Empirical Financial Economics 5. Current Approaches to Performance Measurement Stephen Brown NYU Stern School of Business UNSW PhD Seminar, June 19-21 2006 Overview of lecture Standard approaches Theoretical foundation Practical implementation Relation to style analysis Gaming performance metrics Performance measurement Leeson Market Short-term Investment (S&P 500) Government Managemen Benchmark Benchmark t Average .0065 Return .0050 .0036 Std. .0106 Deviation .0359 .0015 1.0 .0 Beta .0640 Alpha .0025 .0 .0 (1.92) 100% in cash.0 at close of Sharpe Style: Ratio Index .2484 Arbitrage,.0318 trading -1 .0 -0 0 % .5 0 0. % 00 0. % 50 % 1.0 0% 1.5 0% 2. 00 2. % 50 3. % 00 3. % 50 4. % 00 4. % 50 5. % 00 5. % 50 6. % 0 0 6. % 50 % Frequency distribution of monthly returns 35 30 25 20 15 10 5 0 Universe Comparisons 40% Brownian Management 35% S&P 500 30% 25% 20% 15% 10% 5% One Quarter 1 Year 3 Years Periods ending Dec 31 2002 5 Years Total Return comparison Average Return A B C D Total Return comparison Average Return RS&P = 13.68% rf = 1.08% A S&P 500 B C D Treasury Bills Manager A best Manager D worst Total Return comparison Average Return A B C D Sharpe ratio comparison Average Return A B D C Standard Deviation Sharpe ratio comparison Average Return A S&P 500 RS&P = 13.68% B D rf = 1.08% C Treasury Bills ^ σS&P = 20.0%Standard Deviation Sharpe ratio comparison Average Return A S&P 500 RS&P = 13.68% B Sharpe ratio = Average return – rf D C Manager C worst Manager D best Standard Deviation rf = 1.08% Treasury Bills ^ σS&P = 20.0%Standard Deviation Sharpe ratio comparison Average Return A S&P 500 RS&P = 13.68% B D rf = 1.08% C Treasury Bills ^ σS&P = 20.0%Standard Deviation Jensen’s Alpha comparison Average Return A S&P 500 RS&P = 13.68% Jensen’s alpha = Average return – C D B Manager B worst Manager C best {rf + β (RS&P - rf )} rf = 1.08% Treasury Bills βS&P = 1.0 Beta Intertemporal equilibrium model j Max Et U (ct j ) Multiperiod problem: j 0 First order conditions: U (ct ) j Et (1 ri ,t j )U (ct j ) Stochastic discount factor interpretation: 1 Et (1 ri ,t j ) mt , j , U (ct j ) mt , j U (ct ) j mt , j “stochastic discount factor”, “pricing Value of Private Information I1 I 0 Investor has access to information R0 I1 I 0 is givenEby Rwhere Value of t [( R1 R0 )mt ] 1 I1 I0 and are returns on optimal portfolios given and Et [( R1 CAPM R0 )mt ] (Chen 1Knez rft ) 1 Under 1t rft & t ( mt 1996) Jensen’s alpha measures value of private information The geometry of mean variance a 2bE cE 2 ac b2 2 E 1 1/ b 1 x 1/ b 2 0 a b a 1 2 a b Note: returns are in excess of the risk freerf Informed portfolio strategy R1 rf R0 rf Excess return on informed is the return on an optimal strategy where orthogonal portfolio (MacKinlay 1995) Sharpe ratio squared of informed strategy 12 (0 rf )1 (0 rf ) 1 02 2 02 Assumes well diversified portfolios Informed portfolio strategy R1 rf R0 rf Excess return on informed is the return on an optimal strategy where orthogonal portfolio (MacKinlay 1995) Sharpe ratio squared of informed strategy 12 (0 rf )1 (0 rf ) 1 02 2 02 Assumes well diversified portfolios Used in tests of mean variance efficiency of benchmark Practical issues Sharpe ratio sensitive to diversification, but invariant to leverage Risk premium and standard deviation proportionate to fraction of investment financed by borrowing Jensen’s alpha invariant 2 0 to diversification, but sensitive to leverage In a complete market implies Changes in Information Set 1t 1t rft 1t ( mt rft ) How do we measure alpha I1t when information set is not constant? Rolling regression, use subperiods to estimate 1 1 rf 1 ( m rft ) (no t subscript) – Sharpe (1992) macroeconomic variable controls – Ferson and Schadt(1996) Use GSC procedure – Brown and Goetzmann (1997) Use Style management is crucial … Economist, July 16, 1995 But who determines styles? Characteristics-based Styles rjt Jt Jt I t jt jJ Traditional approach … Jt are changing characteristics (PER, Price/Book) It are returns to characteristics Jt Style benchmarks are given by rjt Jt jt jJ Returns-based Styles rjt Jt Jt I t jt jJ Sharpe (1992) approach … Jt are a dynamic portfolio strategy I t are benchmark portfolio returns Jt by Style benchmarks are given rjt Jt jt jJ Returns-based Styles rjt Jt Jt I t jt jJ GSC (1997) approach … jT , Jt vary through time but are fixed forJ style Jt Allocate funds to styles directly using Jt Style benchmarks are given by rjt Jt jt jJ Eight style decomposition 10 0 % 80% 60% 40 % 20 % 0% GSC1 GSC2 GSC3 GSC4 GSC5 GSC6 Ot her Pure Emerging Market Global Macro Event Driven US Equit y Hedge GSC7 GSC8 Pure Propert y Pure Leveraged Currency Non Direct ional/Relat ive Value Non-US Equit y Hedge Five style decomposition 10 0 % 80% 60% 40 % 20 % 0% GSC1 GSC2 Ot her Pure Emerging Market Global Macro Event Driven US Equit y Hedge GSC3 GSC4 GSC5 Pure Propert y Pure Leveraged Currency Non Direct ional/Relat ive Value Non-US Equit y Hedge Style classifications GSC1 GSC2 GSC3 GSC4 GSC5 GSC6 GSC7 GSC8 Event driven international Property/Fixed Income US Equity focus Non-directional/relative value Event driven domestic International focus Emerging markets Global macro Regressing returns on 2 classifications: Adjusted R Year N 1992 149 1993 212 1994 288 1995 405 1996 524 1997 616 1998 668 Average GSC8 classifications 0.3827 0.2224 0.1662 0.0576 0.1554 0.3066 0.2813 0.2246 GSC5 classifications 0.1713 0.1320 0.1040 0.0548 0.0769 0.1886 0.2019 0.1328 TASS 17 classifications 0.4441 0.1186 0.0986 0.0446 0.1523 0.2538 0.1998 0.1874 Variance explained by prior returns-based classifications Year N 1992 198 1993 276 1994 348 1995 455 1996 557 1997 649 1998 687 Average 8 GSC Classifications 0.3622 0.1779 0.1590 0.0611 0.1543 0.2969 0.2824 0.2134 8 Principal Components 0.0572 0.0351 0.0761 0.0799 0.0286 0.0211 0.2862 0.0835 8 Benchmarks (predetermined) 0.1769 0.1748 0.0481 0.0862 0.0691 0.0642 0.2030 0.1175 Variance explained by prior factor loadings Year N 1992 198 1993 276 1994 348 1995 455 1996 557 1997 649 1998 687 Average 8 GSC Classifications 0.2742 0.2170 0.1760 0.0670 0.1444 0.3135 0.2752 0.2096 8 Principal Components 0.1607 0.0928 0.1577 0.0783 0.0888 0.3069 0.3744 0.1799 8 Benchmarks (predetermined) 0.2552 0.0932 0.0700 0.0829 0.0349 0.0899 0.3765 0.1432 Percentage in cash (monthly) 120 % 10 0 % 80% 60% 40 % 20 % 0% 31-Dec-198 9 15-May-1991 26 -Sep-1992 8 -Feb-1994 Examples of riskless index arbitrage … Percentage in cash (daily) 20 0 % 10 0 % 0% -10 0 % -20 0 % -30 0 % -40 0 % -50 0 % -6 0 0 % 31-Dec-198 9 15-May-1991 26 -Sep-1992 8 -Feb-1994 “Informationless” investing Concave payout strategies Zero net investment overlay strategy (Weisman 2002) Uses only public information Designed to yield Sharpe ratio greater than benchmark Using strategies that are concave to benchmark Concave payout strategies Zero net investment overlay strategy (Weisman 2002) Uses only public information Designed to yield Sharpe ratio greater than benchmark Using strategies that are concave to benchmark Why should we care? Sharpe ratio obviously inappropriate here But is metric of choice of hedge funds and derivatives traders We should care! Delegated fund management Fund flow, compensation based on historical performance Limited incentive to monitor high Sharpe ratios Behavioral issues Prospect theory: lock in gains, gamble on loss Are there incentives to control this behavior? Sharpe Ratio of Benchmark 10 0 % 50 % 0% -50 % Benchmark -10 0 % -150 % -20 0 % -50 % 0% 50 % Sharpe ratio = .631 10 0 % Maximum Sharpe Ratio 10 0 % 50 % Benchmark 0% -50 % Maximum Sharpe Rat io St rat egy -10 0 % -150 % -20 0 % -50 % 0% 50 % Sharpe ratio = .748 10 0 % Concave trading strategies 10 0 % 50 % Benchmark 0% Loss Averse Trading (Median) Maximum Sharpe Rat io St rat egy -50 % -10 0 % -150 % -20 0 % -50 % 0% 50 % 10 0 % Examples of concave payout strategies Long-term asset mix guidelines Examples of concave payout strategies Unhedged short volatility Writing out of the money calls and puts Examples of concave payout strategies Loss averse trading a.k.a. “Doubling” Examples of concave payout strategies Long-term asset mix guidelines Unhedged short volatility Writing out of the money calls and puts Loss averse trading a.k.a. “Doubling” Forensic Finance Implications of concave payoff strategies Patterns of returns Forensic Finance Implications of Informationless investing Patterns of returns are returns concave to benchmark? Forensic Finance Implications of concave payoff strategies Patterns of returns are returns concave to benchmark? Patterns of security holdings Forensic Finance Implications of concave payoff strategies Patterns of returns are returns concave to benchmark? Patterns of security holdings do security holdings produce concave payouts? Forensic Finance Implications of concave payoff strategies Patterns of returns are returns concave to benchmark? Patterns of security holdings do security holdings produce concave payouts? Patterns of trading Forensic Finance Implications of concave payoff strategies Patterns of returns are returns concave to benchmark? Patterns of security holdings do security holdings produce concave payouts? Patterns of trading does pattern of trading lead to concave Conclusion Value of information interpretation of standard performance measures New procedures for style analysis Return based performance measures only tell part of the story