INDIAN INSTITUTE OF TECHNOLOGY KANPUR

Lesson 1: Introduction to CAPM

Advanced Algorithmic Trading and Portfolio Management
Introduction

• Assumptions with CAPM

• Capital market line (CML)
• Security market line (SML)
• Fallings of CAPM
• Summary and concluding remarks
 INDIAN INSTITUTE OF TECHNOLOGY KANPUR

Assumptions with CAPM

Capital Asset Pricing Model (CAPM)

The standard form of CAPM equilibrium relation is first shown by

Sharpe, Lintner, and Mossin. Hence, it is also referred to as the
Sharpe–Lintner–Mossin model of CAPM (1960s)
• It is the simplest and most widely employed model of asset pricing
• It has been documented to be extremely efficient in explaining the
observed prices
• It involves some important assumptions
CAPM: Assumptions

No transaction costs: what are these transaction costs?

Securities are infinitely divisible: one can take as small a position as
INR 1.
Prices are given: traders cannot affect prices
Investors are rational: they understand the return distributions and
risk and also process all the available information
CAPM: Assumptions

Unlimited short sales are allowed

Unlimited lending and borrowing is allowed
Uniform expectations: At equilibrium, all the investors have the
same expectation of a security’s return distribution (i.e., expected
return, risk, and correlation structure across securities); they define
the period of equilibrium in a similar manner
All the assets are marketable
 INDIAN INSTITUTE OF TECHNOLOGY KANPUR

A Simple Approach to Understand CAPM

:Part I Capital Market Line (CML)
A Simple Approach to Understand CAPM: CML

Recall our old story of one risky

asset (market portfolio) in the
presence of risk-free lending and
borrowing

Elton, Gruber, Brown, and Goetzmann; Modern Portfolio Theory and Investment Analysis. 9th edition. Chapter 13
Capital Market Line (CML)

• We said (under the assumptions

specified) that all the investors will
hold this portfolio along with the
risk-free asset (investing or
borrowing)
• This line is called the capital
market line (CML)

Elton, Gruber, Brown, and Goetzmann; Modern Portfolio Theory and Investment Analysis. 9th edition. Chapter 13
Capital Market Line (CML)

The equation of this line is as follows

(𝑅ത 𝑀 −𝑅𝐹 )
• 𝑅ത𝑒 = 𝑅𝐹 + 𝜎𝑒 where
𝜎𝑀
subscript “e” denotes an efficient
portfolio
(𝑅ത 𝑀 −𝑅𝐹 )
• The term indicates the
𝜎𝑀
price of risk, i.e., excess returns
per unit of risk

Elton, Gruber, Brown, and Goetzmann; Modern Portfolio Theory and Investment Analysis. 9th edition. Chapter 13
Capital Market Line (CML)

• The combined term

(𝑅ത 𝑀 −𝑅𝐹 )
[ 𝜎𝑒 ] is the total reward
𝜎𝑀
for taking on 𝜎𝑒 risk
• 𝑅𝐹 is simply the risk-free rate
that is the price of time, which
is delaying the consumption
(time value of money)

Elton, Gruber, Brown, and Goetzmann; Modern Portfolio Theory and Investment Analysis. 9th edition. Chapter 13
Capital Market Line (CML)

• Therefore, the equation can be

simply written as follows:
Expected return = Price of time
+ Price of risk × Risk

Elton, Gruber, Brown, and Goetzmann; Modern Portfolio Theory and Investment Analysis. 9th edition. Chapter 13
 INDIAN INSTITUTE OF TECHNOLOGY KANPUR

A Simple Approach to Understanding

the CAPM II: Security Market Line (SML)
Security Market Line (SML)

• Security market line carries all the

securities available in the market
• If all the investors hold well-
diversified portfolios, then only
risk that matters for a security is
beta or market risk
• Imagine five portfolios, i.e., A, B,
C, D, and E, on SML

Elton, Gruber, Brown, and Goetzmann; Modern Portfolio Theory and Investment Analysis. 9th edition. Chapter 13
SML: Arbitrage Portfolio
Investment Expected Return Beta Portfolio
A 10% 1 Efficient
B 12% 1.4 Efficient
D 13% 1.2 Inefficient
E 8% 1.2 Inefficient
C (average of A and B) 11% 1.2 Efficient

Arbitrage portfolio
Sell C −11% −1.2
Buy D 13% 1.2
Expected return 2% 0

Arbitrage portfolio
Buy C 11% 1.2
Sell E −8% −1.2
Expected return 3% 0
Security Market Line (SML)

For a well-diversified portfolio, non-

systematic risk tends to go to zero

• Market risk is the only relevant risk

measured by beta

• The SML shown here plots five

portfolios (A, B, C, D, and E)

• Here, portfolios D and E are anomalous,

i.e., their expected return is not aligned
to the systematic-risk (beta)

Elton, Gruber, Brown, and Goetzmann; Modern Portfolio Theory and Investment Analysis. 9th edition. Chapter 13
Security Market Line (SML)

For a well-diversified portfolio, non-

systematic risk tends to go to zero
• E is overpriced and, therefore,
offers a lower expected return
• D is underpriced and, therefore,
offers a higher expected return

Elton, Gruber, Brown, and Goetzmann; Modern Portfolio Theory and Investment Analysis. 9th edition. Chapter 13
Security Market Line (SML)

For a well-diversified portfolio, non-

systematic risk tends to go to zero
• There is a (partially) riskless
arbitrage opportunity by selling E
and buying D and makes profits
• This will bring securities back to
the SML

Elton, Gruber, Brown, and Goetzmann; Modern Portfolio Theory and Investment Analysis. 9th edition. Chapter 13
Security Market Line (SML)

• SML can be identified using the

two points through which it
passes
• First, the risk-free investment
(beta = 0 and interest rate of
𝑅𝐹 ), and second, the market
portfolio (beta = 1 and interest
rate of 𝑅ത𝑀 )

Elton, Gruber, Brown, and Goetzmann; Modern Portfolio Theory and Investment Analysis. 9th edition. Chapter 13
Security Market Line (SML)

Using these points, we can write

down the equation of SML as
• 𝑅ത𝑖 = 𝑅𝐹 + 𝛽𝑖 (𝑅ത𝑀 − 𝑅𝐹 ) →
CAPM
𝜎𝑖𝑚
• Here, 𝛽𝑖 = 2
𝜎𝑚

Elton, Gruber, Brown, and Goetzmann; Modern Portfolio Theory and Investment Analysis. 9th edition. Chapter 13
 INDIAN INSTITUTE OF TECHNOLOGY KANPUR

Fallings of CAPM
CAPM Assumption Violations

• No transaction costs
• Securities are infinitely divisible
• Prices are given
• Investors are rational
CAPM Assumption Violations

• Unlimited short sales are allowed

• Unlimited lending and borrowing is allowed
• Uniform expectations
• All the assets are marketable
Few Last Words on CAPM

• CAPM appears to hold at an aggregate level

• However, individual investors do hold smaller portfolios, not
similar to market portfolios
• Many CAPM assumptions violate the real-world conditions
• However, there are certain assumptions that can be relaxed and
alternative variants can be derived
Few Last Words on CAPM

• Under the assumptions of CAPM, the only portfolio of risky assets

that investors will hold will be the market portfolio
• In this market portfolio, any security has a proportion that is the
same as the ratio of the market capitalization of that security to
the total market capitalization of that market
Few Last Words on CAPM

• Investors depending upon their risk tolerance will adjust the

proportions of the market portfolio and risk-free asset
• However, we know that individual investors do hold non-market,
smaller portfolios
 INDIAN INSTITUTE OF TECHNOLOGY KANPUR

Summary and Concluding Remarks

Summary and Concluding Remarks

• CAPM is a very simple yet powerful model of equilibrium asset pricing

• CAPM is based on certain assumptions that violate real-life situations
• However, its efficacy lies in its ability to describe the real observed
prices
• All the efficient portfolios lie on the capital market line (CML)
• The CML describes equilibrium prices in terms of price of time and
price of risk
Summary and Concluding Remarks

• Security market line (SML) describes the behavior of all the

securities available in the market at equilibrium
• Essentially, this SML is the equation of CAPM
• It passes through the market portfolio and risk-free security
Summary and Concluding Remarks

• Various assumptions of CAPM violate the real-world scenarios

• CAPM postulates that all individuals should hold the market
portfolio
• However, individual investors hold various small portfolios that are
different from the market portfolio
• The violation of a few assumptions individually may not
necessarily invalidate CAPM
Thanks!
 INDIAN INSTITUTE OF TECHNOLOGY KANPUR

Lesson 2: Single- and Multi-Index

Models
Advanced Algorithmic Trading and Portfolio Management
Introduction

• Single-index models and correlation structure

• Construction of single-index models
• Portfolio characteristics with single-index models
• Estimation of portfolio beta with single-index models
• Single-index models: simple example
Introduction

• A few words on beta

• Introduction to multi-index models
• Multi-index models: expected return and risk
• Summary and concluding remarks
 INDIAN INSTITUTE OF TECHNOLOGY KANPUR

Single-Index Models and Correlation

Structure
Single-Index Models and Correlation Structure

These are the equations corresponding to portfolio returns and standard

deviation
• 𝑅ത𝑃 = σ𝑁 ത
𝑖=1 𝑋𝑖 𝑅𝑖 (1)
• 𝜎𝑝2 = σ𝑁 𝑗=1 𝑘=1(𝑋𝑗 𝑋𝑘 𝜎𝑗𝑘 ) where i≠j (2)
2 2 σ 𝑁 σ𝑁
𝑋 𝜎
𝑖=1 𝑖 𝑖 +
• In order to draw an efficient frontier, three key inputs are required
• Expected returns from each security
• Standard deviations from each security
• Correlations between each possible pair of security
Single-Index Models and Correlation Structure

If an analyst follows 150 stocks, how many estimates he/she requires?

• 150 estimates of expected returns and 150 estimates of standard deviation
but, in addition, she also needs 150 × 149/2 = 11,175 estimates of
covariance (or correlations)
• What if one factor or index affected all these 150 securities?
• That means the observed covariances essentially reflected the correlation
structure between that index and these securities
• This leads to the genesis of single-index models
Single-Index Models and Correlation Structure

Single-index model assumes a single common influence that affects a large

number of securities in a similar manner: 𝑅𝑖 = 𝑎𝑖 + 𝛽i 𝑅𝑚 + 𝑒𝑖 (3)
• This is a more data-driven model
• Researchers in the early days realized that market movements affect a large
number of stocks in a similar manner
• Indices like Nifty affect the returns on a large number of securities
 INDIAN INSTITUTE OF TECHNOLOGY KANPUR

Construction of Single-Index Models

Construction of Single-Index Models

Single-index model: 𝑅𝑖 = 𝑎𝑖 + 𝛽𝑖 𝑅𝑚 + 𝑒𝑖
• Both 𝑅𝑚 and 𝑒𝑖 are random variables
• Random variables are defined by a probability distribution with a mean and
standard deviation
• Mean of 𝑅𝑚 and 𝑒𝑖 are 𝑅ത𝑚 and 0, whereas standard deviations of 𝑅𝑚 and 𝑒𝑖
are 𝜎𝑚 and 𝜎𝑒𝑖 , respectively
• Here, by definition 𝑅𝑚 and 𝑒𝑖 are uncorrelated:
Cov 𝑒𝑖 , 𝑅𝑚 = 𝐸 𝑒𝑖 − 0 𝑅𝑚 − 𝑅ത𝑚 =0
• The model is generally estimated using regression analysis
Construction of Single-Index Models

Single-index model also assumes that 𝑒𝑖 is independent of all 𝑒𝑗 s: More

formally, E(𝑒𝑖 𝑒𝑗 ) = 0

• This means that the only reason two stocks commove is because of market;
no other effects, such as industry
• This is not ensured by the regression analysis
• Thus, the performance of the model depends how good this assumption is

• 𝑅𝑖 = 𝑎𝑖 + 𝛽𝑖 𝑅𝑚 + 𝑒𝑖 under the assumption of single-index model is assumed

to represent the return dynamics for all the stocks, where i = 1,2,3,…,N.
Construction of Single-Index Models

Under the assumption of a single-index model, this equation is assumed to represent

the return dynamics for all the stocks, where i =1,2,3,…,N : 𝑹𝒊 = 𝒂𝒊 + 𝜷𝒊 (𝑹𝒎 ) + 𝒆𝒊
• By the design (or construction) of the regression model. Mean of 𝑒𝑖 , i.e., E(𝑒𝑖 ) = 0.
• By assumption, index (market) is unrelated to the idiosyncratic-specific component
(𝑒𝑖 ), that is, 𝐸 𝑒𝑖 𝑅𝑚 − 𝑅𝑚 = 0
• By assumption, securities are only related to each other through the index (market).
That is, 𝐸 𝑒𝑖 𝑒𝑗 = 0
• By definition, Variance of 𝑒𝑖 = 𝐸 𝑒𝑖 2 2
= 𝜎𝑒𝑖
• By definition, Variance of 𝑅𝑚 = 𝐸 𝑅𝑚 − 𝑅ത𝑚 2 2
= 𝜎𝑚
Construction of Single-Index Models

Now that we have boundary conditions, we can derive the

expressions for expected return, standard deviation, and covariance
• Expected returns:
𝐸 𝑅𝑖 = 𝐸[𝑎𝑖 + 𝛽𝑖 𝑅𝑚 + 𝑒𝑖 ] =𝐸(𝑎𝑖 ) + 𝐸(𝛽𝑖 𝑅𝑚 ) + 𝐸(𝑒𝑖 )
• 𝐸(𝑒𝑖 ) = 0, and that 𝑎𝑖 and 𝛽𝑖 are constants
• 𝐸 𝑅𝑖 = 𝑅ത𝑖 = 𝑎𝑖 + 𝛽𝑖 𝑅ത𝑚
Construction of Single-Index Models

Standard deviation (𝜎𝑖2 )

• 𝜎𝑖2 = 𝐸 𝑅𝑖 − 𝑅ത𝑖 2
= 𝐸[ 𝑎𝑖 + 𝛽𝑖 𝑅𝑚 + 𝑒𝑖 − 𝑎𝑖 + 𝛽𝑖 𝑅ത𝑚 ]2

• 𝜎𝑖2 = 𝐸[𝛽𝑖 (𝑅𝑚 − 𝑅ത𝑚 ) + 𝑒𝑖 ]2 = 𝛽𝑖2 𝐸( 𝑅𝑚 − 𝑅ത𝑚 2

+ 𝐸 𝑒𝑖 2
+ 2𝛽𝑖 𝐸[𝑒𝑖 𝑅𝑚 − 𝑅ത𝑚 ]

• 𝜎𝑖2 = 𝛽𝑖2 𝐸( 𝑅𝑚 − 𝑅ത𝑚 2 + 𝐸 𝑒𝑖 2 because 𝐸 𝑒𝑖 𝑅𝑚 − 𝑅ത𝑚 =0

• 𝜎𝑖2 = 𝛽𝑖2 𝜎𝑚
2 + 𝜎2
𝑒𝑖
 INDIAN INSTITUTE OF TECHNOLOGY KANPUR

Portfolio Characteristics with Single-

Index Models
Portfolio Characteristics with Single-
Index Models
With the assumption that a single-index model holds, let us examine its impact on portfolio
returns and standard deviation
Expected return

𝑖=1 𝑋𝑖 𝑅𝑖 ; substitute the single-index model 𝑅𝑖 = 𝑎𝑖 + 𝛽𝑖 𝑅𝑚

• 𝑅ത𝑝 = σ𝑁 ത ത ത

• 𝑅ത𝑝 = σ𝑁 𝑁 ത
𝑖=1 𝑋𝑖 𝑎𝑖 + σ𝑖=1 𝑋𝑖 𝛽𝑖 𝑅𝑚

Standard deviation
𝑁
𝑖=1 𝑋𝑖 𝜎𝑖 + σ𝑖=1 σ𝑗=1 𝑋𝑖 𝑋𝑗 𝜎𝑖𝑗 ; substituting the expression for variance and
• 𝜎𝑝2 = σ𝑁 2 2 𝑁

i≠j
covariance
𝑁 σ𝑁
• 𝜎𝑝2 = σ𝑁 𝑋 2 2 2
𝛽
𝑖=1 𝑖 𝑖 𝜎𝑚 + σ𝑖=1 𝑗=1 𝑋𝑖 𝑋𝑗 𝛽𝑖 𝛽𝑗 𝜎𝑚
2 + σ𝑁 𝑋 2 𝜎 2
𝑖=1 𝑖 𝑒𝑖
𝑖≠𝑗
Portfolio Characteristics with Single-
Index Models
Assume we have a portfolio of 150 stocks and we require estimates of
(1) 𝑎𝑖 , 𝛽𝑖 , and 𝜎𝑒𝑖 for each of the stock
(2) 𝑅ത𝑚 and 𝜎𝑚
2
for the market
That is, 150 × 3 + 2 = 452 estimates are needed (as compared to 11,485
estimates in the absence of a single-index model)
Portfolio Characteristics with Single-
Index Models
Portfolio expected return
• 𝑅ത𝑝 = σ𝑁 𝑋 𝑎
𝑖=1 𝑖 𝑖 + σ 𝑁 ത
𝑖=1 𝑋𝑖 𝛽𝑖 𝑅𝑚

• 𝛽𝑝 = σ𝑁 𝑋 𝛽
𝑖=1 𝑖 𝑖 ; 𝑎𝑝 = σ 𝑁
𝑖=1 𝑋𝑖 𝑎𝑖

• 𝑅ത𝑝 = 𝑎𝑝 + 𝛽𝑝 𝑅ത𝑚

• Please note that if the portfolio under consideration is the market portfolio,
then 𝑎𝑝 = 0 and 𝛽𝑝 = 1. Then, 𝑅ത𝑝 = 𝑅ത𝑚 . Thus, stocks with 𝛽𝑝 >1 are said to
be riskier than the market, and stocks with 𝛽𝑝 < 1 are said to be less risky
than the market
Portfolio Characteristics with Single-
Index Models
Portfolio standard deviation
𝑁 σ𝑁
• 𝜎𝑝2 = σ𝑁 𝑋 2 2 2
𝛽 𝜎
𝑖=1 𝑖 𝑖 𝑚 + σ 𝑋 𝑋 𝛽 𝛽 𝜎
𝑖=1 𝑗=1 𝑖 𝑗 𝑖 𝑗 𝑚
2 + σ𝑁 𝑋 2 𝜎 2
𝑖=1 𝑖 𝑒𝑖
𝑖≠𝑗

• 𝜎𝑝2 = σ𝑁 𝑁 2 𝑁 2 2
𝑖=1 σ𝑗=1 𝑋𝑖 𝑋𝑗 𝛽𝑖 𝛽𝑗 𝜎𝑚 + σ𝑖=1 𝑋𝑖 𝜎𝑒𝑖

• 𝜎𝑝2 = (σ𝑁 𝑋 𝛽
𝑖=1 𝑖 𝑖 )( σ 𝑁
𝑋 𝛽 )𝜎
𝑗=1 𝑗 𝑗 𝑚
2 + σ𝑁 𝑋 2 𝜎 2
𝑖=1 𝑖 𝑒𝑖

• But (σ𝑁 𝑋 𝛽
𝑗=1 𝑖 𝑖 ) = (σ 𝑁
𝑗=1 𝑋𝑗 𝛽𝑗 ) = 𝛽𝑝
2 + σ𝑁 𝑋 2 𝜎 2
• 𝜎𝑝2 = 𝛽𝑝2 𝜎𝑚 𝑖=1 𝑖 𝑒𝑖
1
• Consider equal investments in the securities so that 𝑋1 = 𝑋2 = ⋯ = 𝑋𝑁 =
𝑁
Characteristics of Single-Index Model

Portfolio standard deviation

2 + σ𝑁 𝑋 2 𝜎 2
• 𝜎𝑝2 = 𝛽𝑝2 𝜎𝑚 𝑖=1 𝑖 𝑒𝑖
1
• Consider equal investments in the securities so that 𝑋1 = 𝑋2 = ⋯ = 𝑋𝑁 =
𝑁
2
𝜎𝑒𝑖
• If there are a large number of securities, then the term , which represents
𝑁
the residual (or specific risk), approaches to zero
• 𝜎𝑝2 = 𝛽𝑝2 𝜎𝑚
2
 INDIAN INSTITUTE OF TECHNOLOGY KANPUR

Estimation of Portfolio Beta with

Single-Index Models
Market Model

• Consider 𝑅ത𝑖 = 𝑎𝑖 + 𝛽𝑖 𝑅ത𝑚 index model. When the assumption of Cov 𝑒𝑖 𝑒𝑗 = 0

is waived, then it becomes the market model
• This allows for comovement across securities because of factors other than
the market
• This means it is a less restrictive form of index model family
• It suggests that there are additional systematic marketwide factors that can
also affect the individual securities
• What can be these factors?
Estimation of Portfolio Beta

We had discussion about the

estimation of betas!
𝑅𝑖𝑡 = 𝑎𝑖 + 𝛽𝑖 𝑅𝑚𝑡 + 𝑒𝑖𝑡
Here, we are fitting a line across the
scatter points of 𝑅𝑖 and 𝑅𝑚
observations, available over time
The slope of this line is the best
estimate of the beta over the period of
examination
Estimation of Portfolio Beta

𝑅𝑖 = 𝑎𝑖 + 𝛽𝑖 𝑅𝑚 + 𝑒𝑖
Here, we are fitting a line across the
scatter points of 𝑅𝑖 and 𝑅𝑚
observations, available over time
The slope of this line is the best
estimate of the beta over the period of
examination
𝜎𝑖𝑚 σ𝑇 ത 𝑖𝑡 )(𝑅𝑚𝑡 −𝑅ത𝑚𝑡 )]
𝑡=1[(𝑅𝑖𝑡 −𝑅
𝛽𝑖 = 2 = 2
𝜎𝑚 σ𝑇 ത
𝑡=1 𝑅𝑚𝑡 −𝑅𝑚𝑡
Estimation of Portfolio Beta

• But beta estimates are also subject to estimation errors

• Also, firm betas change over time (changes in capital structure,
industry, etc.)
• Therefore, analysts estimate betas of industry portfolios
• These are less noisy and more reliable estimates
• The random variation in one security (upwards) and the other
security (downwards) tend to cancel out each other
 INDIAN INSTITUTE OF TECHNOLOGY KANPUR

Single-Index Models: Simple Example

Simple Example

Period 𝑹𝒊𝒕 𝑹𝒎𝒕 ഥ 𝒊𝒕 )(𝑹𝒎𝒕 − 𝑹

(𝑹𝒊𝒕 − 𝑹 ഥ 𝒎𝒕 ) Value
1 10 4 (10 − 8) × (4 − 4) 0
2 3 2 (3 − 8) × (2 − 4) 10
3 15 8 (15 − 8) × (8 − 4) 28
4 9 6 (9 − 8) × (6 − 4) 2
5 3 0 (3 − 8) × (0 − 4) 20
Average 8 4 Total 60
Variance 20.8 2 =8
𝜎𝑚 Covariance (𝜎𝑖𝑚 ) = 60/5 = 12

𝜎𝑖𝑚
• 𝛽𝑖 =
𝜎𝑚2 =1.5, now 𝑎𝑖 = 8 − 1.5 × 4 = 2
Simple Example

Period Ri and 𝜷𝒊 = 𝟏. 𝟓 Rm 𝒆𝒊 = 𝑹𝒊 − 𝒂𝒊 − 𝜷𝒊 𝑹𝒎
1 10 4 10 − 2 − 1.5 × 4 = 2
2 3 2 3 − 2 − 1.5 × 2 = −2
3 15 8 15 − 2 − 1.5 × 8 = 1
4 9 6 9 − 2 − 1.5 × 6 = −2
5 3 0 3 − 2 − 1.5 × 0 − 3 = 1
Average 8 4 0
Variance 20.8 8 2.8

• With 𝑅ത𝑖 = 𝑎𝑖 + 𝛽𝑖 𝑅ത𝑚 , we can estimate 𝑎𝑖 . 8 = 𝑎𝑖 + 1.5 × 4, i.e., 𝑎𝑖 = 2. Now that we have 𝑎𝑖 , we can
estimate the values of 𝑒𝑖 for each period.
• Here, one can confirm that 𝐸 𝑒𝑖 = 0. Also, note that 𝜎𝑒𝑖 2
= 14. Using 𝜎𝑖2 = 𝛽𝑖2 𝜎𝑚
2 + 𝜎 2 , we get the
𝑒𝑖
value 𝜎𝑖 = 1.5 × 1.5 × 8 + 2.8 = 20.8. This variance is the same as that directly calculated from the
2

table.
 INDIAN INSTITUTE OF TECHNOLOGY KANPUR

A Few Words on Beta

A Few Words on Beta

• Beta is a risk measure that is estimated from the relationship

between the return of a security and that of the market
• Some of the well-known fundamental variables that affect the risk
of stock are dividend payout, asset growth, leverage, liquidity,
asset size, and earnings variability
A Few Words on Beta

• Firm beta and dividends: Firms that pay more dividends have
positive future expectations and are considered to be less risky:
low beta
• Firm beta and growth: High-growth firms are generally young
firms with high capital requirements and are considered to be
riskier: high beta
• Firm beta and liquidity: Firms with high liquidity are considered to
be less risky: low beta
A Few Words on Beta

• Size and beta: Large firms are considered to be less risky than
smaller firms: large firms have low beta
• Earnings variability and beta: A firm with high earnings variability
(earnings beta) is considered riskier: positive beta
 INDIAN INSTITUTE OF TECHNOLOGY KANPUR

Introduction to Multi-Index Models

Introduction to Multi-Index models

• An improvement over single-index models is a multi-index model

• These models aim to capture the non-market influences that may
cause securities to move together
• These multi-index models aim to capture the economic factors or
structural groups (e.g., industrial effects)
Introduction to Multi-index models

The generalized multi-index models can be written in the following

form
• 𝑅𝑖 = 𝑎𝑖∗ + 𝑏𝑖1
∗ ∗ ∗ ∗
𝐼1 + 𝑏𝑖2 ∗ ∗
𝐼2 + 𝑏𝑖3 ∗ ∗
𝐼3 + ⋯ + 𝑏𝑖𝐿 𝐼𝐿 + 𝑐𝑖
• What is the interpretation of 𝑎𝑖∗ , 𝑏𝑖1
∗
, 𝑐𝑖 ?
Introduction to Multi-Index Models

The indices (𝐼𝑗∗ s) would capture the influence of market returns, level of interest
rate, and various industry effects.
• However, this model faces one major challenge
• Some of the indices employed in the model may be correlated
• This vitiates the estimation, as the regression estimations of this kind require
the independent variables to be uncorrelated
• When the variables are correlated, it is difficult to segregate their respective
effects (𝑏𝑖𝑗
∗
’s) on the security
Introduction to Multi-Index Models

Researchers often perform a procedure called orthogonalization to

remove the correlated portion from the respective indices and
create orthogonalized indices
• The new transformed equation is provided below
• 𝑅𝑖 = 𝑎𝑖 + 𝑏𝑖1 𝐼1 + 𝑏𝑖2 𝐼2 + 𝑏𝑖3 𝐼3 + ⋯ + 𝑏𝑖𝐿 𝐼𝐿 + 𝑐𝑖
Introduction to Multi-Index Models

Multi-index model: 𝑅𝑖 = 𝑎𝑖 + 𝑏𝑖1 𝐼1 + 𝑏𝑖2 𝐼2 + 𝑏𝑖3 𝐼3 + ⋯ + 𝑏𝑖𝐿 𝐼𝐿 + 𝑐𝑖

• The new indices are so constructed as they have no correlation
• Also, the error term (𝑐𝑖 ) is not correlated with indices, i.e.,
𝐸[𝑐𝑖 𝐼𝑗 − 𝐼𝑗ҧ ] = 0
• However, the economic interpretation of new indices is slightly
difficult
Introduction to Multi-Index Models:
Basic Equation
𝑹𝒊 = 𝒂𝒊 + 𝒃𝒊𝟏 𝑰𝟏 + 𝒃𝒊𝟐 𝑰𝟐 + 𝒃𝒊𝟑 𝑰𝟑 + ⋯ + 𝒃𝒊𝑳 𝑰𝑳 + 𝒄𝒊 ; for all stocks i
= 1, 2, 3,...,N, and indices j = 1,2,3,….,L
By definition
• Residual variance of stock i equals 𝜎𝑐𝑖2
• Variance of index Ij equals 𝜎𝐼𝑗2
Introduction to Multi-Index Models:
Basic Equation
By construction
• Mean of 𝑐𝑖 equals E(𝑐𝑖 ) = 0
• Covariance between indexes j and k equals 𝐸[(𝐼𝑗 − 𝐼𝑗ҧ ) (𝐼𝑘 − 𝐼𝑘ҧ )] = 0

• Covariance between residuals for stock i and index j equals

𝐸[𝑐𝑖 (𝐼𝑗 − 𝐼𝑗ҧ )] = 0

By assumption
• Covariance between 𝑐𝑖 and 𝑐𝑗 is zero, i.e., 𝐸[𝑐𝑖 𝑐𝑗 ] = 0
Introduction to Multi-Index Models: Basic
Equation
By assumption: Covariance between 𝑐𝑖 and 𝑐𝑗 is zero, i.e., 𝐸[𝑐𝑖 𝑐𝑗 ] = 0

• This last assumption suggests that the only reason stocks vary together is
because of their common relationship with the indexes specified in the model
• There is no other reason that two stocks (i,j ) should have a correlation
• However, there is nothing in the model estimation that forces this to be true
• This is only an approximation, and the performance of the model will be as
good as the approximation
 INDIAN INSTITUTE OF TECHNOLOGY KANPUR

Multi-Index Models: Expected Return

and Risk
Multi-Index Models: Expected Return and Risk

Expected return
• 𝑅ത𝑖 = 𝑎𝑖 + 𝑏𝑖1 𝐼1ҧ + 𝑏𝑖2 𝐼2ҧ + ⋯ + 𝑏𝑖𝐿 𝐼𝐿ҧ
Variance of return
• 𝜎𝑖2 = 𝑏𝑖1
2 2 2 2
𝜎𝐼1 + 𝑏𝑖2 2 2
𝜎𝐼2 + ⋯ + 𝑏𝑖𝐿 𝜎𝐼𝐿 + 𝜎𝑐𝑖2
Covariance between security i and j
2 2 2
• 𝜎𝑖𝑗 = 𝑏𝑖1 𝑏𝑗1 𝜎𝐼1 + 𝑏𝑖2 𝑏𝑗2 𝜎𝐼2 + ⋯ + 𝑏𝑖𝐿 𝑏𝑗𝐿 𝜎𝐼𝐿
Multi-Index Models: Expected Return and Risk

To estimate the expected return and risk, the following estimates are required
• 𝑎𝑖 and 𝜎𝑐𝑖
2
for each stock
• 𝑏𝑖𝑘 between each stock and index
• An estimate of index mean (𝐼jҧ ) and variance 𝜎𝐼𝑗
2
of each index
Assuming N securities and L indices, this is a total 2N + LN + 2L estimates
An analyst following 150 stocks having 10 indices, this means 1820 inputs
This structure, although more complex than single-index models, is still less complex
when no simplifying correlation structure is assumed
Multi-Index Models: Expected Return and Risk

• Researchers often derive indices from the available data using quantitative
techniques (e.g., principal component analysis and factor analysis)
• One can add more indices to increase the explanatory power of the model
• However, with more indices model becomes less efficient and more complex
• Therefore, it is a sort of trade-off between the complexity, efficiency, and
explanatory power of the model
Multi-Index Models: 3-Factor Fama–
French Model
𝑅ത𝑖 = 𝑎𝑖 + 𝑏𝑖𝑀 (𝑅ത𝑀 −𝑅ത𝑓 ) + 𝑏𝑖𝑆𝑀𝐵 𝑅ത𝑆𝑀𝐵 + 𝑏𝑖𝐻𝑀𝐿 𝑅ത𝐻𝑀𝐿 Or

𝑅ത𝑖 − 𝑅ത𝑓 = 𝑎𝑖∗ + 𝑏𝑖𝑀 (𝑅ത𝑀 −𝑅ത𝑓 ) + 𝑏𝑖𝑆𝑀𝐵 𝑅ത𝑆𝑀𝐵 + 𝑏𝑖𝐻𝑀𝐿 𝑅ത𝐻𝑀𝐿
ഥ 𝒇 ) 𝐌𝐚𝐫𝐤𝐞𝐭: is the market index indicating the excess returns over risk-free rate
ഥ 𝑴 −𝑹
(𝑹
ഥ 𝑺𝑴𝑩 (small minus big): indicates the excess return on a portfolio of small stocks over
𝑹
large stocks. The excess returns by small stocks capture the fact that they are riskier
than large stocks
ഥ 𝑯𝑴𝑳 (high minus low): indicates the excess return on a portfolio of high book-to-
𝑹
market (BTM) stocks (value stocks) over that of low (BTM) stocks (growth stocks)
 INDIAN INSTITUTE OF TECHNOLOGY KANPUR

Summary and Concluding Remarks

Summary and Concluding Remarks

• Introduction of single- and multi-index models considerably

simplifies security analysis
• In particular, the complex correlation structure between two
securities is replaced by the common influence of the index on
each of the security
• With the application of these index models, portfolio analysis is
considerably simplified
Summary and Concluding Remarks

• Portfolio betas are often less noisy and more informationally

efficient than betas of individual securities
• Construction of multi-index models broadly employ similar
theoretical underpinnings, except that they employ multiple
indices
• Construction of these index models requires certain assumptions,
some of which are held by the assumption, design, or definition of
the respective model
Summary and Concluding Remarks

Some of these key assumptions in these index models are as follows

• Idiosyncratic error terms are not correlated with indices that are more
systematic influence: 𝐸[𝑐𝑖 𝐼𝑗 − 𝐼𝑗ҧ ] = 0

• These indices are not correlated across each other:

𝐸[(𝐼𝑗 − 𝐼𝑗ҧ ) (𝐼𝑘 − 𝐼𝑘ҧ )] = 0
• The error terms are not correlated with each other: 𝐸[𝑐𝑖 𝑐𝑗 ] = 0
Thanks!
 INDIAN INSTITUTE OF TECHNOLOGY KANPUR

Lesson: Arbitrage Pricing Theory (APT)

Advanced Algorithmic Trading and Portfolio Management
Introduction

• Introduction to arbitrage pricing theory (APT)

• A simple proof of APT
• Testing the APT
• APT with CAPM
• Applications of asset pricing models
• Summary and concluding remarks
 INDIAN INSTITUTE OF TECHNOLOGY KANPUR

Arbitrage Pricing Theory (APT)

Arbitrage Pricing Theory (APT)

CAPM had its genesis in the mean-variance analysis

• Investors choose the optimum diversified portfolio on an efficient
frontier based on the expected return and variance analysis
• The arbitrage pricing theory (APT) of Ross (1966, 1977) employs
a multifactor (alternatively called multi-index) approach to explain
the pricing of assets
• It relies on the single/multi-index approach to provide the return-
generating process
Arbitrage Pricing Theory (APT)

Using the return-generating process, APT derives the definition of

expected returns in equilibrium with certain assumptions
• At the heart of this approach is the arbitrage argument (and thus
the name), similar to that employed in the CAPM
• Two items with the same cash flows cannot sell at different prices
• APT is more generic than CAPM in the sense that it does not
assume that only expected return and risk affect the security
prices
Arbitrage Pricing Theory (APT)

The assumption of homogenous expectations remains

• Instead of a mean-variance framework, we make assumptions
about the return-generating process
• APT argues that returns on any stocks are linearly related to a set
of indices
Arbitrage Pricing Theory (APT)

APT argues that returns on any stocks are linearly related to a set of indices

• 𝑅𝑖 = 𝑎𝑖 + 𝑏𝑖1 𝐼1 + 𝑏𝑖2 𝐼2 + … … + 𝑏𝑖𝑗 𝐼𝑗 + 𝑒𝑖 , where

• 𝑎𝑖 is the expected level of return on the stock “i ” if all indices have a value of zero.

• 𝐼𝑗 is the value of the j th index that affects the return on stock i.

• 𝑏𝑖𝑗 is the sensitivity of stock i’s return to the j th index.

• 𝑒𝑖 is a random error term with a mean of zero and variance equal to 𝜎𝑒𝑖
2

• Essentially, the above-mentioned equation describes the process that generates

security returns
Arbitrage Pricing Theory (APT)

APT argues that returns on any stocks are linearly related to a set of indices
• 𝑅𝑖 = 𝑎𝑖 + 𝑏𝑖1 𝐼1 + 𝑏𝑖2 𝐼2 + … … + 𝑏𝑖𝑗 𝐼𝑗 + 𝑒𝑖

• For the above model to be more accurate, the following assumptions are
made

• 𝐸 𝑒𝑖 𝑒𝑗 = 0; for all i and j where i ≠ j

• 𝐸 𝑒𝑖 𝐼𝑗 − 𝐼𝑗ҧ = 0 for all the stocks and indices

• It is an extension of a multi-index family of models

 INDIAN INSTITUTE OF TECHNOLOGY KANPUR

A Simple Proof of APT: Part I

A Simple Proof of APT

Suppose the following two-index model describes the returns

• 𝑅𝑖 = 𝑎𝑖 + 𝑏𝑖1 𝐼1 + 𝑏𝑖2 𝐼2 + 𝑒𝑖 ; also consider that 𝐸 𝑒𝑖 𝑒𝑗 = 0

• Here, each index represents a certain systematic risk

• Now, if the investor holds a well-diversified portfolio, only the systematic risk
– represented by the indices 𝐼1 and 𝐼2 – will matter
• The residual risk captured by 𝜎𝑒𝑖
2
will be close to zero
• The sensitivity of the portfolio to these two components of the systematic risk
is represented by 𝑏𝑖1 and 𝑏𝑖2
A Simple Proof of APT

Consider the three well-diversified portfolios shown below

Portfolio Expected Return (%) 𝑏𝑖1 𝑏𝑖2

A 15 1.0 0.6
B 14 0.5 1.0
C 10 0.3 0.2

• The returns are provided at equilibrium: No arbitrage

• Remember our discussion of CAPM with sensitivity towards a single index (market portfolio)
where all the securities in equilibrium were lying on a straight line (two axes: 𝑅 and 𝑏1 )

• Here, since we have two sensitivities (two betas with respect to each axis), we can safely
assume that these three portfolios will lie on a plane (three axes: 𝑅, 𝑏𝑖1 , and 𝑏𝑖2 )
A Simple Proof of APT

Consider the three well-diversified portfolios shown below

Portfolio Expected Return (%) 𝑏𝑖1 𝑏𝑖2
A 15 1.0 0.6
B 14 0.5 1.0
C 10 0.3 0.2

• The generic equation for a plane is as follows: 𝑅ത𝑖 = 𝜆0 + 𝜆1 𝑏𝑖1 + 𝜆2 𝑏𝑖2

• Can we solve for the values of 𝜆0 , 𝜆1 , and 𝜆2 using the values provided in the
table?
A Simple Proof of APT

We get the following equation: 𝑅ത𝑖 = 7.75 + 5𝑏𝑖1 + 3.75𝑏𝑖2

Portfolio Expected Return (%) 𝑏𝑖1 𝑏𝑖2
A 15 1.0 0.6
B 14 0.5 1.0
C 10 0.3 0.2

• Now, consider a third portfolio E with expected returns of 15%, 𝑏𝑖1 = 0.6 and 𝑏𝑖2 = 0.6

• Compare E with another portfolio D that places one-third in A, B, and C

• What is my expected return and sensitivities of D, and are there arbitrage

opportunities?
A Simple Proof of APT

Solving for D, we get the following values

1 1 1
• 𝑏𝑝1 = ∗ (1.0) + 0.5 + (0.3) = 0.6
3 3 3
1 1 1
• 𝑏𝑝2 = ∗ (0.6) + 1.0 + (0.2)=0.6
3 3 3
1 1 1
• 𝑅ത𝐷 = 15 + 14 + (10)=13
3 3 3

• D has an identical risk profile but offers a lower return

• We could also have computed the expected return on 𝑅ത𝐷 using the equation of the
plane
• 𝑅ത𝐷 = 7.75 + 5𝑏𝐷1 + 3.75𝑏𝐷2 =7.75 + 5 ∗ 0.6 + 3.75 ∗ 0.6 = 13
A Simple Proof of APT

• By the arbitrage argument (or law of one price), two portfolios

with the same risk cannot sell at different prices (or have different
expected returns)
• Arbitrageurs (or investors in general) would buy E and sell D short
• This would guarantee riskless profit (2%)
• This will continue until E falls back on the plane defined by A, B,
and C
A Simple Proof of APT

• Here, the SML at equilibrium has become a three dimensional

plane, defined by the three axis of expected return, 𝑏𝑖1 , and 𝑏𝑖2
• If any security (like E) is undervalued/overvalued, it will be above
or below this plane
• This would lead to an arbitrage opportunity, and such securities
will converge back to this plane
 INDIAN INSTITUTE OF TECHNOLOGY KANPUR

A Simple Proof of APT: Part II

A Simple Proof of APT

The general equation of the plane in return, i.e., 𝑏𝑖1 and 𝑏𝑖2 space, is shown
below
• 𝑅ത𝑖 = 𝜆0 + 𝜆1 𝑏𝑖1 + 𝜆2 𝑏𝑖2
• This is the equilibrium model provided by APT when the returns are generated
by a two-index model
• Here, 𝜆1 and 𝜆2 are the increases in returns for one unit increase in 𝑏𝑖1 and 𝑏𝑖2
• Essentially, 𝜆1 and 𝜆2 reflect the returns for bearing the risks associated with
the indices 𝐼1 and 𝐼2
A Simple Proof of APT

Consider a zero 𝑏𝑖𝑗 portfolio with no sensitivity to either index

• If it has no risk, then it should offer a risk-free return 𝜆0 = 𝑅𝐹
• In case the riskless rates are not available, then instead of 𝑅𝐹 , we
denote it by 𝑅ത𝑍 , i.e., the return on the zero-beta portfolio (what is
a zero-beta portfolio?)
• Imagine a portfolio that mimics index 1 and, therefore, has 𝑏𝑖1 =1
• Also, it is not sensitive to 𝐼2 and, therefore, has 𝑏𝑖2 = 0
A Simple Proof of APT

For this portfolio, the equation [𝑅ത𝑖 = 𝜆0 + 𝜆1 𝑏𝑖1 + 𝜆2 𝑏𝑖2 ] becomes

𝑅ത1 = 𝑅𝐹 + 𝜆1 and 𝜆1 = 𝑅ത1 − 𝑅𝐹
Similarly, 𝜆2 =𝑅ത2 − 𝑅𝐹
The above analysis can be generalized to a j index case shown below
𝑅ത𝑖 = 𝜆0 + 𝜆1 𝑏𝑖1 + 𝜆2 𝑏𝑖2 + ⋯ + 𝜆𝑗 𝑏𝑖𝑗
𝜆0 = 𝑅𝐹 and 𝜆𝑗 = 𝑅ത𝑗 − 𝑅𝐹 where the return-generating process can be
described as
𝑅𝑖 = 𝑎𝑖 + 𝑏𝑖1 𝐼1 + 𝑏𝑖2 𝐼2 + … … + 𝑏𝑖𝑗 𝐼𝑗 + 𝑒𝑖
A Simple Proof of APT

The above analysis can be generalized to a j index case shown below

• 𝑅ത𝑖 = 𝜆0 + 𝜆1 𝑏𝑖1 + 𝜆2 𝑏𝑖2 + ⋯ + 𝜆𝑗 𝑏𝑖𝑗

• 𝜆0 = 𝑅𝐹 and 𝜆𝑗 = 𝑅ത𝑗 − 𝑅𝐹

• The derivation assumes here that both the indices are orthogonal
• In practical situations, there are always correlations between the risk factors
represented by two indices
• Researchers orthogonalize both indices to remove any common component.
In that case, the new indices may not be well-defined
A Simple Proof of APT

The straight line in the CAPM had two axes: return and beta axes
• Thus, two coordinates corresponding to each point denotes a portfolio
• In the context of a two-index APT model, we have one return and two beta
axes (for each index)
• Thus, three coordinates that define the plane also define the efficient
frontiers
• If a point is above (or below) this plane, this means that the security is under
(or over) priced with respect to one or both of these indices
• Thus, it violates the law of one price
A Simple Proof of APT

If the law of one price is violated, then arbitrageurs may conduct risk-less
arbitrage by selling (or buying) the under (or over) priced portfolio and taking a
counter position in the portfolios that are fairly priced
• This will drive the prices of the inefficient portfolio towards this plane, that is,
efficient frontier or efficient plane
• The implication of this riskless arbitrage is that all portfolios in the equilibrium
would lie on this place, that is, an efficient frontier
• That is, in the space defined by three coordinates: expected return, 𝑏𝑖1 , and 𝑏𝑖2
 INDIAN INSTITUTE OF TECHNOLOGY KANPUR

A Few Important Points About APT

A Few Important Points About APT

In the context of CAPM, it was needed to identify the “market portfolio,” and,
therefore, all the risky assets
• While testing CAPM, one can always question whether all the securities are
truly captured in the risky assets
• Therefore, have we achieved the true market portfolio?
• However, in the context of APT, arbitrage conditions can be applied to any
security or portfolio
• Thus, it is not necessary to identify all the risky securities and market
portfolio
A Few Important Points About APT

APT can very well be tested for a small number of stocks, for
example, all the 50 stocks making up the “Nifty-50” index
• Given this advantage with APT, many studies argue that the tests
designed for CAPM are actually the tests of single-factor APT
• Therefore, they utilized a limited number of securities, which
arguably may not capture the entire market
A Few Important Points About APT

• The only caution needed here is that the systematic influences (or
indices/factors) affecting these sets of stocks that are tested for
APT should be adequately described
• This can be an issue when we have a large set of securities.
Then, finding an adequate number of indices (or systematic
influences) may become a challenge
A Few Important Points About APT

APT is extremely general in nature

• It allows us to describe the equilibrium in terms of a single/multi-index model

• However, it does not define what would be the most appropriate multi-index model

• We do not know 𝜆′𝑠 or 𝐼𝑗 s’s

• They are generated from the data available (e.g., through factor analysis)

• For example, what risk factor a given 𝐼𝑗 indicates (inflation risk, market risk, etc.)
that is not provided by the model

• So, one does not have the direct specific economic rationale for a given factor
 INDIAN INSTITUTE OF TECHNOLOGY KANPUR

Testing the APT: Introduction

Testing the APT

The multifactor return-generating process is provided below

• 𝑅𝑖 = 𝑎𝑖 + 𝑏𝑖1 𝐼1 + 𝑏𝑖2 𝐼2 + … … + 𝑏𝑖𝑗 𝐼𝑗 + 𝑒𝑖

• The corresponding APT model is shown below

• 𝑅ത𝑖 = 𝜆0 + 𝜆1 𝑏𝑖1 + 𝜆2 𝑏𝑖2 + ⋯ + 𝜆𝑗 𝑏𝑖𝑗

• In order to test the APT, one has to identify 𝐼𝑗 s, that is, risk factors

• Subsequently, one can define the sensitivity of a given security 𝑏𝑖𝑗 to this risk factor

• Unfortunately, APT does not offer a direct economic rationale or description of 𝐼𝑗 s

• What do we know about 𝑏𝑖𝑗 , 𝐼𝑗 , and 𝜆𝑗 ?

Testing the APT

Each firm has a unique sensitivity 𝑏𝑖𝑗 for each index 𝐼𝑗

• Thus, 𝑏𝑖𝑗 is a security-specific attribute (such as dividend yield) or security-

specific sensitivity to an index
• The value of 𝐼𝑗 is the same for all the securities

• These 𝐼𝑗 s are systematic influences affecting a large number of securities

and, therefore, are the source of covariance between those securities
• 𝜆𝑗 is the extra-expected return required because of the sensitivity of a
security to the j th attribute of the security
Testing the APT

For CAPM 𝑏𝑖𝑗 (sensitivity to the market, beta), 𝐼𝑗 (market index), and 𝜆𝑗 (Rm−Rf) were
well-defined

• For APT, these are not defined in the model

• One has to test the model below with the observed returns

• 𝑅ത𝑖 = 𝜆0 + 𝜆1 𝑏𝑖1 + 𝜆2 𝑏𝑖2 + ⋯ + 𝜆𝑗 𝑏𝑖𝑗

• This requires estimates of 𝑏𝑖𝑗 and 𝜆𝑗

Testing the APT

Most of the APT tests use the following equation on a set of predefined indices to
obtain 𝑏𝑖𝑗
• 𝑅𝑖 = 𝑎𝑖 + 𝑏𝑖1 𝐼1 + 𝑏𝑖2 𝐼2 + … … + 𝑏𝑖𝑗 𝐼𝑗 + 𝑒𝑖
• Then, the following equation is used to obtain the estimates of 𝜆𝑗 s and thus the APT
model
• 𝑅ത𝑖 = 𝜆0 + 𝜆1 𝑏𝑖1 + 𝜆2 𝑏𝑖2 + ⋯ + 𝜆𝑗 𝑏𝑖𝑗
• In this manner, one can keep identifying risk factors until a sizable portion of
expected returns are identified
• Effectively, these are joint tests of APT as well as the factors/influences/portfolios
considered in the model
Testing the APT

Since there is no generalizable theory that explains all the factors, the following
methods are used to provide a broad set of factors in the APT model
1. Factor analysis approach
2. Specifying the attributes of the security
3. Specifying the influences (factors) affecting the return-generating process
4. Specifying a set of portfolios that capture the return-generating process
 INDIAN INSTITUTE OF TECHNOLOGY KANPUR

Testing the APT: Factor Analysis

Testing the APT: Factor Analysis

A slightly purer and advanced method calls for factor analysis of the
security returns
• The analysis determines a specific set of 𝐼𝑗 s and 𝑏𝑖𝑗 s and also aims to
reduce the covariance of the residual returns to as low as possible
• In the factor analysis terminology, 𝐼𝑗 s become the factors and 𝑏𝑖𝑗 s
become the factor loadings
• One can keep adding factors to the model till the ability of the
additional factors to explain the covariance matrix drops below a
certain level
Testing the APT: Factor Analysis

Post factor analysis, the following equation is used to obtain the estimates of 𝜆𝑗s and
thus the APT model
ഥ 𝒊 = 𝝀𝟎 + 𝝀𝟏 𝒃𝒊𝟏 + 𝝀𝟐 𝒃𝒊𝟐 + ⋯ + 𝝀𝒋 𝒃𝒊𝒋
• 𝑹

• The challenges with the factor analysis are discussed as follows

• Like any similar analysis, the estimates of 𝐼𝑗 s and 𝑏𝑖𝑗 s are subject to the error of
the estimate

• The factors produced in the analysis have no meanings

• For example, the signs of factors and three betas (and therefore, the lambdas)
can be reversed with no change in the resulting expected return
 INDIAN INSTITUTE OF TECHNOLOGY KANPUR

Testing the APT: Specifying the

attributes of the Security
Testing the APT

Specifying the attributes of the security

• If we can establish, a priori, that a certain set of attributes of security that affect the
return

• Then, the extra return required on account of these attributes can be measured
through the following equation: 𝑅ത𝑖 = 𝜆0 + 𝜆1 𝑏𝑖1 + 𝜆2 𝑏𝑖2 + ⋯ + 𝜆𝑗 𝑏𝑖𝑗

• Here, 𝑏𝑖𝑗 s would represent the level of an attribute (j ) associated with the security “i
” associated with each characteristic

• 𝜆𝑗 would represent the extra return because of the sensitivity to that characteristics
Testing the APT

Specifying the attributes of the security: 𝑅ത𝑖 = 𝜆0 + 𝜆1 𝑏𝑖1 + 𝜆2 𝑏𝑖2 + ⋯ + 𝜆𝑗 𝑏𝑖𝑗

• “n % increase in dividend of the portfolio is associated with 𝛥% increase in

the expected returns.”
• Once these 𝑏𝑖𝑗 s are directly obtained, risk premiums for these attributes are
computed using the APT model
• These attributes directly affect the expected returns
• Once major firm attributes and the corresponding risk premiums (𝜆s) are
identified, the equation [𝑅ത𝑖 = 𝜆0 + 𝜆1 𝑏𝑖1 + 𝜆2 𝑏𝑖2 + ⋯ + 𝜆𝑗 𝑏𝑖𝑗 ] can be
estimated to define the APT
 INDIAN INSTITUTE OF TECHNOLOGY KANPUR

Testing the APT: Specifying a Set of

Systematic Influences or Portfolios
Testing the APT

Specifying the influences (factors) affecting the return-generating

process
• Another alternative is to determine and pre-decide the set of risk
factors (influences) that affect the return-generating process
• A set of economic variables that affect the cash flows associated
with the security
• For example, inflation, term structure of interest rates, risk premia,
and industrial production
Testing the APT

Specifying the influences (factors) affecting the return-generating process

• Another set of tests involve time-series regressions of the individual portfolios
to examine their sensitivities (𝑏𝑖𝑗 ) towards these macroeconomic variables

• 𝑅𝑖 = 𝑎𝑖 + 𝑏𝑖1 𝐼1 + 𝑏𝑖2 𝐼2 + … … + 𝑏𝑖𝑗 𝐼𝑗 + 𝑒𝑖

• In the second stage, cross-sectional regressions are performed using all the
portfolios to determine the market price of risk (𝜆𝑗 )

• 𝑅ത𝑖 = 𝜆0 + 𝜆1 𝑏𝑖1 + 𝜆2 𝑏𝑖2 + ⋯ + 𝜆𝑗 𝑏𝑖𝑗

Testing the APT

Specifying the influences (factors) affecting the return-generating process

• For example, ONGC will be definitely affected by the crude-oil prices
• So, a crude oil price index or any broad energy index can provide one risk
factor, that is 𝐼𝑗

• Using these indices, the return-generating process can be employed to

estimate the betas (𝑏𝑖𝑗 )

• Once the betas are obtained, the APT model can be used to obtain risk
premiums (𝜆𝑗 : Rj − Rf )
Testing the APT

Specifying a set of portfolios that capture the return-generating process

• Another option is to construct a set of portfolios that capture the influence of
risk factors affecting the return-generating process. For example
• Difference in the returns on small and large stock portfolios

• Difference in returns on the high book-to-market and low book-to-market stocks

• Difference in the returns on long-term corporate and long-term government bonds

 INDIAN INSTITUTE OF TECHNOLOGY KANPUR

APT and CAPM: Single Market Index

APT and CAPM

Does CAPM become inconsistent in the presence of APT?

• We start with a simple single-index case, where this index is a
market portfolio (or market index like Nifty-50)
• The return-generating process is of the following form
• 𝑅𝑖 = 𝑎𝑖 + 𝛽𝑖 𝑅𝑚 + 𝑒𝑖
APT and CAPM

Now, refer to our earlier discussions on APT, where we said that the
above return-generating process could be written in terms of
sensitivities of the securities to index and the price of risk in the
following form
• 𝑅ത𝑖 = 𝜆0 + 𝜆1 𝑏𝑖1 + 𝜆2 𝑏𝑖2 + ⋯ + 𝜆𝑗 𝑏𝑖𝑗 with 𝜆0 = 𝑅𝐹 and 𝜆𝑗 = 𝑅ത𝑗 − 𝑅𝐹
APT and CAPM

𝑅ത𝑖 = 𝜆0 + 𝜆1 𝑏𝑖1 + 𝜆2 𝑏𝑖2 + ⋯ + 𝜆𝑗 𝑏𝑖𝑗 with 𝜆0 = 𝑅𝐹 and 𝜆𝑗 = 𝑅ത𝑗 − 𝑅𝐹

For a single index case, that is, market index, and in the presence of a
risk-free rate, the above expression becomes
𝑅ത𝑖 = 𝑅𝐹 + 𝛽𝑖 (𝑅ത𝑚 − 𝑅𝐹 ): this is the expected return form provided by
CAPM
This suggests when a single-index return-generating process is true
depiction, the CAPM is clearly consistent
But what about multi-indices?
 INDIAN INSTITUTE OF TECHNOLOGY KANPUR

APT and CAPM: Multi-Index

APT and CAPM

The return-generating process in the context of two indices becomes

• 𝑅𝑖 = 𝑎𝑖 + 𝑏𝑖1 𝐼1 + 𝑏𝑖2 𝐼2 + 𝑒𝑖
• The equilibrium model for this return-generating process with a risk-less
asset becomes: 𝑅ത𝑖 = 𝑅𝐹 + 𝜆1 𝑏𝑖1 + 𝜆2 𝑏𝑖2
• Recall that 𝜆𝑗 is the price of risk for a portfolio that has 𝑏𝑖𝑗 =1 for one index
and zero for all the other indices: 𝜆𝑗 = 𝑅ത𝑗 − 𝑅𝐹

• If we say that CAPM holds, it holds for all the securities as well as portfolios
APT and CAPM

If we say that CAPM holds, it holds for all the securities as well as
portfolios
• Therefore, this industry portfolio may have some sensitivity to the
market portfolio, that is, 𝛽𝜆𝑗

• Recall that the risk premium was 𝛽𝑖 (𝑅ത𝑚 − 𝑅𝐹 ) when the sensitivity to
the market was 𝛽𝑖
• Then, the effective risk premium for this index 𝜆𝑗 becomes
𝛽𝜆𝑗 (𝑅ത𝑚 − 𝑅𝐹 )
APT and CAPM

The return-generating process in the context of two indices

becomes
• 𝑅𝑖 = 𝑎𝑖 + 𝑏𝑖1 𝐼1 + 𝑏𝑖2 𝐼2 + 𝑒𝑖
• The equilibrium model for this return-generating process with a
risk-less asset becomes: 𝑅ത𝑖 = 𝑅𝐹 + 𝜆1 𝑏𝑖1 + 𝜆2 𝑏𝑖2 ; 𝜆1 = 𝑅ത1 − 𝑅𝐹
• But if you believe in CAPM, then
• 𝑅ത1 − 𝑅𝐹 = 𝛽𝜆1 (𝑅ത𝑚 − 𝑅𝐹 ) for index 𝐼1 and 𝑅ത2 − 𝑅𝐹 = 𝛽𝜆2 (𝑅ത𝑚 − 𝑅𝐹 )
for index 𝐼2
APT and CAPM

𝑅ത𝑖 = 𝑅𝐹 + 𝜆1 𝑏𝑖1 + 𝜆2 𝑏𝑖2 can be effectively written as

• 𝑅ത𝑖 = 𝑅𝐹 + 𝑏𝑖1 𝛽𝜆1 (𝑅ത𝑚 − 𝑅𝐹 ) + 𝑏𝑖2 𝛽𝜆2 (𝑅ത𝑚 − 𝑅𝐹 )
• 𝑅ത𝑖 = 𝑅𝐹 + (𝑏𝑖1 𝛽𝜆1 + 𝑏𝑖2 𝛽𝜆2 )(𝑅ത𝑚 − 𝑅𝐹 )
• Define 𝛽𝑖 = (𝑏𝑖1 𝛽𝜆1 + 𝑏𝑖2 𝛽𝜆2 )
• Then, we obtain the CAPM form as follows: 𝑅ത𝑖 = 𝑅𝐹 + 𝛽𝑖 (𝑅ത𝑚 − 𝑅𝐹 )
• This can be extended to multiple factors (indices) as well
APT and CAPM

Therefore, the APT solution, even with multiple factors, is consistent

with CAPM
This means that despite the fact that multiple indices (risk factors)
explain the covariance between the returns, the CAPM holds
 INDIAN INSTITUTE OF TECHNOLOGY KANPUR

Application of Asset Pricing Models:

Passive Management
Passive Asset Management

Passive management
• A simple application of APT is to construct a portfolio of stocks
that closely tracks an index
• The index that represents a risk factor (Bank Nifty represents the
risk of banking stocks)
Passive Asset Management

Passive management
• The attempt is made to use a rather lesser number of stocks
• A large number of stocks would incur significant transaction costs
• In order to track the market index (market portfolio), one cannot
hold all the stocks in the markets
Passive Asset Management

Passive management
• One attempts to hold only to the extent the diversifiable risk can
be offset
• Those indices for which portfolio sensitivity is not matched, if
receive unexpected shocks (like oil price shock), may appear as
the residual risk in the model
• That is, our portfolio may be exposed to these changes
Passive Asset Management

Passive management
• The benefit of using multi-indices instead of a single market index
can be explained here as follows
• Consider five indices, including the energy portfolio, banking,
inflation, cyclical stocks, and government bond portfolio
Passive Asset Management

Passive management
• Compare this to holding only the market portfolio (Nifty)
• Both of these strategies will capture the sensitivity to market risk,
as all the portfolios (except government bonds) may reflect, to
some extent, the risk of market
Passive Asset Management

Passive management
• However, if there is a certain oil price shock or unexpected
changes in inflation, the market portfolio with its sensitivity
matched to Nifty may not be very efficient in tracking the index
• This is because one is indifferent to holding stock from different
industries (e.g., oil stocks) in constructing the Nifty, as long as
she is able to replicate a market portfolio with no diversifiable risk
Passive Asset Management

Passive management
However, this portfolio’s sensitivity to oil shocks can be very
different to that of a multi-index (APT) model that is explicitly
matched to the sensitivity of the oil price index
 INDIAN INSTITUTE OF TECHNOLOGY KANPUR

Application of Asset Pricing Models:

Active Management
Active Asset Management

Active management
• In active management, one continuously holds on the market
portfolio and makes calculated bets on different risk factors
• For example, if one believes that oil prices can go up – this means
that currently the stocks that are sensitive to this risk are
underpriced and will go up in future
Active Asset Management

Active management
• Then, one can increase the sensitivity of his portfolio by adding
additional stocks from oil companies and others to the extent that
increases the sensitivity to this risk index
• Once the price increase has materialized, one can go back to
holding the market portfolio by selling the additional stocks and
realizing the gains
 INDIAN INSTITUTE OF TECHNOLOGY KANPUR

Summary and Concluding Remarks

Summary and Concluding Remarks

• While CAPM has its genesis in the mean-variance framework, APT relies on
the arbitrage argument
• APT utilizes the return-generating process provided by the single- and multi-
index models to generate the equilibrium asset pricing model
• Under the APT, risk-less arbitrage drives prices towards the equilibrium
plane
• The equation of this plane is determined by the systematic risk influences
affecting the set of securities under consideration
Summary and Concluding Remarks

• APT can be tested with the help of (a) factor analysis, (b) specifying the
attributes, (c) specifying a set of systematic influences, and (d) specifying a
set of portfolios
• In the presence of APT, CAPM does not necessarily become invalid as long
as the APT factors are influenced by the market factor (have a well-specified
beta with respect to the market factor)
• Some of the most widely employed applications of asset pricing models
include active and passive asset management and factor investing
Thanks!