Faculté des HEC

Université de Lausanne

Master of Science in Finance

SUSTAINABILITY AWARE
ASSET MANAGEMENT

Eric Jondeau
MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 1/48
 SAAM
Lecture 2: Estimating the Inputs

Eric Jondeau

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 2/48
 Objectives of the Lecture
In the first lecture, we have studied static, single-period asset allocation decisions,
assuming expected returns and the covariance matrix are known.

In practice, asset allocation is based on some predictions of expected returns.

In this lecture, we discuss:

- the estimation of the expected returns and covariance matrix

- the alternative investment strategies that can be implemented in practice

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 3/48
 Objectives of the Lecture
Readings:

Dahlquist and Harvey (2001), Global Tactical Asset Allocation, Working Paper.
ð Description of the three levels of asset allocation

Black and Litterman (1992), Global Portfolio Optimization, Financial Analysts Journal
48, 28–43.
ð Seminal description of the Black-Litterman

Idzorek (2004), A step-by-step guide to the Black-Litterman model, working paper.

ð Guide to Black-Litterman approach

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 4/48
 Objectives of the Lecture

è Estimating the Inputs

- Benchmarking versus Active Management

- Black-Litterman Approach

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 5/48
 Expected Returns
Slowly evolving
Portfolio
weights Market Dynamic
Constant

Mode Benchmark Strategic Tactical Market Timing

Information Indexing Unconditional Conditional

Macro-Economics: Macro-Finance:
Micro-Finance:
Model -Growth models -Business cycle
-Market sentiment
-Stationary risk -Time-varying risk
-No risk premium
premium premium
MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 6/48
 Expected Returns
Using Conditional Information
Benchmark allocation: No conditioning information is used.

Strategic allocation: Long-horizon forecasting model is used to predict 5- or 10-year

returns.

Tactical allocation: Conditioning information is always used. Expected short-term

returns, standard deviations, and correlations are the basis for weight changes.

Unconditional approach: Expected returns and the covariance matrix are constant over
time. It assumes that historical average returns are good predictors of future returns.

Conditional approach: We use information available today. The mean changes through
time depending on the values of the information variables.

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 7/48
 Expected Returns
Predictability
Predicting equity returns is a difficult task:

- Returns are highly volatile

- Markets are quite efficient in incorporating information

The value of the firm is the discounted value of the cash flows

∞
CFt+i D
Pt = Et ∑ with Rt+i = R f ,t+i + βi (Rm,t+i − R f ,t+i )
D i
(1+ R ) ! ! ! !#"#$
i=0 t+i risk-adjusted risk
risk-free market premium
discount rate rate exposure

There are at least four fundamental sources of predictability in equity returns:

- Expected cash flows (numerator)

- Risk-free interest rate (denominator)
- Expected firm’s risk exposure (denominator)
- Expected market risk premium (denominator)
MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 8/48
 Expected Returns
Predictability
- Expected cash flows: They often move with the business cycle. Both are slow-
moving and persistent. Short and long horizon predictability.

- Risk-free rate: It mostly reflects expectations about monetary policy.

- Expected firm’s risk exposure: It depends on the industry and on the firm’s capital
structure (leverage). Its changes are slow, persistent, and partly predictable.

- Expected market risk premium: It is also related to the business cycle. During
recessions, the premium must be high to attract investors into the market.

A common component of all four fundamental variables is the business cycle. The
business cycle is persistent, and it is possible to partially predict real economic growth:

- Consumers tend to smooth consumption.

- Investment is sticky, i.e., corporate investment in projects is usually long-term.
- Government expenditures have a low level of variability.
MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 9/48
 Estimating Risk Premia: Unconditional Mean
Given historical prices (including dividends) {𝑃! , 𝑃" , … , 𝑃# }, we compute

𝑃$%" − 𝑃$
𝑅$%" = 𝑡 = 0, … , 𝑇 − 1
𝑃$%"
(
Using the log-risk-free rate, 𝑅&,$ , we obtain excess returns: 𝑅$%" = 𝑅$%" − 𝑅&,$

We then estimate unconditional risk premia as:

1 #
𝜇̂ = 0 𝑅$(
(
𝑇 $)"

Properties:

- easy to compute and update

- assumes constant risk premia
- with iid normal returns, the central limit theorem states that √𝑇(𝜇̂ ( − 𝜇( )~𝑁(0, Σ)

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 10/48
 Estimating Risk Premia: Unconditional Mean
Example:

Assume 𝜇̂ ( = 6%, 𝜎 = 15% and 𝑇 = 120 (10 years of monthly data)

The standard error of the sample mean is

𝑠𝑡𝑑 [𝜇̂ ( ] = 𝜎/√𝑇 = 15%/√120 = 1.4%

This implies a 95% confidence interval of

C𝜇̂ ( ± 2 𝑠𝑡𝑑 [𝜇̂ ( ]E = [3.25% ; 8.75%]

Risk premia are difficult to estimate precisely.

Remark: Most models are designed for daily returns. However, the asset allocation is
often performed at the weekly or monthly frequency (→ Temporal Aggregation).

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 11/48
 Estimating Risk Premia: Linear Conditional Mean
Conditional risk premia for next period can be defined as linear functions of current
macro-economic variables or firm-specific variables:

- Time series:
(
𝜇̂ *,$%" = 𝑎* + 𝑏*+ 𝑍$ t = 1, …, T

where 𝑍$ may include business cycle indicators, such as interest rate, inflation rate, or
dividend yield.

- Cross section:
(
𝜇̂ *,$%" = 𝑎* + 𝑏*+ 𝑍*,$ t = 1, …, T, i = 1, …, N

where 𝑍*,$ may be the market beta 𝛽*,$ .

Remark: For long-term predictions, we can use a Vector AutoRegression (VAR) Model,
which allows us to predict the predictors. See EMF Lecture 4 for VAR models.

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 12/48
 Factor Fishing
- Theory
o Market portfolio (CAPM) or Intertemporal hedge portfolios: portfolios
maximally correlated with changes in investment opportunities (I-CAPM)

- Fundamental factors (accounting-based)

o Size
o Book to market
o Earnings to market
o Cash-flow to market
o Dividend yield
o Industry factors

- Macroeconomic factors
o Default premium
o Term premium
o Industrial production
o Inflation

- Statistical procedure
o Factor analysis or Principal components
MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 13/48
 Factor Zoo

The factor zoo contains 98 tradable and 1 nontradable (No. 97) factors for monthly data
from July 1980 to December 2016. In addition to these publicly available factors, we
follow Fama and French (1993) to construct value-weighted portfolios as factors using
firm characteristics collected in Green et al. (2016). For each characteristic therein, we
sort all stocks into deciles based on their previous year-end values, then build and
rebalance a long-short portfolio (top 30% - bottom 30% or 1-0 dummy difference) every
June. For factor classification, “M” is Momentum, “V” is Value-versus-Growth, “I” is
Investment, “P” is Profitability, “IN” is Intangibles, and “T” is Trading Frictions.

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 14/48
 Factor Zoo

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 15/48
 Factor Zoo

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 16/48
 Factor Zoo

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 17/48
 Factor Zoo

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 18/48
 Factor Zoo

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 19/48
 Guidelines for Backtesting

- Use theory and intuition to guide model specification

- Constrain specification search process before confronting data

o explicitly list all model specifications to be considered
o single-shot model comparison
o stop regardless of outcome

- Check if results are reasonable

o 10% of realized information ratios ³ 1
o information ratios ³ 2 likely due to data mining or mistakes

- Careful out-of-sample analysis

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 20/48
 Estimating Covariance Matrix: Unconditional Moment
Given excess returns, {𝑅"( , … , 𝑅#( }, the sample covariance matrix can be estimated as:

1 # ,
𝜎*, = 𝑉[𝑅*( ] = (
0 L𝑅*,$ − 𝜇̂ *( M
𝑇−1 $)"

1 #
𝜎*- = 𝐶𝑜𝑣C𝑅*( , 𝑅-( E = (
0 L𝑅*,$ − 𝜇̂ *( ML𝑅-,$
(
− 𝜇̂ -( M
𝑇−1 $)"

Properties:

- easy to compute and update

- assumes constant return distribution

However:

- the horizon of the allocation is not necessarily the same as the frequency of the data
- we know that volatilities and correlations are time varying.

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 21/48
 Estimating the Covariance Matrix: Rolling Window
An alternative approach that accounts for time-variability is the use of Rolling Windows

- Consider the first 𝜏 observations (𝜏 ≪ 𝑇)

- Estimate risk premia (𝜇̂ .( ) and the covariance matrix (Σ. ) using 𝑡 = 1, … , 𝜏

1 . 1 .
𝜇̂ .( = 0 𝑅$( Σ. = 0 (𝑅$( − 𝜇̂ .( )(𝑅$( − 𝜇̂ .( )′
𝜏 $)" 𝜏−1 𝑡=1

- Roll the sample by 1 observation

(
- Estimate risk premia (𝜇̂ .%" ) and the covariance matrix (Σ.%" ) using 𝑡 = 2, … , 𝜏 + 1

- …

- Estimate risk premia (𝜇̂ (# ) and the covariance matrix (Σ # ) using 𝑡 = 𝑇 − 𝜏 + 1, … , 𝑇

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 22/48
 Estimating the Covariance Matrix: EWMA
A slightly better estimator is the Exponentially Weighted Moving Average (EWMA)

- Consider a memory parameter 𝜙 (0 ≤ 𝜙 ≤ 1), an initial risk premium (𝜇̂ !( ), and an

initial covariance matrix Σ! . 𝜇̂ !( and Σ! can be based on a pre-sample.

- At date 𝜏, compute the risk premium and the covariance matrix as:
(
𝜇̂ .( = 𝜙 𝜇̂ ./" + (1 − 𝜙)𝑅.(

Σ. = 𝜙Σ./" + (1 − 𝜙)(𝑅.( − 𝜇̂ .( )(𝑅.( − 𝜇̂ .( )′

Remark: Problems related to the number of observations are even more severe.

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 23/48
 Estimating Covariance Matrix: GARCH models
Another alternative approach is to use Multivariate GARCH models:

- models for the covariance matrix: Vech, diagonal vec, or BEKK models
- models for the correlation matrix: CCC, or DCC models.
(See EMF Lecture 12)

or Factor models for the covariance matrix (See SAAM Lecture 4)

Most of these models ensure that the covariance matrix is positive definite.

Remark: For N assets, the number of parameters to estimate is N(N+1)/2, the number of
observations is N T

- For N = 500 stocks and T = 600 months of data (50 years): N(N+1)/2 = 125’250
unique parameters and N T = 300’000 observations, so that each parameter is
estimated with 2.4 observations on average.

- The sample covariance matrix is singular when N > T – 1.

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 24/48
 Objectives of the Lecture

- Estimating the Inputs

è Benchmarking versus Active Management

- Black-Litterman Approach

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 25/48
 Active versus Passive Management
Most asset managers are evaluated based on running the adjusted CAPM regression:

𝑅*,$%" − 𝑅&,$ = 𝛼* + 𝛽* L𝑅0,$%" − 𝑅&,$ M + 𝜀*,$%"

where 𝑅*,$%" is the fund return and 𝑅0,$%" is the return of the benchmark adopted by the
fund

Passive management = beta versus Active management = alpha

We distinguish three levels of asset allocation (Dahlquist and Harvey, 2001):

- Benchmark / Passive management (index funds, passive mutual funds)

- Strategic / Active management (active mutual funds, long-term managers: pension
funds, insurers, sovereign wealth funds)
- Tactical / High-frequency trading (hedge funds)

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 26/48
 The Three Levels of Asset Allocation
Relative importance of passive and active management – AUM

Note: 1 As of end-June for each year. 2 Includes investment fund assets of closed-end funds, hedge funds, insurance
funds, investment trusts and pension funds. Source: BIS Quarterly Review, March 2018, The implications of passive
investing for securities markets.
MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 27/48
 The Three Levels of Asset Allocation
Average daily trading volume (in bln shares)

HFT: high-frequency trading

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 28/48
 The Three Levels of Asset Allocation
Benchmark (Passive) Asset Allocation
Investors exactly replicate the investment weights of a benchmark index.

Example: If the benchmark is MSCI world, the benchmark allocation has the same
weights as this index.

This type of asset allocation is sometimes referred to as indexing. No information is used

other than the usual details of indexing: market weights, delistings, new listings,
buybacks, secondary market offerings, dividends, and warrants.

If the benchmark is broad (MSCI world), this allocation is sometimes referred to as

Equilibrium Asset Allocation because the CAPM implies that investors hold the world
market portfolio in equilibrium, levered up or down to match desired risk aversion.

Remark: If we use historical covariance matrix, we can back out of the optimization the
expected returns that exactly imply the market weights. These expected returns are often
referred to as Equilibrium Expected Returns. (See Black-Litterman approach below)

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 29/48
 The Three Levels of Asset Allocation
Strategic Asset Allocation (SAA)
Investors with long-term horizon (say, 5 years) have a view that certain assets will out-
perform or under-perform and adopt weights that deviate significantly from the index.

Example: The manager’s view is that Japanese government bonds will underperform
(relative to historical performance) over the next 5 years. She would underweight these
bonds relative to the benchmark.

Even if weights are based on 5-year forecasts, it is common practice to update the 5-year
forecasts every year and, hence, rebalance annually.

Deviation from the benchmark introduces tracking error, i.e., the standard deviation of
the differences between the benchmark return and the portfolio return.

Strategic deviations from benchmark will induce a ‘strategic’ tracking error.

Remark: SAA may be implemented using either static models or dynamic models.

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 30/48
 The Three Levels of Asset Allocation
Tactical Allocation (TAA)
Investors take short-term bets, usually 1 month to 1 quarter, and deviate from the strategic
weights. This also induces tracking error. The difference between the strategic and
tactical weights induces ‘tactical’ tracking error.

Because of the frequent changes in investment weights in tactical programs, transactions

costs are important. A common strategy is to minimize transactions costs by using futures
or swaps (instead of stocks or bonds) to manage tactical deviations.

As futures and swaps can be used to manage bets at minimal cost, we could consider a
fourth level of asset allocation – the “High-Frequency Tactical Asset Allocation” (or
Market Timing). It is run on a daily or intraday basis, using quantitative models.

Tactical Asset Allocation (TAA) is the approach to portfolio management of many hedge
funds.

Short-term bets are placed for periods of typically less than a year, but it may also be used
by an investment fund that simultaneously has strategic asset allocation objectives.

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 31/48
 Tracking Error
The difference between the return on the strategic or tactical asset allocation and the
return on the benchmark portfolio is the tracking error:

𝑒*,$%" = 𝑅*,$%" − 𝑅0,$%"

Rationales for Benchmark-Relative Investment:

- Risk perspective: A benchmark anchors the portfolio in risk-return space and thus
gives clients confidence in what risk the investment carries.

- Return perspective: Some claim it is easier to forecast relative returns.

- More plausible: Estimation error is smaller for relative than for absolute return
forecasts.

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 32/48
 Tactical Allocation: Total Risk/Return Perspective
We consider again the optimization problem:

ìï max a ' µ - l2 a ' Sa

a
í t
ïîsubject to a ' e = 1

From the Mutual Fund Separation Theorem, weights of the optimal portfolio are:

S -1 e 1 -1 æ e ' S -1 µ ö 1 -1 æ e ' S -1 µ ö
a = *
+ S çµ - e ÷ = a gmv + S ç µ - e÷
e' S e l
-1
è
-1
e' S e ø l è
-1
e' S e ø

Now, assume that we have a vector of equilibrium returns µ! . We can write

* 1 −1 # e ' Σ−1 (µ − µ! + µ! ) &

α = α gmv + Σ % µ − µ! + µ! − e(
λ $ e' Σ e
−1
'
1 −1 # e ' Σ−1 µ! & 1 −1 # e ' Σ−1 (µ − µ! ) &
= α gmv + Σ % µ! − e ( + Σ % µ − µ! − e(
λ $ e' Σ e ' λ
−1
$ e' Σ e
−1
'

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 33/48
 Tactical Allocation: Total Risk/Return Perspective
Finally, we get

1 Σ −1
!
µ(e '− e !
µ ' Σ)−1
1 Σ −1
( µ −(!
µ )e '− e( µ − !
µ )' Σ −1
)
α * = α gmv + e + e
λ e' Σ e−1
λ e' Σ e
−1
"$$$#$$$% "$$$$$#$$$$$%
αS αT

so that

a * = a gmv + a S + aT

with a S ' e = aT ' e = 0

We have the following interpretations:

- 𝛼123 is the global minimum-variance portfolio (≈ benchmark portfolio)

- 𝛼4 is the strategic portfolio
- 𝛼 # is the tactical portfolio

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 34/48
 Tactical Allocation: Active Risk/Return Perspective
We have assumed that total returns enter the optimization. This situation makes sense for
an investor who cares about the level of wealth. In the asset management industry,
managers are often rewarded according to their performance relative to a benchmark.

Consequently, managers are more likely to focus on active return and risk, rather than on
total return and risk of portfolio.

If only the tactical position matters, we can reformulate the optimization. Let 𝛼0 be the
vector of weights of the benchmark and 𝛼5 the vector of weights of the portfolio. Let 𝛼6
be the vector of tactical bets (or active portfolio bets) on each asset in the portfolio, i.e.,
𝛼6 measures the deviation from the benchmark: 𝛼6 = 𝛼5 − 𝛼0 .

The Tracking Error is defined as: 𝑒$%" = 𝛼6+ 𝑅$%"

Now, the important quantities are

- The expected return (Alpha) relative to benchmark: 𝐸 [𝑒$%" ] = 𝛼6+ 𝜇
- The volatility of the Tracking Error: 𝜎[𝑒$%" ] = [𝛼6+ Σ 𝛼6

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 35/48
 Tactical Allocation: Active Risk/Return Perspective
The objective of TAA is then to minimize the benchmark tracking errors, while at the
same time outperforming the benchmark on average by a certain amount.
The problem to solve is:
8 8
max 𝐸 [𝑒$%" ] − 𝜎 , [𝑒$%" ] = α+6 𝜇 − 𝛼6+ Σ 𝛼6
\ 7! , ,
s. t. α+6 𝑒 = 0 (sum equal to 0)

Proposition: The weights of the active portfolio bets are:

e ' S -1 µ ö 1 S ( µ e '- eµ ') S
-1 -1
1 æ
a *
= S çµ -
-1
e÷ = e
act
l è e' S e ø l
-1
e' S e-1

Properties:
- If the benchmark portfolio is (a gmv + a S ), the total portfolio is a = a gmv + a S + a act .
- In general, α+6 ≠ 𝛼 # . Hence, the TAA under the active risk/return perspective is
usually not mean-variance efficient.
9 [("#$ ] % ?
=!
- The information ratio is defined as 𝐼𝑅5 = =
< [("#$ ] % A=
@= ! !

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 36/48
 Objectives of the Lecture

- Portfolio Optimization

- Benchmarking versus Active Management

è Black-Litterman Approach

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 37/48
 Black-Litterman Approach
Managers may have a passive benchmark (say, the market) but use their own views for
designing an active strategy.

Managers may be reluctant to use historical average returns for their neutral view because

- sample means are too noisy

- sample means are not neutral, as they favor past winners.

Instead, priors (“neutral” views) about risk premia are derived from market equilibrium.

The Black-Litterman approach combines:

- the managers’ subjective views regarding the expected returns

- with the market equilibrium expected returns (prior distribution)
- to form a new estimate of expected returns (posterior distribution)

This approach is an application of the mixed estimation approach: it attempts to improve

estimates of mean returns by combining information from data with the investor’s priors.

We focus on the market equilibrium expected returns

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 38/48
 The Logic
Intuition:
- if you do not have views, you should hold the market portfolio
- your views will tilt the final weights away from the market portfolio, the degree to
which depending on how confident you are about your views.

Market weights: Research: Views

Equilibrium returns with confidence

Black-Litterman model:
consistent expected returns

Portfolio Optimizer:
asset weights

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 39/48
 The Logic

Risk Covariance Market Cap Views Uncertainty

Aversion 𝜆 Matrix S weights wmkt v W
Tapez une équation ici.
\lambda

Implied Equilibrium Return Vector

𝜇(BC*D = 𝜆 Σ 𝑤2E$

Prior Equilibrium Distribution View Distribution

µ ~ MVN ( µequil ,t S) v ~ MVN ( P ´ µ , W)

New Combined Distribution

E[ µ | v] = [(t S)-1 + P ' W-1 P]-1[(t S) -1 µequil + P ' W-1v]

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 40/48
 CAPM Equilibrium Risk Premia
Excess return distribution: Excess returns are drawn from an i.i.d. normal distribution:

𝑅 − 𝑅& ~𝑀𝑉𝑁(𝜇( , Σ) with 𝜇( : risk premium (or expected excess return)

In equilibrium, all investors hold the market portfolio.

Equilibrium risk premia: Premia that induce a risk-averse investor to hold all available
assets in proportion to their current market capitalizations.

With mean-variance utility (Tobin’s approach): 𝑤2E$ vector of market cap weights
" ( (
𝑤2E$ = Σ /" 𝜇(BC*D Þ 𝜇(BC*D = 𝜆 Σ 𝑤2E$
8

What value for 𝜆?

% 7
+ ( + F56" ?789:; ?56"
Suggestion: 𝑤2E$ 𝜇(BC*D =𝜆 𝑤2E$ Σ 𝑤2E$ so that 𝜆 = GF % H
= <
56" A F56" <56"

𝜆 is often calibrated to the historical Sharpe ratio (e.g., 𝜆=3)

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 41/48
 CAPM Equilibrium Risk Premia
Illustration
Monthly returns for 6 portfolios of all AMEX, NASDAQ, and NYSE stocks sorted by
their market capitalization and book-to-market ratio from 1983 to 2003. (Source: Brandt,
2010)

Descriptive statistics on portfolio returns

Size Book to Sample risk Volatility Correlations
market premia (%) (%)
Small Low 5.61 24.56 1
Small Medium 12.75 17.01 0.926 1
Small High 14.36 16.46 0.859 0.966 1
Big Low 9.72 17.07 0.784 0.763 0.711 1
Big Medium 10.59 15.05 0.643 0.768 0.763 0.847 1
Big High 10.44 13.89 0.555 0.698 0.735 0.753 0.913

Remark: Sample risk premia are based on the sample means of portfolio returns, so
there are extremely noisy. The Small-Low portfolio has the lowest risk premium while it
has the highest volatility. Equilibrium risk premia are biased estimates of the risk
premia (because they are based on a model), but they are more precisely estimated.
MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 42/48
 CAPM Equilibrium Risk Premia
Illustration
Equilibrium risk premia implied by market capitalization weights
Equilibrium risk premia (%)
Size Book to Market Historical risk
𝜆 = 1 𝜆 = 2.5 𝜆=5 𝜆 = 7.5
market weight (%) premia (%)
Small Low 2.89 3.07 7.69 15.37 23.06 5.61
Small Medium 3.89 2.21 5.52 11.03 16.55 12.75
Small High 2.21 2.04 5.11 10.22 15.33 14.36
Big Low 59.07 2.62 6.55 13.10 19.64 9.72
Big Medium 23.26 2.18 5.44 10.88 16.32 10.59
Big High 8.60 1.97 4.91 9.83 14.74 10.44

Comments:
- The levels of the risk premia depend on the level of risk aversion, which therefore
needs to be calibrated before using the results in the mixed estimator.
- The implied equilibrium risk premia are quite different from the empirical risk
premia. The two sets of risk premia are negatively correlated in the cross-section (a
correlation coefficient of −0.83).
MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 43/48
 Illustration
Idzorek (2004), AllocationADVISOR and the Black-Litterman Model.

International markets.

Descriptive statistics on asset classes returns

Asset classes Historical Market Risk Risk-free Total
return cap premia rate implied
(%) (%) (%) (%) return (%)
US bonds 6.8 20.2 0.1 4.0 4.1
Global bonds xUSD 4.7 27.9 1.0 4.0 5.0
World Equity xUSD 5.4 22.2 4.0 4.0 8.0
Emerging Equity 6.5 2.33 5.4 4.0 9.4
US Large cap Growth 6.3 12.6 5.1 4.0 9.1
US Large cap Value 9.8 12.6 3.7 4.0 7.7
US Small cap Growth 6.0 1.1 6.1 4.0 10.1
US Small cap Value 12.2 1.1 3.5 4.0 7.5

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 44/48
 Illustration
Efficient frontier based on Historical returns and covariance matrix

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 45/48
 Illustration
Optimal weights based on Historical returns and covariance matrix

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 46/48
 Illustration
Efficient frontier based on implied returns and historical covariance matrix

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 47/48
 Illustration
Optimal weights based on implied returns and historical covariance matrix

MScF (2023-24) Pr. Eric Jondeau – Sustainability Aware Asset Management 48/48