Fund Performance and Top Management Turnover: Evidence from the UK Unit Trust Industry

Fund Performance and Top Management Turnover: Evidence from the UK Unit Trust Industry Zhichao Zhang a, Li Ding a, Si Zhou a a School of Economics, Finance and Business, Durham University Corresponding author, contact address: Durham Business School, Durham University, Mill Hill Lane, Durham, DH1 3LB, UK. Email: si.zhou@durham.ac.uk Electronic copy available at: http://ssrn.com/abstract=1535829

Abstract Managers play a pivotal role in the operation of funds. Using the data from UK unit trusts during 1990 to 2009, we analyse the relation between fund performance and management turnover in the UK fund market. We compare the performance, shifts in the level of risk and changes in investment preferences between the pre- and post-replacement periods. A bootstrapping simulation is applied to further examine fund companies capability of identifying and hiring managers with genuine skills of stock selection. Evidence shows that there exists a close relationship between fund performance and top management turnover. Managers replacement can be predicted by managers underperformance. The managerial mechanism in UK fund companies is found to be capable of identifying poor performance and, based on this, fund companies will try to improve fund performance through managerial replacement. Managers in danger of replacement tend to window-dress their performance record by increasing holdings of smaller size stocks, which increases the risk of the portfolios under their management. Our bootstrapping simulations indicate that, in the UK mutual fund market, fund companies are generally able to screen out lucky managers and replace them with skilful ones. 2 Electronic copy available at: http://ssrn.com/abstract=1535829

Fund Performance and Top Management Turnover: Evidence from the UK Unit Trust Industry Introduction Mutual funds are professionally managed, pooled investment vehicles (Gremillion, 2005). One distinct feature of such collectively managed investments is that, while their managers are appointed because of their specialist financial knowledge and information advantage, they are usually not the major risk-bearers of the fund under their management. This creates agency problems since, without an effective control system, these managers may deviate from the interests of residual claimants or fund investors. The cost thus incurred by the fund could be considerable. To control such agency problems in the decision process, Fama and Jensen (1983) show that it is essential to have an effective corporate governance system within which the internal monitoring mechanism plays a critical role. In the context of mutual funds, the internal monitoring mechanism involves measuring the performance of fund managers and implementing rewards and punishment, including dismissal. This internal monitoring, in tandem with the pressure exerted by the external managerial labour market through investors hunting for better performed managers, drives the managers to satisfy the interests of the investors. The extant literature has devoted great attention to evaluating the performance of mutual funds using various assessment methods. Major research in this field includes Jensen (1968), Fama and French (1993), Carhart (1997), Grinblatt and Titman (1989, 1993), 3 Electronic copy available at: http://ssrn.com/abstract=1535829

Goetzmann and Ibbotson (1994), Wermers (2000) and Pastor and Stambaugh (2002b), among others. In parallel with the rapid development of theoretical models, extensive empirical research has been attempted to test the effectiveness of corporate government in the fund industry, as in Chevalier and Ellison (1999a), Khorana (1996), Hu et al. (2000), Baks et al. (2007), Jin and Scherbina (2005), Tonks (2005) and Shinozawa (2007). These studies examine the empirical efficacy of the fund industry s internal and external controls, with a overwhelming focus on the potential relation between the performance of actively managed funds and the fate of their managers (Baks et al., 2007). Existing research however is mainly concerned with the conventional notion of managers performance, and overlooks the linkage between the managers genuine skills and fund companies decisions. This tends to bias the analysis of the fund industry s governance efficiency. One weakness in the conventional literature is the small sample size. In Khorana (2001) and Gallagher et al. (2006), the design of the analysis of managers turnover follows that of Denis and Denis (1995), which considers a sample of monthly fund returns of only three years in the pre- and post-periods of managerial replacement. Other drawbacks of conventional research involve problems of the non-normality of funds historical returns and benchmark factors. Stocks that have been components of the portfolio of a fund s investment often show non-normality in their returns. This also applies to the returns on the market benchmark deployed in the evaluation procedure. Hence, prior investigations may over-reject or under-estimate the performance of the 4

funds that lie at the extreme tails of the cross-sectional distribution. Moreover, conventional methods do not distinguish between performance which is related to genuine skills (good/bad), and mere sample variation. Without further separating the performance from these two influences, the previous literature seems hardly able to identify whether the managers performance is driven by luck or by poor skill, and whether they should be rewarded or penalized accordingly. As a consequence, this blurs the analysis of the efficiency of corporate governance in the fund industry. The focus of the previous literature has generally been put onto the US mutual fund industry. The UK unit trusts however can provide a weighty case for advancing the study in this field, as the UK fund market is one of the largest in the world and also among the most sophisticated. Findings on the UK case may collaborate with the American studies and are important in their own right in that they may reveal the effectiveness of the internal monitoring mechanism in the UK funds, and hence how well the interests of investors can be protected in the UK unit trusts market. This research aims to better our understanding of the corporate governance in the fund industry by shedding critical light on the effectiveness of the internal monitoring and control system in the UK equivalent of equity mutual funds, equity unit trusts. First, we analyse the relation between top manager turnover and the performance of UK unit trusts. The sample covers the period from 1990 to 2009, and our research focuses on the interaction between managerial replacement and fund performance. Second, a simulation procedure is implemented to further test whether, in their managerial dismissal and appointment, UK fund companies can distinguish between the managers 5

with genuine skills and those who are lucky or poorly skilled. To analyse top manager replacement we use a series of methods, including performance evaluation based on the factor models, percentile ranking, and sample matching test and sensitivity analysis. We divide funds that have experienced top manager replacement into two time intervals or subsamples, namely the pre-replacement period and the post-replacement period. A variety of methods are then applied to measuring factors that affect the performance in the two subsamples. The essence of these methods is to construct comparisons of factors, such as changes in the abnormal performance and shifts in portfolio risks, between the pre- and post-replacement periods so as to identify to what extent the corporate governance system of fund companies is effective in affecting the working and performance of the underlying funds. To achieve the second research objective, i.e. to determine whether fund companies can distinguish between skilled managers and lucky or poorly skilled ones, we conduct the bootstrapping simulation. This methodology is also employed by Kosowski et al. (2006) and Cuthbertson et al. (2008). Research by Bickel and Freedman (1984) and Hall (1986) demonstrates that bootstrapping can improve the approximation of the true distribution of funds abnormal returns by recognizing the thick tails among individual funds. Horowitz (2003) also shows that in Monte Carlo experiments the bootstrapping analysis can reduce the difference between normal and actual probabilities in rejecting a proposed null hypothesis. Moreover, through comparing the actual estimation of abnormal performance with the luck distribution given by bootstrapping simulations, we are able to distinguish the managers with real stock selection skills from those who are simply lucky. Hence, such a method could provide additional evidence on the 6

effectiveness of the internal and external controls in the fund industry. Our results suggest that there exists a close relation between fund performance and top management turnover. We find that managers replacement can be predicted by historical poor performance. Also, results from the matching sample analysis indicate that the magnitude of decrease in the performance of peer funds without replacement is considerably less than that of the funds in the replacement group. Thus, the corporate governance mechanism in fund companies can identify truly poor performance. Fund companies will then try to improve the situation through managerial action. Our results suggest that, in the post-replacement period the previously inferior performance will be improved by the newly installed management. Given the close relation between performance and investors preferences, fund companies are more likely to hire managers with better skills to improve the performance record. In addition, the results of our analysis of performance decomposition show that the potential outgoing managers are likely to window-dress the performance record by placing more weight on smaller size stocks, since stocks of smaller companies may show better earnings potential. However, these stocks may also exhibit more fluctuations in their returns. In the bootstrapping simulation, we find that about 64% of the funds are able to identify lucky managers and replace them with skilled ones. This indicates that lucky managers seems to be a factor that is taken into consideration by fund companies in the performance evaluation procedure. The remainder of this chapter is constructed as follows. The next section will review the theoretical background of our empirical investigation, including a discussion on the limitations of the previous research. Section 3 sets out the main hypotheses of our 7

research. Then, we describe the data and methodology in Sections 4 and 5, respectively. Section 6 is mainly concerned with interpreting the results of the empirical investigation. The final section presents the conclusion. 8

II. Related Literature In recent years, much effort has been devoted to analysing the effectiveness of corporate governance under different mechanisms. Past literature on controlling the operation of corporate managers has investigated the disciplinary forces from the labour market, mergers and acquisitions and external product markets. Major research, such as that of Fama and Jensen (1983) and Shleifer and Vishny (1986), has made crucial contributions to the field. Research of managerial compensation has shed light on the effect of incentives on corporate managerial operation, and on the relation between various stakeholders in terms of the influence on managers behaviour. A large body of evidence has been uncovered by researchers regarding the relationship between firm performance and top management turnover. Major research in this regard includes that of Denis and Denis (1995), Murphy (2000), Engel et al. (2003), Farrell and Whidbee (2003), and Jensen et al. (2004). Most studies document that in ordinary corporations top management turnover can be predicted by poor performance (See Murphy, 2000, Jensen et al., 2004 Kaplan and Minton, 2008). Given the notable difference in the governance structure of mutual fund industries, attempts have been made in the literature evaluating the impact of managerial turnover of fund companies on their underlying funds performance, and examining factors such as managers characteristics, compensation structure and governance mechanism (See Chevalier and Ellison, 1997, 1999a, 1999b, Wermers, 2000, Khorana, 1996, 2001, Elton et al., 2003, Gallagher et al., 2006). The mutual fund industry traditionally pays close attention to the relation between fund performance and quality of human capital. Participants of the fund market including 9

fund investors, financial brokers, trustees and financial media scrutinize behaviours of top managers as an integral element of fund investment. Top managers are held the responsibility for deciding on portfolio composition and implementing trading strategies, especially in actively managed funds (Gallagher et al., 2006). Behind this is the notion that associates fund performance with the performance of fund managers. To ensure the quality of fund managers, and hence performance of the funds, the governance mechanism in fund companies is tasked to make the appropriate appointment or dismissal of subordinated managers. As such, it is essential to understand how fund companies react to underperforming managers and, related to that, how top management turnover impacts fund performance. Khorana (1996, 2001) was among the first to investigate the effectiveness of internal and external governance mechanisms in the US mutual fund industry, through comparison of fund performances between pre- and post- manager replacement periods. His results indicate that on average funds experienced two years of underperformance before the month of replacement. Following the research on CEO turnover by Denis and Denis (1995), Khorana finds significant improvement in poor performing funds after managers replacement. In the empirical investigation, he divides the US equity funds into two sub groups according to their pre-replacement record of performance. The research documents that, in the negative performance sample, alpha decreases in the pre-replacement period and recovers afterwards. In the positive performance sample, the returns exhibit deterioration in the post-replacement period. The outcome implies that the US fund market is effective in penalizing the managers who performed poorly. Chevalier and Ellison (1999b) examine the impacts of turnover with an emphasis on the 10

role of managers age. They suggest that younger managers are more likely to be replaced when a fund s risk taking significantly deviates from the average portfolio risk of its peers. Additional results of the influence of fund management turnover are provided in Hu et al. (2000), Jin and Scherbina (2005), Evans (2006) and Gallagher et al. (2006). Jin and Scherbina (2005) consider the influence of manager replacement on the funds performance from a portfolio holding perspective. Their results demonstrate that the disposition effect, i.e the tendency to hold poor performing stocks for far too long, exists in the behaviour of mutual fund managers. This effect reduces fund performance, which calls for management replacement. Their research suggests that new managers are more likely to change their inherited portfolios, preferring to sell poor performing stocks rather than better ones. As a consequence, regular replacement of fund management is imperative. Managers replacement may take place in the form of their promotion, demotion or early retirement, etc. In practice, the replacement may be triggered by various reasons, which in turn may have different impacts. Previous research has attempted to study these reasons and their impacts on fund performance. Evans (2006) suggests that managers alpha is a factor to be considered in making the decision on promotion or demotion. Baks et al. (2007) find significant evidence of superior performance by promoted managers in a co-integration analysis which compares the performance persistence in pre-promotion (demotion) and post-promotion (demotion) periods. Previous literature has presented a considerable amount of evidence indicating 11

manager dismissal to be negatively associated with fund performance in the pre-replacement period, and improvement in performance after manager replacement. However, it is plausible that this association is caused by luck or accident. A natural question remains as to whether fund companies have made the right decision in dismissing those managers whose under-performance is caused by poor skills and appointing more skilful ones. Much of the previous research on this issue relies on results from conventional OLS regression. These, however, may be biased due to small sample size or non-normality in funds historical returns and benchmark factors. It is possible that managers produce a superior performance due to luck (sample variation) in the performance evaluating procedure, while genuinely skilful managers may have been forced to leave as a result of bad luck. Given this possibility, not all performance improvements can be portrayed as having an effective internal governance mechanism. To test this possibility, Kosowski et al. (2006) introduce the bootstrap simulation into fund performance evaluation to detect genuine stock-picking skills among fund managers. According to Bickel and Freedman (1984) and Hall (1986), bootstrapping would improve approximation of the distribution of funds abnormal returns through recognition of the influence of thick tails in an individual fund s performance. Horowitz (2003) also suggests that, in Monte Carlo experiments, the bootstrapping analysis reduces the difference between normal and true probabilities in rejecting a proposed null hypothesis. Kosowski et al. (2006) apply the bootstrapping method to the performance procedure of US equity mutual funds. They bootstrap the historical returns of 2118 open-end domestic equity mutual funds which survive for at least five years. Their results imply that the funds with performance lying at the right or left tails among the cross-sectional 12

funds are not solely due to luck. They find significant evidence to support skill-led superior performance and performance persistence in US growth-oriented funds. This finding differs from that of conventional evaluation that claims that managers with superior performance are not skilful. Cuthbertson et al. (2008) show that, while most of the poor performing funds of UK domestic equity unit trusts demonstrate bad skills, the funds that that lie between 5% and 10% of the top cross-sectional funds do show significant stock-picking skills. Their results are consistent with that of Kosowski et al. (2006). Extant literature shows that poor performance may trigger managers replacement. Given this possibility, under-performance of the funds where managers have been replaced should centralize to the left tail of the cross-sectional performance distribution. This then makes it possible to implement bootstrapping simulation in skill and luck separation. Such simulation of sample variation will provide us with a better understanding of the manager evaluation mechanism in the fund industry. It can also shed additional light on the real efficacy of internal monitoring in the funds industry. Extensive evidence has been reported in previous research confirming the relationship between top management turnover and fund performance. However, much of such research does not take into consideration non-normality of distribution among the cross-sectional returns. This leads to model misspecification and hence weak estimation results. It also leaves open the question whether the internal governance mechanism of the fund industry is efficient enough to initiate apt replacement. Given the limitations in the previous literature, our research will focus on analysing top 13

management turnover based on the UK unit trusts market. We will design an analysis to focus on the shift of performance, market risk, portfolio risk and factors loading around the turnover. We will also include the bootstrapping method to simulate sample variation, which will provide us with luck distribution for individual funds. Such simulation will allow us to identify the skilful managers and the lucky ones, thereby provide a critical element in testing the efficiency of governance mechanism in those UK unit trusts that are actively managed. 14

III. Methodology and Main Hypotheses 3.1 Performance Evaluation 3.1.1 Objective Adjusted Returns The essence of the method of objective adjusted returns (ORA) is to analyse a firm s performance against a selected benchmark (Morck, Shleifer and Vishny, 1990). This is measured by calculating the excess of the annual holding returns between the individual company and the benchmark. Formally, ORA can be computed as follows: 12 12 OOOOOO = 1 + RR ii,tt 1 1 + RR bb,tt 1 (1) tt=1 tt=1 where R i,t denotes the annual holding returns of an individual firm and R b,t is the annual holding returns on the benchmark. In this research, we deploy the annual total returns of the individual fund as R i,t, and R b,t is defined as the annual total returns of the fund s portfolio on the matched investment objective. The ORA evaluates the managers performance against others in the peer group (Khorana, 2001). With the ORA results, we sort the funds into two groups, i.e. a Positive Group (PG) and a Negative Group (NP), so that we can separate the funds with historical superior performance from the others with poor performance. 3.1.2 Abnormal Performance We evaluate abnormal returns on the basis of factor models. These models include the one factor model, the Fama-French (1993) three factor model, and the four factor model developed by Carhart (1997). Specification of the one factor model is based on the basic 15

CAPM model. The Fama-French (1993) three factor model is an extension of the basic CAPM model by including two more market benchmarks: the size effect and the book-to-market ratio effect. The three factor model is expressed as follows: rr ii,tt = αα ii + ββ ii RRRRRR tt + ss ii SSSSSS tt + h ii HHHHHH tt + εε ii,tt (2) where SML and HML denote the monthly returns of the factor-mimicking portfolio for the size and book-to-market ratio effects, respectively. The four factor model suggested by Carhart (1997) includes the momentum trading strategies in addition to the influences of the three factor model: rr ii,tt = αα ii + ββ ii RRRRRR tt + ss ii SSSSSS tt + h ii HHHHHH tt + mm ii MMMMMM tt + εε ii,tt (3) where MOM t represents the month t return of the one year momentum-mimicking portfolio. Other variables are the same as in the three factor model. We then compare the estimation results of the above unconditional factor models. Existing research suggests that the cross-sectional alphas for all three models are small and negative. However, in our model estimation, the factor loading on the MOM effect is statistically significant, suggesting that it is permissible and indeed desirable to apply the unconditional four factor model in our analysis of manager turnover. 3.1.3 Percentile Ranking To obtain a fuller perspective on the performance of individual funds as against the 16

whole fund industry during the replacement period, we analyse and report the changes in percentile ranking around the top management turnover for every fund within its matched sample. In addition, we include the ranking data from the Morningstar UK database, which give supplementary information on performance evaluation by independent financial media. 3.1.4 Matching Sample Analysis Given that cross-sectional performance may experience mean reversion during a long time interval, and that superior performance may not persist, we follow Denis and Denis (1995) and Khorana (2001) to adopt a matching sample analysis. This will enable us to identify whether the improvement or deterioration in performance in the post-replacement period is due to the managerial efforts of the new managers, or is the result of mean reversion of securities returns. The matching sample analysis is based on the abnormal performance of those funds that do not experience management turnover during the selected time period. We construct a sample of funds that have similar ORA (PG or NG) with each fund in the turnover sample during the time period that matches the replacement period. We then calculate the 36-month abnormal performance during the pre-replacement period and 24-month performance in the post-replacement period for the funds in the matching sample. A particular fund will be used to match the manager-replaced fund only once. In other words, funds that have the same replacement data will share the same matching sample. 3.2 Performance Shift Analysis In order to examine the influence of fund manager turnover on the variation in fund 17

performance, we compare the fund performance within different ORA groups in the preand post-replacement intervals. In addition, we incorporate into our research the analysis of risk shifting and performance decomposition to provide a fuller picture of fund manager turnover and its consequences. First, we sort the funds in the pre-replacement period into two groups, a positive group (PG) and a negative group (NG), according to the sign of their alphas estimated using the four factor model. Then we estimate the annual alpha for each group in both pre- and post-replacement periods. If the fund company is capable of dismissing the less-skilful fund managers and appointing skilled ones, we should expect that for the funds in the NG group the fund performance will show an improvement in the post-replacement period. The PG funds should at least maintain their positive performance after the replacement. A comparison table is compiled to facilitate investigation of this effect. From the table, one may observe changes in the abnormal performance each year for the funds in the sample. Second, we measure the risk shifting. We compare the portfolio risk and funds total risk around the replacement time to estimate the impact of changes in risk taking on performance. This comparison will provide additional evidence on the existence of the window-dressing behaviour of underperforming managers. Finally, we undertake a performance decomposition to obtain evidence that will provide us with information of investment preferences of outgoing and incoming managers. 18

3.3 Sensitivity Analysis To further investigate the relation between fund performance and top management turnover, we construct probability models within a logistic regression framework to ascertain the influence of fund performance and investment preferences on the probability of management turnover. The econometrical form of the model is as follows: llllllllll(pp) = α + β i ΔSMB t + γ i ΔHML t + δ i ΔMOM t + η i ΔBETA t + θ i ΔPR t + ε i,t (4) where ΔSMB t, ΔHML t, ΔMOM t, ΔBETA t and ΔPR t stand for the percentage changes of the size effect, value effect, momentum effect, total risk and portfolio returns, respectively. The term p is the probability of manager replacement. The percentage changes under examination are based on two time intervals, i.e. [-1, +1] and [+1, +2]. We further apply the logistic regression to investigate the relation between the fund performance and the probability of managerial demotion. The econometrical formulation for this testing is specified as follows: 3 3 3 3 llllllllll(pp) = α + β i RAR t j + η i BETA t k + θ i HML t h + θ i SMB t g + ε i,t (5) j=0 k=0 h=0 g=0 where p stands for the probability of demotion. RAR t denotes the risk adjusted returns prior to manager replacement and t is the replaced year. 19

3.4 Bootstrapping Implementation Using conventional econometrical methods in the evaluation of abnormal returns of US and UK funds, most previous research has found no positive abnormal performance (alpha) for the best funds. This raises a question as to the very existence of skilled fund managers, on which Carhart (1997) s evidence is inconclusive, since recent literature shows that superior performance can be found in a short time interval. Abnormal returns often have a non-normal distribution. Conventional statistical methods therefore seem inappropriate to analyse these returns when funds cross-section returns are assumed to be jointly normal. Recent research proposes that implementation of bootstrap analysis may address this problem and separate managers with real stock-picking skills from those who obtain superior performance only by chance. In the management turnover analysis, however, due to drawbacks in conventional econometrical methods, the analyses in previous works have not been particularly successful in providing a clear picture as to whether the abnormal performance is a factor in fund companies decision-making process over the replacement. In light of this, we will employ the bootstrapping method to examine the replacement decisions with a view to finding out whether or not fund companies are able to dismiss non-skilled managers and appoint skilled ones. We follow the procedure of Kosowski et al. (2006) and Cuthbertson et al. (2008) to construct our bootstrapping analysis of manager replacement. The three factor model is chosen as the main estimation model for bootstrapping analysis: rr ii,tt = αα ii + ββ ii RRRRRR tt + ss ii SSSSSS tt + h ii HHHHHH tt + εε ii,tt (8) 20

First, we divide the returns of all the funds with replacement into two groups: the pre-replacement group and the post-replacement group. For both groups, we estimate the abnormal returns αα ii, the factor loadings and the residuals for each fund i through OLS regression. In the estimation, the residual term is given by {εε ii,tt bb, tt = TT ii1, TT iiii }, where T i1 and T in are the beginning and end dates respectively for the returns of fund i, and N is the sample size. These estimations are then saved for the later bootstrapping analysis. With the bootstrapping simulation, we first construct a new sample of length N, {εε ii,tt bb }, through re-sampling the saved residuals, where b stands for the times of re-sampling. Next, we create a time series of monthly returns for each fund i by setting the null hypothesis as αα ii =0, meaning that the manager of fund i has no stock-picking skill. The functional form of the analysis is: rr bb bb ii,tt = αα ii + ββ ii RRRRRR tt + ss ii SSSSSS tt + h ii HHHHHH tt + εε ii,tt (9) where rr bb ii,tt is the artificial returns of fund i with the true abnormal returns equal to zero. We regress rr bb ii,tt on all the factors in the right-hand side of equation (8) to estimate the bootstrapped alphas and the corresponding t-statistics. This procedure is applied to all the funds in the sample, including the funds without replacement. Thus, we can gain the one-time bootstrapped alphas and the t-statistics for all funds. Repeating the above steps 1000 times, we can draw the distribution of bootstrapped alphas for each fund i. This distribution indicates which abnormal performance is only due to sample variation. In analysing top manager turnover, we similarly sort the performance into two groups on the basis of positive and negative ORA. For the funds that have experienced replacement 21

in these two groups, we first estimate the conventional alphas and the t-statistics for each fund with the estimation time being restricted to the fund s pre- and post-replacement periods. Then we rank the alpha and t-statistics for each replaced fund within its matching sample, i.e. the alpha of fund i is at the k th percentile of the fund alphas in its matching sample. As we have already bootstrapped the alpha and t-ratio for each matching sample, we can draw the luck distribution of fund i by using the alphas at the k th percentile among all bootstrapped alphas in its matching sample. We further compare the fund s real alpha with the bootstrapped alpha/t-ratio distribution. If we find that the real alpha lies in the area which is greater than the 5% upper tail point of such bootstrapped alpha distribution, we reject the null hypothesis that the abnormal performance of the fund is due to sample variation, and conclude that the manager of this fund has genuine stock selection skill. 3.5 Main Hypotheses Existing research claims a two-way influence between managerial replacement and fund performance. The dismissal of managers can be predicted by poor performance of the fund, while manager replacement tends to result in performance improvement for the underperforming fund (Khorana, 1996, 2001). Therefore, our first hypothesis is that where funds have historical poor performance, the fund companies will dismiss the top managers. We also expect those replaced managers to be the less-skilful (or unlucky) ones in a bootstrapping exercise. In addition, where funds with outperformance replace their managers, we hypothesize that there is no significant relation between fund performance and the probability of top manager replacement. 22

In the post-replacement period, granted that the dismissal of managers following poor performance is the outcome of the working of an effective governance mechanism, performance of the fund is likely to significantly improve. This is because, with effective governance, fund companies tend to appoint managers with genuine stock selection skills to replace the less-skilful (or the lucky) ones, which is reflected in performance improvement. Therefore, our second hypothesis is that the performance of previously underperforming funds will significantly improve after manager replacement. Existing research also indicates that the performance of previously superior funds may deteriorate after manager replacement. We conjecture that this is because, in the competitive fund market, superior performance cannot last for long. Therefore, we also hypothesize that superior performance will not persist after top management turnover. 23

IV. Data 4.1 Sample Description We start our sample construction with the data of the replacement series, which is collected from the Morningstar UK database using information of the specific day on which the current fund managers first undertook operation of the funds after replacing the former ones. In analysing managers skills in picking stocks, we focus on the performance of UK domiciled equity unit trusts and open-ended investment companies (OEICs), which conduct their investment business primarily in the UK equity market. These funds differ from closed-ended funds in that they can only be traded between the trust companies and investors. Funds with anonymous managers and management groups have been screened out from the sample, since we consider the stock-picking skill of specific managers or a solo manager. By the same reasoning, we also exclude index tracking funds. As a result, we find 386 funds which have experienced top manager replacement in the Morningstar database. From these 386 funds, we then select funds with at least 5 years performance history. Specifically, the sample includes funds with at least 3 years performance history preceding the replacement month and 2 years data in the post-replacement period. This is consistent with established research and has the advantage of avoiding the possible small sample bias. The replacement sample that meets our data screening criteria then comprises 218 funds from four sectors. 1 1 The IMA categorizes UK equity unit trusts and OEICs into four sectors, which we follow. Specifically, our sample includes 120 funds from the UK All Companies sector, 54 funds from the UK Equity Income sector, 5 funds from the 24

The returns data for analysing performance of individual funds are also collected from Morningstar. The monthly returns of all the funds are calculated by dividing changes in monthly net asset value, returns from reinvesting all income, and capital gains distributions by the starting NAV of the month. The returns data are not adjusted by sales charges in the total returns. However, the management fees and administrative fees are deducted from the total returns, so that the performance evaluation can provide evidence on whether the funds are profitable for investors. To evaluate funds abnormal returns we follow standard factor models, for example the Fama and French (1993) three factor model and the Carhart (1997) four factor model. The market benchmark factor in our research is the FTSE all share index. The value factor, also known as the HML factor, is derived by deducting monthly returns of the Morgan Stanley Capital International (MSCI) UK value index from the returns of the MSCI growth index. The size factor is the difference between the monthly returns of the Hoare Govett Small Companies (HGSC) index and the returns on the FTSE 100 index. The mimicking portfolio for the Carhart (1997) momentum effect, or the MOM factor, is constructed by using the total return index data for all UK listed equities. It is given by the monthly average return of a weighted portfolio measured by taking the difference between the returns of top and bottom 30% stocks. Finally, the market excess returns are calculated by using the monthly data of the 3-month UK Treasure Bill. 4.2 The Survivorship Bias Issue To investigate possible survivorship bias in our sample of 218 funds, we carry out a preliminary test to check whether the sample chosen is biased. Table 1 displays the result UK Equity Income & Growth sector. The remaining funds are from the UK Small Companies sector. 25

of the analysis. Table 1 shows that the difference between all funds with replacement experience and the sample funds is mild, particularly in terms of the number of observations. Columns 1 and 6 describe the number of funds which have experienced manager replacement for all investment objectives in the pre- and post-replacement periods. During the period 2007-2009 the difference between all funds and sampled funds is relatively large. However, as Columns 2 and 7 suggest, this is largely because some of the funds from the UK All Companies sector replaced their managers in late 2007, which means that they have less than 2-years return data and so are disqualified from our sample. Table 1 also provides the comparison of abnormal performance based on the three and four factor models. It can be seen that, for all investment objectives across the board, the alphas estimated for funds with ob 1 and with ob 36/24 do not differ significantly. This formally confirms that the sample does suffer from survivorship bias. 26

Table 1 Summary Statistics of Sample In Table 1, funds are sorted into four investment objectives according to IMA. They are also classified into two groups, i.e. all the funds that have experienced managerial replacement (ob 1), and the funds that are included in our sample (ob 36/24). The table also provides the number of observations of the replacement in five time periods. Formal results of estimated alphas using the three factor and four factor models are also reported for each type of fund. Replacement Date (Pre) Number of Funds Alpha (3 Factor Model) Alpha (4 Factor Model) Panel A ob 1 ob 36 ob 1 ob 36 ob 1 ob 36 All Investment Objectives ~1998 6 6-0.6385-0.6385-0.6306-0.6306 1999~2002 75 73-0.4645-0.4485-0.464-0.4598 2003~2006 146 144 0.753 0.756 0.765 0.779 2007~2009 115 107 0.289 0.275 0.277 0.279 UK All Companies ~1998 1 1-0.6114-0.6114-0.6348-0.6348 1999~2002 37 35-0.1817-0.182-0.197-0.1867 2003~2006 81 81-0.0593-0.0593-0.0593-0.0593 2007~2009 85 77-0.0974-0.092-0.091-0.0904 UK Equity Income ~1998 - - - - - - 1999~2002 13 13 0.284 0.284 0.2735 0.2735 2003~2006 41 41-0.222-0.222-0.211-0.211 2007~2009 21 21-0.1324-0.1324-0.1301-0.1301 UK Equity Income & Growth ~1998 - - - - - - 1999~2002 6 6-0.1285-0.1285-0.1307-0.1307 2003~2006 5 3-0.0593-0.0699-0.0603-0.0612 2007~2009 - - - - - - UK Smaller Companies ~1998 5 5-0.2247-0.2247-0.2529-0.2529 1999~2002 19 19-0.0963-0.0963 0.0899 0.0899 2003~2006 19 19 0.1014 0.1014 0.0784 0.0784 2007~2009 9 9 0.2245 0.2245 0.2587 0.2587 27

Replacement Date (Post) Table 1 Summary Statistics of Sample (Continued) Number of Funds Alpha (3 Factor Model) Alpha (4 Factor Model) Panel B ob 1 ob 24 ob 1 ob 24 ob 1 ob 24 All Investment Objectives ~1998 6 6-0.6104-0.6104-0.6065-0.6065 1999~2002 75 75-0.4507-0.4507-0.4101-0.4101 2003~2006 146 146 0.753 0.753 0.702 0.702 2007~2009 115 27 0.319 0.297 0.301 0.265 UK All Companies ~1998 1 1-0.5872-0.5872-0.4793-0.4793 1999~2002 37 37-0.1817-0.1817-0.198-0.198 2003~2006 81 81-0.0478-0.0478-0.0359-0.0359 2007~2009 85 25-0.0743-0.0635-0.0847-0.0727 UK Equity Income ~1998 - - - - - - 1999~2002 13 13 0.2721 0.2721 0.3045 0.3045 2003~2006 41 41-0.2107-0.2107-0.2344-0.2344 2007~2009 21 2-0.1104-0.1012-0.1621-0.1453 UK Equity Income & Growth ~1998 - - - - - - 1999~2002 6 6-0.1024-0.1024-0.1185-0.1185 2003~2006 5 5-0.0364-0.0364-0.0292-0.0292 2007~2009 - - - - - - UK Smaller Companies ~1998 5 5-0.2187-0.2187-0.2798-0.2798 1999~2002 19 19-0.0765-0.0765 0.0569 0.0569 2003~2006 19 19 0.0978 0.0978 0.0876 0.0876 2007~2009 9-0.0345-0.0467-28

V. Empirical Results 5.1 Performance Changes Surrounding Replacement The analysis of top manager turnover in the fund industry focuses on identifying performance changes in the replacement period. For this purpose, we follow the methodology discussed above to collect funds returns for three years before and two years after the replacement. Four evaluation methods - the factor model, total returns measurement, percentile rankings measurement and matched sample performance measurement - are then applied in the evaluation. The ensuing estimation results are classified into those for the positive group (PG) and the negative group (NG). The grouping is based on individual funds objective-adjusted returns in the pre-replacement period. Our results indicate that, while NG funds experience a significant performance decline before replacement, their performance tends to recover post-replacement. As shown in Panel A of Table 2, estimates from the factor models provide more pronounced results than those of the others. It comes as no surprise that, based on the outcome of the three factor model, NG funds abnormal returns drop by an average of 33 basis points in the year [-3,-2], with a further 1 basis point drop in the year [-2,-1]. In the post-replacement period, results from the three factor model suggest a sizable improvement of 60 basis points for the year [0, +1] and an improvement of 72 basis points for the years [0, +2]. However, for the PG group, Panel A indicates only a moderate increase in abnormal returns in the interval of years [-3, -1]. Despite a sharp drop in the replacement month, performance comes back swiftly to the level attained by the previous management in the 29

pre-replacement period. For example, in year [0, +1] the abnormal returns increase from 0.0012 (year 0) to 0.0023 (year +1) and remain almost unchanged in year +2. A possible reason for this lies in the fact that the PG funds had a good performance record before the replacement and so the incoming managers may find it unnecessary to change the composition of the original portfolio. Results from the four factor model show a similar outcome for the NG group, but indicate a rapid decline in the PG group s performance throughout the sample period. Moreover, almost all abnormal returns calculated under the four factor model are consistently and statistically insignificant, which is consistent with the results in previous research on the selection of models on mutual funds. Blake and Timmermann (1998), Quigley and Sinquefield (2000), Tonks (2005) and Cuthbertson et al. (2008) suggest that the momentum variable is not prevalent in the UK fund industry. Bearing this in mind, in what follows we will be mainly concerned with the three factor model. The improvement post replacement is also reflected in the percentile rankings of funds abnormal returns (from lowest to highest) shown in Panel A. It is not surprising to see that the NG group experiences a significant increase in cross-sectional ranking during year [-1, +1], since the ranking is subject to the abnormal return estimated by the three factor model. However, given the result during year [+1, +2], the growth cannot persist. Turning to the measure of total returns, the outcome becomes more fluctuant. This implies that total returns may not be a major concern in fund companies decision on managers replacement. Assuming that in general efficient internal control is in place in the UK fund operations, our results do not agree with the proposal in favour of evaluating fund performance in terms of their total returns. 30

Furthermore, the matched sample performance measurement included in Panel A sheds additional light on the value-added feature of manager replacement. The trend of performance changes of the matched sample in the NG group in the pre- and post-replacement periods closely matches the changes in abnormal returns, which suggests that further analysis of those matched samples would provide additional evidence of whether or not the replacement is a value-added activity. Since no replacement has taken place in the matched sample, if the turnover generated by replacement is less than the performance improvement in the matched sample, we cannot regard such replacement as a value-generated activity. Panel B of Table 2 compares the changes in levels among the four different measurements. The levels of change for the NG replacement group are 201.11% and 200.78% during the intervals [-2, +1] and [-1, +1], respectively, compared with 107.12% and 106.59% for the NG matched sample. Hence, it seems that turnover generated by top manager replacement would add more value to the performance of those funds than that gained by funds with the same pattern of OAR but which choose not to replace their managers. Moreover, the results show a marginal decline of decreasing rate for abnormal returns in the post-replacement period for both groups, especially the NG Group. This outcome is robust when checked with the four factor model. One possible explanation for this effect is the influence of window-dressing by fund managers. Managers under pressure to go may construct more risky portfolios in order to window-dress the performance record before the reporting date, leading to a marginal decrease in each of the measures in the table before the replacement. 31

Table 2 Analysis of Top Management Turnover Reported are the means of the performance of a sample of 218 unit trusts and OEICs in terms of abnormal returns. Medians are in brackets. The results are based on the three factor model, the four factor model, the measure of total returns, percentile rankings and matched sample performance measurement. PG stands for the positive returns group and NG for the group with negative returns. The division of the groups is dependent on individual funds objective-adjusted returns in the three years before the replacement. Panel A reports the level of the index in the sample period and Panel B provides percentage changes across each event window around a replacement date. ** and * indicate statistical significance at 1% and 5% levels respectively, based on the paired T test. 3 Factor Model 4 Factor Model Total Return Ranking Matched Sample Panel A PG NG PG NG PG NG PG NG PG NG 0.0018 0.0011 0.0239 0.1166 0.0082 0.0076 0.5121 0.4937 0.0025-0.0014 year-3 (0.0013) (0.0006) (-0.0029) (-0.0065) (0.0079) (0.0086) (0.5020) (0.4100) (0.0026) (-0.0014) 0.0016-0.0021 0.0096-0.0116 0.0036 0.0036 0.5472 0.3843 0.0022-0.0020 year-2 (0.0012) (-0.0010) (0.0067) (-0.0072) (0.0074) (0.0114) (0.5740) (0.3130) (0.0018) (-0.0018) 0.0026-0.0022 0.0063-0.1026 0.0061 0.0077 0.6164 0.4148 0.0019-0.0022 year-1 (0.0026) (-0.0006) (-0.0128) (-0.016) (0.0119) (0.0126) (0.7300) (0.4950) (0.0017) (-0.0019) 0.0012-0.0037 0.045-0.0504 0.0041 0.0012 0.6170 0.4240 0.0017-0.0023 year0 (0.0013) (-0.0026) (-0.0019) (-0.0379) (0.0142) (0.0031) (0.7289) (0.4137) (0.0015) (-0.0018) 0.0023 0.0022-0.0503-0.0354 0.0073 0.0055 0.6175 0.6068 0.0009 0.0001 year+1 (0.0022) (0.0013) (-0.0373) (-0.0171) (0.0102) (0.0059) (0.6610) (0.6800) (0.0007) (-0.0010) 0.0022 0.0035-0.0529-0.0137 0.0032-0.0068 0.6056 0.5347 0.0010 0.0013 year+2 (0.0013) (0.0039) (-0.0181) (-0.0336) (0.0086) (-0.0089) (0.7070) (0.6170) (0.0004) (0.0013) 32

Table 2 Analysis of Top Management Turnover (Continued) Panel B 3 Factor Model 4 Factor Model Total Return Ranking Matched Sample 100% PG NG PG NG PG NG PG NG PG NG -3 to -2-0.1268-2.9436** -0.5983 1.0995** -0.5669-0.5244 0.0685-0.2216** -0.1560-0.3902 (-0.0917) (-2.7370) (3.3103) (-0.1077) (-0.0684) (0.3347) (0.1434) (-0.2366) (-0.3221) (-0.3312) -3 to -1 0.4165-2.9501** -0.7364 1.8799** -0.2554 0.0094 0.2036** -0.1598-0.2450* -0.5024 (0.9389) (-2.0240) (-3.4138) (-1.4615) (0.5104) (0.4663) (0.4542) (0.2073) (-0.3487) (-0.4003) -3 to +1 0.2411 0.9653-3.1046** 1.3036** -0.1134-0.2785 0.2058** 0.2291** -0.6378** 1.0990** (0.6674) (1.2542) (-11.8621) (-1.6308) (0.2905) (-0.3114) (0.3167) (0.6585) (-0.7225) (0.2551) -2 to +1 0.4212 2.0111** -6.2396** 2.0517** 1.0469 0.5169 0.1285* 0.5789** -0.5708** 1.0712** (0.8357) (2.2978) (-6.5672) (-1.3750) (0.3853) (-0.4841) (0.1516) (1.1725) (-0.5906) (0.4404) -1 to +1-0.1238 2.0078** -8.9841** 0.6550 0.1907-0.2853 0.0018 0.4628** -0.5202** 1.0659** (-0.1400) (3.2014) (-1.9141) (-0.0688) (-0.1456) (-0.5304) (-0.0945) (0.3737) (-0.5739) (0.4680) +1 to +2-0.0257 0.5920* -0.0517 0.6130-0.5616-2.2364-0.0193-0.1189* 0.0760 7.7497* (-0.4282) (2.0232) (0.5147) (-0.9649) (-0.1569) (-2.5085) (0.0696) (-0.0926) (-0.4296) (2.2851) Obs. 127 91 127 91 127 91 127 91 127 91 33

Further evidence of performance switching may be found from the fund rating data. Table 3 summarizes the historical rating data from Morningstar for the unit trusts and OEICs in our sample. With the exception of UK equity income funds, both NG and PG funds receive upgrading of their ratings after the managers are replaced. In comparison, in Table 1, we find that only the NG group shows enhanced performance, as the new manager tends to change the composition of the portfolio constructed by the former management, while the PG group shows no such improvement as it may retain the inherited portfolio structure. Table 3 Rating Analysis of Management Turnover Note: Reported are the mean ratings of unit trusts and OEICs. Medians are in brackets. The data are sourced from Morningstar, and cover the pre- and post- manager replacement periods. The funds are sorted by IMA investment objectives. PG stands for the positive returns group and NG for the group with negative returns. The division of the groups is dependent on individual funds objective-adjusted returns in the three years before the replacement. Morningstar rates funds from 1 (minimum) to 5 (maximum). Rating Whole Sample UK All UK Equity UK Equity UK Smaller Companies Income Income & Growth Companies Year PG NG PG NG PG NG PG NG PG NG -3 to 0 3.1979 2.8734 3.0511 2.8204 3.4572 3.3676 2.4444 2.4444 3.2087 2.6520 (3.0909) (2.9231) (3.0000) (2.9009) (3.5000) (3.3676) (2.8333) (2.4444) (3.2087) (2.6520) 0 to +2 3.4820 2.9147 3.5721 2.9974 3.3493 2.6246 3.7668 3.0564 3.4041 2.6689 (3.4194) (2.8534) (3.6250) (2.9463) (3.4231) (2.6246) (3.8760) (2.2800) (3.4041) (2.6489) 5.2 Risk Characteristics and Decomposition Analysis Another interesting issue to explore is the role that risk characteristics play in the turnover process. Table 4 provides performance evaluations around the replacement, sorted by different investment objectives. For the NG group, fund performance significantly worsens before managers replacement and improves afterwards. Funds in smaller UK companies show a sharper shift in performance relative to all UK companies. As a further reflection of their distinct risk characteristics, abnormal returns of these funds drop by 177 basis points, while performance of funds in all UK 34