Chapter 2: Risk and return: mean–variance analysis and the CAPM

Chapter 2: Risk and return: mean–variance

analysis and the CAPM
Essential reading
Grinblatt, M. and S. Titman Financial Markets and Corporate Strategy. (Boston,
Mass.; London: McGraw-Hill, 2002) second edition [ISBN 0072294337]
Chapters 4 and 5.

Further reading
Brealey, R. and S. Myers Principles of Corporate Finance. (Boston, Mass.; London:
McGraw-Hill, 2003) seventh edition [ISBN 0071151451] Chapters 7 and 8.
Copeland, T. and J. Weston Financial Theory and Corporate Policy.
(Reading, Mass.; Wokingham: Addison-Wesley, 1988) third edition
[ISBN 0201106485] Chapters 6 and 7.
Roll, R. ‘A Critique of the Asset Pricing Theory’s Texts. Part 1: On Past and
Potential Testability of the Theory’, Journal of Financial Economics (1977)
4:129–176.

Introduction
In Chapter 1 we examined the use of present-value techniques in the
evaluation of physical investment projects and in the valuation of primitive
financial assets (i.e. stocks and bonds). A key input into NPV calculations is
the rate of return used in the construction of the discount factor but, thus far,
we have said little regarding where this rate of return comes from. Our
objective in this chapter is to demonstrate how the risk of a given security or
project impacts on the rate of return required from it and hence affects the
value assigned to that asset in equilibrium.
We begin by introducing the basic statistical tools that will be needed in our
analysis, these being expected values, variances and covariances. This
leads to an analysis of the statistical characteristics of portfolios of financial
assets and ultimately to a presentation of the standard mean–variance
optimisation problem. The key result of mean–variance analysis is known as
two-fund separation, and this result underlies the capital asset pricing
model, which we will present next.

Statistical characteristics of portfolios

A portfolio is a collection of different assets held by a given investor. For
example, an American investor may hold 100 Microsoft shares and 650
shares of Bethlehem Steel and therefore holds a portfolio comprising two
assets. The objective of this section is to arrive at the statistical characteristics
of the return on the entire portfolio, given the statistical features of each of
the constituent assets. The key statistical measures used are expected
returns and return variances or standard deviations. The expected return
on a given asset can be thought of as the reward gained from holding it,
whereas the return variance is a measure of total asset risk.
Let us define notation. First, we should clarify the way in which we are
thinking about asset returns. The return on an asset is assumed to be a
random variable with known distributional characteristics. Each individual
asset is assumed to have an expected return of E(r ) and return variance σ2 .
j j
Assets i and j are assumed to have covariance σij . Similarly, we denote the
19
Corporate finance

expected return of the portfolio held as E(Rp) and its variance by σ2. Finally,
P
we assume that an investor can pick from N different stocks when forming
his portfolio.
Returning to the example of the American investor given above, assume that
the market price of Microsoft shares is 130 and that of Bethlehem Steel is
10.1 Hence, given the numbers of each share held, the total value of this 1
These prices are in US cents.
investor’s portfolio is $195. We further assume that the expected returns on
Microsoft and Bethlehem Steel are 10 per cent and 16 per cent respectively,
whereas their variances are 0.25 and 0.49.
We are now in a position to define the share of the entire portfolio value that
is contributed by each individual stockholding. These are referred to as
portfolio weights. The portfolio weight of Bethlehem Steel, for example, is
simply the value of the Bethlehem Steel holding divided by $195 (i.e. 1 3 or
approximately 33.3 per cent). Hence our US investor allocates 1 3 of every
dollar invested to Bethlehem Steel stock.

Activity
Calculate the portfolio weight for Microsoft, using the method presented above.

From the calculations undertaken it is clear that the sum of portfolio weights
must be unity. Each portfolio weight represents the share of total portfolio
value contributed by a given asset. Obviously, aggregating these shares
across all assets held will give a result of unity. Hence, extending the notation
presented above, we denote the portfolio weight on asset i by αi, and the
preceding argument implies that ∑ α i = 1.
i

Our American investor now knows the statistical characteristics of the return
on each of the assets she holds, plus how to calculate the portfolio weight on
each of the assets. What she would really like to know now is how to
construct the return characteristics for the entire portfolio (i.e. she’s
concerned about the risk and reward associated with her entire investment).
In order to do this we will need to introduce some basic properties of
expectations, variances and covariances.

Expectations, variances and covariances

Consider two random variables, x and y. The expected values and variances of these
variables are E(x), E(y), σ 2 and σ 2. The covariance between the random variables is σxy.
x y
Form an arbitrary linear combination of these two random variables and denote it P (i.e.
P = ax + by, where a and b are constants). We wish to know the expected return and
variance of the new random variable P. These are calculated as follows:
2.1 E(P) = aE(x) + bE(y)
2.2 σ P2 = a2σ x2 + b2σ y2 + 2abσxy.
The preceding results are readily extended to the case where more than two
random variables are linearly combined. Consider N random variables denoted xi,
where i runs from 1 to N. Denote their expected values and variances as E(xi) and
σ i2. The covariance between xi and xj is σij. Again we form a linear combination of
the random variables, denoted again by P, using an arbitrary set of constants denoted
αi. The expected value and variance of the random variable P are given by:
N
2.3 Ε (Ρ ) = ∑ a1 Ε ( x1)
i =1

N
2.4 σ 2p = ∑ ai2σ i2 + ∑ a i a j σ ij
i =1 i≠ j

20
 Chapter 2: Risk and return: mean–variance analysis and the CAPM

Given that the returns on individual assets are assumed to be random

variables with known distributional characteristics, the statistical results
given above allow us to calculate portfolio returns and variances very simply.
In addition to the data on Microsoft and Bethlehem Steel provided earlier, we
also need to know the covariance between Microsoft and Bethlehem Steel
returns in order to determine the statistical characteristics of portfolios of
these two assets. However, rather than using covariances, we shall work
throughout the rest of this analysis with correlation coefficients. The
relationship between correlations and covariances is given below.

Covariances and correlations

Assume two random variables, x and y, with variances denoted by σ 2and
x
σ y2 . The covariance between the random variables is σxy. The correlation
coefficient is defined as follows:
σ xy
2.5 ρ xy =
σ xσ y
That is, the correlation between the two random variables is simply the
covariance, divided by the product of the respective standard deviations.
Clearly, knowledge of the correlation and the variances of the two random
variables allows one to retrieve the covariance between the two random
variables.
If we again define a linear combination of the two random variables, P, using
arbitrary constants a and b, the expression for the variance of the linear
combination can be rewritten using the correlation as follows:
2.6 σ p2 = a2σ x2 + b2σ y2 + 2abρxyσxσy
This is a straightforward substitution of equation 2.5 into equation 2.2.
Now we are in a position to calculate the characteristics of our American
investor’s portfolio. Let us take the simplest possible case first and assume
that the returns are uncorrelated (i.e. ρxy = 0). Recalling that the portfolio
2
weights on Microsoft and Bethlehem Steel are 3 and 13 respectively, we
can use equations 2.1 and 2.6 to derive the expected return and variance of
the investor’s portfolio. These calculations yield:

2.7 E( R p ) = 23 (0.1) + 13 (0.16) = 0.12 = 12%

2.8 σ 2p = ( 23 )2 (0.25) + ( 13 )2 (0.49) = 16.6%

Hence, as we would expect, the expected portfolio return lies between the
returns on the individual assets. The portfolio variance, however, is actually
less than that on the return of either of the component assets (i.e. the risk
associated with the portfolio is lower than the risks associated with either
individual asset). This result is one that should be kept in mind and is the
focus of the next section.
Now let’s change our assumption regarding the correlation between the two
asset returns. Assume now that ρxy = 0.5. Obviously, the expected portfolio
return won’t change (as equation 2.1 doesn’t involve the correlation or
covariance at all). The portfolio variance now becomes:
2.9 σ 2p= ( 23 )2 (0.25) + ( 13 )2 (0.49) + 2( 13 )( 23 ) x 0.5 x 0.5 x 0.7 = 24.3%
The portfolio variance has obviously increased, although it is still less than
the return variances of either component assets.

21
Corporate finance

Activity

Assume that ρxy = –0.5. Calculate the portfolio return variance in this case, using the
data on portfolio weights and asset return variances given above.

Now, given the expected returns, return variances and covariances for any set
of assets, we should be able to calculate the expected return and variance of
any portfolio created from those assets. At the end of this chapter, you will
find activities that require you to do precisely this, along with solutions to
some of these activities.

Diversification
A point that we noted from the calculations of expected portfolio returns and
variances above was that, in all of our calculations, the variance of the
portfolio return was lower than that on any individual component’s asset
return.2 Hence, it seems as though, by forming bundles of assets, we can 2
Note that this result does not
eliminate risk. This is true and is known as diversification: through holding hold in general (i.e. it may be the
case that the return variance of a
portfolios of assets, we can reduce the risk associated with our position.
portfolio exceeds the return
Why is this the case? The key is that, in our prior analysis and in real stock variance of one of the component
assets).
return data, the correlations between returns are less than perfect. If two
returns are imperfectly correlated it implies that when returns on the first are
above average, those on the second need not be above average. Hence, to an
extent, the returns on such assets will tend to cancel each other out, implying
that the return variance for a portfolio of these stocks will be smaller than
the corresponding weighted average of the individual asset variances.
To illustrate this point in a general setting, consider the following scenario.
An investor holds a portfolio consisting of N stocks, with each stock having
-1
the same portfolio weight (i.e. each stock has portfolio weight N ). Denote
the return variances for the individual assets by σ 2 where i = 1 to N, and the
i
covariance between returns on assets i and j by σij. Using equation 2.4, the
variance of the investor’s portfolio return can be written as:

2.10 1 N 2 1
σ P2 =
2 ∑ i
σ + 2 ∑ σ ij
N i =1 N i≠ j
Examining the second term of equation 2.10, the existence of N component
assets implies that the summation for all i not equal to j involves N(N – 1)
terms. Obviously the summation in the first term of 2.10 involves N terms.
Hence, defining the average variance of the N assets as σ– and average
2

covariance across all assets as C, 2.10 can be rewritten as:

2.11 σ 2 = N σ 2 + N ( N − 1) C
P 2 2
N N
Equation 2.11 obviously simplifies to the following:
1 2  1
2.12 σ P2 = σ + 1− C
N  N

Now we ask the following question. How does the portfolio variance change
as the number of assets combined in the portfolio increases towards infinity
(i.e. N → ∞). It is clear from 2.12 that, as the number of assets held increases,
the first term will shrink towards zero. Also, as N increases the second term
in 2.12 tends towards C. Together, these observations imply that:
1. The portfolio variance falls as the number of assets held increases.
2. The limiting portfolio return variance is simply the average covariance
between asset returns: this average covariance can be thought of as the

22
 Chapter 2: Risk and return: mean–variance analysis and the CAPM

risk of the market as a whole, with the influence of individual asset return
variances disappearing in the limit.
The moral of the preceding statistical story is clear. Holding portfolios
consisting of greater and greater numbers of assets allows an investor to
reduce the risk he or she bears. This is illustrated diagrammatically in Figure
2.1.

Figure 2.1

Mean–variance analysis
In the preceding two sections, we have demonstrated two important facts:
1. The expected return on a portfolio of assets is a linear combination of the
expected returns on the component assets.
2. An investor holding a diversified portfolio gains through the reduction in
portfolio variance, when asset returns are not perfectly correlated.
In this section, we use these facts to characterise the optimal holding of risky
assets for a risk-averse agent. Our fundamental assumption is that all agents
have preferences that only involve their expected portfolio return and return
variance. Utility is assumed to be increasing in the former and decreasing in
the latter. For illustrative purposes we begin using the assumption that only
two risky assets are available. The results presented, however, generalise to
the N asset case.
To begin, assume there is no risk-free asset. The investor can hence only form
his or her portfolio from risky assets named X and Y. These assets have
2 2
expected returns of E(Rx) and E(Ry) and return variances of σ x and σ y. The
first question the investor wishes to answer is how the characteristics of a
portfolio of these assets (i.e. portfolio expected return and variance) change
as the portfolio weights on the assets change. Given equation 2.6, the answer
to this question is obviously dependent on the correlation between the
returns on the two assets.
First assume the assets are perfectly correlated and, further, assume asset X
has lower expected returns and return variance than asset Y. We form a
portfolio with weights α on asset X and 1 – α on asset Y. Equation 2.6 then
implies that the portfolio variance can be written as follows:
2.13 σ P2 = (ασx + (1 – α)σy)2.

23
Corporate finance

Taking the square root of equation 2.13, it is clear that the portfolio standard
deviation is linear in α. As the portfolio expected return is linear in α, the
locus of expected return–standard deviation combinations is a straight line.
This is shown in Figure 2.2.

Figure 2.2
If the correlation between returns is less than unity, however, the investor
can benefit from diversifying his portfolio. As previously discussed, in this
scenario, portfolio standard deviation is not a linear combination of σx and
σy. The reduction of portfolio risk through diversification will imply that the
mean–standard deviation frontier bows towards the y-axis. This is also
shown on Figure 2.2. The final curve on Figure 2.2 represents the case where
returns are perfectly negatively correlated. In this situation, a portfolio can
be constructed, which has zero standard deviation.

Activities

1. Assuming asset returns are perfectly negatively correlated, use equation 2.6
to find the portfolio weights that give a portfolio with zero standard deviation.
(Hint: write down 2.6 with the correlation set to minus one and a = α and
b = 1 – α. Then minimise portfolio variance with respect to α.)
2. Assume that the returns on Microsoft and Bethlehem Steel have correlation of
0.5. Using the data provided earlier in the chapter, construct the mean–variance
frontier for portfolios of these two assets. Start with a portfolio consisting only
of Microsoft stock and then increase the portfolio weight on Bethlehem Steel by
0.1 repeatedly, until the portfolio consists of Bethlehem Steel stock only.

From here on we will assume that return correlation is between plus and
minus one. The expected return–standard deviation locus for this case is
redrawn in Figure 2.3. In the absence of a risk-free asset, this locus is named
the mean–variance frontier. As our investor’s preferences are increasing in
expected return and decreasing in standard deviation, it is clear that his or
her optimal portfolio will always lie on the frontier and to the right of the
point labelled V. This point represents the minimum-variance portfolio. He
or she will always choose a frontier portfolio at or to the right of V, as these
portfolios maximise expected return for a given portfolio standard deviation.
In the absence of a risk-free asset, this set of portfolios is called the efficient
set.

24
 Chapter 2: Risk and return: mean–variance analysis and the CAPM

Figure 2.3
We can now, given a set of preferences for the investor, find his or her optimal
portfolio. The condition characterising the optimum is that an investor’s
3
indifference curve must be tangent to the mean–variance frontier. Two such 3
In technical terms
optima are identified on Figure 2.3 at R and S. The investor locating at the optimum is
equilibrium point R is relatively risk-averse (i.e. his or her indifference curves characterised by the
marginal rate of
are quite steep), whereas the equilibrium at S is that for a less risk-averse substitution being equal
individual (with correspondingly flatter indifference curves). Figure 2.3 also to the marginal rate of
shows sub-optimal indifference curves for each set of preferences. transformation (i.e. the
slope of the indifference
Hence, as Figure 2.3 demonstrates, in a world of two risky assets and no risk- curve equals the slope
of the frontier).
free asset, the optimal portfolio of risky assets held by an investor depends on
his or her preferences towards risk and return. The same is true when there
are N risky assets available. Figure 2.4 depicts the same type of diagram for
the N asset case.

Figure 2.4
Note that the mean–variance frontier is of the same shape as that in Figure
2.3. However, unlike the two-asset case, the interior of the frontier now
consists of feasible but inefficient portfolios (i.e. those that do not maximise
expected return for given portfolio risk). The mean–variance frontier now
consists of those portfolios that minimise risk for a given expected return,
whereas those portfolios on the efficient set (i.e. on the frontier but to the
right of V) additionally maximise expected return for a given level of risk.

25
Corporate finance

We now re-introduce a risk-free asset to the analysis (i.e. we assume the

existence of an asset with return rf and zero return–standard deviation).
A key question to address at this juncture is as follows. Assume that we form a
portfolio consisting of the risk-free asset and an arbitrary combination of
risky assets. How do the expected return and return–standard deviation of
this portfolio alter as we vary the weights on the risk-free asset and the risky
assets respectively?
Denote our arbitrary risky portfolio by P. We combine P with the risk-free
asset using weights 1 – a and a to form a new portfolio Q. The expected
return and variance of Q are given by:

2.14 E(RQ) = (1 – a)rf + aE(RP) = rf + a[E(RP) – rf]

2.15 σ Q2 = a2σ P2

In order to analyse the variation in the risk and expected return of the
portfolio Q with respect to changes in the portfolio weights, we construct the
following expression:
dE( RQ ) dE( RQ ) / da
2.16 =
dσ Q dσ Q / da
Using equations 2.14 and 2.15 we find that:
dE( RQ ) E( R p ) − r f
2.17 =
dσ Q σP
As this slope is independent of a, the risk–return profile of the portfolio Q is
linear. This is known as the capital market line, and two such CMLs are
shown in Figure 2.5 for two different portfolios of risky assets.
We now have all the components required to describe the optimal portfolio
choice of an investor faced with N risky assets and a risk-free investment.
Figure 2.6 replots the feasible set of risky asset portfolios. The key question to
answer is, what portfolio of risky assets should an investor hold? Using the
analysis from Figure 2.5, it is clear that the optimal choice of risky asset
portfolio is at K. Combining K with the risk-free asset places an investor on a
capital market line (labelled rfKZ), which dominates in utility terms the CML
generated by the choice of any other feasible portfolio of risky assets.4 The
4
optimal portfolio choice and a sub-optimal CML (labelled CML2) are shown That is, choosing portfolio K
places an investor on a CML with
on Figure 2.6 along with the indifference curves of two investors. greater expected returns at each
Recall that we previously defined the efficient set as the group of portfolios level of return variance than does
any other.
that both minimised risk for a given level of expected return and maximised
expected return for a given level of risk. With the introduction of the risk-free
asset, the efficient set is exactly the optimal CML.
The key result that is depicted in Figure 2.6 is known as two-fund
separation. Any risk-averse investor (regardless of his or her degree of risk-
aversion) can form his or her optimal portfolio by combining two mutual
funds. The first of these is the tangency portfolio of risky assets, labelled K, 5
A short sale is the sale of an
and the second is the risk-free asset. All that the degree of risk-aversion asset that one does not actually
dictates is the portfolio weights placed on each of the two funds. The investor own. One borrows the asset in
with the optimum depicted at X on Figure 2.6, for example, is relatively risk- order to complete the
transactions and immediately
averse and has placed positive portfolio weights on both the risk-free asset receives the sale price.
and K. An investor locating at Y, however, is less risk-averse and has sold the Subsequently, one uses the
risk-free asset short in order to invest more in K.
5 proceeds from the sale to
repurchase a unit of the asset,
Two-fund separation is the result that underlies the capital asset pricing model and deliver it to the creditor. If
the price of the asset has
(CAPM), which is developed in the next section. dropped in the interim, one
makes a cash profit.

26
 Chapter 2: Risk and return: mean–variance analysis and the CAPM

Figure 2.5

Figure 2.6

The capital asset pricing model

To begin our derivation of the CAPM, we present the assumptions that
underlie the analysis. These assumptions formalise those implicit in the
preceding section.
• Investors maximise utility defined over expected return and return
variance.
• Unlimited amounts may be borrowed or loaned at the risk-free rate.
• Investors have homogenous expectations regarding future asset returns.
• Asset markets are perfect and frictionless (e.g. no taxes on sales or
purchases, no transaction costs and no short sales restrictions).
We next need to extend slightly our analysis of the previous section in order
to derive the familiar form of the CAPM.

A mathematical characterisation of mean–variance optimisation

Consider Figure 2.6, which graphically identifies the optimal portfolio of
risky assets (K), held by an arbitrary risk-averse investor. The key condition
for optimality is that the capital market line and the mean–variance frontier
are tangent. The following equations give a mathematical description of this
optimality condition.

27
Corporate finance

From equation 2.17, we know that the slope of the capital market line at the
optimum is:
E( R K ) − r f
2.18 .
σK
We also need the slope of the mean–variance frontier at the point of tangency.
To derive this, consider a position (called I) with portfolio weight a in an
arbitrary portfolio of risky assets (called j) and (1 – a) in the optimal
portfolio K. The expected return and standard deviation of this position are:

2.19 E(RI) = aE(Rj) + (1 – a)E(RK)

2.20 σ1 = [a2σ j2 + (1 – a)2σ K2 + 2a(1 – a)σjK]0.5.

Using the same method as shown in equation 2.16 to derive the risk–return
trade-off at the point represented by portfolio I, we get:

2.21 dE( R1 )
= E( Rj ) − E( RK )
da

2.22 dσ 1 = 0.5[ a 2σ 2j + ( 1 − a ) 2 σ K2 + 2a( 1 − a )σ jK ] −0.5 ( 2aσ 2j − 2( 1 − a )σ K2 + 2( 1 − 2a )σ jK ).

da
The slope of the mean–variance frontier at K will be the ratio of 2.21 to 2.22
in the limit as a → 0. Note that equation 2.21 does not depend on a. Taking
the limit of equation 2.22 as a → 0 we get:
1 2
2.23 ( σ jK − σ K )
σK
The slope of the mean–variance frontier at K is the ratio of 2.21 to 2.23 i.e.
σ K E( R )− E( R )
2.24 [ j ] K

σ JK−σ K2
The optimum in Figure 2.6 equates the slope of the mean–variance frontier
at K with the slope of the CML. Hence, equating 2.18 and 2.24 and
rearranging the resulting expression, we arrive at:
σ
2.25 E( R j ) = rf + jK [ E( RK ) − rf ] .
σ K2
2
Defining βj = σjK / σ K, equation 2.26 can be rewritten as:

2.26 E(Rj) = rj + βj[E(RK) – rf].

Equation 2.26 is the standard β-representation of the mean–variance
optimisation problem. The equation translates as follows: the expected
return on a given asset (or portfolio of assets) is equal to the risk-free rate
plus a risk premium multiplied by the asset’s β.6 Assets that have large values 6
The risk premium is defined as
of β will have large expected returns, whereas those with smaller values of β the excess of the expected return
on the tangency portfolio over
will have low expected returns with β defined as the ratio of the covariance the risk-free rate.
of an asset’s returns with those on the market to the variance of the market
return.

Equilibrium and the CAPM

Equation 2.26 is simply derived from mean–variance analysis, and as yet we
have said nothing regarding equilibrium in asset markets. Capital market
equilibrium requires that the demand for risky securities be identical to their
supply. The supply of risky assets is summarised in the market portfolio,
which is defined below.

28
 Chapter 2: Risk and return: mean–variance analysis and the CAPM

Definition
The market portfolio is the portfolio comprising all assets, where the weights
used in the construction of the portfolio are calculated as the market
capitalisation of each asset divided by the sum of market capitalisations
across all assets.

Two-fund separation gives us the fundamental result that all investors hold
efficient portfolios and, further, that all investors hold risky securities in the
same proportions (i.e. those proportions dictated by the tangency portfolio 7
7 All investors perceive
(K)). For demand to be equal to supply in capital markets, it must be the
the same efficient set
case that the market portfolio is constructed with identical portfolio weights. and tangency portfolio
The implication of this is simple: the market portfolio and the tangency due to our assumption
that they have
portfolio are identical. This allows us to express the CAPM in the following homogeneous
form. expectations regarding
asset returns.

The capital asset pricing model

Under the prior assumptions, the following relationship holds for all
expected portfolio returns:
2.27 E(Rj) = Rf + βj [E(rM) – rf],
where E(RM) is the expected return on the market portfolio, and βj is the
covariance of the returns on asset j with those on the market divided by the
variance of the market return.
Equation 2.27 gives the equilibrium relationship between risk and return
under the CAPM assumptions. In the CAPM framework, the relevant
measure of an asset’s risk is its β, and 2.27 implies that expected returns
increase linearly with risk.
To clarify the source of the CAPM equation, note that the identification of the
tangency portfolio and the linear β-representation are implied by
mean–variance analysis. The CAPM then imposes equilibrium on capital
markets and identifies the market portfolio as identical to the tangency
portfolio.

The security market line

Given equation 2.27, the equilibrium relationship between risk and return has
a very simple graphical depiction. In equilibrium expected returns are linear
in β. The expected return on an asset with a β of zero is rf , whereas an asset
with a β of unity has an expected return identical to that on the market.
Plotting this relationship, known as the security market line, we get Figure 2.7.
Comparison of Figures 2.6 and 2.7 implies that, in equilibrium, two assets
with identical expected returns must have identical βs, although their return
variances can differ. The reason that their variances can differ is that a
proportion of asset return variance can be eliminated through diversification.
Agents should not be rewarded for bearing such risk, and hence diversifiable
risk will not affect expected returns. Undiversifiable risk is that which is
driven by variation in the return on the market as a whole, and an asset’s
exposure to such risk is summarised by β. Hence an asset’s β measures its
relevant risk and, via equation 2.27, determines equilibrium expected
returns.
The key message of the preceding paragraph is that β measures asset risk. A
high β asset is risky as it has high returns when market returns are high. An
asset with a low β tends to have high returns when market returns are low.
Hence a low β asset, when included in one’s portfolio, can provide insurance
against low market returns and hence is low risk.

29
Corporate finance

Figure 2.7

Systematic and unsystematic risk

To mathematically illustrate the sources of asset risk we can use the CAPM
equation to decompose the variance of a given asset. Equation 2.27 gives the
equilibrium expected return for asset j. Actual returns on asset j will follow a
similar relationship but will also include a random error term. Denoting this
error by εj we have the following equation:
2.28 rj = rf + βj[rM – rf] + εj.
The variance of the risk-free return is zero by definition. Assuming that β is
j
fixed we can represent the variance of asset j as:
2.29 σ j2 = β j2σ M2 + σ ε2.
The final term on the right-hand side of equation 2.29 is the variance of the
error term and represents diversifiable risk. This source of risk is also known
as unsystematic and idiosyncratic risk. As emphasised previously, this risk is
unrelated to market fluctuations and, therefore, does not affect expected
returns. The first term on the right-hand side of 2.29 represents
undiversifiable risk, also known as systematic risk. This is risk that cannot be
escaped and hence increases equilibrium expected returns.
8
Activities8 You will find the solutions to
these activities on p.32 of this
subject guide.
1. An investor forms a portfolio of two assets, X and Y. These assets have expected
returns of nine per cent and six per cent and standard deviations of 0.8 and 0.6
respectively. Assuming that the investor places a portfolio weight of 0.5 on each
asset, calculate the portfolio expected return and variance if the correlation
between returns on X and Y is unity.
2. Using the data from question 1, recalculate the portfolio expected return and
variance, assuming that the correlation between returns is 0.5.
3. An investor forms a portfolio from two assets, P and Q, using portfolio weights of
one-third and two-thirds respectively. The expected returns on P and Q are five
per cent and seven per cent, and their respective return standard deviations are
0.4 and 0.5. Assuming the return correlation is zero, calculate the expected
return and variance of the investor’s portfolio.
4. Assuming identical data to that in question 3, recalculate the statistical
properties of the portfolio, assuming the return correlation for P and Q is –0.5.

30
 Chapter 2: Risk and return: mean–variance analysis and the CAPM

The Roll critique and empirical tests of the CAPM

The final topic we touch on in this chapter is the empirical validity of the
CAPM. The model of equilibrium expected returns that we have developed in
the preceding sections of this chapter is obviously not guaranteed to hold in
practice, and hence, rather than just blindly accepting its output, we should
examine how it holds up when applied to real data. However, this task brings
us face-to-face with a problem first pointed out by Richard Roll and hence
known as the Roll critique.9 9
See Roll (1977).
The statement of the CAPM is identical to the proposition that the market
portfolio is mean–variance efficient. Hence, Roll pointed out that empirical
tests of the CAPM should seek to examine whether this is indeed the case.
However, he also noted that the market portfolio (or the return on the market)
is not observable to an econometrician, who wishes to conduct a test.
Empirical researchers generally use a broad-based equity index such as the
FTSE-100, S&P-500 or Nikkei 250 to proxy the market. But the true market
portfolio will contain other financial assets (such as bonds and stocks not
included in such indices) as well as non-financial assets such as real estate,
durable goods and even human capital. Hence, the validity of tests of the
CAPM depend critically on the quality of the proxy used for the market
portfolio.
Based on the above, Roll’s critique is simply that, due to the fact that the
market portfolio is not observable, the CAPM is not testable. We can
understand this through the following arguments. First, it might be the case
that the market portfolio is efficient (and hence the CAPM is valid), but our
chosen proxy for the market is not efficient, and hence our empirical test
rejects the CAPM. Second, our proxy for the market might be efficient
whereas the market portfolio itself is not. In this case our test will falsely
indicate that the CAPM is valid. Put simply, the fact that we can’t guarantee
the quality of our proxy for the market implies that we can’t place any faith in
the results that tests based upon it generate, and hence it’s impossible to test
the CAPM.
The Roll critique is clearly damaging in that it implies that we can’t judge the
predictions of the CAPM against reality and trust the results. However, many
researchers have disregarded the prior discussion and estimated the
empirical counterpart of 2.27. From these estimates such researchers pass
judgement on the CAPM.
One prediction of the CAPM upon which some have focused is that the
expected excess return of a given portfolio over the risk-free rate (the risk
premium) is linearly related to β with the slope of this relationship given by
the market risk premium. Historically, however, when one plots actual excess
portfolio returns against their estimated β s one finds that the line traced out
is flatter than one would expect from the theory (i.e. low β portfolios earn
greater expected returns than the CAPM predicts, and high β portfolios earn
10
less). This is clearly evidence against the CAPM.10 See pages 185–186
of Brealey and Myers.
Another prediction of the CAPM, which is also empirically tested, is that β is
the only factor that should cause expected returns to differ (i.e. no other
variable should explain expected returns once we’ve accounted for the effects
of β). Again, the evidence from this line of attack is not good for the CAPM.
Among other factors, firm size, book-to-market ratios, P/E ratios and
dividend yields have all been shown to explain ex-post realised returns (after
accounting for β).

31
Corporate finance

Amalgamating the above evidence implies that, if you are willing to disregard
the Roll critique, you should probably conclude that the CAPM does not
hold. This has led certain authors to investigate other asset-pricing
paradigms such as the APT (which we discuss in the next chapter). An
alternative viewpoint would be to argue that such results tell us little or
nothing about the validity of the CAPM due to the insight of Roll (1977).

Learning outcomes
By the end of this chapter and the relevant reading, you should be able to:
• calculate portfolio expected return and variance from the expected returns and
return variances of constituent assets
• describe the effects of diversification on portfolio characteristics
• derive the CAPM using mean–variance analysis
• point out some theoretical and practical limitations of the CAPM.

Sample examination questions

1. Detail the assumptions that underlie the CAPM and provide a derivation of the
CAPM equation. Support your derivation with graphical evidence. (15%)
2. The returns on ABC stock and on the market portfolio in three consecutive years
are given in the following table:
Year ABC return Market return
1 8% 6%
2 24% 12%
3 28% 15%
Showing all your workings, compute the β for ABC’s equity. (7%)
3. Assume that the risk free rate is five per cent. What is the expected return on
ABC’s stock? (3%)

Solutions to activities
1. The expected return on the equally weighted portfolio is 7.5%. The portfolio
return variance is 0.49, and hence the portfolio return standard deviation is 0.7.
2. Obviously the expected return is the same as in question 1. With correlation of
0.5, the portfolio return variance is 0.37.
3. The expected return on the portfolio is 6.33%, and the portfolio has a return
variance of 0.1289.
4. When the correlation changes to –0.5, the portfolio return variance drops to
0.0844. The expected return on the portfolio doesn’t change from that calculated
in question 3.

32