Online Appendix. Do Funds Make More When They Trade More?

Transcription

1 Online Appendix to accompany Do Funds Make More When They Trade More? Ľuboš Pástor Robert F. Stambaugh Lucian A. Taylor April 4, 2016

2 This Online Appendix presents additional empirical results, mostly robustness results, complementing the results presented in the paper. Most of the results presented here are summarized in the paper. Contents Section 1: Results from the Subperiod Table A1: Turnover-Performance Relation in the Cross Section and Time Series Table A2: Heterogeneity in the Turnover-Performance Relation Table A3: Properties of Fund Turnover and Performance Across Fund Categories Table A4: Correlations of Average Turnover Across Fund Categories Table A5: Commonality in Fund Turnover Table A6: What Explains Turnover? Table A7: Relation Between Fund Performance and Average Turnover Section 2: Alternative Fixed Effects Table A8: Counterpart of Paper s Table 1 with Fund and Benchmark-Month Fixed Effects Table A9: Counterpart of Paper s Table 1 with Fund Manager Fixed Effects Section 3: Placebo Test with Passive Funds Table A10: Counterpart of Paper s Table 1 with Passive Index Funds, Conservative Filter Table A11: Counterpart of Paper s Table 1 with Passive Index Funds, No Expense- Ratio Filter Table A12: Counterpart of Paper s Table 1 with Passive Index Funds, Higher Turnover Filter 2

3 Section 4: Finite-Sample Bias? Table A13: Does Performance Predict Turnover? Figure A1: Simulation Analysis To Assess Finite-Sample Bias Table A14: Turnover-Performance Relation Controlling for Lagged Performance Section 5: Alternative Clustering Table A15: Counterpart of Paper s Table 1, Clustering by Month Table A16: Counterpart of Paper s Table 1, Clustering by Fund and Month Table A17: Counterpart of Paper s Table 1, Clustering by Year Table A18: Counterpart of Paper s Table 2, Clustering by Month Table A19: Counterpart of Paper s Table 2, Clustering by Fund and Month Table A20: Counterpart of Paper s Table 7, Clustering by Month Table A21: Counterpart of Paper s Table 7, Clustering by Fund and Month Section 6: Measuring Turnover Table A22: Robustness to Flow-Induced Turnover Table A23: Counterpart of Paper s Table 1 with Benchmark-adjusted Turnover Table A24: Turnover-Performance Relation with Rescaled Turnover Measure Table A25: Turnover-Performance Relation Controlling for Stock Holdings Section 7: Alternative Benchmark Models Table A26: Counterpart of Paper s Table 1 with Estimated Morningstar Betas Table A27: Version of Table A26 with Betas Conditional on FundTurn Table A28: Counterpart of Paper s Table 1 with Three-Factor Fama-French Benchmark Table A29: Version of Table A28 with Betas Conditional on FundTurn Table A30: Counterpart of Paper s Table 1 with Four-Factor Fama-French-Carhart Benchmark 3

4 Table A31: Version of Table A30 with Betas Conditional on FundTurn Table A32: Counterpart of Paper s Table 1 with Five-Factor Fama-French Benchmark Table A33: Version of Table A32 with Betas Conditional on FundTurn Table A34: Counterpart of Paper s Table 1 with Cremers-Petajisto-Zitzewitz Benchmark Table A35: Version of Table A34 with Betas Conditional on FundTurn Section 8: Alternative Skill Proxies Table A36: Counterpart of Paper s Table 2 with Unadjusted Gross Alpha Table A37: Counterpart of Paper s Table 2 with Adjusted Gross Alpha Table A38: Manager Age, Tenure, and the Turnover-Performance Relation Table A39: Counterpart of Paper s Table 2 with fund-size and expense-ratio terciles computed within style-months Table A40: Comparing Direct-Sold and Broker-Sold Funds Section 9: Additional Results Table A41: Summary Statistics Table A42: Counterpart of Paper s Table 1 with Benchmark-Adjusted Net Returns Table A43: Counterpart of Paper s Table 1 with Annual Data Table A44: Interacting Turnover with Time Since Turnover Table A45: Additional Lags of Turnover Table A46: Counterpart of Paper s Table 1 in Cold-IPO-Market Subperiod ( ) Table A47: Economic Significance Figure A2: Nonlinearities in the Turnover-Performance Relation? 4

5 1. Results for the Recent Subsample Table A1 Turnover-Performance Relation in the Cross Section and Time Series This table is the same as Table 1 in the main paper, but uses data from Month Fixed Effects Fund Fixed Effects No Yes Yes (4.37) (3.74) No (-1.00) (-1.71) 5

6 Table A2 Heterogeneity in the Turnover-Performance Relation This table is the same as Table 2 in the main paper, but uses data from Panel A: Stock Size Categories Small Cap Mid Cap Large Cap Small - Large Controls No (4.23) (1.57) (3.63) (2.25) Yes (2.16) (0.47) (0.85) (1.57) Panel B: Stock Value-Growth Categories Growth Blend Value Growth Value Controls No (3.06) (4.52) (3.88) (-0.07) Yes (0.14) (0.47) (0.12) (0.03) Panel C: Fund Size Categories Small Medium Large Small Large Controls No (5.44) (1.66) (0.97) (2.54) Yes (2.06) (0.47) (0.50) (1.93) Panel D: Fund Expense Ratio Categories High Medium Low High Low Controls No (3.99) (3.34) (2.44) (1.96) Yes (1.35) (0.47) (0.16) (2.25) 6

7 Table A3 Properties of Fund Turnover and Performance Across Fund Categories This table is the same as Table 3 in the main paper, but uses data from Average benchmark- Funds Number Fund turnover (fraction/year) adjusted return (%/month) included of funds Average Volatility Autocorr. Gross Net Panel A: Full Sample All Panel B: Stock Size Categories Small-Cap Mid-Cap Large-Cap Small Large (t-statistic) (4.43) (-0.29) (-0.89) (0.72) (0.58) Panel C: Stock Value-Growth Categories Growth Blend Value Growth Value (t-statistic) (13.37) (7.17) (2.93) (1.25) (1.19) Panel D: Fund Size Categories Small Medium Large Small Large (t-statistic) (5.98) (4.72) (-3.63) (0.57) (0.01) Panel E: Fund Expense Ratio Categories High Medium Low High Low (t-statistic) (6.50) (6.59) (1.13) (0.34) (-2.79) 7

8 Table A4 Correlations of Average Turnover Across Fund Categories This table is the same as Table 4 in the main paper, but uses data from Stock Size S M L Stock Value-Growth G B V Small 1.00 Growth 1.00 Mid Blend Large Value Fund Size S M L Fund Expense Ratio L M H Small 1.00 Low 1.00 Medium Medium Large High

9 Table A5 Commonality in Fund Turnover This is the same as Table 5 in the main paper, but uses data from (1) (2) (3) (4) (5) (6) AvgT urn (14.08) (1.10) (2.55) (2.81) (4.40) (7.63) AvgT urn Stock Size (5.60) AvgT urn Stock V G (2.66) AvgTurn Fund Size (4.97) AvgTurn Fund Exp (4.20) AvgT urnsim (6.10) Observations 217, , , , , ,058 Within-fund R 2 (%)

10 Table A6 What Explains Turnover? This table is the same as Table 6 in the main paper, but uses data from Dependent variable: FundTurn it Dependent variable: AvgTurn t (1) (2) (3) (4) (5) (6) (7) (8) Sentiment t (2.53) (2.02) (4.47) (3.62) V olatility t (5.19) (4.31) (3.93) (5.34) Liquidity t (-4.02) (-4.60) (-2.62) (-3.84) Business Cycle t (-2.96) (-5.31) Market Return t (-1.22) (-1.06) Time Trend t (-0.87) (-1.38) (-1.96) (-1.54) (-1.73) (-2.24) (-2.71) (-2.60) R R 2 R 2 (trend only) Observations

11 Table A7 Relation Between Fund Performance and Average Turnover This table is the same as Table 7 in the main paper, but uses data from (1) (2) (3) (4) (5) (6) AvgTurn i,t (3.29) (3.22) (2.27) AvgTurnSim i,t (3.89) (3.74) (2.06) FundTurn i,t (4.37) (3.43) (4.50) (4.09) Observations 221, , , , , ,058 11

12 2. Alternative Fixed Effects Table A8 Counterpart of Paper s Table 1 with Fund and Benchmark-Month Fixed Effects This table is the same as Table 1 in the main paper, except we replace month fixed effects with benchmark-month fixed effects. Benchmark-Month Fixed Effects Fund Fixed Effects No Yes Yes (6.63) (7.96) No (1.92) (1.48) 12

13 Table A9 Counterpart of Paper Table 1 with Fund Manager Fixed Effects This table is the same as Table 1 in the paper, except it replaces fund fixed effects with fund manager fixed effects. Data on manager identities are from Morningstar. If a manager manages multiple funds in a given month, the same manager s fixed effect appears in multiple fund-month observations in that month. If a fund has multiple managers in a given month, we follow a simple seniority-based approach to assign the manager fixed effects. When there are multiple managers, we define the fund s manager to be the person who arrived at the fund first. If there is a tie, we take the person who stays at the fund longest. Month Fixed Effects Fund Manager Fixed Effects No Yes Yes (6.30) (6.38) No (1.92) (1.61) 13

14 3. Placebo Test with Passive Funds Table A10 Counterpart of Paper s Table 1 with Passive Index Funds, Conservative Filter This table is the same as Table 1 in the main paper, except we use data on mutual funds that Morningstar classifies as index funds. To remove active funds that may be accidentally classified as passive funds, we exclude funds with turnover greater than 100% per year and funds with expense ratio greater than 1% per year. There are 12,520 observations in each regression. Month Fixed Effects Fund Fixed Effects No Yes Yes (-0.51) (-1.07) No (0.05) (0.01) 14

15 Table A11 Counterpart of Paper s Table 1 with Passive Index Funds, No Expense-Ratio Filter This table is the same as the previous table, except we remove the expense-ratio filter to create the sample of passive funds. There are 13,590 observation in each regression. Month Fixed Effects Fund Fixed Effects No Yes Yes (-0.44) (-1.15) No (0.34) (0.40) 15

16 Table A12 Counterpart of Paper s Table 1 with Passive Index Funds, Higher Turnover Filter This table is the same as Table A10, except we apply a 500% turnover filter instead of a 100% filter to create the sample of passive funds. There are 12,669 observations in each regression. Month Fixed Effects Fund Fixed Effects No Yes Yes (-0.61) (-1.38) No (0.27) (0.16) 16

17 4. Finite-Sample Bias? Table A13: Does Performance Predict Turnover? This table shows results from regressing fiscal-year fund turnover on contemporaneous and lagged fund performance. We work at the fiscal-year frequency so the timing of turnover and performance overlaps. The dependent variable is F undt urn(i, t) [FYR], defined here as fund i s turnover in fiscal year t. GrossR(i, t) [FYR] is the fund s benchmark-adjusted gross return during fiscal year t. GrossR(i, t 1) [FYR] is the fund s benchmark-adjusted gross return during the 12-month period prior to GrossR(i, t) [FYR]. All regressions include fund fixed effects and cluster by fund to account for persistence in turnover. (1) (2) (3) GrossR(i,t) [FYR] (0.55) (-0.19) GrossR(i,t-1) [FYR] (-1.88) (-1.80) Observations

18 9 x R(t+1) on v(t) R(t+1) on v(t), R(t), and R(t 1) 7 Mean slope coefficient on v(t) Simulated sample length (years per fund) Fig. A1. Simulation to Assess Finite-Sample Bias. This figure shows results from simulations used to quantify potential finite-sample bias in our main turnover-performance regressions. The figure also shows how controlling for lagged performance eliminates the bias. We simulate many samples, each of which has τ yearly observations. We plot τ on the horizontal axis. Simulated returns R are i.i.d. normal with mean zero and standard deviation 10.3% per year, which matches the empirical volatility. Simulated turnover follows v(t) = R(t 1) + u(t), where u(t) is i.i.d. normally distributed with mean zero and standard deviation The numbers and 0.44 are the slope and residual volatility, respectively, from an actual regression of annual FundTurn it on lagged GrossR i,t 1, both measured at the fiscal-year frequency, and fund fixed effects. The dashed red line shows the slope from a regression of R(t + 1) on v(t), averaged across simulated samples. The solid black line shows the slope from a regression of R(t +1) on v(t), R(t), and R(t 1), averaged across simulated samples. 18

19 Table A14: Turnover-Performance Relation Controlling for Lagged Performance The dependent variable is GrossR it, the fund s benchmark-adjusted gross return in month t. The first column matches the top-left cell of our paper s Table 1. The second column is the same but only uses observations with non-missing values of GrossRFY Ri, t 1 and GrossRFY Ri, t 2. Abusing notation slightly, GrossRFY R i,t 1 is the fund s benchmarkadjusted gross return during the fiscal year that coincides with the timing of FundTurn i,t 1. GrossRFY R i,t 2 is the fund s benchmark-adjusted gross return during the 12-month period prior to GrossRFY R i,t 1. Column 2 shows that we lose some observations when we require the additional regressors, and the smaller subsample has a slightly weaker, but still strong, turnover-performance relation. All regressions include fund fixed effects and cluster by Sector month. (1) (2) (3) FundTurn i,t (6.63) (5.56) (5.47) GrossRFY R i,t (0.99) GrossRFY R i,t (-4.33) Observations

20 5. Alternative Clustering Table A15: Counterpart of Paper Table 1, Clustering by Month This table is the same as the paper s Table 1, except we compute t-statistics differently. Whereas Table 1 in the paper clusters by Sector month, this table clusters by month. Month Fixed Effects Fund Fixed Effects No Yes Yes (4.32) (4.66) No (1.10) (0.86) 20

21 Table A16: Counterpart of Paper Table 1, Clustering by Fund and Month This table is the same as the paper s Table 1, except we compute t-statistics differently. Whereas Table 1 in the paper clusters by Sector month, this table clusters by fund and month. Month Fixed Effects Fund Fixed Effects No Yes Yes (3.96) (4.22) No (1.08) (0.84) 21

22 Table A17: Counterpart of Paper Table 1, Clustering by Year This table is the same as the paper s Table 1, except we compute t-statistics differently. Whereas Table 1 in the paper clusters by Sector month, this table clusters by calendar year. Month Fixed Effects Fund Fixed Effects No Yes Yes (3.65) (3.36) No (0.81) (0.61) 22

23 Table A18 Counterpart of Paper s Table 2, Clustering by Month This table is the same as Table 2 in the paper, except we compute t-statistics clustering by month instead of by sector month. Panel A: Stock Size Categories Small Cap Mid Cap Large Cap Small - Large Controls No (4.85) (2.70) (4.17) (3.79) Yes (3.14) (1.20) (0.98) (2.88) Panel B: Stock Value-Growth Categories Growth Blend Value Growth Value Controls No (3.58) (4.83) (4.39) (0.44) Yes (1.25) (1.20) (1.55) (0.21) Panel C: Fund Size Categories Small Medium Large Small Large Controls No (5.67) (2.83) (1.04) (3.36) Yes (2.67) (1.20) (0.50) (2.20) Panel D: Fund Expense Ratio Categories High Medium Low High Low Controls No (4.74) (3.29) (2.08) (3.93) Yes (2.43) (1.20) (0.72) (3.29) 23

24 Table A19 Counterpart of Paper s Table 2, Clustering by Fund and Month This table is the same as Table 2 in the paper, except we compute t-statistics clustering by fund and month instead of by sector month. Panel A: Stock Size Categories Small Cap Mid Cap Large Cap Small - Large Controls No (3.81) (2.64) (3.73) (2.84) Yes (2.66) (1.11) (0.85) (2.24) Panel B: Stock Value-Growth Categories Growth Blend Value Growth Value Controls No (3.30) (4.23) (4.10) (0.41) Yes (1.22) (1.11) (1.40) (0.19) Panel C: Fund Size Categories Small Medium Large Small Large Controls No (4.96) (2.49) (1.00) (2.91) Yes (2.32) (1.11) (0.46) (1.93) Panel D: Fund Expense Ratio Categories High Medium Low High Low Controls No (3.99) (3.15) (1.97) (3.08) Yes (2.21) (1.11) (0.63) (2.79) 24

25 Table A20 Counterpart of Paper s Table 7, Clustering by Month This table is the same as Table 2 in the paper, except we compute t-statistics clustering by month instead of by sector month. (1) (2) (3) (4) (5) (6) AvgTurn i,t (1.37) (1.36) (1.03) AvgTurnSim i,t (2.47) (2.17) (2.70) FundTurn i,t (4.32) (4.69) (4.61) (4.83) Observations

26 Table A21 Counterpart of Paper s Table 7, Clustering by Fund and Month This table is the same as Table 2 in the paper, except we compute t-statistics clustering by fund and month instead of by sector month. (1) (2) (3) (4) (5) (6) AvgTurn i,t (1.37) (1.36) (1.02) AvgTurnSim i,t (2.46) (2.15) (2.60) FundTurn i,t (3.96) (4.08) (4.27) (4.38) Observations

27 6. Measuring Turnover 27

28 Table A22 Robustness to Flow-Induced Turnover The dependent variable in each regression is GrossR i,t, fund i s benchmark-adjusted gross return in month t. Column 1 reproduces the results with fund fixed effects from the paper s Table 1. We consider two measures of lagged flow-induced turnover, denoted FlowTurn1 i,t 1 and FlowTurn2 i,t 1. Both measures are computed from the monthly series of fund size and fund returns, and both are measured over the same 12-month period as FundTurn i,t 1. FlowTurn1 i,t 1 is the sum of the absolute values of the 12 monthly dollar flows, divided by the average fund size during the 12-month period. FlowTurn2 i,t 1 is the smaller of two sums, one of all positive dollar flows and one of all negative flows during the 12-month period, divided by average fund size. FlowV ol i,t 1 is the standard deviation of the 12 monthly flows, measured as fractions during the same time window as FundTurn i,t 1. We winsorize all flow measures at the 1st and 99th percentiles. All regressions include fund fixed effects. t-statistics are computed as in Table 1. Data are from (1) (2) (3) (4) (5) (6) (7) (8) FundTurn i,t (6.63) (5.08) (5.03) (5.96) FlowTurn1 i,t (1.11) (0.83) FlowTurn2 i,t (1.35) (0.72) FundTurn i,t 1 FlowTurn2 i,t (4.96) FlowV ol i,t (0.53) (0.16) Observations

29 Table A23: Counterpart of Paper Table 1 with Benchmark-Adjusted Turnover This table is the same as the paper s Table 1, except we replace FundTurn i,t 1 with benchmark-adjusted turnover, computed as FundTurn i,t 1 minus the median turnover computed across all index funds in the same Morningstar Category (e.g. Large Blend) as fund i and whose turnover is measured in the same time window as FundTurn i,t 1. Month Fixed Effects Fund Fixed Effects No Yes Yes (4.01) (3.88) No (1.06) (0.90) 29

30 Table A24: Turnover-Performance Relation with Rescaled Turnover Measure Column 1 reproduces the results with fund fixed effects from the paper s Table 1. Column 2 is the same but uses a re-scaled version of FundTurn i,t 1. Whereas FundTurn i,t 1 divides min(buys, Sells) by the fund s average size during the previous fiscal year, the re-scaled version divides by the fund s size at the beginning of the previous fiscal year. FundTurn i,t 1 and its re-scaled versions are both winsorized at the 1st and 99th percentiles. We lose some observations in the second column because of missing fund size. Both regressions include fund FEs. (1) (2) FundTurn i,t (6.63) Re-scaled FundTurn i,t (11.68) Observations 285, ,071 30

31 Table A25: Turnover-Performance Relation Controlling for Stock Holdings Column 1 reproduces the results with fund fixed effects from the paper s Table 1. Column 2 controls for Percent Stock, the percent of the fund s assets held in long stock positions. To remove extreme outliers, we set Percent Stock equal to 100% when it exceeds 100%; this change affects fewer than 1% of observations. Percent Stock is from Morningstar and is typically reported quarterly. We use the most recent portfolio report date before month t, but we do not look back more than 12 months. All regressions include fund FEs. (1) (2) FundTurn i,t (6.63) (6.83) Percent Stock (2.67) Observations 285, ,897 31

32 7. Alternative Benchmark Models Table A26 Counterpart of Paper s Table 1 with Estimated Morningstar Betas This table is the same as Table 1 in the main paper, except it replaces GrossR with Morningstar-adjusted returns with estimated betas. Specifically, the dependent variable equals the fund s gross return minus the product of Morningstar s benchmark return and the fund s estimated beta against that benchmark. If the fund has fewer than 24 monthly observations, we use the average estimated beta across all funds in the same Morningstar category. Month Fixed Effects Fund Fixed Effects No Yes Yes (7.67) (7.62) No (1.48) (1.26) 32

33 Table A27 Version of Table A27 with Betas Conditional on FundTurn This table is the same as the previous table, except we allow estimated betas to vary over time with lagged fund turnover. For each fund we regress the fund s excess gross return on the benchmark return and its interaction with lagged F undt urn. The dependent variable in this table equals the estimated intercept plus residual from those regressions. If the fund has fewer than 24 monthly observations, we use the average estimated slopes across all funds in the same Morningstar category. Month Fixed Effect Fund Fixed Effect No Yes Yes (7.49) (7.39) No (1.21) (1.01) 33

34 Table A28 Counterpart of Paper s Table 1 with Three-Factor Fama-French Benchmark This table is the same as Table 1 in the main paper, except it replaces GrossR with threefactor Fama-French adjusted gross returns. Specifically, the dependent variable equals the fund s gross return minus the risk-free rate minus the three Fama-French excess factor returns (MktRf, HML, SMB) times their respective estimated betas. Beta estimates are fundspecific. If the fund has fewer than 24 monthly observations, we use the average estimated betas across all funds in the same Morningstar category. Month Fixed Effect Fund Fixed Effect No Yes Yes (7.09) (8.27) No (1.27) (1.41) 34

35 Table A29 Version of Table A28 with Betas Conditional on FundTurn This table is the same as the previous table, except we allow estimated betas to vary over time with lagged fund turnover. For each fund we regress the fund s excess gross return on the benchmark returns and their interaction with lagged F undt urn. The dependent variable in this table equals the estimated intercept plus residual from those regressions. If the fund has fewer than 24 monthly observations, we use the average estimated slopes across all funds in the same Morningstar category. Month Fixed Effect Fund Fixed Effect No Yes Yes (5.28) (6.32) No (0.44) (0.55) 35

36 Table A30 Counterpart of Paper s Table 1 with Four-Factor Fama-French-Carhart Benchmark This table is the same as Table 1 in the main paper, except it replaces GrossR with fourfactor-adjusted gross returns. Specifically, the dependent variable equals the fund s gross return minus the risk-free rate, minus the three Fama-French excess factor returns (MktRf, HML, SMB) times their respective estimated betas, minus the momentum excess return times the fund s momentum beta. Beta estimates are fund-specific. If the fund has fewer than 24 monthly observations, we use the average estimated betas across all funds in the same Morningstar category. Month Fixed Effect Fund Fixed Effect No Yes Yes (7.89) (8.87) No (-0.01) (-0.13) 36

37 Table A31 Version of Table A30 with Betas Conditional on FundTurn This table is the same as the previous table, except we allow estimated betas to vary over time with lagged fund turnover. For each fund we regress the fund s excess gross return on the benchmark returns and their interaction with lagged F undt urn. The dependent variable in this table equals the estimated intercept plus residual from those regressions. If the fund has fewer than 24 monthly observations, we use the average estimated slopes across all funds in the same Morningstar category. Month Fixed Effect Fund Fixed Effect No Yes Yes (5.18) (5.92) No (-1.36) (-1.60) 37

38 Table A32 Counterpart of Paper s Table 1 with Five-Factor Fama-French Benchmark This table is the same as Table 1 in the main paper, except it replaces GrossR with fivefactor Fama-French adjusted gross returns. Specifically, the dependent variable equals the fund s gross return minus the risk-free rate minus the five Fama-French excess factor returns (MktRf, HML, SMB, RMV, CMA) times their respective estimated betas. Beta estimates are fund-specific. If the fund has fewer than 24 monthly observations, we use the average estimated betas across all funds in the same Morningstar category. Month Fixed Effect Fund Fixed Effect No Yes Yes (5.93) (7.17) No (3.15) (3.71) 38

39 Table A33 Version of Table A32 with Betas Conditional on FundTurn This table is the same as the previous table, except we allow estimated betas to vary over time with lagged fund turnover. For each fund we regress the fund s excess gross return on the benchmark returns and their interaction with lagged F undt urn. The dependent variable in this table equals the estimated intercept plus residual from those regressions. If the fund has fewer than 24 monthly observations, we use the average estimated slopes across all funds in the same Morningstar category. Month Fixed Effect Fund Fixed Effect No Yes Yes (5.07) (6.06) No (2.65) (3.16) 39

40 Table A34 Counterpart of Paper s Table 1 with Cremers-Petajisto-Zitzewitz Benchmark This table is the same as Table 1 in the main paper, except it replaces GrossR with gross returns adjusted using the modified Fama-French factors of Cremers, Petajisto, and Zitzewitz (2013). Specifically, the dependent variable equals the fund s gross return minus the risk-free rate minus the three excess factor returns (S5RF, R3VR3G, and R2S5) times their respective estimated betas. Beta estimates are fund-specific. If the fund has fewer than 24 monthly observations, we use the average estimated betas across all funds in the same Morningstar category. Month Fixed Effect Fund Fixed Effect No Yes Yes (8.84) (9.34) No (2.05) (2.05) 40

41 Table A35 Version of Table A34 with Betas Conditional on FundTurn This table is the same as the previous table, except we allow estimated betas to vary over time with lagged fund turnover. For each fund we regress the fund s excess gross return on the benchmark returns and their interaction with lagged F undt urn. The dependent variable in this table equals the estimated intercept plus residual from those regressions. If the fund has fewer than 24 monthly observations, we use the average estimated slopes across all funds in the same Morningstar category. Month Fixed Effect Fund Fixed Effect No Yes Yes (7.51) (7.93) No (1.07) (0.97) 41

42 8. Alternative Skill Proxies Table A36 Counterpart of Paper s Table 2 with Unadjusted Gross Alpha This table is the same as Table 2 in the paper, except it replaces the fund s expense ratio with the fund s unadjusted gross alpha, defined as the fund s full-sample average GrossR. This change mainly affects Panel D, but it also affects the other panels via the change in control variables. Panel A: Stock Size Categories Small Cap Mid Cap Large Cap Small - Large Controls No (5.87) (2.88) (4.72) (3.92) Yes (3.65) (1.68) (1.39) (2.76) Panel B: Stock Value-Growth Categories Growth Blend Value Growth Value Controls No (5.06) (5.45) (4.66) (0.54) Yes (1.55) (1.68) (1.75) (-0.15) Panel C: Fund Size Categories Small Medium Large Small Large Controls No (7.56) (3.74) (1.46) (4.11) Yes (3.49) (1.68) (0.81) (4.11) Panel D: Fund Unadjusted Gross Alpha Categories High Medium Low High Low Controls No (6.93) (4.35) (1.44) (4.78) Yes (2.77) (1.68) (1.00) (3.01) 42

43 Table A37 Counterpart of Paper s Table 2 with Adjusted Gross Alpha This table is the same as Table 2 in the paper, except it replaces the fund s expense ratio with the fund s adjusted gross alpha, the measure of fund skill from Pastor, Stambaugh, and Taylor (2015). This change mainly affects Panel D, but it also affects the other panels via the change in control variables. Panel A: Stock Size Categories Small Cap Mid Cap Large Cap Small - Large Controls No (5.87) (2.88) (4.72) (3.92) Yes (3.28) (1.73) (2.21) (1.97) Panel B: Stock Value-Growth Categories Growth Blend Value Growth Value Controls No (5.06) (5.45) (4.66) (0.54) Yes (1.13) (1.73) (2.11) (-1.08) Panel C: Fund Size Categories Small Medium Large Small Large Controls No (7.56) (3.74) (1.46) (4.11) Yes (3.61) (1.73) (0.85) (3.50) Panel D: Fund Adjusted Gross Alpha Categories High Medium Low High Low Controls No (6.90) (5.03) (-0.29) (5.79) Yes (2.67) (1.73) (-0.52) (4.17) 43

44 Table A38 Manager Age, Tenure, and the Turnover-Performance Relation The dependent variable is GrossR it, fund i s benchmark-adjusted gross return in month t. Column 1 matches the top-left cell in Table 1 in the paper. Avg. Mgr. Tenure is the average number of years managing the fund, computed across the fund s managers in the given month. Avg. Mgr. Age is the average age across fund s managers in the given month. Manager tenure and age are highly correlated within funds, so we avoid including both in the same regression. All regressions include fund fixed effects and cluster by Sector*month. Data on fund managers are from Morningstar. (1) (2) (3) (4) (5) FundTurn(i,t-1) (6.63) (6.02) (7.43) (5.47) (4.82) Avg. Mgr. Tenure (-5.49) (-1.40) FundTurn(i,t-1)*Avg.Mgr.Tenure (-4.53) Avg. Mgr. Age (-2.72) (0.37) FundTurn(i,t-1)*Avg.Mgr.Age (-3.40) Observations

45 Table A39 Counterpart of Paper s Table 2 with Fund-Size and Expense-Ratio Terciles Computed Within Style-Months This table is the same as Table 2 in the paper, except instead of computing fund size and expense ratio terciles within months, it computes them within style-months. Style here is Morningstar s Category, e.g., Large-Cap Growth. This change mainly affects Panels C and D, but it also affects Panels A and B via the change in control variables. Panel A: Stock Size Categories Small Cap Mid Cap Large Cap Small - Large Controls No (5.87) (2.88) (4.72) (3.92) Yes (4.00) (1.42) (0.69) (3.57) Panel B: Stock Value-Growth Categories Growth Blend Value Growth Value Controls No (5.06) (5.45) (4.66) (0.54) Yes (1.49) (1.42) (1.59) (0.11) Panel C: Fund Size Categories (Within Style-Month) Small Medium Large Small Large Controls No (7.10) (3.81) (2.04) (3.32) Yes (2.83) (1.42) (1.03) (2.64) Panel D: Fund Expense Ratio Categories (Within Style-Month) High Medium Low High Low Controls No (7.62) (3.58) (3.37) (3.48) Yes (2.67) (1.42) (1.29) (2.25) 45

46 Table A40: Comparing Broker-Sold and Direct-Sold Funds This table compares the turnover-performance slope across direct-sold and broker-sold funds in our sample. This table is the same as Table 2 Panel D in our paper, except we replace expense-ratio terciles with three distribution-channel categories: direct-sold, broker-sold, and not sure. Following Sun (2014), we say that a share class is broker-sold if it has a non-zero front load, a non-zero back load, or a 12b-1 fee exceeding 25 bps; otherwise, the share class is direct sold. Data on loads and 12b-1 fees are from Morningstar. Similar to Del Guercio and Reuter (2014), we classify a fund as broker-sold (direct-sold) if at least 75% of its assets are broker-sold (direct-sold) on average over time. If 25 75% of the fund s assets are broker-sold on average, we categorize the fund s distribution channel as not sure. In our full sample, 32% of fund-month observations are for broker-sold funds, 59% are for direct-sold funds, and 9% are not sure. Direct-Sold Not Sure Broker-Sold Direct - Broker Controls No (6.28) (2.83) (4.07) (1.43) Yes (1.63) (1.17) (1.80) (-0.35) 46

47 9. Additional Results Table A41 Summary Statistics Variable N Mean Stdev P1 P25 P50 P75 P99 Fund Return 314, Benchmark-adjusted return 314, Expense ratio 314, GrossR 314, FundTurn 285, Sentiment 296, Volatility 314, Liquidity 314, Business Cycle 314, Lagged Mkt. Return 314,

48 Table A42 Counterpart of Paper s Table 1 with Benchmark-Adjusted Net Returns This table is the same as Table 1 in the main paper, except we replace benchmark-adjusted gross fund returns (GrossR) with benchmark-adjusted net fund returns. Month Fixed Effects Fund Fixed Effects No Yes Yes (6.45) (6.56) No (1.48) (1.14) 48

49 Table A43 Counterpart of Paper s Table 1 with Annual Data This table is the same as Table 1 in the main paper, except for four changes: the unit of observation is the fund-fiscal year; GrossR is the fund s annual benchmark-adjusted gross return; we replace month FEs with fiscal year FEs; and we cluster by Morningstar sector fiscal year. There are 16,629 observations in each regression. Fiscal Year Fixed Effects Fund Fixed Effects No Yes Yes (5.56) (4.95) No (2.76) (2.53) 49

50 Table A44 Interacting Turnover with Time Since Turnover The dependent variable is GrossR i,t. Columns 1 and 3 match the bottom row of Table 1 in the main paper. NmonthsSince i,t is the number of months elapsed between the end of 12-month period used to measure FundTurn i,t 1 and the end of month t. Its value ranges from 1 to 12. (1) (2) (3) (4) FundTurn i,t (6.63) (5.03) (6.77) (5.71) FundTurn i,t 1 NmonthsSince i,t (-1.25) (-1.24) Observations Fund FEs Yes Yes Yes Yes Month FEs No No Yes Yes 50

51 Table A45 Additional Lags of Turnover The dependent variable is GrossR i,t. FundTurn i,t 1 is fund turnover in the previous fiscal year, FundTurn i,t 2 is fund turnover two fiscal years ago, and FundTurn i,t 3 is fund turnover three fiscal years ago. The regression includes fund fixed effects. t-statistics are computed as in Table 1. Data are from FundTurn i,t (3.84) FundTurn i,t (-0.89) FundTurn i,t (0.39) Observations

52 Table A46 Counterpart of Paper s Table 1 in Cold-IPO-Market Subperiod ( ) This table is the same as Table 1 in the main paper, but uses data from Month Fixed Effects Fund Fixed Effects No Yes Yes (3.21) (3.23) No (-1.95) (-2.67) 52

53 Table 47: Economic Significance This table computes the economic significance of the turnover-performance slopes. To conform as closely as possible to our turnover-performance regressions, we compute the withinfund standard deviation of each turnover variable using monthly data, and we require that GrossR is non-missing. The last column indicates the source of the estimated slope coefficient. All slopes are from regressions of GrossR it on lagged turnover. Within-fund Economic significance Variable Stdev. Slope =Stdev*Slope*1200 Source of slope FundTurn i,t Table 7 column Table 7 column 6 AvgTurn i,t Table 7 column Table 7 column 6 AvgTurnSim i,t Table 7 column Table 7 column 6 53

54 De meaned GrossR(i,t) De meaned FundTurn(i,t 1) 95% CI lpoly smooth: Demeaned GrossR(i,t) Figure A2. Nonlinearities in the Turnover-Performance Relation? We perform a kernel-weighted local polynomial regression of GrossR i,t on FundTurn i,t 1, both de-meaned at the fund level. We plot the smoothed values with 95% confidence bands. We use the ruleof-thumb bandwidth, epanechnikov kernel, and local-mean smoothing. We drop the bottom 1% of fund-demeaned FundTurn observations (values less than -106%) and the top 1% (values greater than 155%). 54