Lecture 3: The Single-Index Model

Jean-Gabriel Attali1
1 Leonard de Vinci Graduate School of Engineering (ESILV)

J.G. Attali (ESILV) Single-Index Model Academic year 2023-2024 1 / 22

Table of Contents

1 The Single-Index Model

Introduction to Factor Models
Specification
Estimation

2 The Model in Practice

Numerical Examples : Individual Stocks
Numerical Examples : Portfolios
Conclusion

J.G. Attali (ESILV) Single-Index Model Academic year 2023-2024 2 / 22

 The Single-Index Model Introduction to Factor Models

Table of Contents

1 The Single-Index Model

Introduction to Factor Models
Specification
Estimation

2 The Model in Practice

Numerical Examples : Individual Stocks
Numerical Examples : Portfolios
Conclusion

J.G. Attali (ESILV) Single-Index Model Academic year 2023-2024 3 / 22

 The Single-Index Model Introduction to Factor Models

Introduction

Universe
1 risk-free asset and n risky assets.

I Inputs Estimation of MPT requires a large number of observations :

1 n returns to be estimated.
n(n + 1)
2 independent parameters to fill the covariance matrix.
2
I Factor Models lead to a more robust estimation of the covariance
matrix.
I Main idea is to decompose risk into systematic risk and specific (or
idosyncratic) risk.
I Also simplifies the understanding (if realistic).

J.G. Attali (ESILV) Single-Index Model Academic year 2023-2024 4 / 22

 The Single-Index Model Introduction to Factor Models

The Single-Factor Model

I A Single-Factor Model is defined by:

Ri = αi + βi F + εi
where
R = (R1 , · · · , Rn )0 is the vector of risky-asset returns.
F is a macroeconomic factor (e.g. GDP growth)
βi is the response of return i to the common factor F .
εi is the firm specific.

Hypothesis:
(F , ε1 , · · · , εn ) are independent variables

J.G. Attali (ESILV) Single-Index Model Academic year 2023-2024 5 / 22

 The Single-Index Model Specification

Table of Contents

1 The Single-Index Model

Introduction to Factor Models
Specification
Estimation

2 The Model in Practice

Numerical Examples : Individual Stocks
Numerical Examples : Portfolios
Conclusion

J.G. Attali (ESILV) Single-Index Model Academic year 2023-2024 6 / 22

 The Single-Index Model Specification

The Single-Index Model

I The macroeconomic factor F is generally unobservable (Lag, low
frequency, computational errors, etc.).
I The idea behind the Single-Index Model is to use a/the market index
as a proxy for F .
I The Single-Index Model is defined by:

Ri = αi + βi RIndex + εi
where
RIndex is the return of a (Benchmark) Index (S&P 500, Euro Stoxx 50,
etc.)
βi is the response of an individual security’s return to the benchmark.
εi still is the firm specific.

Hypothesis
(RIndex , ε1 , · · · , εn ) are independent variables
J.G. Attali (ESILV) Single-Index Model Academic year 2023-2024 7 / 22
 The Single-Index Model Specification

Computation of Returns and Covariance Matrix

I Beta-relationship for expected returns:

E(Ri ) = αi + βi E(RIndex )
I Covariance matrix of risky-assets in the Single-Index Model:

2
Σ = σIndex ββ 0 + Σε
| {z } |{z}
Systematic Risk Specific Risk

where
β = (β1 , · · · , βn )0
 2 
σ (ε1 ) (0)
Σ =
 .. 
ε . 
(0) σ 2 (εn )

J.G. Attali (ESILV) Single-Index Model Academic year 2023-2024 8 / 22

 The Single-Index Model Specification

Diversification in the Index Model

I Variance of a Portfolio of risky-assets with vector-weights X :

n
!2 n
X X
0 2 0 0 0 2
X ΣX = σIndex X ββ X + X Σε X = σIndex βi Xi + σ 2 (εi )Xi2
i=1 i=1

I Variance of the Equal-Weighted Portfolio:

n
!2 n
!
2 2 1 X 1 1 X
σEW = σIndex βi + σ 2 (εi )
n n n
| i=1
{z } | i=1{z }
β̄ 2 σ̄ 2

Diversification of Specific Risks

1 2
When n gets large, σ̄ → 0
n
J.G. Attali (ESILV) Single-Index Model Academic year 2023-2024 9 / 22
 The Single-Index Model Estimation

Table of Contents

1 The Single-Index Model

Introduction to Factor Models
Specification
Estimation

2 The Model in Practice

Numerical Examples : Individual Stocks
Numerical Examples : Portfolios
Conclusion

J.G. Attali (ESILV) Single-Index Model Academic year 2023-2024 10 / 22

 The Single-Index Model Estimation

Regression Equation

I Single Index-Model:

Ri = αi + βi RIndex + εi
where ε ∼ N (0n , Σε ).

Estimation and Interpretation

• Feasible-GLS to estimate the parameters
• R2 to measure the goodness of fit
• t-Student for significance

Robustness of Estimation
Non-normality of returns may be an issue for small samples

J.G. Attali (ESILV) Single-Index Model Academic year 2023-2024 11 / 22

 The Single-Index Model Estimation

Number of estimates
I Initial number of estimates:
n expected returns.
n(n+1)
2 independent parameters in the covariance matrix.
I Number of paramaters to estimate in the Index-Model:
2
RIndex and σIndex .
n intercepts αi .
n sensitivity βi .
n variances σ 2 (εi ).

n(n + 1)
Downsizing from n + to 3n + 2!
2

Examples
• 122 instead of 860 for French equities
• 1502 instead of 125750 for US equities
J.G. Attali (ESILV) Single-Index Model Academic year 2023-2024 12 / 22
 The Model in Practice Numerical Examples : Individual Stocks

Table of Contents

1 The Single-Index Model

Introduction to Factor Models
Specification
Estimation

2 The Model in Practice

Numerical Examples : Individual Stocks
Numerical Examples : Portfolios
Conclusion

J.G. Attali (ESILV) Single-Index Model Academic year 2023-2024 13 / 22

 The Model in Practice Numerical Examples : Individual Stocks

US Equity Market (β > 1)

I Stock: Apple.
I Market Capitalization: 3000b$.
I Index of the Model: S&P 500.
I Estimation made on weekly returns over 4 years.

Statistics

βApple = 1.096 significant

αApple = 0.260 not significant
R2 = 60.7%

J.G. Attali (ESILV) Single-Index Model Academic year 2023-2024 14 / 22

 The Model in Practice Numerical Examples : Individual Stocks

US Equity Market (β < 1)

I Stock: Coca-Cola.
I Market Capitalization: 256b$.
I Index of the Model: S&P 500.
I Estimation made on weekly returns over 4 years.

Statistics

βCoke = 0.804 significant

αCoke = −0.109 not significant
R2 = 51.1%

J.G. Attali (ESILV) Single-Index Model Academic year 2023-2024 15 / 22

 The Model in Practice Numerical Examples : Individual Stocks

French Equity Market (β > 1)

I Stock: LVMH.
I Market Capitalization: 386be
I Index of the Model: CAC 40.
I Estimation made on weekly returns over 4 years.

Statistics

βLVMH = 1.060 significant

αLVMH = 0.147 not significant
R2 = 68.6%

J.G. Attali (ESILV) Single-Index Model Academic year 2023-2024 16 / 22

 The Model in Practice Numerical Examples : Individual Stocks

French Equity Market (β < 1)

I Stock: Air Liquide.

I Market Capitalization: 90be
I Index of the Model: CAC 40.
I Estimation made on weekly returns over 4 years.

Statistics

βAir Liquide = 0.676 significant

αAir Liquide = 0.128 not significant
R2 = 49.3%

J.G. Attali (ESILV) Single-Index Model Academic year 2023-2024 17 / 22

 The Model in Practice Numerical Examples : Portfolios

Table of Contents

1 The Single-Index Model

Introduction to Factor Models
Specification
Estimation

2 The Model in Practice

Numerical Examples : Individual Stocks
Numerical Examples : Portfolios
Conclusion

J.G. Attali (ESILV) Single-Index Model Academic year 2023-2024 18 / 22

 The Model in Practice Numerical Examples : Portfolios

US Equity Market (β > 1)

I Fund: JP Morgan Large Cap Growth.
I AUM (Asset Under Management): 77b$.
I Index of the Model: S&P 500.
I Estimation made on daily returns over 4 years.

Statistics

βJPM = 1.059 significant

αJPM = 0.001 not significant
R2 = 60.7%

J.G. Attali (ESILV) Single-Index Model Academic year 2023-2024 19 / 22

 The Model in Practice Numerical Examples : Portfolios

US Equity Market (β < 1)

I Holding: Berkshire Hataway.
I Market Capitalization: 830b$.
I Index of the Model: S&P 500.
I Estimation made on daily returns over 4 years.

Statistics

βBuffett = 0.77 significant

αBuffett = 0.025 not significant
R2 = 60.7%

J.G. Attali (ESILV) Single-Index Model Academic year 2023-2024 20 / 22

 The Model in Practice Conclusion

Table of Contents

1 The Single-Index Model

Introduction to Factor Models
Specification
Estimation

2 The Model in Practice

Numerical Examples : Individual Stocks
Numerical Examples : Portfolios
Conclusion

J.G. Attali (ESILV) Single-Index Model Academic year 2023-2024 21 / 22

 The Model in Practice Conclusion

Comments and Drawbacks

I Index-Model is very useful to practitioners:

It gives a better understanding of the market.
It reduces the numbers of parameters required for estimation.
I The hypothesis of normality of returns is questionable. One can
generalize to non-normal returns but:
Difficulty to write the Maximum Likelihood Program.
Numerical Implementation needed.
t-Student and R2 not valid for interpretation for small samples.
I Still no guidelines for expected returns (cf Lecture 6).
I Market Structure may be much more complex (cf Lecture 5).
I Lecture 4 to follow:
Is the Benchmark (S&P 500, etc.) the best regressor?
What about the α ?

J.G. Attali (ESILV) Single-Index Model Academic year 2023-2024 22 / 22