The Role of Work Experience in the Effect of Education. on Mutual Fund Performance

Transcription

1 The Role of Work Experience in the Effect of Education on Mutual Fund Performance Author: Raphaël LOUTER Anr Supervisor: Dr. Alberto MANCONI Second reader: Dr. Michel VAN BREMEN Master Thesis Tilburg School of Economics and Management Master in Finance August 22, 2014

2 Preface This M.Sc. thesis represents the culmination of the work I have done in the context of the Master in Finance at Tilburg University. Research project ''The horse or the jockey? Fund vs. Fund Manager Characteristics and Mutual Fund Performance'' has been the main inspiration for my subject. Under the supervision of Alberto Manconi and in collaboration with four fellow students I have been working on a unique dataset that contains biographical information on fund managers. My participation in this project enabled me to use this dataset for my thesis. I am very proud to defend my work. Likewise, I want to thank a number people who have been particularly important for the realization of this thesis. First of all, I want to express my gratitude to my supervisor Alberto Manconi, he provided me excellent guidance and advice, which was always clear and valuable. His insightful comments and suggestions made me utilize and consider different points of view. In addition, I want to thank Michel van Bremen for his role as second reader. Second, I want to thank my fellow students Bart Kiesewetter, Mark van Kruijsbergen, Paul van Laarhoven and Richard Jansen for the collaboration during the data collection. Although the process was tough from time to time, I am proud of the final result. Last but not least, I want to thank my close friends, my mother and my girlfriend for their moral support. I hope you will enjoy reading this thesis. Tilburg, August 2014 Raphaël Louter 1

3 The Role of Work Experience in the Effect of Education on Mutual Fund Performance Raphaël Louter August 22, 2014 Abstract The aim of this thesis is to give an insight in the manager education-performance relation using a sample of single-manager US domestic-equity mutual funds over the period In line with the studies of Golec (1996), Chevalier and Ellison (1999a) and Gottesman and Morrey (2006) I find that more qualified managers show better riskadjusted performance. Managers holding an economic bachelor s degree, a bachelor s degree from a high-ranked university, an MBA degree or Ph.D. outperform their peers on average. In addition, I find that more work experience has a mitigating effect on this education-performance relation for managers holding an economic bachelor s degree or an MBA degree, perhaps suggesting that work experience compensates for lack of these forms of education. 2

4 Table of Contents Preface... 1 Abstract... 2 Table of Contents Introduction Theory Development Manager Education and Fund Performance Manager Education and Turnover Ratio Manager Education and Expense Ratio Manager Age and Fund Performance Data Data Description and Variables Econometric Approach Results Sorting Manager Education and Fund Performance Manager Education and Fund Characteristics Manager Education and Fund Performance, Controlling for Fund Characteristics Work Experience and the Effect of Education on Fund Performance Baseline Robustness and Discussion Conclusion References Appendix A: List of Constructed Variables Appendix B: Tables and Figures Table 1 Summary Statistics Table 2 Descriptive Statistics by Age Group

5 Table 3 Correlations Table 4 Manager Characteristics and Mean Performance: Figure 1 Proportions of Educated Growth Managers per Return Quintile Figure 2 Proportions of Educated Growth-Income Managers per Return Quintile Table 5 Manager Characteristics and Mutual Fund Performance: Table 6 Manager Characteristics and Mutual Fund Characteristics: Table 7 Manager Characteristics and Mutual Fund Performance with Fund Controls Table 8 Relative Differences in the Effect of Education on Performance Table 9 Work Experience and the Effect of Education on Fund Performance Table 10 Work Experience and the Effect of Education on Turnover and Expenses Table 11 Baseline Robustness Summary MBA and Bachelor Major

6 1 Introduction Investment-management ability is more rationally predictable than is often assumed: It s not just luck and mojo and intangible; it s teachable in a degree program and deductible from a resume Matt Levine, Bloomberg, November 8, 2013 Fund managers make investment decisions based on their personal preferences (Golec, 1996). Besides the personal preferences there are also other manager characteristics such as skill, education and work experience which might have an impact on the abilities of the managers and thus an impact on the investment decision (Chevalier and Ellison, 1999a). There is a wide variety in the educational backgrounds of mutual fund managers (Del Guercio and Reuter, 2014). From an investor s perspective it is desirable to know to what extent these educational differences have an impact on performance and to see if better, higher and more specialized education creates value. The education of a fund manager is a factor that might influence the decision whether or not to invest in a specific fund. In addition, for fund companies hiring managers and for those who invest in mutual funds, it is useful to know if and how this education-performance relation changes when mutual fund managers gain more experience, given that the main components of a resume are education and work experience. Several studies try to address the question whether manager characteristics lead to better mutual fund performance, and several papers conclude that outperformance is partly due to manager specific variables, in special, educational characteristics. Shukla and Singh (1994) find that CFA-chartered managers outperform other funds as a group. Golec (1996) finds that mutual fund performance, risk and fees are significantly impacted by manager characteristics. He finds that one can expect better risk-adjusted performance from younger managers with MBA degrees who have longer tenure at their funds. Chevalier and Ellison (1999a) document that the mean SAT score of undergraduate institutions is positively related to fund performance. Gottesman and Morey (2006) find that managers which attended better undergrad education and hold an MBA have systematically higher risk-adjusted excess returns. Thus, previous studies point out that education has a significant impact on fund performance; however, does education matter more for those managers who just started? It 5

7 might be possible that education matters less and less when a manager gains expecience. Can one really deduct performance from a resume? This thesis zooms in on the impact of manager education on mutual fund performance at different age levels. The main question is whether education has a persistent and constant impact on mutual fund performance as work experience increases. Therefore I formulate the following research question: Does experience moderate the effect of education on mutual fund manager performance? This thesis makes use of a sample of unique, hand-collected data regarding mutual fund manager information, combined with data from the CRSP Survivorship-Bias Free US Mutual Fund Database on actively managed single-manager US equity-domestic mutual funds over the period I use manager age as a proxy for experience. I split my sample into two subsamples, young (less experienced) and old (more experienced) managers. I define young managers as managers with an age under the median age (46) of all managers in my sample. Old managers are defined as managers with an age above the median age. The education-related manager characteristics of interest are the ranking of the undergraduate institution where a manager obtained a bachelor s degree and whether a manager is holding an economic bachelor s degree, MBA, CFA designation or a Ph.D. I find that more and better educated mutual fund managers outperform their less educated colleagues. Additionally, the outperformance due to an economic bachelor s degree or MBA is more pronounced among younger managers than among older ones (respectively and percentage point on an annual four-factor alpha). This might suggest that holding an MBA or the major of undergraduate education matters less and less as a mutual fund manager gains more experience. In addition, I find small evidence that fund manager education leads to lower fund expenses and portfolio turnover. There is no consistent evidence that more experience moderates this education effect on fund expenses. However, there seem to be indications that portfolio turnover becomes less negatively related to education for those managers holding an MBA degree and a bachelor s degree from a high-ranked undergraduate institution. Age initially has a negative effect on performance, I find small proof that this relation becomes less negative as managers gain more work experience. 6

8 This thesis makes two main contributions to the literature. These contributions are to the literature on the influence of manager characteristics on investor behavior. Existing studies of Golec (1996), Chevalier and Ellison (1999a) and Gottesman and Morrey (2006) focus on the relation between mutual fund performance and manager characteristics. The shared view is that quantity and quality of mutual fund manager education does have an impact on fund performance. These studies focus, among other things, on the impact of the average composite SAT score of the undergraduate institution, MBA degree, CFA designation and Ph.D. on performance. First, in addition to the impact of an MBA degree, CFA-charter, Ph.D. and the ranking of the undergraduate institutions, this thesis looks at the major of the bachelor s degree of a fund manager and the effect on performance. I find that the major also plays a role in fund performance. A manager with an economic bachelor s degree outperforms its peers on average. Second, in contrast to the papers of Golec (1996), Chevalier and Ellison (1999a) and Gottesman and Morrey (2006), this thesis delves into the relation between education and fund performance by adding an extra dimension, namely age (as an estimate for work experience). I find small evidence that education causes better performance for managers with less work experience; the education-performance relation is lower for more experienced managers holding an economic bachelor s degree or MBA degree. The remainder of the thesis is organized as follows. The second chapter gives a short introduction to the existing empirical findings, the theory and the hypotheses. In the third chapter I describe my data and the econometric methods used to do the empirical study. Chapter four contains the results derived from the empirical study. The fifth and last chapter concludes the findings of previous chapters and tries to answer the main question of this thesis. 7

9 2 Theory Development This chapter discusses the existing empirical evidence, the theory and my hypotheses. Section 2.1 is devoted to the relation between manager education and performance. Sections 2.2 and 2.3 focus respectively on the education-turnover and education-expenses relation. Age has also an important role in the theory, as I use it as proxy for work experience, therefore section 2.4 deals with the effects of age on performance. 2.1 Manager Education and Fund Performance A handful of academic studies examine the relation between mutual fund performance and managerial characteristics that may indicate knowledge, effort or ability. The most recurring managerial characteristics in these studies are manager education, age, tenure and gender. Shukla and Singh (1994) investigate whether a CFA designation leads to better performance. They study a selection of equity mutual funds over the period and find that CFA-chartered managers outperform manager without a CFA designation. Likewise, Golec (1996) analyses the impact of manager characteristics on fund performance, risks and fees. He finds that risk-adjusted performance, risks and fees are significantly impacted by manager characteristics. Younger managers with an MBA, who have longer tenure at their funds, perform better. Moreover, Chevalier and Ellison (1999a) find, using a sample of single-manager managed growth and growth and income funds over the period , that managers who attended undergraduate institutions with higher SAT scores have systematically higher riskadjusted performance; they find that the effect of an MBA degree is insignificant and unrelated to mutual fund performance. Gottesman and Morey (2006) contradict the findings of Chevalier and Ellison (1999a). Using the same proxy for SAT scores, they find SAT is not significantly related to performance on a consistent basis. Nevertheless, they do find that the mean GMAT score of the MBA program is positively and significantly related to fund performance. Additionally, they find that a CFA and Ph.D. are unrelated to performance. Andreu and Puetz (2012) show that managers holding both an MBA degree and a CFA designation have less extreme and more persistent performance results. When comparing the performance of managers with both degrees to managers who obtained only one of these degrees they find no statistical difference. 8

10 Consistent with the existing literature, I assume that education is positively related to fund performance. A degree indicates more knowledge, which in turn might lead to better performance. When a manager gets older and more experienced, lack of education might be compensated by the extra experience and the network of the manager. Thus, one could assume that the education impact on performance is bigger for young managers, fresh from college, equipped with the newest theories. This theory helps me to formulate the following hypothesis: H1 Education is positively related to fund performance; this relation is less positive for older managers. 2.2 Manager Education and Turnover Ratio Besides the manager characteristic-performance relation, several studies additionally describe the relation between fund turnover and fund manager characteristics. The turnover ratio measures how active a fund trades (Golec, 1996). The opinion under scholars, with respect to this manager education-turnover relation seems to be fairly consistent. Golec (1996) finds that funds with managers holding an MBA from high-gmat programs tend to have lower turnover than funds with managers from low-gmat programs. Chevalier and Ellison (1999a) look at the impact of SAT scores on turnover ratio and find that the mean SAT of the undergraduate institution of the manager is negatively related to turnover. Likewise, Gottesman and Morey (2006) discover that holding an MBA significantly reduces fund turnover. The thread of poor performance compels mutual fund managers to trade more actively, however this active trading could hurt or improve the performance (Wu, 2014). One believes that portfolio performance often suffers due to excessive churning (Chevalier and Ellison, 1999a). From a first intuition this makes sense, given the fact that expenses, fees and turnover costs get deducted from shareholder returns (Golec, 1996). The empirical evidence regarding the turnover-performance relation however is not that consistent (Wu, 2014). Where Grinblatt and Titman (1994) suggest that turnover is positively related to the ability of fund managers to earn positive abnormal returns, finds Carhart (1997) that turnover has a negative impact on performance. Ippolito (1989) instead finds that portfolio turnover is unrelated to fund performance. 9

11 Consistent with previous literature (i.e. Golec, 1996; Chevalier and Ellison, 1999a and Gottesman and Morey, 2006), I predict that more qualified managers have lower portfolio turnover. Due to their education they are better capable of picking the right stocks. However the turnover-education relation might mitigate when managers become more experienced (older) assuming that work experience compensates for the lack of education. Therefore I formulate the following hypothesis: H2 Education is negatively related to fund turnover; this relation is less negative for older managers. 2.3 Manager Education and Expense Ratio Expense ratio might also be related to manager education. Prior literature indicates that there is a negative coherence between education and fund expenses. Golec (1996) for instance, finds that the number of years of education and expenses are negatively related to fund expenses. He interprets this finding with the theory that education might indicate that a manager is better capable of doing his own research and keeping expense ratio low. Chevalier and Ellison (1999a) and Gottesman and Morey (2006) document that the mean SAT of the undergraduate institution attended by the managers and the expense ratio of funds are negatively and significantly related. Additionally, Gottesman and Morey (2006) find that managers with MBAs from lower GMAT programs have higher expenses. Even though one expects expenses might cause higher performance due to extra money spend on investment research, Carhart (1997) finds the opposite. He finds that expenses are negatively related to performance. Golec (1996) and Chevalier and Ellison (1999a) also find a negative relation between fund expenses and performance. However, Gottesman and Morey (2006) find expenses unrelated to performance. Prior literature with respect to the relation between education and expenses as well as education and performance support the theory that more qualified managers might spend less on investment investigation to define a good or bad investment decision. Due to education they are able to define good investment opportunities themselves and signal bad holdings on time. Therefore educated managers should have lower expenses. More work experience decreases the effect of education on expense ratio; as I expect that experience makes better education more superfluous. This theory forms the basis for hypothesis 3. 10

12 H3 Education is negatively related to fund expenses; this relation is less negative for older managers. 2.4 Manager Age and Fund Performance In the existing literature the opinions seem unanimous about the impact of age on performance. Golec (1996) documents that younger managers perform better than older managers. Chevalier and Ellison (1999a) and Gottesman and Morey (2006) share this view; they also find that manager age is negatively related to fund performance. Chevalier and Ellison (1999b) wrote a second paper where they find that younger managers hold less unsystematic risk and hold less extreme portfolios. Mutual fund management is a demanding job for which one needs stamina. One claims that aging has a negative impact on stamina and thus results in poorer performance (Golec, 1996). Likewise, older managers might have more job security which makes them less eager to create higher returns. More progressed in the career, on an older age, the impact of age on performance might become less negative. Older managers might have more job security, however the total amount of years of experience starts to add up and pay off in a positive way. Besides that, over the years managers build up a bigger network. This network may, for instance, result in better information sources, preferred positions in trades or improved access to IPOs (Chevalier and Ellison, 1999a). Therefore, I formulate the following hypothesis: H4 Manager age has a negative impact on fund performance; this relation is less negative for older managers. 11

13 3 Data This chapter discusses the data and the models I use in this study. Section 3.1 discusses in detail the data preparation, the composition of the various variables and the descriptive statistics, as well as the correlation matrices. Section 3.2 elaborates the econometric models I use to test the various hypotheses drawn in the previous chapter. 3.1 Data Description and Variables Starting point of the panel dataset are all available fund manager names of those funds, managed by a fund manager that had sole responsibility of a fund. We obtain these names from the CRSP Survivorship-Bias Free US Mutual Fund Database, hereinafter referred to as CRSP Mutual. This leaves us with a total of 3,915 fund manager names. For each of these names we search manually, online, for corresponding manager information. 1 The online search focuses on information with respect to: year of birth, gender, tenure, education, birthplace, hobbies, marital status and number of job changes of the fund managers. Most of the manager information is available on LinkedIn, company websites, the websites of: Morningstar, Business Week, Forbes, Bloomberg and online SEC filings. We check the data with a 100% overlap, to filter out errors and to make sure data is noted down correctly. Other manager specific information I obtain from CRSP Mutual, such as fund name, the date a manager started and ended working for a fund and the headquarter location of the fund. The next step focuses on fund information data. I employ CRSP Mutual for fund information regarding monthly fund returns as well as information on total net assets (TNA), portfolio turnover, fund expenses, Lipper classifications and fund objective codes for each fund. In addition, I retrieve information with respect to start en quit dates of the fund managers at the funds. I use the Fama-French factors database for information regarding risk factors, which I employ to compute performance variables. I obtain the following factors: the monthly market return which is defined as the value weighted return on all NYSE, AMEX and NASDAQ stocks, the risk-free interest rate which is defined as the one month treasury bill rate, high minus low (HML) measured by the difference in returns between high and low book-to- 1 I collected the data regarding the biographical information on fund managers together with Bart Kiesewetter, Mark van Kruijsbergen, Paul van Laarhoven and Richard Jansen under the supervision of Dr. Alberto Manconi. 12

14 market ratio firms, small minus big (SMB) measured by the difference in returns between small and large stocks and momentum (UMD) data defined as the difference in return between past winners and past losers. After I obtained the separate databases, I merge the manager information database and fund information database, based on fund number, year and month. Each set of fund information of year is linked to the manager information of the manager in charge of the fund at December 31 of year (Chevalier and Ellison, 1999a). Funds with less than 12 monthly return observations per year I omit from the data. The variables are computed in the following ways: 1. Alphas are computed using factor models. The four-factor alphas (Carhart, 1997) are computed using regression model 1., (1) where, is the return for fund at time, is the risk-free interest rate. is the alpha for fund, is the value weighted return on the NYSE, AMEX and NASDAQ stocks in excess of the risk-free rate.,, are the small minus big, high minus low and momentum factors as defined earlier in this section. I also compute alphas using a one-factor model (Jensen, 1969) and a three-factor model (Fama and French, 1993), where only the first one or the first three risk factors in equation 1 are controlled for. Equal to Chevalier and Ellison (1999a) I use an estimation period of one year with monthly return data. I multiply the monthly alpha by twelve to obtain a one year alpha. Since alpha is measured as a percentage, I change it in a percentage number. 2. Benchmark-adjusted returns are calculated by taking funds annual fund return and subtracting the average annual return on funds with the same Lipper category code. 3. Beta is the coefficient of the one-factor model regressions of fund return minus the risk-free rate on return on the market minus the risk-free rate. Beta is computed with one year of monthly return data. A longer estimation period gives more data but there is a bigger chance that the firms risk characteristics may change over time (Damodaran, 2011). 13

15 4. Total Net Assets is used as proxy for fund size and is computed by the natural logarithm of the average 12-month TNA of each fund, expressed in millions. 5. Expense ratio is directly obtained from CRSP Mutual and defined as the ratio of total investment that shareholders pay for the fund s operating expenses, which include 12b-1 fees Turnover ratio is directly obtained from CRSP Mutual and defined as the minimum of aggregated sales or aggregated purchases of securities, divided by the average 12- month TNA of the fund. 7. Manager Age is measured in years and defined as year of the observation minus year of birth of fund manager. 8. Manager Tenure is measured in years and determined by the year of the observation minus the year the manager started working at a fund minus one. 9. Manager Bachelor Rank is a dummy variable that is equal to one if a fund is managed by a manager that went to a high-ranked school. To define high-ranked school I make use of average SAT scores of students who enter their first year of college. This data is available on the website of The Center of Measuring University Performance (MUP). The MUP reports median SAT scores of 770 major research universities in the US over the period I add up all the median SAT scores per university for the eight years of available data and define the twenty-five universities with the highest sum of median SAT scores as the high-ranked schools. The dummy equals zero for non-high-ranked school managers. 10. Manager Major is a dummy variable that indicates if a fund is managed by a manager who attended an economic major at an undergraduate institution. The manager data includes the name of the bachelor s degree of each manager; I define managers with an economic major as managers with a bachelor s degree in finance, accounting, 2 12b-1 fees are defined by CRSP Mutual as marketing and distribution costs. 3 I use the time period to take in account any big changes in the ranking of universities. Of course the question arises whether SAT scores from 2003 are representative for managers who obtained their bachelor s degree 20 years ago. However, the rankings of undergraduate institutions in the 2003 are very similar to the rankings in the early 1980s. Gottesman and Morey (2006) find a correlation of 86.15% between the 1983 and 2003 SAT rankings. 14

16 economics, econometrics, business, banking or a hybrid of aforementioned. 4 I assign a dummy that is equal to one if a manager holds an economic bachelor s degree, I assign a zero otherwise. 11. Other dummy variables are Manager Gender (1=female, 0=otherwise), Manager MBA (1=manager has MBA degree, 0 otherwise), Manager CFA (1=manager has CFA designation, 0=otherwise), Manager Ph.D. (1=manager holds Ph.D., 0=otherwise) and fund objectives where growth is the baseline objective. After creating the variables I turn my data into yearly data. I keep all data over the period This period includes mainly bull markets, however it also includes a temporarily bear market in (Meric et al., 2002). A period with only bull markets might influence the results. 5 For optimal comparability across funds I solely use domestic-equity growth and growth-income funds equal to Chevalier and Ellison (1999a). I omit all other funds, which include bond funds, international funds, sector funds and cap-based funds because these type of funds trade and hold minimum amounts of domestic equities (Wermers, 2000). Index funds are also left out of the sample, since this thesis focuses on actively managed mutual funds. To eliminate outliers I winsorize all variables at a 99% level, this modifies a proportion of the observations in each tail. Previous adjustments leave me with a panel dataset of 11,777 observations over the period , containing manager characteristics and performance information on 1,058 fund managers. Table 1 reports the summary statistics on the various manager and fund characteristics. 6 The oldest manager in my sample is 78 years old the youngest is 28. The table reports that the mean manager age in my sample is More than half of the funds in the sample are managed by managers that hold an MBA degree, 64%. The same is observable for manager holding a CFA designation, 54.6%. The average overall performance of the funds in the sample seems to be negative with an average annual four-factor alpha of minus 1.287%, statistically significant at a 1% level (not reported in the table). 4 Data regarding different bachelor s degrees in the US I obtained from the website of the National Science Foundation: 5 Gottesman and Morey (2006) argue that the time period Chevalier and Ellison (1999a) use covers a relatively bullish market ( ). He states that their findings are the result of an error of the time period they use. 6 All tables are included in Appendix B at the end of this thesis. 7 This is slightly higher than the mean manager age in the samples of Chevalier and Ellison (1999a) and Golec (1996), which respectively have samples with a mean manager age of and

17 Equal to Chevalier and Ellison (1999b), I use manager age as proxy for experience. I determine the median age of all observations, which is equal to 46. I split my sample in two groups based on this median age. I consider all managers older than the median age old (experienced). Managers younger than the median age I consider young (inexperienced). Table 2 shows and compares the means of both subgroups per manager and fund characteristic, furthermore the medians are shown. Over the period older managers seem to have more tenure at their fund on average, this difference statistically significant at 1% level. A higher proportion of young managers seem to have an economic major (46.3%) compared to old managers (34.3%), this difference is statistically significant at 1% level. The difference might be explained by the fact that fund managers start their career in another field and make the switch to fund management later in their career, as a specialist in a certain industry. The proportion of older managers who obtained a Ph.D. is on average twice as high as the proportion in the sample of young managers, 2.4% against 4.8%, this difference is also statistically significant at a 1% level. This could be explained with the theory that Ph.D. managers work in research in the beginning of their career. A first look at the averages of the risk-adjusted performance variables shows that young managers have better performance over the period compared to old managers over the same period. This difference is statistically significant at the 1% level for benchmark-adjusted return and for three-factor- and four-factor alphas. To check if there are any signs of multicollinearity the correlation matrices of the explanatory variables are presented in Table 3. The table shows the correlations of the pooled dataset as well as the correlations of the subsamples of old and young managers. There are no indications of multicollinearity, since the highest correlation is -0.36, representing the correlation between fund size and expense ratio. This negative relation is not surprising, given the fact that costs become relatively lower as fund size increases, as a result of scale economies (Golec, 1996). 3.2 Econometric Approach In order to answer the hypotheses I make use of several econometric techniques. The main statistical tool used in this thesis is the OLS method. This method helps me to estimate the coefficient of the linear relationship between the dependent and independent variables. In this 16

18 thesis I elaborate four multiple linear regression models to test the various hypotheses. Furthermore, I use several alternative specifications of regression equation 4 to test for robustness of the results. Education and fund performance To test the effect of education on fund performance, I use the OLS model as specified in regression equation 2. Besides the education variables I control for fund and manager specific characteristics, as specified below., (2) denotes the performance, specified as either benchmark-adjusted return, CAPM-, thee-factor-, or four-factor alpha of fund at time, is a vector containing the manager education variables: Manager Bachelor Rank, a dummy variable which is one if manager went to a high-ranked undergraduate institution, zero otherwise, Manager Major, a dummy variable which is one if manager obtained an economic bachelor s degree, zero otherwise, Manager MBA is a dummy equal to one if manager holds an MBA degree, zero otherwise, Manager CFA and Manager Ph.D. are dummies that are one if manager has respectively a CFA designation or Ph.D. degree, zero otherwise. represents a set of controls, including the following manager and fund characteristics: manager age, manager tenure, manager gender, log NTA, lagged expense ratio, lagged turnover ratio and fund objective. Because fund turnover ratio and expense ratio can simultaneously influence performance I use the lagged expense and turnover ratios (Gottesman and Morey, 2006). I control for fund objective since requirements regarding return might be recorded in a funds charter (Golec, 1996). All variables are defined as described above, furthermore all variables are described in detail in Appendix A. The standard errors are robust, clustered around funds and I use pooled data. Education and fund characteristics The OLS model I employ to get a better sight on the causal relation between manager education and fund characteristics is presented in equation 3. The fund characteristics of interest are systematic risk (beta), fund expenses, portfolio turnover, and fund size., (3) 17

19 refers to the fund characteristic of fund at time is a vector containing the variables Manager Bachelor Rank, Manager Major, Manager MBA, Manager CFA and Manager Ph.D. is a set of controls, including: manager age, manager tenure, manager gender, fund objective and size. The control variable size is left out in the model specified with fund size a dependent variable. All variables are defined as in section 3.1; furthermore Appendix A contains a detailed overview of all variables. The regression makes use of pooled data and standard errors are clustered around funds. Work experience and the effect of education on mutual fund performance The OLS models as presented in equation 4 and 5 aim to give more insight in the change in the education-performance, education-turnover ratio and education-expenses relations, when managers get more experienced. The basis of both models is formed by equation 2 and 3, additionally a dummy is included which indicates whether a manager is experienced (old) as well as a full set of interaction terms. (4) {, and, are specified as in regression equation 2. is a vector of interaction terms between the each of the education variables: Manager Bachelor Rank, Manager Major, Manager MBA, Manager CFA, Manager Ph.D. and the dummy variable for fund at time. identifies a vector with interaction terms between each of the controls: manager age, manager gender, manager tenure, fund size, lagged turnover ratio, lagged expense ratio and fund objective and the dummy variable All variable definitions are displayed in section 3.1, furthermore a list with a detailed overview of all variables is available in Appendix A. The standard errors are clustered around funds and the model makes use of the pooled dataset. 18

20 (5) { where is defined as either turnover ratio or expense ratio. refers to a vector with the five education variables and refers to a vector with the usual control variables both equal to the variables in regression equation 3. symbolizes a vector of interaction terms between all manager education variables and the aged dummy. Likewise, is a vector of all controls, interacted with dummy. The model makes use of the pooled dataset, standard errors are clustered around funds. 19

21 4 Results This chapter presents the results of the various tests. I start with sorting my data in section 4.1. The following four sections present the regression results of regression equation 2 to 5. Section 4.2 shows a basic regression of manager characteristics on performance, without fund characteristics as controls. Section 4.3 shows the regression results of manager characteristics on fund characteristics. Section 4.4 and 4.5 discuss whether work experience moderates the impact on fund performance, expenses and turnover. Finally, section 4.6 deals with the performed robustness tests and refutation of alternative explanations of my findings. 4.1 Sorting To get a first sense of the effect of education on fund performance I sort all observations into groups by experience, fund objective code and respectively by Bachelor Rank, Bachelor Major, MBA, CFA and Ph.D. and compare the mean annual benchmark-adjusted returns over the period , as shown in table 4. After sorting, the average returns show that over the period educated managers outperform less educated fund managers. We observe this outperformance in all groups, except for young managers who obtained a CFA designation. The strongest indications of this outperformance we observe for young managers, managing growth funds and holding an MBA degree. These managers outperform their non- MBA holding peers in the same group by percentage points annually. The meancomparison test shows a statistical significance of 1%. Old growth fund managers who did not attend high-ranked undergraduate institution perform percentage points worse than their colleagues in the comparison group who attended a high-ranked school, statistically significant at a 1% level. The findings are consistent with the theory that education enhances fund performance. However there seems no direct indication that for experienced managers the education becomes a less dominant performance driver. Next, I sort the benchmark-adjusted returns by age group and objective code and subdivide the returns into quintiles. The first quintile represents the 20% lowest returns over the period for each subgroup. The fifth quintile represents the 20% highest returns 20

22 over the same period. For each quintile I compare the proportions of educated managers. Figure 1 and Figure 2 show the proportions of educated managers per quintile. Given the theory that education leads to knowledge and skills which in turn leads to better performance one expects a higher proportions of educated managers in the top quintiles. However, there seems to be no clear pattern between the proportions of educated managers and the quintiles of benchmark-adjusted returns in any of the groups. Note that Ph.D. is left out since the proportions of managers holding a Ph.D. are so small that comparison across quintiles is hardly possible. 4.2 Manager Education and Fund Performance The next step in my analysis is the causal impact of education on performance. I use regression equation 2 to regress four performance measures on the manager education, the only controls in this first pooled data regression are the fund objective dummy with growth funds as baseline, manager age, manager gender and tenure at fund. The regression results are displayed in table 5. What immediately strikes is that the coefficients of four out of five education variables show a positive sign; furthermore these are statistically significant, except for the CFA dummy which is statistically insignificant. When we look at the economic impact of the variables we see that a manager who obtained a bachelor at a high-ranked undergraduate institution outperforms percentage point on an annual four-factor alpha. Obtaining an economic bachelor improves the same riskadjusted performance measure with percentage point and managers holding an MBA outperform their non-mba colleagues with percentage point, ceteris paribus. We observe the biggest economic impact for managers holding a Ph.D., the coefficient indicates that the performance is superior to non-ph.d. managers with percentage point. Age shows to be negatively related to performance, when a manager gets 10 years older, the four-factor alpha decreases with percentage point annually. Gender seems to be unrelated to performance. Tenure at fund is negatively related to performance, however only statistically significant in the three-factor alpha specification. The results are consistent with the theory that education has a positive impact on performance due to extra skill and knowledge. Both education quality (high-ranked 21

23 undergraduate institution) and education quantity (i.e. economic bachelor, MBA degree or Ph.D.) seem to have a positive impact. The negative age-performance relation is consistent with the theory that more experience leads to more job security, which might make fund managers less eager to outperform. 4.3 Manager Education and Fund Characteristics The results in the previous section might be the result of cross-sectional differences in manager behavior (Chevalier and Ellison, 1999). Therefore, this section deals with the impact of education on fund characteristics. The characteristics of interest are systematic risk, expenses, portfolio turnover and fund size. I employ regression equation 3 with these four fund characteristics as dependent variables and the education variables as explanatory variables, controlling for manager age, tenure at fund, manager gender, fund objective and fund size. The latter control variable is left away in the fund size regression. Table 6 shows the coefficients of the fund characteristics regressions. None of the education variables has a significant effect on systematic risk. Gender and tenure are also unrelated to beta. Age alone appears to have a significant impact. One year increase in manager age results in a decrease of beta with This reflects the fact that the average beta of funds managed by young managers in my sample is 1.010, the mean beta among funds managed by old managers is Both Golec (1996) and Chevalier and Ellison (1999a) find that funds managed by managers that hold an MBA have higher betas. My findings are more consistent with the findings of Gottesman and Morey (2006) who find that education is unrelated to systematic risk. This difference might be explained by the theory that the studies of Chevalier and Ellison (1999a) and Golec (1996) covered a relatively bullish market where qualified fund managers might take a more risky position (Gottesman and Morey, 2006). As predicted, there are some indications that expense ratio is negatively related to manager education; three out of five education variables are negative and statistically significant. A manager holding an MBA has on average a percentage point lower expense ratio, statistically significant at a 5% level. A bachelor s degree from a high-ranked undergraduate institution decreases expense ratio with percentage point, significant at a 1% level. Also managers holding a Ph.D. have significant lower expense ratios. 22

24 These results are consistent with the theory that good educated managers have a benefit from their education. They are well able to determine what good and bad investment decisions are, so that they have less consultancy fees, this might bring the expense ratio down. Turnover ratio shows no consistent relation to education. The coefficients of manager major and Ph.D. are both negative and significant; CFA however shows a positive sign and is also significant. The theory suggests that turnover ratio is negatively related to education, since abundant stirring in a portfolio may lead to poorer results (Chevalier and Ellison, 1999a). Albeit, I find no convincing proof in the signs of the coefficients although there are some indications. I find a strong relation between size and tenure at fund; one year extra tenure at a fund increases the logarithm of the total size of the company by Managers stay longer at big funds, perhaps suggesting that managers start at small funds and job-hop to bigger funds where they have more career perspective and thus stay longer. The negative coefficient of the variable Manager Age confirms this theory; however this coefficient is not statistically significant. 4.4 Manager Education and Fund Performance, Controlling for Fund Characteristics Now that we have a better insight in the behavioral differences among fund managers I include the fund characteristics in the performance regression as controls. These extra control variables take into account possible endogeneity, as for example fund size might explain better performance because there might be better access to information resources. Likewise, high fund expenses and low turnover might explain performance because more money is allocated to do investment research (Chevalier and Ellison, 1999a). I run the performance regression as in equation 2 with the pooled data. In addition, I run the regressions with subsamples old and young managers separately. Table 7 shows the coefficients of these performance regressions. The coefficients of the education variables show that the extra controls have a small impact on the statistical significance of the estimates of the education variables as presented in table 7. We observe no sign changes in the coefficients. Nevertheless, the economic impact of the coefficients decreases. This indicates that some of the performance, originally explained by the education of the manager, is actually explained by fund characteristics. 23

25 Looking at the four-factor alpha regression in panel A we observe that the economic impact of a bachelor s degree obtained at a high-ranked university drops from to percentage point, when including the extra controls. An MBA degree increases the CAPM alpha percentage point annually. The coefficient of this variable lost its statistical significance in the four-factor alpha regression. Contrasting these findings to the findings of Chevalier and Ellison (1999a), they also report MBA positive but insignificant in the fourfactor alpha specification. The three-factor alpha of a Ph.D. manager is on average percentage points higher compared to a non-ph.d. manager, ceteris paribus. Again a CFA designation seems to be unrelated to fund performance. Turnover ratio and expense ratio seem to be negatively related to fund performance. This is consistent with the theory, that excessive churning in portfolios leads to poorer performance. Likewise, funds with fewer expenses perform better, which is consistent with the theory that if the fund manager is better able to pick the stocks with fewer costs this will lead to better results. The manager education-performance relation of the subsamples of young and old managers tells us a different story. For inexperienced managers in Panel B, the main statistically significant education variables are bachelor major and MBA, which have respectively a and percentage point impact on an annual four-factor alpha. Furthermore age is negatively related to performance in three out of four performance regressions. The subsample of old managers in Panel C shows that education ranking is positive and statistically significant in all performance regressions. When an old manager attended a high-ranked undergraduate institution this increases its four-factor alpha with percentage point. We cannot draw any conclusions about the difference between the estimates of subsamples of old and young manager; therefore I test the differences in the coefficients of these subsamples in the next section. 4.5 Work Experience and the Effect of Education on Fund Performance In this section I test whether the effect of education on performance is significantly lower for more experienced managers. I compare the coefficients of experienced and inexperienced managers (as shown in table 7, Panel B and C) with each other with a Chow test to find out 24

26 more about the statistical significance of the differences. The Chow tests are shown in Table 8. Comparing the manager education-return relationship for young and old managers, I find a significant difference with respect to the variable Manager Major. The impact of an economic major from an undergraduate institution on performance is significantly lower for older managers. I find this statistical significance for three out of four performance regressions. I find small indications of differences in the coefficients of the variable Manager MBA. The impact of an MBA degree on performance seems to be significantly lower in two out of four performance regressions for more experienced manager. The impact of a CFA degree shows a significant increase in the first two performance regressions, indicating that the impact of a CFA designation becomes stronger for older managers. To get a better overview on the economic impact of the changes in the effect of education on performance when managers become more experienced, I employ the linear regression with interaction terms as I describe in regression equation 4, the results are displayed in table 9. For young managers an economic bachelor s degree is positively related to four-factor alpha, with an economic impact of percentage point annually. When a manager is old this impact decreases significantly with percentage point. This result is consistent with the theory that managers leaving college with an economic major have an advantage in terms of financial knowledge over managers with non-economic majors. This advantage decreases over the years as industry knowledge and network might compensate for lack of education. The variable Manager MBA shows a pattern equal to the variable Manager Major, positive and statistically significant for young managers, with an economic impact of percentage point on an annual four-factor alpha. This impact is statistically significantly lower ( percentage point) for old managers in the four-factor alpha performance specification. The theory that managers with MBAs benefit more from it when they just leave grad school seems to be consistent with the results, as managers get older the benefits of the MBA degree become less important as some of the theories they learned become outdated and as network and industry experience start to play a more important role. The coefficient of the interaction term has a positive sign and is statistically significant in two out of four performance regressions. When a manager is old this 25