in expense ratios that are inversely related to the level of assets under management. 19
  22. 22. An examination of consumer inßows to the various fractiles provides some interesting results. While managers with good track records have only slightly higher net returns than other managers, these “star” managers attract much higher levels of cash inßows. For example, the top quintile of managers experience an average yearly inßow equal to 25 percent of the beginning-of-year TNA of their funds, while the manager of the average fund attracts only 17 percent. This Þnding indicates that consumers appear to prefer to invest their money in a fund managed by a “star,” independent of the immediate past net return of the fund. Finally, the panel shows the net return alphas of each equal-weighted fractile of funds. Both the Carhart and Ferson-Schadt alphas are insigniÞcant for all fractiles, except for the fractile of funds that are managed by the managers with the very worst track records. These managers continue to perform poorly, underperforming their benchmarks by about two percent, on average over all test years. C The Impact of Managerial Replacement on Fund Characteristics and Per- formance As discussed by Baks (2001), the replacement of a manager provides a unique opportunity to study the impact of the manager on the performance of a fund, independent of the fund’s other characteristics. In this section, we examine the characteristics and returns of funds during the periods immediately before and after a lead manager is replaced. Each year, we separate funds into those having a lead manager change during the year, and those with no change in lead manager. Then, we measure the returns and characteristics of the equal-weighted portfolio of funds in each group, during the year of the potential change, and during the three following years. Table V presents the results of this test. Panel A shows that managers are replaced during years when their stockpicking talents are signiÞcantly worse than those of all other managers. SpeciÞcally, the characteristic selectivity measure during the year that the manager is replaced is insigniÞcant, compared to a measure of 0.5 percent (which is statistically signiÞcant) for all funds with no managerial change. Further, the arrival of a new fund manager is very good news for a fund: the new manager brings stockpicking talents that are statistically indistinguishable from the talents of all other managers during the three years following the managerial change. Panel B, which measures the net returns of funds with managerial turnover vs. all other funds, shows similar results—underperformance during the 20
  23. 23. year of manager replacement, followed by improved performance during the following years. Finally, Panel C shows that the appears to reduce the level of trading, relative to the replaced manager. In particular, the average portfolio turnover drops from a level of 99 percent, during the managerial replacement year, to 89 percent during the third year following the managerial replacement. These results indicate that the manager who is replaced may be engaging in heavy trading during the Þnal year of her tenure at a fund in an attempt to “gamble” as a last resort. D Multivariate Regressions Our results of the prior sections indicate correlations between stockpicking talents and manager characteristics—speciÞcally, managerial experience, career stockpicking record, and managerial re- placement. In this section, we test whether our prior sorting results, which test the relation between current talent and manager characteristics, still hold in a multivariate setting. Here, we conduct Fama-McBeth (1973)-type multivariate regressions to conduct these tests. For each year, starting with 1986 and ending with 2000, we run a cross-sectional regression of a fund’s CS measure, averaged across all four quarters of that year, on the manager’s level of experience and the manager’s stockpicking track-record (CST ), both measured at the end of the prior year. A dummy variable is also added to indicate whether the manager is replaced during the prior year. We then average the coefficient estimates over all years, and report this average, as well as the time-series t-statistic. The resulting regressions (1) and (2) in Table VI show that experience, alone, does not explain future stockpicking success, but that the career stockpicking track-record does. Regression (3) includes both experience and career track-record as regressors, and shows that the track-record remains signiÞcant, controlling for any correlation between these two variables. Thus, the Þrst three regressions conÞrm the results of our sorting tests of the prior sections: the past stockpicking record of a manager helps to predict her future stockpicking success, but experience does not matter, either alone or in combination with the track-record. In the fourth regression, we add a manager replacement dummy that equals one, if a manager is replaced during the prior year. This speciÞcation shows that managerial turnover does not help to explain future stockpicking success, when experience and stockpicking talent are included as regressors. In the last section, we found that manager replacement, alone, is a strong predictor of improved fund performance. Thus, manager replacement provides predictive power only because it 21
  24. 24. serves as a proxy for career stockpicking record—that is, a new manager enters a fund with a strong track record, and it is that variable that predicts the future success of the manager. This Þnding is consistent with Khorana (1996), who Þnds that new fund managers have substantially better records than the managers that they replace. In Table VII, we repeat these Fama-McBeth regressions on growth-oriented funds only. For example, the cross-sectional regression for 1986 includes only managers of funds having a self- declared investment objective of either “aggressive growth” or “growth” at the end of 1985. The results show some interesting contrasts with the full-sample results of Table VI. SpeciÞcally, man- agerial experience provides signiÞcant explanatory power, by itself (regression (1)), in combination with stockpicking record (regression (3)) and in combination with both stockpicking record and the replacement dummy (regression (4)). Thus, experience appears to a strong inßuence in pick- ing growth stocks, perhaps because it is much more difficult to accurately forecast the growth in earnings of growth stocks, relative to value stocks. Growth-oriented managers may either develop specialized skills over time, or, alternatively, they develop valuable relationships with corporate managers that give them access to private information on future earnings. Table VIII repeats these regressions on income-oriented funds, which are deÞned as funds having a self-declared investment objective of “growth and income,” “income,” or “balanced” at the end of the year prior to the regression year. For these managers, none of the variables are signiÞcant— experience, track record, or managerial replacement. Although this Þnding is somewhat surprising, it is consistent with prior work by Chen, Jegadeesh, and Wermers (2000), who show that income- oriented funds exhibit no abnormal returns, while growth-oriented funds do. Thus, income-oriented funds appear to provide style-based return premia, but nothing else. E The Role of Managerial Risk-Aversion Our Þnal tests explore whether fund managers with lower levels of risk-aversion are better able to exploit their stockpicking talents (if any) to generate higher average levels of fund performance. Regression (5) in Tables VI, VII, and VIII add two proxies for managerial risk tolerance to address the role of this characteristic in generating performance. The Þrst proxy, “Career Risk Tolerance,” is the standard deviation of the manager’s S&P500-adjusted monthly return over her career, prior to a given year, while the second proxy, “Career Aggressiveness,” is the turnover ratio of all funds managed, averaged over the manager’s career prior to the given year. 22
  25. 25. Table VI, regression (5) shows that these risk tolerance proxies are not signiÞcant inßuences on future performance. However, the career CST record is now insigniÞcant, which indicates that there is substantial multicollinearity between risk tolerance and track-record. Indeed, in unreported tests, we Þnd cross-sectional Pearson correlations of 0.25 and 0.11 (both signiÞcant at the one percent level) between the CST measure and the career risk tolerance and career aggressiveness variables, respectively. These correlations are measured for all managers at the end of 1999. A different result holds for managers of growth-oriented funds, as shown in regression (5) of Table VII. Here, the inßuence of experience and track-record remain after adding the risk-tolerance proxies. However, the risk-tolerance variables are still insigniÞcant for these managers, which indicates that any inßuence of risk-tolerance on performance is already captured by the experience or CST track-record variables. V Conclusion In this paper, we have presented evidence on the role of mutual fund managers in generating mutual fund performance. This topic has received relatively little attention in the academic literature, with the exception of Chevalier and Ellison (1999) and Baks (2001). Our study uses the longest cross- sectional database of fund managers available to date, extending from 1985 to 2000, and includes both the stockholdings, net returns, and other characteristics of each managed fund. This database allows us to investigate several issues of interest regarding the role of managers, including the importance of experience and past track record in generating future performance. We Þnd that experience is an important indicator of stockpicking talent, but only for growth- oriented fund managers. The stockpicking track record of a fund manager, however, is a stronger indicator of manager talent for all types of fund managers. Thus, manager talent strongly persists. We also Þnd that the replacement of a manager is good news for a fund, as the pre-replacement performance of the fund is reliably lower than its counterpart funds, while the post-replacement performance is statistically indistinguishable from the counterpart performance. However, the signiÞcance of this variable disappears, once we include both the stockpicking track record and manager replacement in a multivariate regression setting. Our study, while providing new insight on the performance and performance persistence issues that have been a focus of academic research for decades, also opens up possible new studies on the behavior of fund managers. Our database allows the study of these behavioral issues though an 23
  26. 26. analysis of the stock trades of fund managers having various characteristics. We believe that this is an important new direction for future research. 24
  27. 27. Bibliography [1] Baks, Klaas, 2001, “On the Performance of Mutual Fund Managers,” Working Paper, The Wharton School. [2] Carhart, Mark, 1997, “On Persistence in Mutual Fund Performance,” Journal of Finance, Volume 52, pp. 57-82. [3] Chevalier, Judith, and Glenn Ellison, 1999, “Are Some Mutual Fund Managers Better than Others? Cross-Sectional Patterns in Behavior and Performance,” Journal of Finance, Volume 54 (3), pp. 875-899. [4] Collins, Sean, and Phillip Mack, 1999, “Some Evidence on Scope, Scale, and X-Efficiency in U.S. Mutual Funds,” Reserve Bank of New Zealand Working Paper. [5] Daniel, Kent, Mark Grinblatt, Sheridan Titman, and Russ Wermers, 1997, “Measuring Mutual Fund Performance with Characteristic-Based Benchmarks,” Journal of Finance, Volume 52, pp. 1035-1058. [6] Daniel, Kent, and Sheridan Titman, 1997, “Evidence on the Characteristics of Cross Sectional Variation in Stock Returns,” Journal of Finance, 1997, Volume 52, pp. 1-33. [7] Grinblatt, Mark and Sheridan Titman, 1989, “Mutual Fund Performance: An Analysis of Quarterly Portfolio Holdings,” Journal of Business, 62, pp. 394-415. [8] Grinblatt, Mark and Sheridan Titman, 1993, “Performance Measurement without Bench- marks: An Examination of Mutual Fund Returns,” Journal of Business, 66, pp. 47-68. [9] Jegadeesh, Narasimhan, and Sheridan Titman, 1993, “Returns to Buying Winners and Selling Losers: Implications for Stock Market Efficiency,” Journal of Finance, Volume 48, pp. 65-92.
  28. 28. [10] Keim, Donald B. and Ananth Madhavan, 1997, “Transactions Costs and Investment Style: An Inter-Exchange Analysis of Institutional Equity Trades,” Journal of Financial Economics, Volume 46 (3), pp. 265-292. [11] Malkiel, Burton G., 1995, “Returns from Investing in Equity Mutual Funds, 1971-1991,” Jour- nal of Finance, 50, pp. 549-572. [12] Sefcik, Stephan E. and Rex Thompson, “An Approach to Statistical Inference in Cross- Sectional Models with Security Abnormal Returns as Dependent Variable,” 1986, Journal of Accounting Research, Volume 24, pp. 316-334. [13] Stoll, Hans R., 1995, The importance of equity trading costs: Evidence from securities Þrms’ revenues, in Robert A. Schwartz, ed.: Global Equity Markets: Technological, Competitive, and Regulatory Challenges (Irwin Professional Publishing, New York). [14] Wermers, Russ, 2000, “Mutual Fund Performance: An Empirical Decomposition into Stock- Picking Talent, Style, Transactions Costs, and Expenses,” Journal of Finance, Volume 55, pp. 1655-1695. [15] Wermers, Russ, 2002a, “A Matter of Style: The Causes and Consequences of Style Drift in Institutional Portfolios,” University of Maryland Working Paper. [16] Wermers, Russ, 2002b, “Predicting Mutual Fund Returns,” University of Maryland Working Paper.
  29. 29. Table I: Summary Statistics of Mutual Funds and Fund Managers This table presents the summary statistics of mutual funds and fund managers in our fund-manager sample between 1985 and 2000 (inclusive). Our mutual fund data are drawn from the merged CDA–CRSP mutual fund database (CDA–CRSP). An early version of CDA–CRSP is used in Wermers (2000), which also contains a detailed description of the construction of CDA–CRSP. The fund manager data are collected from three most used mutual fund data sources: the Morningstar Principia Pro (January 2001), the CRSP Mutual Fund Data Base (2000Q3), and Wiesenberger. Panel A reports the number of mutual funds existing during 1985, 1990, 1995, and 2000, as well as during the whole sample period, 1985–2000, for the whole fund universe as well as each of the following four self-declared investment objective categories—aggressive growth (AG), growth (G), growth and income (GI), income or balanced (I or B). Self-declared investment-objective data are collected from CDA. A fund’s investment objective for a reporting period is the one first reported in the period. Panel B presents the counts of lead managers and the average number of funds lead managed by a lead manager during 1985–2000 as well as during 1985, 1990, 1995, and 2000. In case of team management, the lead manager is defined as the active manager who starts to manage the fund earliest. To calculate the number of funds lead managed by a fund manager during a reporting period, we divide the total number of fund-months she lead manages by the total number of months when she is a lead manager. The average number of funds lead managed by a fund manager is calculated by taking the cross-sectional average across all managers in a group. A fund manager is counted as a lead manager of AG funds for a given period if she is the lead manager of at least one AG fund during the period. Managers of G, GI, and I or B funds are similarly defined. Since some managers lead-manage multiple funds with different investment objectives for a given period, the sum of the numbers of lead managers of AG, G, GI, and I or B funds may be greater than the total number of lead managers. Panel C reports the number of funds missing managers. The 1985 (1990, 1995, 2000) column in Panel C reports the funds that exist in 1985 (1990, 1995, 2000) but do not have managers matched in 1985 (1990, 1995, 2000). The 1985–2000 column in Panel C reports the funds that exist at one point of time during 1985–2000 but do not have a matched manager throughout the sample period. The percent of funds missing managers is calculated as the the number of funds missing managers to the number of funds in the same period. Panel D provides a comparison of median total net assets (TNA) and mean excess returns between the funds that are matched with a manager and the funds that do not have any matched fund manager. The 1985 (1990, 1995, 2000, 1985–2000) column in Panel D reports the funds existing in 1985 (1990, 1995, 2000, 1985–2000). A fund’s total net assets in 1985, 1990, 1995, or 2000 is defined as its year-end TNA. A fund’s total net assets over 1985–2000 is the time-series average of its monthly total net assets between 1985 and 2000. The median TNAs are expressed in millions of year 2000 dollars. The excess return of a fund for a given year is computed by subtracting the annual S&P 500 return from the fund’s annual net return. The excess return of a fund over 1985–2000 is the time-series average of annual excess returns in the period it exists in 1985–2000. The mean excess returns of funds are expressed in percent. Panel A: Counts of Mutual Funds Investment Objective 1985 1990 1995 2000 1985–2000 All Funds 352 621 1513 1683 2272 AG 83 118 164 129 217 G 151 294 943 1072 1507 GI 93 142 258 330 366 I or B 25 67 148 152 182
  30. 30. Panel B: Counts of Lead Managers of Mutual Funds 1985 1990 1995 2000 1985–2000 Avg. No. Avg. No. Avg. No. Avg. No. Avg. No. of Funds of Funds of Funds of Funds of Funds Lead Lead Lead Lead Lead N Managed N Managed N Managed N Managed N Managed All Managers 202 1.27 334 1.38 1116 1.43 1256 1.57 2229 1.34 Managers of AG Funds 48 1.52 87 1.64 165 1.88 126 2.18 372 1.62 Managers of G Funds 102 1.42 169 1.46 763 1.54 879 1.70 1661 1.40 Managers of GI Funds 64 1.41 78 1.51 245 1.70 323 1.91 563 1.54 Managers of I or B Funds 20 1.70 41 1.65 143 1.91 145 2.19 289 1.62 Panel C: Counts of Mutual Funds Missing Managers 1985 1990 1995 2000 1985–2000 Percent of Percent of Percent of Percent of Percent of N Total Funds N Total Funds N Total Funds N Total Funds N Total Funds All Funds 116 33.0 102 16.4 93 6.1 80 4.8 142 6.3 AG 32 38.6 14 11.9 4 2.4 3 2.3 15 6.9 G 47 31.1 49 16.7 64 6.8 52 4.9 88 5.8 GI 32 34.4 32 22.5 13 5.0 13 3.9 23 6.3 I or B 5 20.0 7 10.4 12 8.1 12 7.9 16 8.8 Panel D: Comparison of Funds Reporting Lead Managers and Funds Missing Managers 1985 1990 1995 2000 1985–2000 Mean Mean Mean Mean Mean Median Excess Median Excess Median Excess Median Excess Median Excess TNA Return TNA Return TNA Return TNA Return TNA Return All Funds 174.4 -3.85 108.8 -2.69 134.5 -7.46 244.6 8.43 109.0 -3.19 Funds Reporting Lead Managers 187.8 -2.83 138.0 -2.46 143.9 -7.36 249.7 8.54 117.3 -2.97 Funds Missing Managers 131.4 -6.13 45.3 -3.98 64.5 -9.14 179.6 6.19 28.7 -7.30
  31. 31. Table II: Summary Statistics of Lead Manager Characteristics This table presents the summary statistics of lead manager characteristics, including experience, track record, risk attitude, and aggressiveness, with current fund as well as over career, at the beginning of 1985, 1990, 1999, and 2000, as well as for 1985–2000. The “Career” (“Current Fund”) experience of a lead manager is defined as the time elapsed since she first becomes a fund manager (since she becomes the lead manager of the current fund). In calculating the rest of manager characteristics with “Current Fund,” we start from when the manager becomes the lead manager of the fund. To compute the “Career” measures, we start from when the fund manager first becomes a lead manager. Three proxies are employed to measure a fund manager’s track record: excess return (time-series average monthly net return in excess of the S&P 500 return), objective-adjusted return (time-series average monthly objective- adjusted return), and stockholding characteristics-based DGTW measure following Daniel, Grinblatt, Titman, and Wermers (1997). We use the standard deviation of monthly excess return and monthly objective-adjusted return to proxy for the risk attitude of a fund manager. The aggressiveness of a fund manager is proxied by the time-series average turnover ratio of the managed fund(s). A fund’s turnover ratio is defined as the lesser of its securities sales and purchase divided by the average monthly total net assets. A fund manager’s characteristics for 1985–2000 is her characteristics when she leaves the sample (either on the sample end date December 31, 2000 or when she departs the last fund managed by her). The experience is expressed in years while all the track record and risk variables are annualized and expressed in percent. Aggressiveness is also annualized.
  32. 32. 1985 1990 1995 2000 1985–2000 Mean Median Mean Median Mean Median Mean Median Mean Median Experience (In Years) With Current Fund 6.3 4.7 5.2 3.1 4.4 2.1 4.4 2.2 4.8 3.3 Career 7.4 6.1 6.2 3.9 6.5 4.9 7.6 6.2 7.7 6.1 Track Record (Excess Return, % Per Year) With Current Fund 4.86 5.96 0.55 0.97 2.76 2.44 0.90 -1.45 1.00 0.20 Career 4.86 5.98 0.74 1.18 2.77 2.46 0.06 -1.59 0.55 0.10 Track Record (Objective-Adjusted Return, % Per Year) With Current Fund 1.03 0.76 0.30 0.97 0.63 0.38 1.10 0.40 -0.13 -0.12 Career 0.86 0.47 0.69 0.76 0.50 0.34 0.87 0.40 0.14 0.15 Track Record (DGTW, % Per Year) With Current Fund 1.39 1.87 0.22 0.71 0.47 0.53 0.58 -0.11 1.70 0.52 Career 0.90 1.60 0.50 0.88 0.52 0.57 0.58 -0.01 1.19 0.63 Risk Attitude (Std. Dev. of Excess Return, % Per Year) With Current Fund 8.46 7.79 7.86 7.26 6.89 6.16 10.60 9.18 11.51 9.26 Career 8.61 7.79 8.08 7.48 7.12 6.37 10.23 8.63 11.16 9.05 Risk Attitude (Std. Dev. of Objective-Adjusted Return, % Per Year) With Current Fund 7.18 6.66 6.54 5.66 5.51 4.86 9.08 8.03 10.02 8.24 Career 7.27 7.23 6.31 5.60 5.72 4.90 8.91 7.66 9.59 7.85 Aggressiveness With Current Fund 0.74 0.59 0.80 0.64 0.86 0.66 0.90 0.70 0.95 0.74 Career 0.72 0.57 0.80 0.66 0.86 0.66 0.89 0.69 0.93 0.74
  33. 33. Table III A Decomposition of Returns for Experienced vs. Inexperienced Managers A decomposition of mutual fund returns and costs is provided below for the merged manager, CDA holdings, and CRSP mutual fund characteristics/net returns databases. At the end of each calendar year, starting December 31, 1985 and ending December 31, 1999, we rank all mutual funds in the merged database that existed during the entire prior 12-month period (and had a complete data record during that year) on the level of experience of the lead fund manager (the months of career experience, with any fund, of the manager starting at a given fund at the earliest date) at the end of that year (the “ranking year”). Then, fractile portfolios are formed, and we compute average return measures (e.g., net returns) for each fractile portfolio during the following year (the “test year”). In computing the average return measure for a given test year, we Þrst compute quarterly buy-and-hold returns for each fund that exists during each quarter of the test year, regardless of whether the fund survives past the end of that quarter. Then, we compute the equal-weighted (EW) average quarterly buy-and-hold return across all funds for each quarter of the test year. Finally, we compound these returns into an annual return that is rebalanced quarterly. Panel A presents several characteristics of these sorted fractiles during the Þrst year following the ranking year: the number of funds in each fractile, the average career experience of the lead fund manager, the average total net assets of funds, the coefficients from a regression of the EW-average excess net return on the four Carhart factors, and the EW-average (over all event years): career aggressiveness of the lead manager (the average portfolio turnover level over all funds managed over her career), portfolio turnover level, lead manager turnover level (percentage of lead managers that are replaced), and active style drift (the sum of the absolute values of the active style movements in the three style dimensions of the fund over the test year). Panel B presents a decomposition of fund returns and costs during the test year. SpeciÞcally, the panel presents the EW-average annual: characteristic selectivity measure (CS), estimated transactions costs, expense ratio, net reported return, fund inßows, Carhart net return alpha, and Ferson and Schadt net return alpha. Both panels of this table present test year statistics, averaged over all test years. In forming all portfolios in this table, we limit our analysis to funds having a self-declared investment objective of “aggressive growth,” “growth,” “growth and income,” “income,” or “balanced” at the beginning of the test year. Panel A. Fractile Characteristics (Test Year) Ranking Variable = Experience Career Avg Career Portfolio Manager Avg Experience TNA Aggress. Turnover Turnover ASD Fractile No (Months) ($millions) RMRF SMB HML PR1YR (%/yr) (%/yr) (%/yr) (Style #) Top 5 % (Most Experienced) 45 343 2,214 0.87∗∗∗ 0.09∗∗∗ -0.12∗∗∗ 0.07∗∗∗ 57.4 64.6 32.4 0.75 Top 10 % 89 297 1,704 0.87∗∗∗ 0.12∗∗∗ -0.07∗∗∗ 0.06∗∗∗ 59.7 66.3 20.7 0.71 Top 20 % 178 243 1,434 0.86∗∗∗ 0.15∗∗∗ -0.05∗∗∗ 0.05∗∗∗ 62.1 66.3 17.1 0.69 2nd 20 % 178 127 1,186 0.88∗∗∗ 0.20∗∗∗ -0.05∗∗∗ 0.04∗∗∗ 70.8 71.8 17.7 0.63 3rd 20 % 178 82 838 0.92∗∗∗ 0.21∗∗∗ -0.04∗∗ 0.03∗∗∗ 76.3 76.5 19.2 0.66 4th 20 % 178 54 559 0.94∗∗∗ 0.23∗∗∗ -0.05∗∗∗ 0.05∗∗∗ 84.5 82.4 18.8 0.69 Bottom 20 % 178 29 400 0.93∗∗∗ 0.25∗∗∗ -0.06∗∗∗ 0.06∗∗∗ 90.8 90.4 20.7 0.76 Bottom 10% 89 22 375 0.95∗∗∗ 0.25∗∗∗ -0.06∗∗∗ 0.05∗∗∗ 94.2 92.7 28.8 0.79 Bottom 5% (Least Experienced) 45 18 321 0.98∗∗∗ 0.27∗∗∗ -0.05∗∗ 0.06∗∗∗ 98.6 99.0 34.8 0.84 All Funds 890 108 963 0.90∗∗∗ 0.21∗∗∗ -0.05∗∗∗ 0.05∗∗∗ 76.7 77.4 18.7 0.69
  34. 34. Table III (continued) Panel B. Performance Attribution (Test Year) Ranking Variable = Experience Avg Execution Net Avg TNA CS Costs Expenses Return Inßows αN et Carhart αNet F erson−Schadt Fractile No ($millions) (%/yr) (%/yr) (%/yr) (%/yr) (%/yr) (%/yr) (%/yr) Top 5 % (Most Experienced) 45 2,214 1.9 1.0 1.2 14.9 18.9 -0.3 -0.5 Top 10 % 89 1,704 1.5 1.0 1.2 14.7 18.4 -0.4 -0.3 Top 20 % 178 1,434 1.2 0.9 1.2 14.7 17.1 -0.04 -0.4 2nd 20 % 178 1,186 0.7 0.9 1.2 14.2 15.5 -0.4 -0.6 3rd 20 % 178 838 0.9 0.8 1.2 14.9 18.2 -0.1 -0.3 4th 20 % 178 559 1.0 0.9 1.3 14.7 16.9 -0.6 -0.7 Bottom 20 % 178 400 1.0 1.0 1.4 15.2 17.4 -0.3 -0.4 Bottom 10% 89 375 1.1 1.0 1.4 15.4 16.4 -0.3 -0.6 Bottom 5% (Least Experienced) 45 321 1.0 1.0 1.3 15.4 11.7 -0.8 -0.9 Top-Bottom 5% 45 – 0.9 0.01 -0.2∗ -0.5 7.2∗∗ 0.6 0.4 Top-Bottom 10% 89 – 0.4 -0.02 -0.2∗∗ -0.8 2.0 -0.1 0.3 Top-Bottom 20% 178 – 0.2 -0.07∗∗ -0.2∗∗ -0.5 -0.3 0.2 0.003 All Funds 890 963 0.9 0.9 1.3 14.8 17.0 -0.3 -0.5 ∗ SigniÞcant at the 90% conÞdence level. ∗∗ SigniÞcant at the 95% conÞdence level. ∗∗∗ SigniÞcant at the 99% conÞdence level.