Chapter 3: The arbitrage pricing theory

Chapter 3: The arbitrage pricing theory

Aim of the chapter

The aim of this chapter is to derive arbitrage pricing theory, an alternative
to the capital asset pricing model, enabling us to price financial assets.

Learning objectives
At the end of this chapter, and having completed the essential reading and
activities, you should be able to:
• understand single-factor and multi-factor model representations
• derive factor-replicating portfolios from a set of asset returns
• understand the notion of arbitrage strategies and that well-functioning
financial markets should be arbitrage-free
• derive arbitrage pricing theory and calculate expected returns using the
pricing formula.

Essential reading
Grinblatt, M. and S. Titman Financial Markets and Corporate Strategy. (Boston,
Mass.; London: McGraw-Hill, 2002) Chapter 6.

Further reading
Brealey, R., S. Myers and F. Allen Principles of Corporate Finance. (Boston,
Mass.; London: McGraw-Hill, 2008) Chapter 9.
Chen, N-F. ‘Some Empirical Tests of the Theory of Arbitrage Pricing’, The
Journal of Finance 38(5) 1983, pp.1393–1414.
Chen, N-F., R. Roll and S. Ross ‘Economic Forces and the Stock Market’, Journal
of Business 59, 1986, pp.383–403.
Copeland, T., J. Weston and K. Shastri Financial Theory and Corporate Policy.
(Reading, Mass.; Wokingham: Addison-Wesley, 2005) Chapter 6.

Overview
The arbitrage pricing theory is an alternative paradigm used to calculate
equilibrium expected returns on financial assets. As its name suggests,
it rests on the notion that well-functioning financial markets should
be arbitrage-free. This, using a factor model of asset returns, implies
restrictions on the relationships between asset returns and generates an
equilibrium pricing relationship.

Introduction
The arbitrage pricing theory (APT) developed in this chapter gives an
alternative to the CAPM as a method to compute the expected returns on
stocks. The basis for the APT is a factor model of stock returns, and we will
define and discuss these models first. From there we will demonstrate how
to derive expected returns using the idea that the returns on stocks, which
are exposed to a common set of factors, must be mutually consistent, given
each stock’s sensitivity to each factor.

37
92 Corporate finance

To give structure to what we mean by ‘mutually consistent’, we need to define

the notion of an arbitrage. An arbitrage strategy can be one of two types.
• It could involve investment in a set of assets (both buying and selling)
that yields an immediate, positive cash inflow (i.e. the receipts from our
sales exceed the cost of our purchases) and, further, is guaranteed not to
make a loss tomorrow. Faced by an investment strategy with this pay-
off structure, any investor who prefers more to less wealth would try to
invest on an infinite scale.
• It could be an investment strategy that is costless today but guarantees
positive future returns. This is akin to receiving something for nothing,
and again, sensible investors would capitalise on the possibility by
investing as much as possible.
The idea that underpins the APT is that investment situations, such as those
described above, should not be permitted in well-functioning financial
markets. Then, if financial markets do not permit the existence of arbitrage
strategies, this places restrictions on the relationships between the expected
returns on assets given the factor structure underlying returns. We will
explain further in later sections of this chapter.

Single-factor models
Before using the notion of absence of arbitrage to provide pricing relations,
we need a basis for the generation of stock returns. Within the context of
the APT, this basis is given by the assumption that the population of stock
returns is generated by a factor model. The simplest factor model, given
below, is a one-factor model:
ri = αi + βi F + εi E(εi) = 0. (3.1)
In equation 3.1, the returns on stock i are related to two main components:
1. The first of these is a component that involves the factor F. This factor
is posited to affect all stock returns, although with differing sensitivities.
The sensitivity of stock i’s return to F is βi. Stocks that have small values
for this parameter will react only slightly as F changes, whereas when βi
is large, variations in F cause very large movements in the return on stock
i. As a concrete example, think of F as the return on a market index (e.g.
the S&P-500 or the FTSE-100), the variations in which cause variations in
individual stock returns. Hence, this term causes movements in individual
stock returns that are related. If two stocks have positive sensitivities to
the factor, both will tend to move in the same direction.
2. The second term in the factor model is a random shock to returns, which
is assumed to be uncorrelated across different stocks. We have denoted
this term εi and call it the idiosyncratic return component for stock i. An
important property of the idiosyncratic component is that it is also assumed
to be uncorrelated with F, the common factor in stock returns. In statistical
terms we can write the conditions on the idiosyncratic component as follows:

Cov(εi , εj) = 0 i≠j Cov(εi , F) = 0 i

An example of such an idiosyncratic stock return might be the unexpected
departure of a firm’s CEO or an unexpected legal action brought against the
company in question.
The partition of returns implied by equation 3.1 implies that all common
variation in stock returns is generated by movements in F (i.e. the
correlation between the returns on stocks i and j derives solely from F). As
the idiosyncratic components are uncorrelated across assets they do not
bring about covariation in stock price movements.
38
 Chapter 3: The arbitrage pricing theory

Application exercise
Consider an economy in which the risk-free rate of return is 4% and the expected rate of
return on the market index is 9%. The variance of the return on the market index is 20%.
Two portfolios A and B have expected return 7% and 10%, and variance 20% and 50%,
respectively.
a) Work out the portfolios’ beta coefficients.
According to the CAPM:
E(rA) = rF + βA [E(rM) – rF]
and
E(rB) = rF + βB [E(rM) – rF].
Hence:
βA = [E(rA) – rF]/[E(rM) – rF] = (7% − 4%)/(9% − 4%) = 0.6
βB = [E(rB) – rF]/[E(rM) – rF] = (10% − 4%)/(9% − 4%) = 1.2.
b) The risk of a portfolio can be decomposed into market risk and idiosyncratic
risk. What are the proportions of market risk and idiosyncratic risk for the two
portfolios A and B?
From the market model:
rA = αA + βA rM + εA
rB = αB + βB rM + εB
with cov(rM, εA) = cov(rM, εB) = 0.

It hence follows that the variance of portfolio A’s returns, σ2A, has two
components, systematic and idiosyncratic risk:
σ2A = β2A σ2M + σ2εA.

Similarly:
σ2B = β2B σ2M + σ2εB.

The proportion of systematic risk for A is hence

β2A σ2M / σ2A = (0.6)2*20%/20% = 36%.

The proportion of idiosyncratic risk for A is hence

1 − [β2A σ2M / σ2A] = 64%.

The proportion of systematic risk for B is hence

β2B σ2M / σ2B = (1.2)2*20%/50% = 58%.

The proportion of idiosyncratic risk for B is hence

1 − [β2B σ2M / σ2B] = 42%.

Portfolio B is much riskier than portfolio A as the variance of its returns is 50%
compared with 20% for A. The main reason why it is riskier is that it is much
more sensitive to the return of the market index than portfolio A as its beta is
1.2 compared with 0.6 for portfolio A.
c) Assume the two portfolios have uncorrelated idiosyncratic risk. What is the
covariance between the returns on the two portfolios?
Cov(rA,rB) = Cov(αA +βA rM + εA, αB +βB rM + εB) = βA βB σ2M = 0.6*1.2*20% = 14%.

The returns of portfolios A and B are hence (positively) correlated even though
their idiosyncratic return components are not. These returns are positively
correlated because they are positively correlated with the returns of the market
index.
39
92 Corporate finance

Multi-factor models
A generalisation of the structure presented in equation 3.1 posits k factors
or sources of common variation in stock returns.
ri = αi + β1iF1 + β2iF2 + .... + βkiFk + εi E(εi) = 0. (3.2)
Again the idiosyncratic component is assumed uncorrelated across stocks
and with all of the factors. Further, we’ll assume that each of the factors
has a mean of zero. These factors can be thought of as representing news
on economic conditions, financial conditions or political events. Note that
this assumption implies that the expected return on asset i is just given by
the constant in equation 3.2 (i.e. E(ri) = αi). Each stock has a complement
of factor sensitivities or factor betas, which determine how sensitive the
return on the stock in question is to variations in each of the factors.
A pertinent question to ask at this point is how do we determine the return
on a portfolio of assets given the k-factor structure assumed? The answer
is surprisingly simple: the factor sensitivities for a portfolio of assets are
calculable as the portfolio weighted averages of the individual factor
sensitivities. The following example will demonstrate the point.

Example
The returns on stocks X, Y, and Z are determined by the following two-factor model:
rX = 0.05 + F1 – 0.5F2 + εX

rY = 0.03 + 0.75 F1 + 0.5F2 + εY

rz = 0.04 + 0.25 F1 – 0.3F2 + εz
Given the factor sensitivities in the prior three equations, we wish to derive the factor
structure followed by an equally weighted portfolio of the three assets (i.e. a portfolio
with one-third of the weights on each of the assets). Following the result mentioned
above, all we need to do is form a weighted average of the stock sensitivities on the
individual assets. Subscripting the coefficients for the equally weighted portfolio with a p
we have:
1
αp = (0.05 + 0.03 + 0.04) = 0.04
3
β1p = 1 (1 + 0.75 – 0.25) = 0.5
3
1
β2p = (–0.5 + 0.5 –0.3) = –0.1;
3
and hence; the factor representation for the portfolio return can be written as:
rp = 0.04 + 0.5F1 – 0.1F2 + εp
where the final term is the idiosyncratic component in the portfolio return.

Activity
Using the data given in the previous example, compute the return representation for a
portfolio of assets X, Y and Z with portfolio weights –0.25, 0.5 and 0.75.

An important implication of the result is the following. Assume a two-

factor model, and also assume that we are given the factor representations
for three stocks. I can construct a portfolio of these three assets, which has
any desired set of factor sensitivities through appropriate choice of the
portfolio weights.1 What underlies this result? Well, to illustrate let’s use 1
In general, if I have
the data from the prior example. Assume I wish to construct a portfolio a k-factor model I will
need k+1 stocks to
with a sensitivity of 0.5 on the first factor and a sensitivity of 1 on the
do this.
second factor. Denoting the portfolio weights on the individual assets by
ωX, ωY and ωZ it must be the case that:

40
 Chapter 3: The arbitrage pricing theory

ωX + 0.75ωY – 0.25ωZ = 0.5 (3.3)

–0.05ωX + 0.5ωY – 0.3ωZ = 1. (3.4)
Finally, it must also be the case that the portfolio weights add up to unity,
so we must also satisfy the following equation:
ωX + ωY + ωZ = 1.
Equations 3.3, 3.4 and 3.5 are three equations in three unknowns, and
we can find values for the portfolio weights which satisfy all three
simultaneously. This illustrates the fact that (as the portfolio factor
sensitivities were arbitrarily set at 0.5 and 1) we can derive any constellation
of factor sensitivities. A particularly interesting case is when the portfolio is
sensitive to one of the factors only. We call this a factor-replicating portfolio
and discuss it below.

Broad-based portfolios and idiosyncratic returns

In what follows we will assume that the basic securities that we’re going
to work with are themselves broad-based portfolios. The reason for this
is that it allows us to lose the idiosyncratic risk terms associated with
single stocks. Why is this the case? Well, consider the idiosyncratic risk
term for an equally weighted portfolio of 100 stocks. Call the ith idiosyncratic
term εi and assume that all idiosyncratic terms have variance σ2. The variance
of the idiosyncratic element of the portfolio return is then:
Var (ε ) = Var ( ∑ εi ) = 1 Var (∑ ε ) = 100 = σ 2 = σ 2 .
100 100

P
i =1 100 10000 i =1
i 10000 100
Note that, under these assumptions the variance of the idiosyncratic portfolio
return is only one-hundredth of the variance of any individual asset’s
idiosyncratic return. In a general case, where one forms an equally weighted
portfolio of n assets, the variance of the idiosyncratic term for the portfolio
return is n-1σ2. This is a diversification result just like those we used in
Chapter 2. The fact that the idiosyncratic returns are uncorrelated with one
another means that their influence tends to disappear when one groups
assets into large portfolios.

Factor-replicating portfolios
An important application of the technology developed previously in this
chapter is the construction of factor-replicating portfolio. A factor-replicating
portfolio is a portfolio with unit exposure to one factor and zero exposure to
all others. For example, the portfolio replicating factor 1 in model 3.2 would
have β1 = 1 and βj = 0 for all j = 2 to k.

Activity
Assume that stock returns are generated by a two-factor model. The returns on three
well-diversified portfolios, A, B and C, are given by the following representations:
rA = 0.10 F1 – 0.5F2
rB = 0.08 + 2F1 + F2
rC = 0.05 + 0.5F1 + 0.5F2.
Determine the portfolio weights you need to place on A, B and C in order to construct
the two factor-replicating portfolios plus a portfolio which has zero exposure to both
factors. What are the expected returns of the factor-replicating portfolios and what is the
expected return of the risk-free portfolio?

41
92 Corporate finance

The question to ask at this point is: why bother constructing factor-
replicating portfolios? The reason is as follows. Suppose I want to build a
portfolio that has identical factor exposures to a given asset, X. Assume a
two-factor world and that asset X has exposure of 0.75 to factor 1 and –0.3
to factor 2. Assume also that I know the two factor-replicating portfolios.
Building a portfolio with the same factor exposures as X is now simple.
Construct a new portfolio, Y, which has portfolio weight 0.75 on the
replicating portfolio for the first factor, portfolio weight –0.3 on the
replicating portfolio for the second factor and the rest of the portfolio
weight (i.e. a weight of 1 – 0.75 + 0.3 = 0.55) on the risk-free asset.
Via the results on the factor representations of a portfolio of assets and
the definition of a factor-replicating portfolio it is easy to see that Y is
guaranteed to have identical factor exposures to X.
The replication in the preceding paragraph forms the basis for the APT. For
absence of arbitrage we require all assets with identical factor exposures
to earn the same return. If they did not, then we would have the chance to
make unlimited amounts of money. For example, assume that the expected
return on the replicating portfolio Y was greater than that on asset X. Then
I should short X and buy Y. The risk exposures of the two portfolios are
identical and hence risks cancel out and I am left with an excess return
that is riskless (i.e. an arbitrage gain).
In order to progress, let us introduce some notation. Denote the risk-
free rate with rf. Denote the expected return on the ith factor-replicating
portfolio with rf + λi such that λi is the risk premium associated with the
ith factor. Again, for simplicity, assume that the world is generated by
a two-factor model, and assume that I wish to replicate asset X, which
has sensitivity β1X to the first factor and β2X to the second factor. Finally,
we will assume that the primary securities being worked with are well-
diversified portfolios themselves. Hence, we will ignore any idiosyncratic
risk in this derivation.
Using the prior argument, to replicate asset X’s factor sensitivities,
we construct a portfolio with weight β1X on the first factor-replicating
portfolio, weight β2X on the second factor-replicating portfolio and weight
1 – β1X – β2X on the risk-free asset. The expected return of the replicating
portfolio is hence:
β1X (rf + λ1) + β2X (rf + λ2) + (1 – β1X – β2X) rf = rf + β1X λ1+ β2X λ2. (3.6)
Hence, using our factor-replicating portfolios we can write the expected
return on a portfolio which replicates X’s factor exposures as the risk-free
rate plus each factor exposure multiplied by the risk premium on the
relevant factor-replicating portfolio.

The arbitrage pricing theory

Consider an arbitrary asset. The previous sub-section tells us that it’s
simple to replicate this asset’s risk (i.e. its factor exposures) using factor-
replicating portfolios. The key to the APT is that absence of arbitrage
requires that such a pair of portfolios must have identical expected returns
in a financial market equilibrium. If they did not, it would be possible to
make unlimited amounts of money without incurring any risk.
This implies that the expected return on asset X, rX, must be identical to
the expression arrived at in equation 3.6, that is:
E(rX) = rf + β1X λ1+ β2X λ2. (3.7)

42
 Chapter 3: The arbitrage pricing theory

Equation 3.7 is the statement of the APT. The expected return on a financial
asset can be written as the risk-free rate plus sum of the asset’s factor
sensitivities multiplied by the factor-risk premiums (which are invariant
across assets). If such an expression does not hold at all times, arbitrage
opportunities exist. Note the assumptions that are required to achieve this
result. First, we require that asset returns are generated by a two-factor (or
in general k-factor) model. Second, we assume that arbitrage opportunities
cannot exist. Lastly, we assume that enough assets are available such that
firm-specific risk washes away when portfolios are formed.

Example
In the previous two-factor example, we determined the expected returns on the two
factor-replicating portfolios. Denoting the expected return on the ith factor-replicating
portfolio by E(ri) we have:
E(r1) = 8.29% E(r2) = 1.71% E(r3) = 5.14%.
Hence, the premiums associated with the two factors are:
λ1 = 8.29 – 5.14 = 3.15%, λ2 = 1.71 – 5.14 = 3.43%.
This implies that the expected return on any asset in this world can be written as:
E(ri) = 5.14 + 3.15β1i – 3.43β2i.
To check that this works, substitute portfolio C’s (for example) factor sensitivities into the
preceding expression. This gives:
E(rC) = 5.14 + 3.15 (0.5) – 3.43 (0.5) = 5%,
and hence, agrees with the expected return implied by the original representation for
asset C. Check that the expected returns on assets A and B also come out correctly.

To analyse an arbitrage opportunity that might arise in markets, attempt

the following Activity.

Activity
Assume that a new well-diversified portfolio, D, is added to our world. This asset has
sensitivities of 3 and –1 to the two factors and an expected return of 15 per cent.
Using the equilibrium expected return equation given above, derive the equilibrium
expected return on an asset with identical factor exposures to D. Is there now an
arbitrage opportunity available? If so, dictate a strategy that could be employed to exploit
the arbitrage opportunity.

Summary
The APT gives us a straightforward, alternative view of the world from
the CAPM. The CAPM implies that the only factor that is important
in generating expected returns is the market return and, further, that
expected stock returns are linear in the return on the market. The APT
allows there to be k sources of systematic risk in the economy. Some may
reflect macroeconomic factors, like inflation, and interest rate risk, whereas
others may reflect characteristics specific to a firm’s industry or sector.
Empirical research has indicated that some of the well-known empirical
problems with the CAPM are driven by the fact that the APT is really the
proper model of expected return generation. Chen (1983), for example,
argues that the size effect found in CAPM studies disappears in a multi-
factor setting. Chen, Roll and Ross (1986) argue that factors representing
default spreads, yield spreads and GDP growth are important in expected
return generation. Work in this area is still progressing.

43
92 Corporate finance

A reminder of your learning outcomes

Having completed this chapter, and the essential reading and activities,
you should be able to:
• understand single-factor and multi-factor model representations
• derive factor-replicating portfolios from a set of asset returns
• understand the notion of arbitrage strategies and that well-functioning
financial markets should be arbitrage-free
• derive arbitrage pricing theory and calculate expected returns using the
pricing formula.

Key terms
arbitrage pricing theory
factor-replicating portfolio
factor sensitivity
multi-factor model
single-factor model

Sample examination question

1. Assume that stock returns are generated by a two-factor model. The
returns on three well-diversified portfolios, A, B and C, are given by the
following representations:
rA = 0.10 + F1
rB = 0.08 + 2F1 – F2
rC = 0.05 – 0.5F1 + 0.5F2
a) Discuss what the factor representations above imply for the
variation and comovement in the three stock returns. Show how the
returns of the stocks should be correlated between themselves.
b) Find the portfolio weights that one must place on stocks A, B and
C to construct pure tracking portfolios for the two factors (i.e.
portfolios in which the loading on the relevant factor is +1 and the
loadings on all other factors are 0).
c) If one was to introduce a new portfolio, D, with loadings of +1 on
both of the factors, what would the expected return on D have to be
to rule out arbitrage?
d) Explain the concepts of idiosyncratic risk and factor risk in the APT.
What role does diversification play in the APT?

44