This Draft: November 20, 2006

Transcription

1 Managerial Career Concern and Mutual Fund Short-termism Li Jin Harvard Business School Boston, MA and Leonid Kogan Sloan School of Management Massachusetts Institute of Technology Abstract: This Draft: November 20, 2006 Mutual fund investors reward short run performance with large inflows. Fund managers facing strong performance-related flows are shown to focus more on short horizon investments. Further tests of causality suggest that fund manager s short investment horizons are caused by their investors short horizons, but not the other way around. Fund managers subject to severe short-term tend to invest in larger stocks and growth stocks. Controlling for size, book-to-market and momentum effects, the stocks invested by higher turnover fund managers tend to generate lower return going forward, consistent with a hypothesis that these fund managers are inflating the demand for shorter-horizon investment opportunities and neglecting longer-horizon investment opportunities. The portfolio formed based on shorting the high turnover stocks and longing the lowest turnover stocks generates substantial abnormal return during 1992 to Keywords: Short-termism, career concern, mutual fund, flow performance relation, arbitrage strategy.

2 In our business, there's tremendous pressure for performance.... That potentially drives dysfunctional behavior and way too short-term behavior. - Duncan Richardson, portfolio manager and chief of research staff at Eaton Vance Management, quoted in Boston Globe, December 22, Over the last several decades, the mutual fund industry has supplanted the banking sector as the prime channel through which households invest. 1 A concurrent substantial increase in the trading volume of stocks is often attributed to churning of institutional investors whose average holding period of stocks has steadily decreased, in many cases to less than a year. 2 Pursuit of short-term trading profits over long-term gains has been alleged to account for corporations preoccupation with myopic goals and consequent wide fluctuations in individual stock prices, especially in volatile markets. 3 This paper analyzes churning by exploring how fund managers career concerns might affect investment behavior. Existing evidence suggests that new money flows to mutual funds chase recent fund performance, even though there is no consensus that fund managers can persistently outperform 4. As investors move money in response to shortterm fund performance, fund managers might feel the heat of having to perform in the short run and shun investments that pay off only in the long run. Appendix A presents an illustrative model whereby delegated portfolio managers facing short evaluation horizons might optimally elect to hold assets for only short periods of time. 5 1 Zheng (1999) reports that mutual funds at the end of the first quarter of 1998 managed about $3.3 trillion in assets, exceeding total bank savings deposits. According to the Investment Company Institute, the proportion of equity investment directly managed by households has declined, from approximately 63% in 1973 to 38.1% in 2003, and the number of stock mutual funds in the United States increased, from 399 in January 1984 to 4,601 in December According to Verter (2003), on a value-weighted basis the annual share turnover of the average NYSE/AMEX listed company climbed from approximately 13% in 1965 to more than 90% in 1999, a sevenfold increase. 3 For the former see Stein (1988, 1989), Jacobs (1991), and Porter (1992); for the latter see Dennis and Strickland (2002) and Griffin, Harris, and Topaloglu (2003). 4 In fact, part of the reason why funds might not be able to consistently outperform could be exactly due to the flow induced diseconomy of scale, as suggested by Berk and Green (2004). 5 A more rigorous model can be developed along the lines of Shleifer and Vishny (1997). 1

3 Casual empiricism suggests that pressure of investor redemption is a serious concern for many types of money managers. Hedge funds and private equity funds, for example, routinely adopt stringent minimum holding period restrictions to protect themselves from premature termination of long-term and illiquid investment projects. There is also some evidence that managerial career concern and agency problem might make mutual fund managers less willing to go against their clients will as reflected in the way investors allocate new fund flows. If managers chose to ignore the clients will and the subsequent performance is not good, managers might lose their clients 6. This paper tests whether mutual fund managers subject to severe pressure from their clients for short-term performance would correspondingly make shorter-horizon investment decisions, using data on actively managed mutual funds investing in domestic stocks. We first construct measures of investor impatience, which is how quickly the money flows into and out of the fund in response to the short-term performance of the fund. If the flow is very sensitive to short-term performance, then the fund investors are not really staying for the long term. We then measure the fund managers investment horizon by the average holding period of the equity portfolio in the fund, and the turnover of the portfolio. We demonstrate empirically that when the investment flow to a fund is more sensitive to recent performance, fund managers are in turn more focused on short horizon investing, as evidenced by shorter average holding periods of stocks and higher turnover of the fund. We need to first address a simple alternative explanation of the findings. Managers would turn over their portfolios more quickly when flow sensitivity to performance is higher if they are (mechanically) selling to meet redemption needs. Although selling for redemption might seem to plausibly explain the finding, a manager meeting redemption needs would probably sell the most liquid assets first, which are cash and cash equivalents, not the equity portfolio. In our empirical analysis we focus on funds with 6 Frazzini and Lamont (2005) find that mutual fund managers often heed to the implicit will of less sophisticated principles (mutual fund investors) in their asset allocation decisions, and that such decisions are on average losing money. This happens even though the fund managers have demonstrated (weak) stock picking ability. 2

4 substantial cash holdings (at least 10% cash or cash equivalents) so that mechanical redemption cannot be a first order reason for our findings for this sample. 7 After presenting evidence that managerial and investor short-termism are positively correlated with each other, we consider an alternative interpretation of the finding. Rather than managers responding to fund investor short-termism, it could be the other way round. Maybe investors chase managers with a certain trading style and some investors just happen to like managers who trade a lot. If these investors, for unknown reasons, also like to change managers often, it might be the case that hot money self-sorts itself into funds with shorter investment horizons. To address this possibility we develop tests to sort out causality. Our results indicate that investor short-termism causes managerial short-termism, but not the other way round. One direct consequence of money manager short-termism is the higher turnover of portfolios, and thus higher transaction costs. Edelen (1999) provides empirical evidence of the impact of higher turnover on portfolio performances. In addition, indirectly, fund managers facing higher short-term performance pressure will be unwilling to invest in stocks that are less liquid, or stocks that could deviate from their fundamental values for sustained period of time. This could generate supply-demand imbalance of certain assets. In particular, stocks that are over-demanded by short-termist money managers might see their prices bid up, and the future return depressed, whereas stocks that are shunned by these money managers might see their price depressed, and thus their future expected return increased. Arbitraging away such apparent mispricing could be difficult, if money managers feel that it might take a long time for the price to eventually correct itself, and in the interim the price could even deviate further away from the fundamental value, such that the arbitragers could be forced to close the position with a loss. We analyze the characteristics and future performance of stocks held by fund managers with different holding horizons. Stocks held by the mutual funds with higher turnovers 7 To ensure that the mechanical redemption need is not mistaken for managers conscientious response to investor short-termism, our analysis directly controls for volatility of fund flows. 3

5 (and shorter holding horizons) tend to be bigger, with lower book-to-market ratio. They also tend to have lower performance going forward, both the raw returns and at the characteristics adjusted returns. The underperformance is economically as well as statistically significant, suggesting that mutual fund short-termism is hurting fund performance. The paper is organized as follows. Section 1 reviews related literature. Explanations of data, measures of variables, and descriptive statistics are provided in Section 2. Section 3 presents regression results that show that mutual fund investors short-term behavior impels fund managers to have short investment horizons. Section 4 contains results showing the characteristics of stocks held by fund managers with different holding horizons, as well as the performance of these stocks going forward. Conclusions and implications are offered in Section Literature and theoretical background A substantial literature documents the relationship between fund flow and performance. Brown, Harlow, and Starks (1996), Chevalier and Ellison (1997), Gruber (1996), Ippolito (1992), and Sirri and Tufano (1998), among others, have demonstrated that new money inflows to mutual funds respond to recent fund performance, especially significant outperformance. Berk and Green (2004) and Lynch and Musto (2003) provide theoretical justification for the sensitivity of flow to performance in a rationale equilibrium context. Brown, Harlow, and Starks (1996), Chevalier and Ellison (1997, 1999a), and Khorana (1996) analyze the career concerns of fund managers and their implications for risk taking. These researchers document strong short-term performance-related turnover of fund managers, particularly younger fund managers. Researchers have also analyzed the relation between level of fund openness and fund returns. Stein (2004) argues that open-end mutual funds are ill-positioned to take on longterm arbitrage activities. [Open-end] funds, he maintains, will stick primarily to short- 4

6 horizon strategies and earn low excess returns. In so doing, they will leave large longhorizon mispricings such as the internet bubble mostly untouched because attacking such mispricings aggressively would require a closed-end structure. Empirical work by Edelen (1999) and Rakowski (2003) suggests that providing liquidity is costly and thus lowers fund performance. Edelen (1999), for example, finds that a unit of liquiditymotivated trading, defined as an annual rate of trading equal to 100% of fund assets, is associated with an estimated 1.5%-2% decline in abnormal returns of mutual funds. In this paper we explicitly test whether fund investment horizons are shortened when mutual funds degrees of openness are higher. We analyze how cross-sectional variations in managers investment horizons relate to the level of short-termism, measured by the sensitivity of fund flow to short-term performance. When a fund is largely dominated by hot money in pursuit of short-term profit, managers cannot afford to ignore short-term performance. Investments with long horizons, even if they promise to earn positive risk adjusted returns, tend to be avoided because they are less liquid, and their prices could deviate even further from the fundamental values for long periods of time, even if they eventually converge to fundamental values, as shown in Shleifer and Vishney (1990, 1997). This would also be consistent with the empirical literature that documents that liquidity might be priced; relevant studies include Amihud and Mendelson (1986), Brennan and Subrahmanyam (1996), Brennan, Chordia, and Subrahmanyam (1998), Pastor and Stambaugh (2003), among others. There is also an extensive literature on how investor short-termism occasions myopic investment decisions by corporate managers. Theoretical work includes that of Bebchuk and Stole (1993), Bolton, Scheinkman, and Xiong (2004), Shleifer and Vishny (1990, 1997), and Stein (1988, 1989). Empirical work includes that of Bushee (1998), Meulbroek et al. (1990), and Verter (2003). This paper moves one level up in the food chain in understanding why firm investors might exhibit short-termism. Institutional investors such as mutual funds hold increasingly large shares of public corporations, and managers likely react to the pressure from institutional investors. This study also provides solid empirical evidence of the theoretical models of myopic investment. Compared with 5

7 corporate managers, fund managers are under greater pressure to perform in the short term. Short-term underperformance can lead not only to the firing of fund managers, but also to partial liquidation of funds through fund outflow. Additionally, compared to corporate investments, investments by mutual funds are much easier to gauge, typically being in publicly traded securities for which continuous observations of market price are available. We can also observe holding horizons, as funds report their holdings regularly Data, measures of variables, and descriptive statistics In this section we identify the sources of our data and define the variables used and means of measuring them Data Our empirical analysis relies primarily on the CRSP survivorship bias free mutual fund database, which provides monthly open-end mutual fund data from 1961 to 2003, including information about fund investment style, monthly returns, and total net assets. Because information about fund investment styles becomes available only in 1992, we perform the analyses using data from 1992 to We adopt as our fund investment style indicator the Strategic Insight objective code, which is more complete and contains more detailed style information. 9 We emphasize fund styles with significant holdings in U.S. equities. Our fund style classification is more refined than those used in existing empirical studies, which typically consist of crude classifications of a few investment style categories. All fund investment styles included in this analysis are identified in 8 That fund holdings are constantly marked to market might lead some to perceive short-termism to be less severe. We do not think so, inasmuch as the existence of a market price does not guarantee efficient pricing. As long as it is possible for short-term price to deviate substantially from long-term fundamental values, fund managers will continue to be concerned about the impact of short-term performance. 9 CRSP provides several fund objective codes, the most recent being ICDI s Fund Objective Codes and Strategic Insight Fund Objective Codes. The raw data include 32,930 instances in which the ICDI objective code is missing but the Strategic Insight objective code is not, and 806 cases in which the Strategic Insight objective code is missing but the ICDI objective code is not. The ICDI objective code covers 23 different investment types, the Strategic Insight objective code more than 100 investment types. 6

8 Appendix B. Our sample is restricted to actively managed funds that have at least 10% of their holdings in cash or cash equivalents to avoid confusing mechanical selling in response to redemption need with conscientious choice of investment horizon in response to fund flows. The CRSP mutual fund data is supplemented with detailed fund level stock holdings data from the CDA/Spectrum mutual fund holdings database, which contains (up to) quarterly stock holding data for all registered mutual funds filed with the SEC, from 1980 to The data sets are linked using a file provided by Wharton Research Data Services (WRDS). For tests of causality between investor and investment short-termism we use an additional database that identifies each fund manager s date of birth or year of graduation from college. This data set is used to construct measures of manager age via a procedure described by Chevalier and Ellison (1999a). Whereas we use all available information to construct measures such as flow-toperformance sensitivity, in the regression analysis we perform analysis at the annual frequency because of the possibility that seasonality that might affect fund flows, fees, and performance. Many investors, moreover, implicitly have an annual horizon for planning and tax purposes. The existing literature (e.g., Brown, Harlow, and Starks, 1996; Chevalier and Ellison, 1997) supports this choice of frequency. As a robustness check, we also perform the analysis at the quarterly frequency. Section 4 discusses the robustness of results. The CRSP survivorship bias free mutual fund data delineate individually the different classes for each fund. Fund classes differ primarily by fee structure, but by definition they share the same managements and holdings. We consequently hand match and merge funds that differ only in fund class Definition of new money inflow 7

9 Following the existing literature (e.g., Sirri and Tufano, 1998, and Nanda, Wang, and Zheng, 2004b), new fund inflow is defined as the additional money attracted by a fund in a given month, measured by the dollar change in Total Net Assets (TNA) net of the assets price appreciation. Assuming that new money is invested at the end of each month, the cash flow for fund i in month t is given by: Newmoney i t = TNA i t - TNA i,t-1 (1 + R i t ) (1) where R it is the rate of return of fund i in month t. Normalizing the variable Newmoney by fund-level TNA at the beginning of the month gives the following measure for fund-level new money growth. Newmoneygrowth = it Newmoney it (2) TNAit, Measuring performance Fund performance in a period is calculated as the risk-adjusted returns. Specifically, we adjust the risk by (1) a raw market return, (2) a CAPM one-factor model, (3) a four-factor model proposed by Carhart (1997), and (4) the median return for the fund style. The results are not sensitive to these adjustment methods. For brevity, the results reported in this paper use CAPM adjusted excess returns. For each fund, in each month, we estimate the following market model: R it - R ft = α i + β i R mft +ε it (3) where R it is fund return in the month, R ft is the return on one month T-bills, R mft is the market risk premium (taken to be the difference between the value-weighted market index return and the one month T-bill rate), and β i is the beta of fund i. We estimate (3) 8

10 using the previous 12 monthly observations of fund return and market return to estimate the fund alpha, and use the alpha as our short-term performance measure Measuring investor short-termism We measure investor short-termism by the sensitivity of fund flows to recent performance. Following the existing literature, we run a regression of new money inflows on past one year performance. Consistent with the literature, to account for the documented convexity in the flow performance relationship (stellar performance draws more new money inflow), we account for the positive and negative past performance separately in our regression specification as below. New_money_growth i = α + β 1 Fund_underperformance +β 2 Fund_outperformance (4) We run this regression for each of the funds in our sample using up to 36 monthly observations of flow and performance, up to the end of the previous year. We then take the average of β 1 and β 2 as our estimate of the fund flow performance sensitivity, β. If there are multiple fund classes for a fund, we value-weight the fund class β using the TNA of the fund classes. Since the raw measures of β are heterogeneous, exhibiting fat tails and high kurtosis, we use the percentile rank of β as our flow performance sensitivity measure in the empirical analysis below. For robustness checks we construct an additional measure of investor short-termism by the goodness of fit of the inflow performance relationship, measured by the regression R 2 in equation (4). Being bounded between 0 and 1, this measure has the additional benefit of being more homogeneous Measuring investment short-termism Our first measure of fund managers investment horizons is the average remaining holding periods of the securities in the fund. Having detailed holding information for 9

11 each of the stocks in a fund portfolio, we estimate the time period in which the stocks are bought and sold and calculate, for all stocks currently in the portfolio at any given time, the remaining time periods until they are sold. 10 We then take the value-weighted average of these remaining holding periods. 11 Acknowledging the foregoing to be an ex post measure of average holding period and that an ex ante measure would be more appropriate, there is no compelling reason to believe the differences between ex-ante and ex-post measures to be systematic. We therefore believe that the ex post measure adds noise, but not bias, to the regressions. As another robustness check, we constructed an estimated version of the holding horizon for which the explanatory variables are fund age, fund size, lagged variable of the turnover of the fund, and number of stocks in the fund. The fitted value of the average holding horizon is an ex ante measure. The results in the paper using the ex ante measure of average holding horizon are qualitatively similar. In the regression analyses, average remaining holding period is sometimes abbreviated as maturity. Our second measure of investment short-termism, fund level turnover as reported in the CRSP survival bias free database, is intuitively the percentage of fund holdings that changes hands in a given period. The higher the turnover, the shorter a fund s average investment horizon. To properly account for the impact of changes in fund flow, the turnover ratio of the fund in CRSP is defined (following a standard definition in the field, e.g., Grinold and Kahn, 1995) as the minimum of aggregate purchases or sales of securities divided by a fund s average total net assets. 12 This measure is by definition fund specific. 10 For stocks held until the last period of the available data (Quarter 4 of 2003) we assume a remaining duration of another four quarters. This assumption, albeit ad hoc, is consistent with an observed average turnover over the full sample of mutual funds of about once per year. Robustness checks using two or six quarters of remaining holding periods do not yield significantly different results. We also tried to do analysis eliminate the last year observations so as to have a more complete transaction history. The regression results on the reduced data set show a similar pattern. 11 In some instances, assumptions must be made about how fund managers sell stocks, for example, whether first in first out (FIFO) or last in first out (LIFO). These turn out to not matter much as the resulting measures are at least 95% correlated. For brevity, the results reported in the paper reflect the LIFO assumption. Various ways to construct the average remaining holding period measure are detailed in Appendix C. 12 A seemingly more straightforward measure, [absolute(buy)+absolute(sell)]/holdings, or one-half thereof to account for round trip trading, is subject to the criticism that if the fund experiences large cash inflows 10

12 Summary statistics for the variables used in the paper are provided in Table 1. [Insert Table 1 about here.] Table 2 presents the correlations between the measures of investor and managerial shorttermism. [Insert Table 2 about here.] All measures of short-termism are positively correlated. Flow-to-performance sensitivity is, for example, positively correlated with turnover, the sensitivity of flow to performance negatively correlated with average remaining holding period (which is an inverse measure of fund level investment short-termism). Similarly, the measure of new money growth volatility, a measure of investor short-termism, is positively correlated with turnover and negatively correlated with average remaining holding period. All correlations are statistically significant at the 1% level. If the measures of short-termism reflect fundamental characteristics of the funds rather than noise, they are likely stable, subject to change only slowly over time. If, on the other hand, the measures reflect estimation error they will likely jump around over time. One way to check this is via the transition probability matrix. We rank the measure of shorttermism (such as turnover) into ten deciles for each period, then tabulate the next year s ranking of the same measure to calculate the transition probabilities. To ensure that we are not merely capturing the style-related stickiness of short-termism, all the measures are ranked within each fund style and each year. Table 3 reports transition probabilities of turnover and flow-to-performance sensitivity, which measure, respectively, short-termism at the managerial and investor levels. the measure can be arbitrarily high without any selling. For example, a fund that doubles its existing holdings will be calculated as having a turnover of 100% even though it might never sell any stocks. Thus might a rapidly growing fund be misclassified as a high turnover fund. 11

13 [Insert Table 3 about here.] The measures of short-termism recorded in Table 3 are far from random. One year on, the probability of staying in the same horizon decile remains considerably higher than the probability of switching to most other deciles. The pattern is particularly striking at the two extreme deciles: extremely high (low) turnover funds exhibit a strong tendency to maintain extremely high (low) turnover. For example, the probability that a fund in the bottom (top) turnover decile one year will continue to be in the same decile one year later is 43% (55%). Similarly, the probability that a fund in the bottom (top) decile of flow to performance sensitivity will continue to be in the bottom (top) decile one year later is 40% (25%). These are all significantly higher than the 10% benchmark that would result were the measures entirely noise. Other measures of investor and managerial shorttermism exhibit similar patterns. Thus, measures of managerial and investor shorttermism likely reflect important information about fund characteristics. 3. Regression analysis For most of the empirical analyses we estimate a pooled regression with panel-corrected standard errors (PCSE) and an AR (1) assumption about the error terms. The PCSE specification enables us to accommodate panel data with heteroskedasticity as well as autocorrelation and cross-correlation of the error terms (Beck and Katz, 1995). We also report clustered regression at the fund level, controlling for fund style and year fixed effects. According to Petersen (2005), of all standard methods commonly used in the literature this method best addresses the cross-correlation, serial correlation, heteroskedasticity, and fixed effects common in financial panel data Establishing the positive relation between investor and managerial short-termism We begin our regression analysis by examining how flow-to-performance sensitivity affects managerial (investment) short-termism measured as either the average remaining 12

14 holding period or turnover, controlling for other fund attributes. We add the following control variables. 1) Past year performance, as this might affect both investor and managerial short-termism. As demonstrated by Brown, Harlow, and Starks (1996) and Chevalier and Ellison (1997), past fund performance might influence the degree of aggressiveness with which fund managers churn their portfolios. Better performing managers who feel relatively more secure in their jobs might be more willing to invest for the longer run. Investors in better performing funds might have greater trust in their managers and thus not leave so quickly when performance is poorer. 2) The size of the fund, measured by the natural log of total net assets. Larger funds are harder to turn over and also more likely, over time, to have some settled investors and, thus, lower flow to performance sensitivity. 3) Monthly fund flow volatility, seasonally adjusted, estimated using the previous three years monthly flow data. This controls specifically for the redemption related story discussed earlier. If funds facing more volatile inflows mechanically shorten the investment horizon to meet redemption need, the effect is likely captured by this control rather than by flow-to-performance sensitivity. Other control variables include fixed effect in fund style and calendar year. Specifically, we use fund-level information to estimate the following pooled regression with panel-corrected standard errors. Managerial_short_termism i,t = α + β 1 flow_performance_sensitivity i,t-1 + β 2 fund performance i,t-1 + β 3 log (TNA) i,t-1 + β 4 volatility of fund flow i,t-1 (5) Here, i is the index for the fund and t the index for time period. Managerial_short_termism is measured by either maturity or turnover. The regression results are presented in Table 4. Columns (1) and (2) report the results of the PCSE regression that permits the error terms to follow an AR (1) process over time. Column (1) reports the regression result of turnover on the sensitivity of flow to performance, Column (2) the regression result of average remaining holding period on the sensitivity of flow to performance. As a comparison, columns (3) and (4) report in the 13

15 same order the results of the clustered regressions, controlling for fixed effects of fund style and calendar year. [Insert Table 4 about here.] The coefficient estimates reported in Table 4 for β 1 are significantly positive when managerial short-termism is measured by turnover (1.31 and 1.34 with t-statistics of 3.82 and 4.84 for columns (1) and (3), respectively) and significantly negative when managerial short-termism is measured by average remaining holding period (-1.03 and with t-statistics of and for columns (2) and (4), respectively). Both results are consistent with the intuition that increased investor short-termism reduces investment horizons for mutual funds. Economically, moving from the lowest rank of flow-to-performance sensitivity to the highest rank will increase turnover and decrease maturity by about 10%. Coefficient estimates on the control variables are largely consistent with those reported in the previous literature. Turnover decreases, and remaining holding period increases, with past performance. Both are statistically significant. Consistent with intuition, fund size significantly reduces turnover, although its impact on remaining holding horizon is less than significant. As expected, higher volatility of fund flow increases fund turnover and decreases maturity, holding all else constant. However, controlling for the impact of fund flow volatility, flow sensitivity to performance still incrementally affects turnover and maturity. We rerun the tests using as an alternative measure of investor level short-termism the goodness of fit of the flow performance regression. The goodness of fit of the flow-toperformance relation measures how well past performance explains the new money inflow and, thus, the responsiveness of investors to short-term performance. The results of these tests and of regressions run using both the PCSE and clustered regression are reported in Table 5. 14

16 [Insert Table 5 about here.] High goodness of fit of the flow performance relationship (regression R 2 of the flowperformance relationship) increases turnover and decreases average remaining holding period, as reported in Table 5. In summary, empirical evidence suggests a positive and significant relation between investor short-termism and managerial short-termism. Higher flow-to-performance sensitivity is associated with shorter fund investment horizons. In deriving the results, we control for fund investment style and time fixed effects as well as for other factors that might affect fund-level investment horizons Causality tests Although we have established a positive relation between investor short-termism and fund manager short-termism, we are not certain of the direction of causality. If shorttermism on the part of fund managers attracts short-term oriented investors, the causality might go the other way round. We employ a simultaneous equations approach to address the causality question. Assuming that both investor-level short-termism and investment short-termism can be determined endogenously and potentially by each other, we look for instrumental variables that enable us to differentiate the causality and use a three-stage least squares method to estimate the relation between investor and managerial shorttermism. We choose manager age to be an instrument for managerial short-termism. As demonstrated by Chevalier and Ellison (1999a), younger managers, being more subject to career concerns and thus under greater pressure to perform in the short run, are likely to be less willing to hold longer-horizon investments. Manager age should therefore be positively related to investment horizon. There is, on the other hand, no compelling reason to believe ex ante that manager age is directly related to the flow-to-performance 15

17 sensitivity of funds. 13,14 We follow the procedure in Chevalier and Ellison (1999a) to calculate manager age, assuming managers to have been 21 upon graduating from college. When reported, we use managers birth years to directly calculate manager age. 15 For funds that have multiple managers we take the average age of the managers that can be identified. Manager age is constructed from a separate data set provided by Morningstar. The sample size for this test is smaller than for the previous ones, not all managers in the previous regression analysis being identified in the Morningstar manager name database. Sirri and Tufano (1998), Jain and Wu (2000), and Huang, Wei, and Yan (2005) show that intensifying marketing activities generates fund flow, but the additional fund flow is more likely from hot money that comes in just for performance and will likely leave if disappointed. In the absence of any compelling reason to link marketing expense to fund managers investment horizons we choose marketing expenses, proxied by total fund fees, to be the instrumental variable for flow-to-performance sensitivity. 16 We follow the procedure in Sirri and Tufano (1998) in calculating total fund fees to be expense plus 1/7 of load, load being amortized without discounting over seven years, the average holding period for an equity fund in their data. 13 There is a concern that learning about manager ability could be involved, that investors react more to younger managers performance because there is more to learn about their abilities. In practice, however, investors are more likely to learn about funds than about fund managers. Funds are more stable than their managers, who often move in and out of funds. Although some star managers might attract investors, most departing managers probably don t see a large number of investors following them. Existing fund rating agencies as well as funds themselves typically report fund historical performance rather than fund managers track records. Investors are thus more likely to buy into funds than into the fund managers behind them. 14 It is possible that manager age is related to fund age, and previous literature has documented that fund age is related to flow to performance sensitivity. To address this concern, as a robustness check, we also use as an instrument the residual of a regression of manager age on fund age. Specifically, we take the manager age that is orthogonal to fund age as an instrument. The results do not change qualitatively. 15 Chevalier and Ellison (1999a) consider and reject as an alternative measure of manager experience the tenure of fund managers, as reported by Morningstar, because the measure is less meaningful and much noisier. 16 Although in theory it is possible that a fund manager runs an integrated strategy to manage investing horizons through both marketing and investment channels, talks with real world fund managers reveal that in practice this would be difficult as most funds have distinct divisions of labor, and manufacturing and distribution are separate channels. 16

18 An additional instrumental variable for flow-to-performance sensitivity can be constructed by looking at the composition of investors in the fund. Since fund age is often longer than the tenure of the fund manager, the composition of investor base in the fund is, from the manager s perspective, largely pre-determined. Such composition might affect the level of investor short-termism. In particular, institutional investors such as pension funds have been shown to be more performance-sensitive than other types of investors (James and Karceski 2005, Del Guercio and Tkac 2002). James and Karceski (2005), for example, find cash flows into institutional funds with high minimum investment requirements to be significantly more sensitive to risk-adjusted measures of performance than flows into small institutional or retail funds. The precise composition of fund investor base being unobservable, we follow James and Karceski (2005) in obtaining from Morningstar, for each fund and each year, the minimum initial investment requirements information. We then construct the natural log of the minimum initial investment requirement as our proxy for institutional investor concentration. 17 The higher the natural log of minimum initial investment requirement, the more likely the fund is to be populated by larger institutional investors. 18 The simultaneous equations to be estimated are: Managerial short-termism = f 1 (investor short-termism, manager age, other controls); and Investor short-termism = f 2 (managerial short-termism, marketing expense, institutional investor concentration, other controls). The results of the causality test are presented in Table 6. Investor short-termism does, indeed, affect investment short-termism. Higher flow-to-performance sensitivity significantly decreases average remaining holding period and significantly increases fund 17 If the minimum initial investment is missing we take it to be zero. We change all the zeros to ones in order for the log measure to be well defined. Removing the zero minimum initial investment sample altogether would substantially reduce the sample size, but would not change the results qualitatively. 18 Morningstar data also allows us to construct an institutional fund dummy, whereby institutional funds are so classified if they have the word institution in the name or have a minimum initial investment requirement of at least $100,000. We feel that such an institutional dummy is cruder than the measure we use. Replacing our measure with the dummy does not change any of our results qualitatively. 17

19 turnover, controlling for endogeneity of the measures of investor short-termism. The results are statistically as well as economically significant. There is, on the other hand, no significant evidence that investment short-termism causes investor short-termism. [Insert Table 6 about here.] Another test of causality can be conducted using the sample of managers who switch from one fund to another. If managerial short-termism attracts investor short-termism, we would expect that after the manager settles down at the new fund investor short-termism might change and should be predicted by the old trading style of the incoming manager. On the other hand, if managerial short-termism is affected by investor shorttermism, the level of managerial short-termism will likely change when the same manager moves to a new fund, in which case the level of existing investor short-termism in the new fund should predict the level of managerial short-termism for the incoming manager. An extensive search of the historical Morningstar database yields 86 cases in which fund managers moved during our sample period and for which we have matching observations of other fund characteristics. We perform OLS analyses to test two hypotheses. 19 Hypothesis 1: That the new funds historical level of investor short-termism predicts new managers investment short-termism, after we account separately for the level of managerial short-termism in the old fund. This, if true, suggests that investor shorttermism incrementally drives managerial short-termism. Hypothesis 2: That old funds level of managerial short-termism predicts new funds level of investor short-termism, after we account for the historical level of investor short-termism in the new fund. This, if true, suggests that managerial short-termism 19 OLS is used as opposed to other more sophisticated econometrics analyses, and we could not control for year or fund style fixed effect because of the limited sample size. 18

20 incrementally drives investor short-termism, after controlling for historical level of investor short-termism. To ensure that we are not merely capturing the unstableness of the transition period when a manager moves, we measure the old fund investment horizon using (up to) the previous three years average level of turnover (or maturity) and exclude the year the manager leaves the old fund. Similarly, we measure the new fund investment horizon and new fund investor characteristics using (up to) three years average levels and we exclude the year the manager joins the new fund. To exclude the transition period, we exclude the year if it is less than six months from the time the manager joins or leaves a fund. The new test results are reported in Table 7. Panel A tests Hypothesis 1. A higher level of investor short-termism in the new fund will predict a higher level of new fund turnover, even after controlling for the managers previous fund turnover. The coefficient of is statistically significant, with a t-statistic of Thus, investor short-termism drives managerial short-termism. On the other hand, higher level of investor short-termism in the new fund will predict a lower level of new fund average maturity, even after controlling for the managers previous fund turnover. The coefficient of is again significant, with a t-statistic of Managers track records of turnover (or maturity) only weakly predict their future levels of turnover (or maturity), once investor shorttermism and other factors are controlled for. In both regressions, the coefficients on the old fund managerial short-termism are positive but statistically insignificant. Panel B tests Hypothesis 2. New funds investor short-termism is significantly predicted by the lagged levels of new fund investor short-termism. Managerial short-termism in the old fund, measured by either turnover or maturity, does not have additional explanatory power on new fund investor short-termism, once everything else is controlled for. [Insert Table 7 about here.] 19

21 Taken together, the results suggest that when managers switch funds the managerial short-termism in the new fund will be influenced by the historical level of investor shorttermism in the new fund. The new level of investor short-termism in the new fund, however, is not significantly affected by the managers previous track record of turnover or maturity. We used flow-performance sensitivity as the measure of investor short-termism. Similar results obtain if we use the regression goodness of fit of the flow-performance relation as the measure of investor short-termism Robustness Checks The performance measure on which we focus in the paper is the CAPM risk adjusted return. Other benchmarks include (1) raw market return, (2) the three factor Fama-French model, (3) the four-factor model proposed by Carhart (1997), and (4) median return for the fund style. The results in the paper are not sensitive to the performance benchmark we chose. There has been evidence that flow performance relationship might be non-linear, and the slope could be steeper for out-performance than for under-performance (as in Lynch and Musto 2003). The main results in the text allows for asymmetric flow-performance relationship by separately estimating the two slopes and taking the average to estimate the fund level flow to performance sensitivity. Alternatively, many existing researches propose using a quadratic function to fit the flow performance relationship. We also tried that, by adding higher order terms to the flow-performance relation, such as quadratic and cubic terms. The resulting flow-to-performance sensitivity is constructed as the first derivatives of flow to performance, evaluated at the median performance of the fund over the past year. The subsequent tests give qualitatively similar results. In the main results the flow to performance sensitivity measures are constructed by regressing current period new money growth on past fund performance, without further 20

22 controlling for other factors that might also affect the flow. Alternatively, control variables can be added to the regression specification. Following the existing literature, we include as control variables the log of fund size and expense ratio. 20 Using the resulting measure of flow-to-performance sensitivity does not qualitatively change our test results. The main results in the text are obtained using annual data of the short-termism measure. The results of an analysis that used quarterly data are not qualitatively different. The main results reported in the paper are for cash-rich mutual funds. Repeating the tests using the full sample of mutual funds yields qualitatively similar results. The results are thus not confined to cash-rich funds. 4. Characteristics and future performance of stocks held by funds with higher and lower short-termism Mutual fund managers who face severe investor short-termism might have two reactions. First, they might have to work harder, as they do not have an entrenched position which they could rely on. Compared to established managers who might be able to enjoy better life, the younger and less established managers have to be extremely diligent in finding investment opportunities and act on these. This would contribute to higher turnover, but it is generally good for the fund performance. Second, the fund managers under pressure to perform (or else face the severe consequence of large redemption) might be forced to concentrate on short-term investment opportunities, i.e., opportunities that are more likely to yield positive performance in the short run. These managers do not have the luxury to wait out for investment opportunities that might have a duration longer than their own career. This second reaction is necessary out of managerial career concern, but 20 In such settings, flow-to-performance sensitivity continues to be defined as in equation (4), but the goodness of fit of flow performance relation is now measured as the incremental R 2, namely, the difference between R 2 with and without the return variables in a regression setting in which all other variables of interest are already controlled for. 21

23 it hurts performance. Compared to the more secured fund managers who could in theory choose to invest in every opportunity, the insecure fund managers have to limit their own investment opportunity set, and shy away from the really long term investment opportunities, even if such opportunities have higher expected returns than those shorterterm investment opportunities. In equilibrium, the number of money managers capable of investing in long-term, illiquid opportunities are less than the number of money managers capable of investing in shorter-term opportunities, the long-term investments will on average command a higher expected return, as a reward for patience. Such reward is out of the reach of insecure money managers. It is in this sense that manager insecurity hurts fund performance. Direct test of whether insecure money managers do better or worse than secured ones might be less conclusive, as it would run into the afore-mentioned two reactions that offset each other. We will conduct tests focusing on the second effect, namely, the assets shun away by insecure managers tend to have higher expected returns, thus failure to include these in fund portfolios hurt performance. We first document the characteristics of stocks held by fund managers with different holding horizons. During our sample period of 1992 to 2003, for each quarter, we first compute the aggregate measure of fund short-termism for each stock in the following way: we first collect the annualized turnover of the previous calendar year for each of the non-index equity fund reported by CRSP mutual fund database. We then calculate the weighted average turnover of these non-index funds for each of the stocks, where the weights are the number of shares held by each of these funds in the current quarter. This measures the aggregate turnover of all non-index mutual funds in the stock. A higher measure indicates that the holding funds tend to churn the stock more quickly, whereas a lower measure indicates that the average holding funds tend to sit on the stock for longer period of time. Similarly, we construct the aggregate holding horizon of all non-index funds in each stock. The higher average holding horizon indicates that the funds in aggregate sit on the stock for longer period of time. 22