Investor Attrition and Mergers in Mutual Funds

Transcription

1 Investor Attrition and Mergers in Mutual Funds Susan E. K. Christoffersen University of Toronto and CBS Haoyu Xu* University of Toronto First Draft: March 15, 2013 ABSTRACT: We explore the properties of equity mutual funds which have experienced a large loss of assets. Borrowing from the mortgage literature, these funds are likely to be left with a group of price insensitive investors and are likely to exhibit different flow characteristics than funds in their early growth phases. We show that funds with higher attrition rates exhibit lower net flows and lower overall flow-performance sensitivities. In addition, we find some evidence that managers respond to attrition with increases in fees and risk-taking above current levels. We also examine fund mergers and whether the stability of a self-selected investor base makes for an attractive acquisition. Consistent with this intuition, high attrition funds are more likely to become merger targets even after controlling for poor performance. Characteristics of the investor base are therefore important determinants of merger acquisitions. JEL Classification: G11, G23 Keywords: Flow-performance relation, mutual funds, attrition, mergers * Contact: Haoyu Xu, haoyu.xu10@rotman.utoronto.ca, Rotman School of Management, 105 St. George Street, Toronto, ON. All errors are our own. 1

2 Understanding the flow-performance relation for mutual funds has been extensively studied in the literature. The seminal papers by Sirri and Tufano (1998) and Ippolito (1992) identify a convex relation which pose potential risk-taking incentives of mutual fund managers (Chevalier and Ellison, 1997, and Brown, Harlow, and Starks, 1996) or even incentives to manipulate performance through fees or other tactics (Christoffersen, 2001, and Carhart, Kaniel, Musto, Reed, 2002). Bad performance does not appear to be punished with investors leaving the fund. Lynch and Musto (2003) develop a rational model to explain this convexity where investors recognize that managers may change their strategies after bad returns, so investors rationally stay in a fund knowing that the manager will adjust their strategy. However, Christoffersen and Musto (2002) and Elton, Gruber, and Busse (2004) show that bad performance looking forward is all but guaranteed for a subset of money and index funds because of excessively high fees so there appears to be more to the story which cannot only be explained by changes in strategy. Christoffersen and Musto (2002) and Berk and Tonks (2007) shed some light on this puzzle by identifying that investor heterogeneity might explain the lack of sensitivity with some investors facing either behavioral or other constraints which prevent them from responding to poor performance. Even if there is heterogeneity across investors, the observed convex relation would only arise if some individuals were non-randomly assigned to different funds. To this end, Christoffersen and Musto (2002) draw from the mortgage literature and first introduce a retention measure for money market mutual funds which measures the current size of the fund relative to its maximum historical size. Funds which have experienced a large outflow of investors are presumably left with a non-random sample of price-insensitive investors. Despite its intuitive appeal, little is known about the attrition measure and to what extent it can help identify funds with different performance sensitivities. The measure so far has only 2

3 been applied to money funds and used to explain changes in money fund pricing around mergers and the wide dispersion in money fund fees. The purpose of this article is to explore properties of attrition amongst equity funds and to answer some remaining questions: (1) Does attrition capture cross-sectional differences in performance sensitivities across funds? (2) Do managers alter their behavior when managing a fund that has undergone extensive attrition? (3) Is the concentration of price-incentive investors an attractive asset for mutual funds given the longterm prospects of the investor horizon? In short, the answer to all three questions is yes. In addressing all these questions, one important contribution of the paper is to properly identify causality as a history of bad performance will likely result in both a reduction in fund size from low performance as well as cause price-sensitive investors to leave. It is the latter consequence which is of greater interest as it is the self-selection of investors that helps identify a fund with less performance sensitivity. For all of the analysis, we therefore define raw attrition as one minus the ratio of current asset size to the historical maximum size and as a first step try to estimate attrition levels with past returns, past risk-taking measures, and past expenses. The residuals from this estimation are used to measure attrition which cannot be explained by past performance and therefore better identifies funds which undergo a substantial amount of self-selection not arising from the mechanical loss in assets from low returns. With regards to the first question, we estimate flow-performance relations as have been done previously in the literature but instead separate funds into those which have experienced high attrition and those which have not. As expected, those funds which have experienced a high amount of attrition also exhibit a significantly flatter flow-performance relation. This finding is separate from the well-known finding of Chevalier and Ellison (1997) that older funds exhibit 3

4 much less sensitive flow-performance relations. Although old funds have more likely undergone greater amounts of attrition and are left with a core group of stable investors, both age and attrition independently dampen the flow-performance relation. We also see that funds with higher attrition show significantly less sensitivity to expenses and category flows as one would expect if attrition captures a set of captive investors. Looking forward, an attrition measure could also be helpful for researchers and practitioners needing to distinguish funds with different performance sensitivities. The only other way to get fund-level performance sensitivities is to estimate time series of the flow-performance relation for each fund but this method is fraught with problems. First, it assumes that performancesensitivity is constant through time for the mutual fund, which may not be the case. Second, it relies on a time series of data so eliminates young funds. Instead, the attrition measure is easy to measure fund-by-fund, changes over time, and does not rely on a large time-series of data. The second question considers whether managers change their behavior in response to the composition of their investor base. With a stable group of investors, there is some concern that managers may increase risk-taking or may increase expenses knowing the investors won t leave. While we find strong evidence that higher expenses and risk-taking lead to greater levels of attrition, we also find some evidence in the other direction. After carefully controlling for the current levels of risk-taking and fees, we observe future levels of fees and risk to increase with attrition so managers seem to respond to the composition of its investor base. The final analysis of the paper assesses the importance of these investors to mutual funds by looking at whether a fund which has experienced significant amount of attrition is also more likely to be merged. One can imagine that a fund with a self-selected group of price-insensitive 4

5 investors is an attractive acquisition given the perpetuity of fees one can draw from these investors. While prior literature has identified bad performance as a reason to merge (Jayaraman, Khorana, and Nelling, 2002), the attractiveness of self-selected group of investors has not been explored. Even after controlling for past performance and other factors identified as influencing merger decisions, attrition is a very significant determinant of the likeliness the fund will be merged. We estimate that the unconditional probability of a fund being targeted for a merger is 5.17% and this probability increases by 0.7% as attrition increases by 10%. The remainder of the paper is divided as follows. Section I outlines the data used. Section II discusses the attrition measure and factors important to it. Section III presents the main results relating attrition to flow performance, managerial behavior, and mergers. Finally, Section IV concludes. I. Data The data we use in this study is the CRSP Survivor-Bias Free Mutual Fund Database. This database provides daily returns of mutual funds, total assets under management, fee structure and other characteristics. We use daily returns to estimate annual beta of each fund, but the CRSP mutual fund daily returns are not available prior to So our sample is from 1998 to We restrict our analysis to actively-managed U.S. domestic equity funds. Index funds are excluded from our sample, which are identified using index fund flag and ETF flag variables in CRSP database. Many funds have multiple share classes where the same fund is offered with various fee structures. We aggregate share classes so each observation is at a fund-year level rather than by 5

6 share class. To match share classes, we use WRDS MFLINKs file (Wermers, 2000) and for the funds which are not covered by MFLINKs file, we use CRSP_portno variable in CRSP mutual fund database as a portfolio identifier or hand match. Fund size is simply the summation of the total net asset (TNA) of each share class and is expressed in millions of dollars in the estimation. The expense ratios, returns, and turnover rates of funds are calculated as the TNA-weighted average from each share class and are expressed in decimal form where 5% is represented as There are situations that a new share class is introduced many years after the fund portfolio is created. For such cases, we use the birth date of the oldest share class as the starting date to determine the age of the fund and age is expressed in years. Fund flows use the total net assets determined at the end of each year but remove the effects of returns on assets, assuming investors hold the fund over the year as assumed in Sirri and Tufano (1998). NetFlows, = TNA, TNA, (1 + r, ) TNA, where r i,t+1 is the net return of the portfolio at t+1. Net flows are expressed in decimals as with expense ratios, returns, and turnover rates. For all the estimations we control for category flows, expenses, and risk-taking where a category average is calculated as an asset-weighted average of funds in the category. We use Lipper category classifications to determine categories and the sample includes eight different Lipper classifications. 1 We proxy for fund risk-taking using market betas from a four-factor model (Carhart, 1997). Betas are estimated using daily returns over the calendar year. 1 The eight category classifications from Lipper that are included in our sample are Capital Appreciation (CA), Equity Income (EI), Growth (G), Growth and Income (GI), Long/Short Equity (LSE), Mid-Cap (MC), Micro-Cap (MR), and Small-Cap (SG) 6

7 II. Measuring Attrition One of the contributions of this paper is to analyze more closely the cross-sectional relevance of differences in attrition rates across funds. The concept of attrition was first introduced in the mortgage literature (i.e. Schwartz and Torous, 1989) where a recent decline in interest rates will cause some to refinance leaving a pool of mortgages that is less susceptible to refinancing risk going forward. Like Christoffersen and Musto (2002), we apply this concept to mutual funds by measuring the current size of a fund to its maximum historical size. Raw attrition is therefore calculated as Attrition 1 Q i, t MAX i, t where Q i,t is the total net assets for each fund i at the end of each year t, and MAX i,t is the historical maximum fund size over all months from its birth date. We calculate the raw attrition rate of each fund at the end of year t and require the funds in our sample to be a minimum of one year. Although the analysis focuses on data from 1998 to 2011, all the attrition measures are determined over the full history of the fund. Table 1 presents the average attrition rates across all funds between 1998 to 2011 and we observe average attrition ranges from 0.13 to 0.53 across the years. Attrition increases in years where there are much higher levels of negative returns such as in This highlights the importance of controlling for the cumulated returns since attrition levels are jointly affected by returns of the portfolio as well as individual investor decisions to enter and leave the fund. 7

8 Table 2 presents the correlations of raw attrition with control variables used in the analysis. Similarly, Table 3 groups the data into five attrition quintiles and presents average levels of size, age, returns, and betas for each of the quintile groups. In both tables, we observe that high levels of attrition correlate strongly with funds that have experienced poor performance, higher fees, higher levels of risk-taking (measured by beta), and larger outflows. In addition, older and smaller funds tend to be ones which have experienced more attrition. To get an idea of the magnitude of these relations, Table 3 shows that fees in a fund with 9% attrition average 1.26% while fees for a fund with 75% attrition are significantly higher at 1.45%. While correlations are indicative of cross-sectional relations with attrition, the next sections tries to identify causality by first identifying what variables are important contributors to attrition and later what effects attrition has on decisions of the manager. A. Explaining Attrition Our measure of raw attrition is influenced by both the performance of the fund and the decisions of price-sensitive investors to leave the fund. If we want to isolate the second effect of investors leaving the fund, we need to control for historical performance. In a panel regression, Table 4 estimates those factors influencing attrition where standard errors are clustered by fund and the cumulated performance of the fund is included. 2 Cumulated performance is calculated as the geometric accumulation of monthly returns over the period since the fund reached its maximum size. Not surprisingly, cumulated returns enter significantly negative. The coefficient on cumulated returns in the first column of Table 4 suggests that if cumulated returns increase by 2 For all estimations, we also estimate using Fama-Macbeth methodology. The results are similar so are not reported in the tables. 8

9 10%, then attrition is expected to decline by 2.3%. While the relation is significantly negative, it is not one-to-one which likely results from performance having effects not only on the value of current assets but also on the decisions of investors to enter and leave. The level of attrition of a mutual fund may also be influenced by other observable features of the funds such as risk-taking and expenses and even the amount of trading. Investors may leave funds with high past expenses or ones with more risk-taking. The remaining columns in Table 4 support this intuition where attrition increases with higher lagged fees and higher levels of past risk-taking. When we place all three variables in a regression to explain attrition, they all continue to enter very significantly. Even funds with higher levels of lagged turnover relate to higher attrition rates as investors leave funds with higher trading levels. For our analysis going forward, we use the residuals from the attrition regression to capture the excess attrition that is not explained by lagged cumulated returns, lagged expenses, lagged beta, and lagged turnover rates. Excess attrition captures funds whose assets have been depleted from their historical maximum for reasons which do not arise mechanically from any of these factors, especially poor performance and high fees which directly affect assets. Those investors who remain in a high attrition fund have identified themselves as ones who are likely to stay in the future. Therefore, the hypothesis we want to next explore is whether funds experiencing a lot of excess attrition exhibit weak flow-performance sensitivity. III. Results Having identified fund specific factors influencing attrition rates, we turn our attention to estimating different effects of attrition by first considering its influence on the flow-performance relationship, then how it might affect managerial decisions to alter risk-taking or fees, and lastly 9

10 its importance in fund mergers. For all the estimations, we report estimates of large panel regressions with robust standard-errors that cluster by fund. In unreported results, we also estimate all the regressions using Fama-MacBeth and sign and significance of the results are similar. A. Flow-performance relation The first part of our analysis determines if attrition provides a proxy for the performancesensitivity of investors in a specific fund. Numerous studies have estimated the flowperformance relation of a large cross-section of funds and in some cases, the funds are grouped into different subsamples, such as old and young funds (Chevalier and Ellison, 1997) to identify differences in sensitivities across funds. However, measures to identify performance-sensitivity of a specific fund is limited to time-series estimation which assumes that performance-sensitivity of a fund remains the same through time and is constrained by the time series data available for the fund. A fund-specific measure of flow-performance sensitivity could be immensely helpful to researchers and practitioners wanting an easy way to assess the stability of investors in a fund and the likeliness these investors will respond to future downturns in the fund s performance. If high attrition does in fact identify funds with a group of individuals who are likely to stay in the future, then we expect higher attrition rates to correspond with lower flow-performance relations. This could arise for several reasons. First, those investors who decide to stay in a fund when all others are leaving after poor performance are identifying themselves as being insensitive to returns. Similarly even if performance improves with attrition, investors who decide to invest in a fund that has recently experienced a history of lagging performance and subsequent loss of investors signal a strong interest in the fund and are unlikely to place great 10

11 importance on returns of the fund. Table 5 presents the results with the first four columns using raw attrition rates and the last four using the excess residuals from Table 4. As with other studies we observe a strong positive performance-flow relation where performance is a cross-sectional ranking of a fund ranging from zero to one. The first two columns divide the funds into those with high and low attrition (where high and low are determined using the median attrition level). The high attrition funds exhibit significantly lower sensitivity to performance with an overall sensitivity of 0.3 compared to 0.65 in the low attrition group. The third column provides a pooled regression across all funds but includes an interaction term between performance and attrition. We observe a significant negative relation on the interaction term. So higher attrition rates lower the performance-flow relation as predicted. Given the estimated flow-performance relation of 0.76 when attrition is zero, the coefficient on Attrition x Performance suggests that if attrition increased by 10% then the flow-performance relation would decline by 0.078, or in other words it would lose approximately 10% of its flowperformance sensitivity. Interestingly, the coefficient on Attrition is also negative. This suggests that higher levels of attrition have not only lower flow-performance sensitivities but also lower overall levels of flows. The coefficient can be interpreted as those funds with 10% more attrition having 1.2% lower levels in overall net flows. One would expect lower levels of flows to accompany funds which have higher attrition because these funds have less incentive to expend resources to attract new investors knowing the current investors are likely to stay. In addition, we have already shown in Table 4 that funds experiencing large amounts of attrition are ones with higher fees, worse historical performance, and higher risk so it is not surprising that these same funds are not attracting large new inflows. 11

12 Another dimension that one might expect to see differences in high and low attrition funds is the importance of category flows. Category flows measure average flows into a fund s category. A fund which has recently experienced a large amount of attrition is likely less correlated with the average investor in the fund category since investors in a high-attrition fund have selfselected to stay in this fund even after poor performance. This intuition is supported in the analysis where we observe that category flows are a much more significant predictor of a fund s individual flow in the subset of low attrition funds. The coefficient on category flows for low attrition funds is 20 times larger than that estimated amongst the high attrition funds. We also observe significant differences in the sensitivity to expenses between the low and high attrition groups. In the high attrition funds the sensitivity to expenses is which is almost half the sensitivity observed in the group of low attrition funds, -10. The high attrition funds have self-selected a group of investors who are insensitive to both expenses and performance. Note the lack of performance sensitivity is not simply arising from high attrition identifying funds with low performance. In the lasts four columns of the table, we repeat the same regressions as in the first four columns but this time replace raw attrition with excess attrition from the residuals in Table 4. We observe the same negative relation when interacting Residuals x Performance where the residuals remove any direct relation between returns and attrition measures. All other relations that we noted above with category flows and expenses also remain when including residuals rather than raw attrition rates. It therefore appears that attrition is in fact a good proxy for the future expected performance-sensitivities of investors on many different dimensions. 12

13 Earlier work has found that the flow-performance relation for older funds is flatter than for younger funds (Chevalier and Ellison, 1997). Because attrition is highly correlated with age we want to rule out the possibility that attrition is simply proxying for age. We therefore include a final regression in the fourth and eighth columns which includes an additional interaction term Performance x Age. In both specifications with raw and excess attrition, the age interaction term enters significantly negative; however, it does not diminish the economic or statistical significance of attrition and its interaction with performance. B. Managerial behavior Because managers can easily observe the historical changes in a fund s size and its ability to retain investors after bad performance, there is an open question whether this changes the manager s behavior. A manager may choose to increase risk or similarly increase expenses knowing the current investors are unlikely to leave and new investors are unlikely to be affected by their actions. We explore both possibilities in Tables 6 and 7. Table 6 first considers risk-taking incentives by looking at changes in market beta as a function of past attrition rates. As documented in Table 4, higher risk levels contribute to higher attrition as investors leave funds taking on greater market risk. In contrast, Table 6 shows that after controlling for past risk levels there is weak evidence that managers of high attrition funds exhibit higher propensities to take on risk. The coefficient on attrition and the residual variable are marginally positively significant once lagged beta and category beta are included. Therefore while higher risk associates with high attrition funds, the causality seems to go from risk-levels causing attrition rates with only some evidence of attrition contributing to excess risk taking. 13

14 Table 7 repeats the same exercise for expenses but finds an extremely significant positive effect of attrition on subsequent fees set. As found with money market funds (Christoffersen and Musto, 2002), equity fund managers seem to respond to a higher concentration of captive investors by increasing fees. Economically the coefficient.002 suggests that a fund with an attrition rate that increases from the 0 to 70% will on average respond by increasing its annual expense ratio by 14bp. C. Mergers Numerous papers have considered reasons for funds to merge. Jayaraman, Khorana, and Nelling (2002) identify that targets which are merged typically have lower returns, higher fees, and small size prior to the merger and after the merger returns improve and fees decline. Performance seems to be particularly bad when a fund is acquired by another fund in the same family which suggests the fund family may be merging the funds to hide a history of bad performance. Christoffersen and Musto (2002) provide an explanation for the favorable change in target fees around a merger and show that the high fees in target money market funds reflect the demand characteristics of these funds. When high attrition funds with less price-sensitive investors merge with those with lower fees, the resulting fees reflect the combined demand characteristics of the two funds. Others (for example, Ding, 2006) suggest that the improvement in fees arises from efficiency gains from improved economies of scale. Fund strategy also seems to be an important factor when deciding whether to merge. Namvar and Phillips (2013) find that funds with similar book-to-market ratios and holdings benefit from merging with funds with similar characteristics as these funds incur lower costs in turning over portfolios of newly acquired funds to align strategies. Better governance seems not only to be 14

15 important at finding funds with more similar strategies, but also at undertaking within-family mergers that are beneficial for target shareholders. For instance, Khorana, Tufano, and Wedge (2007) show how director salaries and independence can influence the outcome and opposition of within-family mergers. While governance, poor performance, high fees, and fund strategy have all been explored as factors influencing the decision to acquire a fund, there has been no discussion on whether the composition of investors in the fund serves as a positive reason to merge. High attrition identifies sticky investors who are very valuable to fund companies earning a percent of assets in perpetuity. Could this be a reason for a fund to be targeted for a merger? Tables 8 and 9 explore attrition as a reason to merge. In Table 8, the dependent variable in a probit regression is a dummy variable taking the value one if the fund is a target fund and zero otherwise. Variables such as the target expenses, past year returns, cumulated returns since its historical maximum size, current size, and age are included as control variables in addition to attrition with the first two columns using raw attrition and the third and fourth columns including excess attrition from Table 4. In all specifications, high attrition funds are significantly more likely to be targeted for a merger. Economically, the likeliness of a merger increases by 0.7% as the attrition level increases by 10%. Given the unconditional probability of a fund being a merger target is 5.17%, an increase of 0.7% is a significant increase in probability. In addition to this, cumulated returns, last year s performance, fees, and fund size enter significantly and with signs consistent with prior studies. However, the effect of attrition on the likeliness of a merger is separate from all these influences that have been studied previously. We also include year dummies in the second and fourth columns to control for periods of time when merger activity may be particularly 15

16 active. There are no mergers in 2011 so the number of observations is dropped for that year when dummy variables are added; however, the addition of year dummies does not otherwise change the results. For comparison, Table 9 tests whether attrition explains the likeliness of a fund becoming an acquirer. In contrast to a target firm, the characteristics of the acquirer s investor base should not be a consideration of its decision to acquire another. We include these regressions to provide an additional test that the attrition variable is not simply picking up the likeliness of a merger. As expected, attrition is insignificant in determining whether a fund will act as an acquirer. Larger funds and ones with higher fees are more likely to serve as acquirers but funds which have experienced more attrition show no evidence of this influencing their decision to acquire another fund. Overall, the results show that the type of investors in a fund is an important consideration for fund companies when deciding to acquire. Performing badly has an upside of self-selecting a group of stable investors which become an attractive asset to acquire. IV. Summary Mutual funds attract a heterogeneous group of investors with different risk-aversions, investment horizons, knowledge, investment objectives, and behavioral biases. The literature has in general treated these investors as homogeneous with some exceptions. For example, Johnson (2004) documents the costs of mixing short-term investors with long-term investors. One of the reasons for the literature not exploring the effects of investor heterogeneity on mutual funds and managerial behavior is we know little about the composition of the investor base and how this differs in the cross-section of mutual funds. 16

17 Attrition is one measure which helps identify funds with a different set of investor demand characteristics. Christoffersen and Musto (2002) first applied this measure to explain changes in money market fees around fund mergers. This study expands the earlier research by estimating the effects of attrition in equity funds and exploring three additional questions. First, it finds that higher attrition corresponds to lower flow-performance sensitivity. Consistent with the lower performance sensitivities, we also find that higher attrition funds exhibit lower sensitivities to fees and category flows. While there is an extensive literature on fund flow estimation and its effects on managerial behavior, this study highlights that flow-performance sensitivities differ in the cross-section of funds and attrition provides an easy way to make this distinction. Our second set of tests focus on how managerial behavior might be affected by differences in the investor composition of a fund. We consider two possibilities: that managers might increase risk-taking or increase fees with a more captive investor base. There is weak evidence that risktaking increases with higher attrition. In contrast, even after controlling for current fee levels, attrition seems to statistically predict increases in fees in the following year. Managers of equity funds seem to recognize the ability to price discriminate and charge higher fees to a group of self-selected investors. Our final tests focus on the importance of attrition in predicting mutual fund mergers. While prior research has highlighted that mergers help fund families hide poor performance and benefit from economies of scale, we offer a new determinant of fund mergers: the price-sensitivity of the investor base. High attrition strongly predicts if a fund will become a merger target, even after controlling for poor performance. In contrast, the same pattern is not observed for acquirers. Attrition therefore identifies a desirable attribute about the investors in a target fund and is an important consideration when deciding to acquire investment assets. 17

18 References Berk, Jonathan B., and Ian Tonks, 2007, Return persistence and fund Flows in the worst performing mutual funds, Working Paper No , NBER. Brown, K. C., Harlow, W. V., Starks, L. T., Of tournaments and temptations: an analysis of managerial incentives in the mutual fund industry, Journal of Finance 51(1), Carhart, Mark M., 1997, "On Persistence in Mutual Fund Performance, Journal of Finance 52, Carhart, M. M., Kaniel, R., Musto, D. K., Reed, A. V., Leaning for the tape: evidence of gaming behavior in equity mutual funds, Journal of Finance 57, Chevalier, J., Ellison, G., 1997, Risk Taking by Mutual Funds as a Response to Incentives, Journal of Political Economy, Vol. 105, No. 6, Christoffersen, Susan.E.K., 2001, Why Do Money Fund Managers Voluntarily Waive Their Fees? Journal of Finance, 56, Christoffersen, Susan E. K. and David K. Musto., 2002, Demand Curves and the Pricing of Money Management, Review of Financial Studies 15, Ding, Bill, 2006, Mutual fund mergers: A long-term analysis, Working Paper, SUNY Albany. Elton, Edwin J., Martin J. Gruber, and Jeffrey A. Busse., 2004, Are Investors Rational? Choices among Index Funds, Journal of Finance 59, Ippolito, R.A., 1992, Consumer Reaction to Measures of Poor Quality: Evidence from the Mutual Fund Industry, Journal of Law & Economics, Vol. 35, No. 1, Khorana, A., Peter Tufano, and Lei Wedge, 2007, Board Structure, Mergers and Shareholder Wealth: A Study of the Mutual Fund Industry, Journal of Financial Economics 85(2), Namvar, Ethan and Blake Phillips, 2013, Commonalities in investment strategy and the determinants of performance in mutual fund mergers, Journal of Banking and Finance 37 (2), Jayaraman, N., Khorana, A., and Nelling,E., 2002, Analysis of the Determinants and Shareholder Wealth Effects of Mutual Fund Mergers, Journal of Finance 57, Lynch, A.W., Musto, D.K., 2003, How Investors Interpret Past Fund Returns, Journal of Finance, Vol. 58, No. 5,

19 Schwartz, E. S., and Torous, W. N., 1989, Prepayment and the Valuation of Mortgagebacked Securities, Journal of Finance, 44, Sirri, Erik R. and Peter Tufano, 1998, Costly Search and Mutual Fund Flows, Journal of Finance 53, Wermers, Russ, 2000, Mutual fund performance: An empirical decomposition into stock-picking talent, style, transactions costs, and expenses, Journal of Finance 55,

20 Table 1 Attrition Rate Summary Statistics, The table summarizes the fund attrition rate from 1998 to Attrition rate of fund i in year t is (1 TNA, /Historical MaxTNA, ), where TNA, is fund i s total net assets at the end of year t, Historical MaxTNA, is the maximum TNA the fund i recorded since its birth to the end of year t. The table reports the mean, median, minimum and maximum attrition rate in each year from Panel A includes all funds in our sample. Panel B only includes the funds with attrition rate larger than 0. Panel A: All Fund Years Year Count Mean Median Min Max Panel B: Fund Years with Attrition Rate > 0 Year Count Mean Median Min Max

21 Table 2 Fund Characteristics Correlation The table presents the correlation between attrition rate and other fund characteristics. It includes all the funds in our sample from Attrition rate of fund i in year t is (1 TNA, /Historical MaxTNA, ), where TNA, is fund i s total net assets at the end of year t, Historical MaxTNA, is the maximum TNA the fund i recorded since its birth to the end of year t. Size, in millions of dollars, is measured at the end of the year t. Age is measured as year t minus the year of birth. Beta is the market beta measured by Carhart 4-factor model using fund daily returns. Expense ratio is the weighted average of all share classes expenses. Returns are the funds gross returns, which is the reported return plus expense. Flow is measured as NetFlows, =,, (, ),, where r i,t is the fund i return in year t. Attrition Size Age Beta t Beta t+1 Expense Return t Flow t Attrition Size Age Beta t Beta t Expense Return t Flow t

22 Table 3 Characteristics of Funds with Different Attrition Rates The table reports average characteristics of funds assigned at the end of year t into groups on the basis of the attrition rate. At the end of each year, funds are sorted on attrition rates and grouped into five quintiles. The characteristics in each group are the averages of all years. Beta is estimated using Carhart 4-factor model in year t+1. Size is in millions. Expenses are shown in percent. Rank Number Attrition Size Annual Return Expense Age Beta t % % % % %

23 Table 4 Attrition Rate and Accumulated Return The table reports coefficients, t-statistics, and R 2 values from regression : Attrition, = α + β Accumulated Return, + β Beta, + β Expense, + Turnover, + ε. Attrition rate of fund i in year t is (1 TNA, /Historical MaxTNA, ), where TNA, is fund i s total net assets at the end of year t, Historical MaxTNA, is the maximum TNA the fund i recorded since its birth to the end of year t. Accumulated Return of fund i in year t is the total return from the time the fund reaches Historical MaxTNA till the end of year t. The standard errors in the panel regression are clustered by fund and t-statistics are provided below each coefficient in brackets. Accumulated Beta t-1 Expense t-1 3 Variables 4 Variables Return t Accumulated Return t (-29.12) (-27.38) (-21.45) Beta t (11.17) (9.62) (10.89) Expense t (10.09) (7.86) (7.23) Turnover t (7.41) Constant (69.77) (19.68) (54.60) (27.35) (18.61) R Number of Fund Years

24 Table 5 Attrition Rates, Residuals, and Fund Flows The table reports coefficients, t-statistics, and R 2 from regressions of fund flows in year t+1. Fund flow in year t+1 is measured as NetFlows, =,, (, ),. The performance is measured following Sirri and Tufano (1998). A fund s performance is its percentile performance relative to other funds with same investment objective in the same year, and ranges from 0 to 1. In panel A, sample is divided into two groups according to attrition rate. Attrition rate of fund i in year t is (1 TNA, /Historical MaxTNA, ), where TNA, is fund i s total net assets at the end of year t, Historical MaxTNA, is the maximum TNA the fund i recorded since its birth to the end of year t. In panel B, the sample is divided into two groups according to residuals. Residual is from the regression in Table 4. Category fund flow is the net flow to all funds with same investment objective in the same year, measured as a fraction of the category size. The standard errors in the panel regression are clustered by fund and t-statistics are provided below each coefficient in brackets. Panel A: Group By Attrition Rate Panel B: Group By Residual High Low All High Low All Performance t (6.94) (9.25) (13.13) (11.67) (7.75) (9.43) (13.39) (9.41) Attrition t (-1.47) (-2.09) Attrition t Performance t (-5.73) (-4.59) Residual t (-1.57) (-2.30) Residual t Performance t (-5.13) (-3.81) Age t Performance t (-2.87) (-2.96) Expenses t (-2.72) (-2.54) (-3.80) (-3.75) (-3.01) (-2.87) (-4.64) (-4.57) Log Fund Size t (-10.58) (-10.50) (-15.42) (-15.31) (-10.30) (-10.76) (-15.42) (-15.32) Age t (-1.27) (-0.25) (0.24) (2.55) (-1.16) (-0.17) (0.36) (2.69) Category Fund Flow t+1 (2.31) (18.52) (15.73) (15.71) (2.28) (18.45) (15.73) (15.71) Constant (5.47) (6.02) (7.78) (5.82) (5.19) (6.31) (8.38) (5.39) R Number of Fund Years

25 Table 6 Attrition Rates, Residuals, and Fund Betas The table reports coefficients, t-statistics, and R 2 from regressions Panel A: Beta = α + β Attrition + β Exp + β Return + β Size + β Age + β Category Beta + β Beta Panel B: Beta = α + β Residual + β Exp + β Return + β Size + β Age + β Category Beta + β Beta Beta is the market beta measured by Carhart 4-factor model using fund daily returns. Attrition rate of fund i in year t is (1 TNA, /Historical MaxTNA, ), where TNA, is fund i s total net assets at the end of year t, Historical MaxTNA, is the maximum TNA the fund i recorded since its birth to the end of year t. Residuals are from the 3-variable regression in table 4. Return is the fund s the annual gross return in year t. Category Beta is estimated using category daily returns, which is the TNA weighted average over funds with same investment objective. The standard errors in the panel regression are clustered by fund and t-statistics are provided below each coefficient in brackets. Panel A: Attrition Panel B: Residuals Attrition t (6.36) (5.46) (1.69) Residual t (1.90) (1.48) (1.89) Expense t (-1.58) (-1.99) (-2.02) (-1.20) (-1.67) (-1.50) Return t (1.28) (1.19) (1.78) (0.79) (0.70) (1.72) Log Fund Size t (3.12) (3.22) (2.51) (2.00) (2.19) (2.56) Age t (-3.15) (-2.47) (-1.06) (-1.86) (-1.06) (-0.98) Category Beta t+1 (6.96) (2.75) (7.25) (2.76) Beta t (16.48) (16.76) Constant (38.94) (5.34) (2.56) (40.79) (5.50) (2.64) R Number of Fund Years

26 Table 7 Attrition Rates, Residuals, and Expense Ratios The table reports coefficients, t-statistics, and R 2 from regressions Panel A: Expense = α + β Attrition + β Category Expense + β Expense + β Return + β Size + β Age Panel B: Expense = α + β Residual + β Category Expense + β Expense + β Return + β Size + β Age Expense t+1 is the fund s annual expense ratio at the end of year t+1. Category Expense t+1 is the TNA weighted average expense ratio over funds with same investment objective at the end of year t+1. Attrition rate of fund i in year t is (1 TNA, /Historical MaxTNA, ), where TNA, is fund i s total net assets at the end of year t, Historical MaxTNA, is the maximum TNA the fund i recorded since its birth to the end of year t. Residuals are from the 3-variable regression in table 4. The standard errors in the panel regression are clustered by fund and t- statistics are provided below each coefficient in brackets. Panel A: Attrition Attrition t (4.01) (3.27) (6.81) Panel B: Residual Residual t 3.86E E E-04 (0.86) (0.26) (6.36) Category Expense t+1 (9.16) (3.36) (9.52) (3.47) Expense t (45.62) (45.88) Return t E-04 (2.11) (3.03) (1.12) (1.57) (2.55) (0.70) Log Size t E-04 (-6.54) (-6.62) (-5.98) (-6.78) (-6.81) (-6.03) Age t -7.2E E E E E E-06 (-1.64) (-0.50) (0.27) (-0.57) (0.48) (0.55) Constant (36.93) (12.30) (2.95) (33.38) (12.37) (3.62) R Number of Fund Years

27 Table 8 Attrition Rate and Merger Targets The table reports coefficients and z-stats from probit regressions Panel A: Target, = α + β Attrition, + β Age, + β Return, + β Size, + β AccumulatedReturn, + β Expense, + β YearDummy Panel B: Target, = α + β Residual, + β Age, + β Return, + β Size, + β AccumulatedReturn, + β Expense, + β YearDummy Target i,t+1 is a binary variable. It equals 1 if fund i is acquired by another fund in year t+1, and equals 0 otherwise. YearDummy is a set of dummy variables which identify each year separately. The residual in Panel B is from the regression in Table 4. Panel A: Attrition Panel B: Residual Attrition t (11.45) (11.33) Residual t (11.11) (11.02) Age t (-3.13) (-1.27) (-2.97) (-1.11) Return t (0.14) (-2.88) (0.08) (-2.88) Size t (-6.28) (-6.82) (-6.37) (-6.91) Accumulated Return t (-0.15) (-0.95) (-2.90) (-3.52) Expense t (2.92) (2.19) (3.57) (2.79) Constant (-16.70) (-10.40) (-13.74) (-7.68) Year Dummy No Yes No Yes Pseudo R Number of Fund Years

28 Table 9 Attrition Rate and Merger Acquirers The table reports coefficients and standard errors from regressions Panel A: Acquirer, = α + β Attrition, + β Age, + β Return, + β Size, + β AccumulatedReturn, + β Expense, + β YearDummy Panel B: Acquirer, = α + β Residual, + β Age, + β Return, + β Size, + β AccumulatedReturn, + β Expense, + β YearDummy Acqurier i,t+1 is a binary variable. It equals 1 if fund i acquires another fund in year t+1, and equal 0 otherwise. YearDmmy is a set of dummy variables which identify each year separately. The residual in Panel B is from the regression in Table 4. Panel A: Attrition Panel B: Residual Attrition t (1.13) (0.80) Residual t (0.86) (0.52) Age t (-0.62) (0.25) (-0.50) (0.36) Return t (-0.93) (0.35) (-0.97) (0.31) Size t (6.75) (6.71) (6.66) (6.63) Accumulated Return t (-1.53) (-0.45) (-1.84) (-0.62) Expense t (3.15) (2.91) (3.21) (2.97) Constant (-18.15) (-14.98) (-19.22) (-15.34) Year Dummy No Yes No Yes Pseudo R Number of Fund Years