Staying the Course Mutual Fund Investment Style Consistency

Staying the Course: Mutual Fund Investment Style Consistency and Performance Persistence Keith C. Brown The University of Texas W. Van Harlow Fidelity Investments Federal Reserve Bank of Atlanta Financial Markets Conference April 15, 2004 0

Research Premise Does Investment Style Consistency Impact Performance? Lower Style Consistency Cap: Small to Large (%) Higher Style Consistency Value to Growth (%) 1

Why Style Consistency. Might Matter n Fund Outflows Due to Style Drift Ø n Higher Consistency = Lower Turnover? Ø n Possibility of Lower Transaction Costs and Expense Ratios Style Timing Might be a “Loser’s Game” Ø n Inability of Plan Sponsors to Identify Manager’s Style Analog to Difficulty of Successful Tactical Asset Allocation Style Consistency as a Possible “Signal” of Superior Manager Performance 2

Simple Evidence Peer Group Large Value Large Blend Large Growth Mid Value Mid Blend Mid Growth Small Value Small Blend Small Growth Style Consistency Lower Higher Lower Higher Lower Higher Average Annual Return (1991 -2000) Higher Returns for More Style Consistent Funds 11. 10% 13. 05% 16. 69% 20. 04% 18. 55% 19. 86% 17. 30% 13. 58% 12. 95% 12. 86% 13. 90% 15. 44% 15. 83% 16. 65% 14. 28% 15. 62% 12. 78% 14. 21% 3

Complicating Factors Peer Group Large Value Large Blend Large Growth Mid Value Mid Blend Mid Growth Small Value Small Blend Small Growth Median Annual Higher Returns Style Median Fund Return for More Consistency (1991 -2000) Style Consistent Funds Turnover Expense Ratio Lower Higher Lower Higher Lower Higher 11. 10% 13. 05% 16. 69% 20. 04% 18. 55% 19. 86% 17. 30% 13. 58% 12. 95% 12. 86% 13. 90% 15. 44% 15. 83% 16. 65% 14. 28% 15. 62% 12. 78% 14. 21% 47. 50% 45. 50% 77. 00% 38. 00% 60. 50% 63. 00% 60. 00% 63. 00% 39. 59% 115. 00% 76. 00% 50. 00% 44. 82% 84. 50% 47. 00% 89. 00% 78. 00% 1. 22% 1. 02% 1. 25% 0. 93% 1. 36% 1. 07% 1. 40% 1. 16% 1. 41% 1. 23% 1. 40% 1. 29% 1. 39% 1. 15% 1. 50% 1. 12% 1. 46% 1. 33% 4

Complicating Factors Peer Group Large Value Large Blend Large Growth Mid Value Mid Blend Mid Growth Small Value Small Blend Small Growth Median Annual Higher Returns Style Median Fund Return for More Consistency (1991 -2000) Style Consistent Funds Turnover Expense Ratio Lower Higher Lower Higher Lower Higher 11. 10% 13. 05% 16. 69% 20. 04% 18. 55% 19. 86% 17. 30% 13. 58% 12. 95% 12. 86% 13. 90% 15. 44% 15. 83% 16. 65% 14. 28% 15. 62% 12. 78% 14. 21% 47. 50% 45. 50% 77. 00% 38. 00% 60. 50% 63. 00% 60. 00% 63. 00% 39. 59% 115. 00% 76. 00% 50. 00% 44. 82% 84. 50% 47. 00% 89. 00% 78. 00% 1. 22% 1. 02% 1. 25% 0. 93% 1. 36% 1. 07% 1. 40% 1. 16% 1. 41% 1. 23% 1. 40% 1. 29% 1. 39% 1. 15% 1. 50% 1. 12% 1. 46% 1. 33% 5

Past Literature n Investment Style Appears to Matter Ø Ø Ø n Fund Objectives : Mc. Donald (JFQA, 1974); Malkiel (JF, 1995) Security Characteristics : Basu (JF, 1977); Banz (JFE, 1981); Fama and French (JF, 1992; JFE, 1993) Style Premiums : Capaul, Rawley, Sharpe (FAJ, 1993); Lakonishok, Shleifer, Vishny (JF, 1994); Fama and French (JF, 1998); Chan and Lakonishok (FAJ, 2004); Phalippou (Working Paper, 2004) Style Definitions : Roll (HES, 1995); Brown and Goetzmann (JFE, 1997) Style Rotation : Barberis and Shleifer (JFE, 2003) Fund Performance Persistence Ø Ø Ø Classic Study: Jensen (JF, 1968) Hot & Icy Hands : Grinblatt and Titman (JF, 1992); Hendricks, Patel, Zeckhauser (JF, 1993); Brown and Goetzmann (JF, 1995); Elton, Gruber, Blake (JB, 1996), Ibbotson and Patel (Working Paper, 2002) Accounting for Momentum : Jegadeesh and Titman (JF, 1993); Carhart (JF, 1997); Wermers (2001) Conditioning Information : Ferson and Schadt (JF, 1996), Christopherson, Ferson, and Glassman (RFS, 1998) Persistence & Style : Bogle (JPM, 1998); Teo and Woo (JFE, forthcoming) 6

Research Design Does Style Consistency Impact Performance? n Use alternative definitions of style consistency n Control for other factors affecting performance Ø Alpha persistence Ø Expense ratio Ø Turnover Ø Fund size Ø Active/passive management 7

Measuring Investment Style & Style Consistency: Two Approaches n Holdings-Based Measures: Daniel, Grinblatt, Titman, and Wermers (JF, 1997) Ø Ø n Pros: Direct Assessment of Manager’s Selection and Timing Skills; Benchmark Construction Around Security Characteristics Cons: Unobservable or Observed with Considerable Lag; “Window Dressing” Problems Returns-Based Measures: Ø Ø Sharpe (JPM, 1992) Pros: Direct Observation of “Bottom Line” to Investor; Measured More Frequently and Over Shorter Time Intervals than Holdings Cons: Indirect Measure of Managerial Decision-Making 8

Returns-Based Measures of Investment Style Consistency n Model Based : Ø Define a style factor model: K Rjt = [ bj 0 + Σbjk. Fkt ]+ ejt K=1 [1 – R 2] represents portion of return not related to style N n Benchmark Based : Δjt Ø = Σ xji Rjit - Rbt = Rjt - Rbt i=1 Active Net Returns: TE = σΔ√P where P is the return periods per year 9

Testable Hypotheses n Hypothesis #1 : Style-consistent (i. e. , high R 2, low TE) funds have lower portfolio turnover than style-inconsistent (i. e. , low R 2, high TE) funds. n Hypothesis #2 : Style-consistent funds have higher total and relative returns than style-inconsistent funds. n Hypothesis #3 : There is a positive correlation between the consistency of a fund’s investment style and the persistence of its future performance 10

Data n Survivorship-bias free database of monthly returns for domestic diversified equity funds for the period 1988 -2000 n Morningstar style classifications (large-, mid-, small-cap; value, blend, growth) n Mutual Fund characteristics for the period 1991 -2000 (e. g. , expense ratio, turnover, total net assets) n Require three years of prior monthly returns to be included in the analysis on any given date n No sector funds; analyze with and without index funds (i. e. , active vs. passive management) 11

Number of Funds with Three Years of Returns (Table 1) Year 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 Large Value 135 140 156 169 215 273 350 410 504 564 Mid Mid Small Large Blend Growth Value 163 172 184 203 245 314 382 446 584 729 118 120 126 139 178 233 297 355 425 549 60 60 65 67 69 87 102 127 167 199 47 49 54 54 62 71 99 104 125 138 79 78 78 82 106 150 183 221 289 333 25 28 31 38 47 62 79 97 121 162 Small Blend Growth 29 30 30 37 52 71 97 123 147 194 42 44 49 59 78 113 152 206 262 309 12

Average Fund Characteristics: 1991 -2000 (Table 2) Average Turnover Average Expense Ratio Average Fund Firm Size ($mm) Large Value 67. 57% 1. 38% 25, 298 Large Blend 69. 14% 1. 22% 44, 611 Large Growth 92. 93% 1. 45% 45, 381 Mid Value 84. 73% 1. 43% 5, 731 Mid Blend 79. 39% 1. 45% 6, 782 Mid Growth 132. 96% 1. 55% 4, 917 Small Value 61. 43% 1. 48% 643 Small Blend 82. 17% 1. 50% 1, 283 119. 89% 1. 64% 1, 057 Peer Group Small Growth 13

Methodology n n Use two alternative returns-based definitions of style consistency Ø Goodness-of-fit from a multivariate factor model (i. e. , R 2) Ø Tracking error relative to peer-group specific benchmarks Evaluate the impact of style consistency on performance by using a tournament-based methodology (Brown, Harlow, Starks (JF, 1996)) Ø Relative performance within a peer group is the focus Ø Avoids the usual model specification issues Ø Controls for cross-sectional differences in consistency measures 14

Methodology n Multivariate Performance Attribution Model Rt = a + b 1 R 1 t + b 2 R 2 t +. . . + b. N RNt + et , where a = the risk-adjusted excess return (alpha); Rt = the excess return of a fund in month t; . . =. 1 … N); Rkt = the excess return of factor k in month t (k b k = the beta of factor k (k = 1. . …. N); et = the tracking error in month t; n Factor Models Ø EGB Four Factor - Elton, Gruber and Blake (JB, 1996) Ø Modified EGB with Five Factors (adding EAFE factor) Ø FF Three Factors - Fama and French (1993) Ø FFC Four Factors - Carhart (1997) n Use R 2 and alpha from the model 15

Methodology (Figure 1) Examples from Multivariate Factor Model R 2 = 0. 78 Cap: Small to Large (%) R 2 = 0. 92 Value to Growth (%) 16

Methodology (Table 3) 17

Methodology Estimate Model Evaluate Tournament Performance Time 36 Months 3 Months (12 Months) n Use past 36 months of data to estimate model parameters n Evaluate performance in tournament n Ø Standardized returns within each peer group on a give date to allow for time-series and cross-sectional pooling Ø Peer rankings Ø Above median performance Roll the process forward one quarter (one year) and estimate all parameters again, etc. 18

Univariate Analysis (Table 4) Correlation with R² FFC Four-Factor Model (1991 -2000) Period 1991 -2000 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 Fund Turnover -0. 216 (0. 000) -0. 185 (0. 000) -0. 246 (0. 000) -0. 195 (0. 000) -0. 260 (0. 000) -0. 277 (0. 000) -0. 240 (0. 000) -0. 180 (0. 000) -0. 166 (0. 000) -0. 246 (0. 000) -0. 233 (0. 000) Fund Expense Ratio -0. 318 (0. 000) -0. 254 (0. 000) -0. 305 (0. 000) -0. 330 (0. 000) -0. 410 (0. 000) -0. 369 (0. 000) -0. 394 (0. 000) -0. 345 (0. 000) -0. 329 (0. 000) -0. 313 (0. 000) -0. 250 (0. 000) Actual Fund Return 0. 029 (0. 000) 0. 034 (0. 411) 0. 108 (0. 006) -0. 058 (0. 128) 0. 159 (0. 000) 0. 240 (0. 000) 0. 291 (0. 000) 0. 265 (0. 000) 0. 089 (0. 000) -0. 088 (0. 000) 0. 044 (0. 030) Tournament Fund Return 0. 110 (0. 000) 0. 031 (0. 449) 0. 110 (0. 006) -0. 054 (0. 160) 0. 170 (0. 000) 0. 278 (0. 000) 0. 301 (0. 000) 0. 329 (0. 000) 0. 147 (0. 000) -0. 082 (0. 000) 0. 035 (0. 083) Tournament Return Ranking 0. 092 (0. 000) 0. 057 (0. 170) 0. 094 (0. 018) -0. 031 (0. 417) 0. 077 (0. 037) 0. 236 (0. 000) 0. 241 (0. 000) 0. 240 (0. 000) 0. 141 (0. 000) -0. 043 (0. 058) 0. 025 (0. 217) 19

Multivariate Analysis (Table 5 A) 3 -Month Future Returns (1991 -2000) Variable Intercept FF Three-Factor Model Parameter Prob Estimate FFC Four-Factor Model Parameter Prob Estimate 0. 000 1. 000 0. 058 0. 000 0. 011 Consistency (R²) Turnover Expense Ratio 0. 034 0. 032 (0. 068) 0. 000 0. 033 (0. 082) 0. 000 Assets (0. 011) 0. 012 (0. 008) 0. 093 Alpha 20

Multivariate Analysis (Table 5 B) 12 -Month Future Returns (1991 -2000) Variable Intercept FF Three-Factor Model Parameter Prob Estimate FFC Four-Factor Model Parameter Prob Estimate 0. 000 1. 000 0. 060 0. 000 0. 038 0. 000 Consistency (R²) Turnover Expense Ratio 0. 081 0. 060 (0. 134) 0. 000 0. 077 0. 062 (0. 145) 0. 000 Assets (0. 021) 0. 022 (0. 019) 0. 038 Alpha 21

Fama-Mac. Beth Cross-Sectional Analysis n Use past 36 months of data to estimate model parameters n Run a sequence of cross-sectional regressions of future performance against fund characteristics and model parameters (alpha and R 2 ) n Average the coefficient estimates from regressions across the entire sample period n T-statistics based on the time-series means of the coefficients 22

Fama-Mac. Beth. Cross-Sectional Analysis (Table 6) 3 -Month Future Returns (1991 -2000) Variable Alpha Consistency (R²) Turnover Expense Ratio Assets FF Three-Factor Model Parameter Prob Estimate 0. 087 0. 000 FFC Four-Factor Model Parameter Prob Estimate 0. 040 0. 029 0. 067 0. 000 0. 068 0. 000 0. 001 (0. 099) 0. 018 0. 970 0. 000 0. 030 23

Multivariate Analysis (Table 7) Summary of Style Consistency Parameters for Individual Style Groups (12 -Month Future Returns) + *** + + *** +* + + ** + _ + *** Note: Significant at the * 10% level; ** 5% level; *** 1% level 24

Logit Analysis for Above-Median Performance (Table 8) 12 -Month Future Returns FFC Four-Factor Model (1991 -2000) Variable Intercept Alpha Consistency FF Three-Factor Model Parameter Estimate Prob 0. 005 0. 788 0. 048 0. 029 0. 115 0. 000 FFC Four-Factor Model Parameter Prob Estimate 0. 004 0. 043 0. 115 0. 821 0. 039 0. 000 0. 304 0. 064 Turnover Expense Ratio 0. 093 (0. 194) 0. 000 Assets (0. 022) 0. 257 0. 098 (0. 200) (0. 020) 0. 008 0. 548 0. 024 Consistency*Alpha 25

Logit Analysis for Above-Median Performance (Table 9 A) Probability Implications for the FFC Four-Factor Model Assuming average characteristics for expense ratio, turnover and assets (1991 -2000) Consistency (RSQ): Standard Deviation Group ALPHA: -2 (Low) -1 0 +1 +2 (High) (High – Low) -2 (Low) 0. 4467 0. 4631 0. 4796 0. 4962 0. 5127 0. 0660 -1 0. 4453 0. 4678 0. 490 3 0. 5129 0. 5355 0. 0902 0 0. 4440 0. 4725 0. 5010 0. 5296 0. 5580 0. 1140 +1 0. 4427 0. 4771 0. 5118 0. 5463 0. 5804 0. 1377 +2 (High) 0. 4414 0. 4818 0. 5225 0. 5628 0. 6024 0. 1610 (High – Low) -0. 0053 0. 0187 0. 0429 0. 0666 0. 0897 26

Logit Analysis for Above-Median Performance (Table 9 B) Probability Implications for the FFC Four-Factor Model Assuming average characteristics turnover and assets but – 2 std for expense ratio (1991 -2000) Consistency (RSQ): Standard Deviation Group ALPHA: -2 (Low) -1 0 +1 +2 (High) (High – Low) -2 (Low) 0. 5464 0. 5628 0. 5790 0. 5951 0. 6110 0. 0646 -1 0. 5451 0. 5674 0. 5895 0. 6111 0. 6324 0. 0873 0 0. 5438 0. 5720 0. 5998 0. 6269 0. 6533 0. 1095 +1 0. 5425 0. 5766 0. 6100 0. 6425 0. 6736 0. 1312 +2 (High) 0. 5412 0. 5812 0. 6202 0. 6577 0. 6933 0. 1522 (High – Low) -0. 0053 0. 0184 0. 0412 0. 0626 0. 0824 27

Active versus Passive Multivariate Analysis Three-Month Future Returns (1991 -2000) Variable Intercept Alpha Consistency (R²) All Funds Parameter Prob Estimate 0. 000 1. 000 Excluding Index Funds Parameter Prob Estimate 0. 000 1. 000 0. 011 0. 030 0. 011 0. 000 0. 012 0. 030 0. 010 0. 000 Turnover Expense Ratio 0. 033 (0. 082) 0. 000 0. 034 (0. 080) 0. 000 Assets (0. 008) 0. 093 (0. 007) 0. 124 28

Alternative Consistency Measure Tracking Error as a Measure of Style Consistency n Analysis using tracking error produces virtually identical results R 1000 V R 1000 G RMid. V RMid. G R 2000 V R 2000 G 29

Trading Strategies Returns of Low and High Expense Ratio Quintiles (1991 -2000) 5. 00 4. 50 Lo EXPR = 15. 58% Hi EXPR = 13. 44% 3. 50 Hi EXPR Annual Return Difference = 2. 14% 3. 00 2. 50 2. 00 1. 50 06 00 20 06 12 99 19 98 19 97 06 12 19 96 12 19 95 19 12 06 95 19 94 19 12 06 94 19 93 19 06 12 93 19 92 19 12 06 92 19 91 06 19 91 19 90 12 1. 00 19 Growth of a $1 4. 00 Date 30

Trading Strategies (Figure 2 A) Style Consistency Implications for Returns of Low and High Expense Ratio Quintiles (1991 -2000) 5. 00 Hi RSQ: Lo EXPR 4. 50 Lo EXPR Hi RSQ: Lo EXPR = 15. 79% Lo RSQ: Hi EXPR = 13. 10% 3. 50 Hi EXPR Lo RSQ: Hi EXPR Annual Return Difference = 2. 69% 3. 00 2. 50 2. 00 “Consistency Premium” = 0. 55% 1. 50 12 06 06 00 20 99 19 06 12 98 19 06 12 97 19 19 97 12 06 96 19 06 12 12 95 19 94 12 06 06 19 94 19 93 19 12 06 12 93 19 92 19 19 91 06 91 19 90 12 1. 00 19 Growth of a $1 4. 00 Date 31

Trading Strategies Returns of Low and High Expense Ratio and Alpha Quintiles (1991 -2000) 5. 00 4. 50 Hi EXPR: Lo ALPHA = 11. 64% 3. 50 Annual Return Difference = 3. 94% 3. 00 Hi EXPR: Lo ALPHA 2. 50 2. 00 1. 50 12 06 06 00 20 99 19 06 12 98 19 06 12 97 19 19 97 12 06 96 19 06 12 12 95 19 94 12 06 06 19 94 19 93 19 12 06 12 93 19 92 19 19 91 06 91 19 90 12 1. 00 19 Growth of a $1 Lo EXPR: Hi ALPHA = 15. 58% 4. 00 Date 32

Trading Strategies (Figure 2 B) Style Consistency Implications for Returns of Low and High Expense Ratio and Alpha Quintiles (1991 -2000) 5. 00 Hi RSQ: Lo EXPR: Hi ALPHA 4. 50 Lo RSQ: Hi EXPR: Lo ALPHA = 10. 14% 3. 50 Annual Return Difference = 5. 94% 3. 00 Hi EXPR: Lo ALPHA 2. 50 Lo RSQ: Hi EXPR: Lo ALPHA 2. 00 1. 50 “Consistency Premium” = 2. 00% 12 06 06 00 20 99 19 98 19 97 19 06 12 96 19 06 12 95 19 12 94 06 19 94 19 06 12 12 93 19 92 19 91 06 19 91 19 90 12 1. 00 19 Growth of a $1 Lo EXPR: Hi ALPHA Hi RSQ: Lo EXPR: Hi ALPHA = 16. 08% 4. 00 Date 33

Consistency Premiums by Style Groups 3. 07 % 0. 85 % 1. 89 % 2. 40 % 0. 54 % 0. 19 % (1. 80 %) 7. 16 % 4. 60 % 34

Conclusion n Funds with more style consistency within a peer group tend to have better performance, ceteris paribus, during the sample period Ø Findings robust with respect to two alternative definitions of consistency (and four factor models for one definition of consistency) Ø Results are not related to active/passive management issues n Style consistency effect appears to be separate from past alpha and expense ratios in explaining future performance n Results are robust within sample period and across fund types n Although not reported, analysis of performance back to 1981 (not entirely survivorship-bias free) produces identical results to the 1991 -2000 analysis 35

Extensions and Implications n n Need to Extend Analysis through 2003 : Same Behavior in “Down” Markets? Consistency as a “Signal” of Persistence : Easier to Identify Good Managers? Consistency and Governance : Manager Evaluation Relative to Peer Group; Manager Compensation; Single vs. Team-Managed Funds Consistency and Regulation : Easier to Assess Whether Fund Prospectus Objectives and Constraints are Satisfied? 36