Benchmarking money manager performance: Issues

1
Benchmarking money manager performance Issues
evidence

• Louis K. C. Chan
• University of Illinois Urbana-Champaign
• March 2006

2
Objectives

• The evaluation and attribution of investment
  performance is crucial for investment research
  and practice
• Money manager performance
• Results of investment strategies trading rules
• Effects of managerial decisions on shareholder
  wealth
• Academic and practitioner research has produced a
  large array of methods for evaluating and
  attributing investment performance

3
Objectives

• Question are conclusions sensitive to the choice
  of evaluation and attribution methods? why?
• We compare the results from various methods
  applied to common samples
• Set of active institutional money managers
• Passive indexes

4
Evaluating method performance

• Many widely-used methods draw on evidence from
  asset pricing studies that size, value/growth
  describe much of the variation in returns
  (notably Fama and French (1992), Fama and French
  (1993))
• We concentrate on benchmarking methods based on
  size, value/growth
• Characteristic-matched control portfolios
• Time-series factor model regressions
• Effective asset mix regressions
• Cross-sectional regressions on characteristics
• 1998 2000 market boom as stress test of
  benchmarking methods

5
Evaluating manager performance

• Much previous work on evaluating performance of
  mutual and closed-end funds (e.g. Jensen (1968),
  Elton et al. (1993), Malkiel (1995), Gruber
  (1995), Carhart (1997), Daniel et al. (1997),
  Kothari and Warner (2001), etc.)
• Managers of pension plan equity assets are just
  as important, but much less previous research
  (see LSV 1992, Coggin et al. 1993)

6
A first look characteristic-matched portfolios
vs. 3 factor model
7
(No Transcript)
8
Benchmark details

• Benchmarks vary according to
• Characteristics or loadings
• Measuring size, value/growth style
• Treating size, value/growth effects separately
• Portfolio weighting scheme
• Frequency of benchmark reconstitution

9
Benchmark details

• Characteristics versus loadings
• Predict benchmark return using portfolios
  attributes (size, book-to-market ) or
  predict benchmark return using portfolios
  loadings on factors
• Some evidence that attributes predict returns
  better than loadings (Daniel and Titman 1997)
• Data on holdings not generally accessible

10
Building performance benchmarks

• Measuring size, value/growth style
• Size market capitalization (float?)
• Value/growth orientation usually measured by
  book-to-market ratio (book value of equity
  divided by market value of equity)
• Book value of equity does not record value of
  intangible assets includes goodwill from
  acquisitions

11
Building performance benchmarks

• Treating size, value/growth effects separately
• E.g. independent 2-way sorts by size, BM
• In one-way sorts by book-to-market equity large
  stocks typically are classified as growth
• Under an independent size/BM sort procedure
  large-cap managers, regardless of large
  value/large growth style, will tend to be
  compared against a growth benchmark

12
Building performance benchmarks

• Weighting scheme for stocks in benchmark
• Equal-weighting
• Value-weighting
• Benchmark reconstitution frequency
• Over time benchmark becomes more heterogeneous
  and may no longer correspond to managed
  portfolios features

13
Data

• Holdings and returns every quarter for 199
  portfolios offered by money managers to clients,
  1989Q1 - 2001Q4
• Domestic U.S. equity portfolios only
• Different styles (large/mid/small,
  value/blend/growth)
• Some selection bias

14
Results outline

• Performance relative to benchmarks based on
  characteristics
• Overall active manager sample
• Classified by investment style
• Diagnostics
• Performance relative to benchmarks based on
  loadings
• Overall active manager sample
• Classified by investment style
• Diagnostics

15
Performance measures

• Abnormal return portfolios return minus return
  on benchmark portfolio
• Tracking error volatility standard deviation of
  quarterly difference between portfolios return
  and benchmarks return

16
(No Transcript)
17
(No Transcript)
18
(No Transcript)
19
Benchmark performance
20
Benchmark performance
21
Benchmark comparisons
22
(No Transcript)
23
(No Transcript)
24
Performance based on regression benchmarks

• Three factor model excess return is
• ( rpt rft ) benchmark return
• benchmark return is from fitted regression
• ß(rmt rft ) s SMBt h HMLt

25
Regression-based benchmark details

• Exposures estimated
• over full period (including the quarter when we
  measure performance)
• or leaving out the quarter when we measure
  performance
• Measuring size, value/growth factors
• High versus low book-to-market
• Other indicators of value/growth orientation

26
Building regression-based benchmarks

• 3 factor model accounts for size, value/growth
  separately
• E.g. benchmark return for small value manager
• return for market exposure
• plus return for smallness
• plus return for value
• Benchmark credits manager for smallness even
  though small stocks performance is because
  small growth does better than small value

27
Regression-based benchmarks

• Alternative compare manager to a selection of
  passive benchmarks (effective asset mix
  regressions)
• rpt a w1LGt w2LVt
• w3MCGt w4MCVt
• w5SGt w6SVt ?pt
• w1, ,w6 portfolio weights (between 0 and 1, add
  up to 1)

28
Building regression-based benchmarks

• Another widely-used alternative each stocks
  predicted return is from a cross-sectional
  regression using stock characteristics, industry
  dummy variables
• rit a ß1X1i ß2X2i

29
(No Transcript)
30
(No Transcript)
31
(No Transcript)
32
Regression-based benchmark comparisons
33
Regression-based benchmark comparisons
34
(No Transcript)
35
Conclusions

• Benchmarking methods that appear similar on the
  surface can lead to very different conclusions
  about investment performance
• Popular methods (characteristic-matched reference
  portfolios, 3 factor time series regression
  models, cross-sectional regression) have
  disappointing ability to track managed active
  portfolios and passive benchmarks

36
Conclusions

• Methods based on within-size classifications, use
  multiple measures of value-growth orientation,
  improve ability to track managed and passive
  portfolios
• Given the fragility in reliably separating skill
  from style, detailed decomposition and
  attribution of performance should be treated with
  caution