SCALE AND SKILL IN ACTIVE MANAGEMENT Ľuboš Pástor University of Chicago, NBER, CEPR National Bank of Slovakia Robert F. Stambaugh University of Pennsylvania, NBER Lucian A. Taylor University of Pennsylvania Journal of Financial Economics, forthcoming in 2015
Motivation Fund performance depends on skill as well as scale To learn about skill, we must understand scale Nature of returns to scale in active fund management? Fund level? Fund size This fund s performance Perold and Solomon (1991), Berk and Green (2004) Evidence: Chen et al. (2004), Bris et al. (2007), Yan (2008), Ferreira et al. (2013), Reuter and Zitzewitz (2013) Industry level? Industry size All funds performance Pástor and Stambaugh (2012) Evidence:?
Main Results Scale: Strong evidence of decreasing returns to scale at industry level Stronger for high-turnover, high-volatility, and small-cap funds Mixed evidence of decreasing returns to scale at fund level Insignificant after removing econometric biases Skill: Active funds have become more skilled over time Yet their performance has not improved Negative age-performance relation A fund s performance decreases over its lifetime Younger funds outperform older funds
Narrative New funds tend to be more skilled than existing funds Education? Technology? Given their better skill, new funds tend to outperform initially As these funds grow older, their performance suffers Because industry keeps growing ( more skilled competition)
Methodology Three methods for estimating fund-level returns to scale: 1. Pooled OLS: R it = a + βq it 1 + ε it Biased: omitted variable (skill) 2. OLS with fund fixed effects: R it = a i + βq it 1 + ε it Biased: Corr(q it,ε it ) > 0 3. Recursive demeaning: new procedure Unbiased
Details of Bias in OLS FE OLS FE is same as first demeaning R it, q it 1 at fund level, and then running OLS: R it = R it 1 T i T i s=1 R is q it 1 = q it 1 1 T i T i s=1 q is 1, R it = β q it 1 + ε it Problem: ε it, q it 1 are likely negatively correlated (via q it ) β FE β = t,i q2 i,t 1 1 t,i q i,t 1 ε it < 0, in expectation
Details of Recursive Demeaning Model: R it = a i + βq it 1 + ε it Step 1: Recursively forward-demean all variables. For example, q it 1 = q it 1 1 T i t + 1 T i s=t q is 1 Model becomes R it = βq it 1 + ε it. Estimate using IV. Step 2: Instrument for q it 1 by recursively backward-demeaned size: q it 1 = q it 1 1 t 1 t 1 q is 1 s=1 q it 1 is likely correlated with q it 1 but not with ε it IV s relevance & exclusion conditions should hold
Simulation Exercise Goal: Illustrate the bias in OLS estimators, lack of bias in RD Step 1: Simulate data where we know true β: R it = a i + βq it 1 + ε it q it 1 = c + γr q it + v it it 1 Step 2: Using simulated data, obtain β OLS, β FE, and β RD Repeat 10,000 times
Simulation Results Mean estimated β β OLS, no FE OLS with FE RD 0 0.84-0.38 0.00-1 0.37-1.73-0.87-3 -0.78-4.28-3.00-10 -7.38-10.82-10.02 Fraction reject the null (β = 0) at 5% level β OLS, no FE OLS with FE RD 0 1.00 0.99 0.06-1 0.90 1.00 0.14-3 0.99 1.00 0.20-10 1.00 1.00 1.00
Sample Data: CRSP and Morningstar, 1979 2011 Check accuracy across databases (return, size, expense ratio) Only domestic active equity mutual funds with size $15 million Final sample: 350,000 monthly observations of 3,126 funds Main sample: 1993 2011 Extended sample: 1979 2011 Noisier data but very similar results, same conclusions
Main Variables GrossR: Fund return gross of fees, minus benchmark return E.g., for Large Growth, benchmark is Russell 1000 Growth Index FundSize = Fund s AUM today Total mkt.cap. today Total mkt.cap. in Dec. 2011 IndustrySize = Funds total AUM today Total mkt.cap. today
Sample Size Over Time 2000 1800 Has return Also has exp. ratio and benchmark data Also has FundSize 1600 1400 Number of funds 1200 1000 800 600 400 200 0 Jan 1980 Jan 1990 Jan 2000 Jan 2010 Main sample: March 1993 December 2011 Extended sample: January 1979 December 2011
Industry Size over Time 0.2 0.18 0.16 IndustrySize (fraction of CRSP) 0.14 0.12 0.1 0.08 0.06 0.04 0.02 0 Jan 1980 Jan 1990 Jan 2000 Jan 2010
Decreasing Returns to Scale at Fund Level? Dependent variable: GrossR FundSize -0.0137 (-1.87) Constant 0.000503 (2.18) Observations 275847 Estimator OLS no FE
Decreasing Returns to Scale at Fund Level? Dependent variable: GrossR FundSize -0.0137-0.168 (-1.87) (-9.38) Constant 0.000503 (2.18) Observations 275847 275847 Estimator OLS no FE OLS FE
Decreasing Returns to Scale at Fund Level? Dependent variable: GrossR FundSize -0.0137-0.168-0.220 (-1.87) (-9.38) (-0.62) Constant 0.000503 (2.18) Observations 275847 275847 270556 Estimator OLS no FE OLS FE RD
Decreasing Returns to Scale at Industry Level? Dependent variable: GrossR IndustrySize -0.0169-0.0326-0.0326 (-1.93) (-3.60) (-2.49) Constant 0.00304 (2.18) Observations 283046 283046 283046 Estimator OLS no FE OLS FE RD
Fund- vs. Industry-level Returns to Scale Dependent variable: GrossR FundSize -0.0147-0.148-0.425 (-2.02) (-9.09) (-1.25) IndustrySize -0.0165-0.0295-0.0277 (-1.90) (-3.27) (-2.14) Constant 0.00300 (2.09) Observations 275847 275847 270556 Estimator OLS no FE OLS FE RD
Industry Size: Just a Time Trend? Dependent variable: GrossR IndustrySize -0.0326 (-3.60) Time Trend -10.26 (-2.99) Observations 283046 283046
Industry Size: Just a Time Trend? Dependent variable: GrossR IndustrySize -0.0326-0.0852 (-3.60) (-3.04) Time Trend -10.26 23.89 (-2.99) (2.21) Observations 283046 283046 283046
A Closer Look at Industry Size Dependent variable: GrossR Average Fund Size -3.862-8.885 (-3.03) (-3.56) Number of Funds 0.450-4.031 (0.83) (-3.23) Observations 283046 283046 283046
A Closer Look at Industry Size Dependent variable: GrossR IndustrySize -0.115 (-2.60) Average Fund Size -3.862-8.885 4.315 (-3.03) (-3.56) (0.73) Number of Funds 0.450-4.031 8.493 (0.83) (-3.23) (1.61) Observations 283046 283046 283046 283046
Determinants of the Size-Performance Relation Dependent variable: GrossR (1) (2) (3) (4) (5) (6) (7) (8) FundSize -0.0987 0.0228-0.316 0.271 0.318 (-0.66) (0.03) (-0.30) (0.42) (0.49) FundSize*1(SmlCap) 0.273-1.402-0.959 (0.13) (-0.70) (-0.49) FundSize*Std(AbnRet) -10.40-29.83-30.19 (-0.28) (-0.94) (-0.94) FundSize*Turnover 0.207 0.0588 0.0360 (0.21) (0.20) (0.12) IndustrySize -0.0120 0.0248 0.00541 0.0450 0.0194 (-3.04) (2.92) (1.11) (2.35) (0.68) IndustrySize*1(SmlCap) -0.0348-0.0340-0.0360 (-2.67) (-1.33) (-1.41) IndustrySize*Std(AbnRet) -2.137-2.013-2.010 (-4.51) (-2.19) (-2.19) IndustrySize*Turnover -0.0287-0.0250-0.0249 (-4.45) (-2.57) (-2.56) Fund age 0.000151 (1.23)
Estimating Skill Our measure of skill: Gross alpha when F undsize = IndustrySize = 0 (Average benchmark-adjusted return on the fund s first dollar invested, with no other funds in the industry) We measure fund skill by a i in GrossR it = a i + b i FundSize it 1 + c i IndustrySize it 1 + ε it We model the slopes as b i = β 0 + β 1 X i and c i = γ 0 + γ 1 X i, where X i includes all fund characteristics from previous table
Distribution of Fund Skill over Time 1 90th pctl. Fund FE (% per month) 0.5 0 75th pctl. Mean Median 25th pctl. 10th pctl. 0.5 Jan 1980 Jan 1990 Jan 2000 Jan 2010
Average Fund Performance over Time 1 Average return (% per month) 0.5 0 0.5 Jan 1980 Jan 1990 Jan 2000 Jan 2010
Industry Size over Time 0.2 0.18 0.16 IndustrySize (fraction of CRSP) 0.14 0.12 0.1 0.08 0.06 0.04 0.02 0 Jan 1980 Jan 1990 Jan 2000 Jan 2010
Average Fund Performance over Time 1 GrossR GrossR adjusted for IndustrySize Average return (percent per month) 0.5 0 0.5 1980 1990 2000 2010
Fund Age vs. Performance Prediction: Fund s skill constant Industry-level DRTS Industry size Performance over fund s life
Fund Age vs. Performance: Age Fixed Effects GrossR it = a i + β 1 1 {age=1} +... + β 20 1 {age=20} + ε it Age fixed effects in GrossR (percent per month) 0.5 0.4 0.3 0.2 0.1 0 Estimates 95% confidence intervals 0.1 0 5 10 15 20 Fund age (years)
Fund Age vs. Performance: Continuous Age Dependent variable: GrossR Fund age -0.000123-0.000102 (-3.00) (-2.37) Observations 283046 248050 Fund ages All 3 years
Fund Age vs. Performance: Continuous Age Dependent variable: GrossR Fund age -0.000123 0.000283-0.000102 0.000281 (-3.00) (2.19) (-2.37) (2.19) IndustrySize -0.0845-0.0799 (-3.02) (-2.86) Observations 283046 283046 248050 248050 Fund ages All All 3 years 3 years
Learning on the Job? We modify our skill measure to allow learning on the job As before, skill is alpha when F undsize = IndustrySize = 0 But now, Skill it = a i + b FundAge it GrossR it = a i + b FundAge it + FundSize it 1 (β 0 + β 1 X i ) + IndustrySize it 1 (γ 0 + γ 1 X i ) + ε it
Distribution of Fund Skill, With Learning on the Job Fund skill (% per month) 1.5 1 0.5 0 90th pctl. 75th pctl. Mean Median 25th pctl. 10th pctl. 1980 1990 2000 2010
Age-based Investment Strategies Average portfolio return Average differences F - test Fund age [0, 3] (3, 6] (6, 10] >10 [0,3] - (>10) (3,6] - (>10) (6,10] - (>10) p-value Avg. GrossR 0.084 0.056 0.020 0.012 0.072 0.043 0.008 0.014 (2.33) (1.45) (0.55) (0.30) (2.85) (2.48) (0.52) Avg. NetR -0.005-0.052-0.084-0.083 0.077 0.031-0.001 0.008 (-0.15) (-1.38) (-2.29) (-2.07) (3.10) (1.79) (-0.08)
Robustness Our conclusions are robust to Controlling for business cycle variables Controlling for F amilysize Trimming extreme outliers in FundSize Different functional forms for FundSize Alternate benchmark-adjustments Fama-French Morningstar benchmark with estimated betas
Conclusions Scale: Strong evidence of decreasing returns to scale at industry level Stronger for high-turnover, high-volatility, and small-cap funds Mixed evidence of decreasing returns to scale at fund level Insignificant after removing econometric biases Skill: Active funds have become more skilled over time Yet their performance has not improved Negative age-performance relation A fund s performance decreases over its lifetime Younger funds outperform older funds