Risk Adjusted Performance Analysis

RISK-ADJUSTED
PERFORMANCE ANALYSIS
Andreas Steiner
Zurich
May 2001
Date: 01.05.98
Produced by: Andreas Steiner Slide 1
CONTENTS
1. INTRODUCTION 4. M – MEASURES
2. FRAMEWORK 1. M2
2. M3
3. DEFINITION AND
APPLICATION OF 5. RAPP
CLASSICAL MEASURES
1. Sharpe Ratio 6. RELATIONSHIP BETWEEN
THE MEASURES
2. Treynor Ratio
3. Information Ratio 7. DISCUSSION
Date: 31.03.2001
INTRODUCTION (1/2)
Common wisdom today: Performance is only a biased and noisy

signal for the quality of asset management.
BIAS: Risk-return trade off
NOISE: Skill versus luck
Risk-adjusted performance analysis is about quantifying and

analyzing unbiased performance. It can also be used to distinguish
skill from luck.
This presentation wants to summarize the best practice concepts

and methods in risk-adjusted performance analysis. It is of a
descriptive nature.
Date: 31.03.2001
INTRODUCTION (2/2)
CONSULTANTS SWITZERLAND
Quantitative performance analysis as a criterion for manager selection has been
practiced for about 5 years → a new toy
Most often requested statistics: Sharpe and Information Ratio
STRUCTURED ALPHA™ (Watson Wyatt)

Alpha: Net Fund Return – Net Benchmark Return.
Net = Fees & switching costs
Sigma: Tracking Error = Standard Deviation of Alpha
→ Financial Factors summarized in…

Investment Efficiency: Net Alpha / TE = IR. Used to rank managers
Theta: Non-financial factors are of importance to trustees: ‘Sleep Well’ factors (loss
aversion), ‘Seems Good’ factors (brand names)
According to WW, the relative influence of financial and Theta factors is 50/50.
Source: ‘Global Industry Survey’, WW, 1999.
There exists a trade off between financial and Theta factors. That’s why you need
WW’s consulting service…
Date: 31.03.2001
FRAMEWORK
1. CLIENT PREFERENCES 2. BENCHMARKING
Client likes return, dislikes risk*: Client chooses benchmark and
sets targets/limits for alpha, beta
U = U ( µ P ,σ P ) a.s.o. at inception
∂U ∂U
dU = dµ P + dσ P Portfolio Mgt controls alpha, beta
∂µ P ∂σ P and beta after inception
∂U
>0 3. INDEX MODELS
∂µ P
∂U µ P − rf = α + β ⋅ ( µ B − rf ) + ε
<0
∂σ P Validity of index models to analyze
performance largely depends on the
*risk is usually defined as the implementation of benchmarking!
second moment of the return
distribution.
Date: 31.03.2001
SHARPE RATIO (1/2) – DEFINITION
µ
µ P − rf
S=
σP
µP
µ P ...Portfolio Return
r f ...Riskfree Rate rf
σ P ...Portfolio Volatility
σP σ
Date: 31.03.2001
SHARPE RATIO (2/2) - APPLICATION
MEASUREMENT INTERPRETATION
Annualized portfolio return, portfolio Summary of the first two moments of

volatility the portfolio excess return
distribution. Model-free
Annualized risk-free rate
– Choice is important because it can Suitable for comparisons across
change ranking asset classes
– Problematic in an international Target in Mean-Variance
context Optimization
Aggregation Does not assume a benchmark.
– No straight-forward adding-up Implicit benchmark is risk-free rate.
because of covariance effects
between volatilities Statistical hypothesis testing: test for
non-zero performance
Are negative values ambiguous?
t-Stat = S * sqrt(T)
µ P − rf µ P − rf
+ −
σP σP
Date: 31.03.2001
TREYNOR RATIO (1/2) - DEFINITION
µ
µ P − rf
T=
βP
µP
µ P ...Portfolio Return
r f ...Riskfree Rate rf
β P ...Portfolio Beta
βP β
σ σ σ PB ...Covariance
β = PB2 = ρ PB P
σB σB ρ PB ...Correlation
Date: 31.03.2001
TREYNOR RATIO (2/2) - APPLICATION
Annualized portfolio return, Accounts for systematic and

annualized risk-free rate unsystematic risk (CAPM-based):
Only systematic risk is
Estimation of beta can be distorted considered.
by market timing. Extensions:
Squared regression, H/M regression Comparison across different asset
classes problematic (beta is
Aggregation: Straight-forward. Beta dependent on benchmark)
of aggregate is weighted sum of
constituent’s betas Choice of benchmark affects
ranking
Date: 31.03.2001
INFORMATION RATIO (1/3) - DEFINITION
αP α
IR =
TE P
αP
αP ...Portfolio Alpha
TE P ...Porfolio Tracking Error
TE P TE
Active Portfolio Return: Alpha
Average annual performance
Jensen’s Alpha
Choice should be consistent to choice of TE definition…
Date: 31.03.2001
IR (2/3) - TRACKING ERROR DEFINITIONS
µP = α + β ⋅ µB + ε
σ 2
σ PB σP
σ P2 = β 2 ⋅ σ B2 + σ ε2 = ρ PB
2
⋅ P2 ⋅ σ B2 + σ ε2 with… β= = ρ
σB σ B2 PB
σB
σ P2 = ρ PB
2
⋅ σ P2 + σ ε2 TEP = σ P ⋅ 1 − ρ PB
2
= σε
…Residual risk = Risk uncorrelated with BM.
TEP = Var (rP − rB ) …Standard deviation of performance
µ P − µ B = α + β ⋅ µ B + ε − µ B = α + (β − 1) ⋅ µ B + ε
Var( µ P − µ B ) = (β − 1)2 ⋅ σ B2 + σ ε2 → For beta ≠ 1, the Stdev(perf) is always

larger than residual risk
→ Stdev(perf) depends on benchmark volatility
Date: 31.03.2001
IR (3/3) - APPLICATION
Best practice in CH: TE as Summary statistic: Active return /

annualized volatility of active risk trade off, efficiency ratio
performance. Alpha as average
annualized performance. Fundamental Law of Active Mgt:
Measurement of Alpha & TE with IR ex ante = IC x BR

index or factor models makes IR
IC: Information Coefficient
dependent on model specification
Corr(Forecast r, Actual r)
errors.
BR: Breadth of strategy
# of independent bets taken
Statistical hypothesis testing: Non-

zero alpha signals
t-Stat = IR * sqrt(T)
Generally not consistent with MVO…
Date: 31.03.2001
M MEASURES - M2 (1/2)
σ µ
µ RAP = B (µ P − r f ) + r f µP
σP
µ RAP
µ RAP ...Risk - Adjusted Return
µB
σP ...Portfolio Volatility
σB ...Benchmark Volatility rf
µf ...Portfolio Return
σB σP σ
rf ...Riskfree Rate
→ Performance is volatility-adjusted by
σB
...Leverage Factor d
leveraging the fund with risk-free- σP
investments so that the resulting
volatility equals the benchmark volatility. µ RAP = d ⋅ µ P + (1 − d ) ⋅ rf
Date: 31.03.2001
The difference between M2 can be interpreted intuitively: Unit of
measurement is % → Risk expressed in units of return
M2 rankings are independent of the chosen benchmark (benchmark risk

as a scaling factor)
The M2 measure is a transformed Sharpe Ratio and therefore consistent

with MPT
σB
µ RAP = (µ P − rf ) + rf = σ B ⋅ S + rf
σP
M2 ranking equals Sharpe Ratio ranking
Drawback: Correlation risk (timing, selection) is neglected…
Date: 31.03.2001
µCAP = a ⋅ µ P + b ⋅ µ B + (1 − a − b ) ⋅ rf
2
TEPB → M3 cannot be illustrated graphically
ρ PB = 1−
2 ⋅ σ B2 in an elegant way (three dimensions)
→ Performance is correlation-adjusted
by leveraging the fund with active,
σ B ( 1 − ρ PB ) passive and risk-free funds so that (1)
2
a= the resulting volatility equals

σ P ( 1 − ρ PB 2 ) benchmark volatility and (2) the TE
equals the Target TE
( 1 − ρ PB )
2
b = ρ PB − ρ PB
( 1 − ρ PB )
2
Date: 31.03.2001
M3 is ‘volatility-risk- and-correlation-risk’-adjusted-performance
M3 rankings differ from M2 and rankings
If no target tracking error exists, a = 0 and M3 will equal M2
M3 can be used in a forward looking sense: It can provide ex ante

guidance how to structure portfolios with TE restrictions (given the
stability of distributional characteristics in the future)
Drawback (of all RAP measures): Timing and selection activities

are not decoupled.
Date: 31.03.2001
RAPP (1/2) …Risk-Adjusted Performance and Positioning Index
∂U ∂U
U ( µ P ,σ P ) ≈ U ( µ B ,σ B ) + dµ + dσ
∂µ ∂σ
U ( µ P ,σ P ) − U ( µ B ,σ B )
RAPP ≡ ≈ α + λ ⋅ TE
∂U
∂µ µ UP
∂U µP
λ= ∂σ ...Risk Aversion
∂U
∂µ
µB
α ≈ dµ UB
TE ≈ dσ
σB σP σ
Date: 31.03.2001
RAPP (2/2)
The RAPP concept is very flexible (TE targets, for example)
Utility functions are considered at least problematic by many

economists, especially in decision making under risk (‘Homo
Oeconomicus’ debate, Behavioral Finance)
To implement RAPP, the marginal utilities of parameters (risk

aversion, for example) have to be quantified. RAPP ranking will
depend on these marginal utilities.
Aggregation across asset classes is achieved by measuring

everything in terms of utilities. A new aggregation problem is
introduced: aggregating client preferences.
Non-financial aspects are neglected. Considering the importance of

such factors: Is it worth developing and maintaining an internal RAP
measure?
Date: 31.03.2001
RELATIONSHIP BETWEEN MEASURES
550
Sharpe/M2 Max
500
Treynor/Alpha
450
Max
Index of Chain-Linked Total Returns
400
MinVar
350
TE Min
300
250
IR Max
200
M3 Max
150
100
RAPP Max
50
92
92
93
93
94
94
95
95
96
96
97
97
98
98
99
99
00
00
01
MSCI World AC
l
l
n
n
Ju
Ju
Ju
Ju
Ju
Ju
Ju
Ju
Ju
Ja
Ja
Ja
Ja
Ja
Ja
Ja
Ja
Ja
Ja
Months 01.02.93 - 31.03.01
Markets: S&P 500, DJ Euro STOXX 50, SPI, MSCI Japan, FTSE 100
Observations:
- RAP strategies are highly correlated
- The ex ante / ex post choice of RAP targets creates significant incentives
Date: 31.03.2001
DISCUSSION
IT‘S YOUR
TURN...
Date: 31.03.2001

Risk Adjusted Performance Analysis

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Risk Adjusted Performance Analysis

Uploaded by

Copyright:

Available Formats

RISK-ADJUSTED

Common wisdom today: Performance is only a biased and noisy

BIAS: Risk-return trade off

NOISE: Skill versus luck

Risk-adjusted performance analysis is about quantifying and

This presentation wants to summarize the best practice concepts

 Most often requested statistics: Sharpe and Information Ratio

STRUCTURED ALPHA™ (Watson Wyatt)

 Sigma: Tracking Error = Standard Deviation of Alpha

→ Financial Factors summarized in…

 Annualized portfolio return, portfolio  Summary of the first two moments of

 Annualized portfolio return,  Accounts for systematic and

…Residual risk = Risk uncorrelated with BM.

TEP = Var (rP − rB ) …Standard deviation of performance

Var( µ P − µ B ) = (β − 1)2 ⋅ σ B2 + σ ε2 → For beta ≠ 1, the Stdev(perf) is always

 Best practice in CH: TE as  Summary statistic: Active return /

 Measurement of Alpha & TE with IR ex ante = IC x BR

 Statistical hypothesis testing: Non-

 Generally not consistent with MVO…

 M2 rankings are independent of the chosen benchmark (benchmark risk

 The M2 measure is a transformed Sharpe Ratio and therefore consistent

 Drawback: Correlation risk (timing, selection) is neglected…

a= the resulting volatility equals

 M3 rankings differ from M2 and rankings

 If no target tracking error exists, a = 0 and M3 will equal M2

 M3 can be used in a forward looking sense: It can provide ex ante

 Drawback (of all RAP measures): Timing and selection activities

 Utility functions are considered at least problematic by many

 To implement RAPP, the marginal utilities of parameters (risk

 Aggregation across asset classes is achieved by measuring

 Non-financial aspects are neglected. Considering the importance of

You might also like

Most often requested statistics: Sharpe and Information Ratio

Sigma: Tracking Error = Standard Deviation of Alpha

Annualized portfolio return, portfolio Summary of the first two moments of

Annualized portfolio return, Accounts for systematic and

Best practice in CH: TE as Summary statistic: Active return /

Measurement of Alpha & TE with IR ex ante = IC x BR

Statistical hypothesis testing: Non-

Generally not consistent with MVO…

M2 rankings are independent of the chosen benchmark (benchmark risk

The M2 measure is a transformed Sharpe Ratio and therefore consistent

Drawback: Correlation risk (timing, selection) is neglected…

M3 rankings differ from M2 and rankings

If no target tracking error exists, a = 0 and M3 will equal M2

M3 can be used in a forward looking sense: It can provide ex ante

Drawback (of all RAP measures): Timing and selection activities

Utility functions are considered at least problematic by many

To implement RAPP, the marginal utilities of parameters (risk

Aggregation across asset classes is achieved by measuring

Non-financial aspects are neglected. Considering the importance of