JWPR026-Fabozzi c01 June 22, 2008 6:54

CHAPTER 1

Portfolio Selection
FRANK J. FABOZZI, PhD, CFA, CPA
Professor in the Practice of Finance, Yale School of Management

HARRY M. MARKOWITZ, PhD

Consultant

FRANCIS GUPTA, PhD

Director, Research, Dow Jones Indexes

AL
Some Basic Concepts 4 Portfolio Diversification 8
Utility Function and Indifference Curves 4 Portfolio Risk and Correlation 9
The Set of Efficient Portfolios
RI
The Effect of the Correlation of Asset Returns on
TE
and the Optimal Portfolio 4 Portfolio Risk 9
Risky Assets versus Risk-Free Assets 4 Choosing a Portfolio of Risky Assets 9
Measuring a Portfolio’s Expected Return 5 Constructing Efficient Portfolios 10
MA

Measuring Single-Period Portfolio Return 5 Feasible and Efficient Portfolios 10

The Expected Return of a Portfolio of Risky Choosing the Optimal Portfolio in the Efficient Set 11
Assets 5 Index Model’s Approximations to the Covariance
ED

Measuring Portfolio Risk 6 Structure 12

Variance and Standard Deviation Single-Index Market Model 12
as a Measure of Risk 6 Multi-Index Market Models 13
HT

Measuring the Risk of a Portfolio Comprised Summary 13

of More than Two Assets 8 References 13
IG

Abstract: The goal of portfolio selection is the construction of portfolios that maximize
R

expected returns consistent with individually acceptable levels of risk. Using both
PY

historical data and investor expectations of future returns, portfolio selection uses
modeling techniques to quantify “expected portfolio returns” and “acceptable levels
of portfolio risk,” and provides methods to select an optimal portfolio. It would not
CO

be an overstatement to say that modern portfolio theory has revolutionized the world
of investment management. Allowing managers to quantify the investment risk and
expected return of a portfolio has provided the scientific and objective complement to
the subjective art of investment management. More importantly, whereas at one time
the focus of portfolio management used to be the risk of individual assets, the theory
of portfolio selection has shifted the focus to the risk of the entire portfolio. This theory
shows that it is possible to combine risky assets and produce a portfolio whose expected
return reflects its components, but with considerably lower risk. In other words, it is
possible to construct a portfolio whose risk is smaller than the sum of all its individual
parts.

Keywords: portfolio selection, modern portfolio theory, mean-variance analysis, utility

function, efficient portfolio, optimal portfolio, covariance, correlation,
portfolio diversification, beta, portfolio variance, feasible portfolio

3
JWPR026-Fabozzi c01 June 22, 2008 6:54

4 Portfolio Selection

In this chapter, we present the theory of portfolio selection u3

as formulated by Markowitz (1952). This theory is also
u2
referred to as mean-variance portfolio analysis or simply
mean-variance analysis. u1

Utility
SOME BASIC CONCEPTS increases
Portfolio theory draws on concepts from two fields: finan-
cial economic theory and probability and statistical theory.
u3 u

Expected return
This section presents the concepts from financial economic
theory used in portfolio theory. While many of the con-
cepts presented here have a more technical or rigorous ′
u2
definition, the purpose is to keep the explanations simple
and intuitive so the reader can appreciate the importance u1
and contribution of these concepts to the development of
modern portfolio theory.

Utility Function and Indifference Curves

In life there are many situations where entities (that is,
individuals and firms) face two or more choices. The eco- Risk
nomic “theory of choice” uses the concept of a utility func-
tion developed by von Neuman and Morgenstern (1944), Figure 1.1 Indifference Curves
to describe the way entities make decisions when faced
with a set of choices. A utility function assigns a (numeric)
value to all possible choices faced by the entity. The higher obtain the same level of utility, the investor requires a
the value of a particular choice, the greater the utility de- higher expected return in order to accept higher risk.
rived from that choice. The choice that is selected is the one For the three indifference curves shown in Figure 1.1,
that results in the maximum utility given a set of (budget) the utility the investor receives is greater the further the
constraints faced by the entity. indifference curve is from the horizontal axis, because that
In portfolio theory too, entities are faced with a set of curve represents a higher level of return at every level of
choices. Different portfolios have different levels of ex- risk. Thus, for the three indifference curves shown in the
pected return and risk. Also, the higher the level of ex- exhibit, u3 has the highest utility and u1 the lowest.
pected return, the larger the risk. Entities are faced with
the decision of choosing a portfolio from the set of all
possible risk/return combinations: where return is a de- The Set of Efficient Portfolios
sirable which increases the level of utility, and risk is an and the Optimal Portfolio
undesirable which decreases the level of utility. There-
Portfolios that provide the largest possible expected re-
fore, entities obtain different levels of utility from different
turn for given levels of risk are called efficient portfolios.
risk/return combinations. The utility obtained from any
To construct an efficient portfolio, it is necessary to make
possible risk/return combination is expressed by the util-
some assumption about how investors behave when mak-
ity function. Put simply, the utility function expresses the
ing investment decisions. One reasonable assumption is
preferences of entities over perceived risk and expected
that investors are risk averse. A risk-averse investor is an
return combinations.
investor who, when faced with choosing between two in-
A utility function can be expressed in graphical form by
vestments with the same expected return but two different
a set of indifference curves. Figure 1.1 shows indifference
risks, prefers the one with the lower risk.
curves labeled u1 , u2 , and u3 . By convention, the horizontal
In selecting portfolios, an investor seeks to maximize
axis measures risk and the vertical axis measures expected
the expected portfolio return given his tolerance for risk.
return. Each curve represents a set of portfolios with dif-
Alternatively stated, an investor seeks to minimize the
ferent combinations of risk and return. All the points on a
risk that he is exposed to given some target expected re-
given indifference curve indicate combinations of risk and
turn. Given a choice from the set of efficient portfolios,
expected return that will give the same level of utility to a
an optimal portfolio is the one that is most preferred by the
given investor. For example, on utility curve u1 , there are
investor.
two points u and u
, with u having a higher expected re-
turn than u
, but also having a higher risk. Because the two
points lie on the same indifference curve, the investor has
an equal preference for (or is indifferent to) the two points, Risky Assets versus Risk-Free Assets
or, for that matter, any point on the curve. The (positive) A risky asset is one for which the return that will be re-
slope of an indifference curve reflects the fact that, to alized in the future is uncertain. For example, an investor
JWPR026-Fabozzi c01 June 22, 2008 6:54

INVESTMENT MANAGEMENT 5

who purchases the stock of Pfizer Corporation today with In shorthand notation, equation (1.1) can be expressed
the intention of holding it for some finite time does not as follows:
know what return will be realized at the end of the hold-


G
ing period. The return will depend on the price of Pfizer’s Rp = wg Rg (1.2)
stock at the time of sale and on the dividends that the com- g=1
pany pays during the holding period. Thus, Pfizer stock,
and indeed the stock of all companies, is a risky asset. Equation (1.2) states that the return on a portfolio (Rp )
Securities issued by the U.S. government are also risky. of G assets is equal to the sum over all individual assets’
For example, an investor who purchases a U.S. govern- weights in the portfolio times their respective return. The
ment bond that matures in 30 years does not know the portfolio return Rp is sometimes called the holding period
return that will be realized if this bond is held for only return or the ex post return.
one year. This is because changes in interest rates in that For example, consider the following portfolio consisting
year will affect the price of the bond one year from now of three assets:
and that will impact the return on the bond over that
year. Market Value at
There are assets, however, for which the return that will the Beginning of Rate of Return Over
be realized in the future is known with certainty today. Asset Holding Period Holding Period
Such assets are referred to as risk-free or riskless assets. 1 $6 million 12%
The risk-free asset is commonly defined as a short-term 2 8 million 10%
obligation of the U.S. government. For example, if an 3 11 million 5%
investor buys a U.S. government security that matures
in one year and plans to hold that security for one year,
then there is no uncertainty about the return that will be The portfolio’s total market value at the beginning of the
realized. (Note: Here “return” refers to the nominal return. holding period is $25 million. Therefore,
The “real” return, which adjusts for inflation, is uncertain.) w1 = $6 million/$25 million = 0.24, or 24% and R1 = 12%
The investor knows that in one year, the maturity date of w2 = $8 million/$25 million = 0.32, or 32% and R2 = 10%
the security, the government will pay a specific amount to w3 = $11 million/$25 million = 0.44, or 44% and R3 = 5%
retire the debt. Notice how this situation differs for the U.S.
government security that matures in 30 years. While the Notice that the sum of the weights is equal to 1. Substi- tut-
1-year and the 30-year securities are obligations of the U.S. ing into equation (1.1), we get the holding period portfolio
government, the former matures in one year so that there return,
is no uncertainty about the return that will be realized. In
contrast, while the investor knows what the government R p = 0.24(12%) + 0.32(10%) + 0.44(5%) = 8.28%
will pay at the end of 30 years for the 30-year bond, he Note that since the holding period portfolio return is
does not know what the price of the bond will be one year 8.28%, the growth in the portfolio’s value over the holding
from now. period is given by ($25 million) × 0.0828 = $2.07 million.

The Expected Return of a Portfolio

MEASURING A PORTFOLIO’S of Risky Assets
EXPECTED RETURN Equation (1.1) shows how to calculate the actual return of a
We are now ready to define the actual and expected return portfolio over some specific time period. In portfolio man-
of a risky asset and a portfolio of risky assets. agement, the investor also wants to know the expected (or
anticipated) return from a portfolio of risky assets. The
expected portfolio return is the weighted average of the
Measuring Single-Period Portfolio Return expected return of each asset in the portfolio. The weight
The actual return on a portfolio of assets over some spe- assigned to the expected return of each asset is the per-
cific time period is straightforward to calculate using the centage of the market value of the asset to the total market
following: value of the portfolio. That is,

R p = w1 R1 + w2 R2 + . . . + wG RG (1.1) E(R p ) = w1 E(R1 ) + w2 E(R2 ) + . . . + wG E(RG ) (1.3)

The E( ) signifies expectations, and E(RP ) is sometimes
called the ex ante return, or the expected portfolio return
where over some specific time period.
Rp = rate of return on the portfolio over the period The expected return, E(Ri ), on a risky asset i is calculated
Rg = rate of return on asset g over the period as follows. First, a probability distribution for the possible
wg = weight of asset g in the portfolio (that is, market rates of return that can be realized must be specified. A
value of asset g as a proportion of the market value probability distribution is a function that assigns a proba-
of the total portfolio) at the beginning of the period bility of occurrence to all possible outcomes for a random
G = number of assets in the portfolio variable. Given the probability distribution, the expected
JWPR026-Fabozzi c01 June 22, 2008 6:54

6 Portfolio Selection

Table 1.1 Probability Distribution for the Rate of Return for used a variety of definitions to describe risk. Markowitz
Stock XYZ (1952, 1959) quantified the concept of risk using the well-
known statistical measures of variances and covariances.
n Rate of Return Probability of Occurrence
He defined the risk of a portfolio as the sum of the
1 12% 0.18 variances of the investments and covariances among the
2 10 0.24 investments. The notion of introducing the covariances
3 8 0.29 among returns of the investments in the portfolio to mea-
4 4 0.16 sure the risk of a portfolio forever changed how the in-
5 −4 0.13 vestment community thought about the concept of risk.
Total 1.00

Variance and Standard Deviation

value of a random variable is simply the weighted av-
erage of the possible outcomes, where the weight is the as a Measure of Risk
probability associated with the possible outcome. The variance of a random variable is a measure of the
In our case, the random variable is the uncertain return dispersion or variability of the possible outcomes around
of asset i. Having specified a probability distribution for the expected value (mean). In the case of an asset’s return,
the possible rates of return, the expected value of the rate the variance is a measure of the dispersion of the possible
of return for asset i is the weighted average of the possi- rate of return outcomes around the expected return.
ble outcomes. Finally, rather than use the term “expected The equation for the variance of the expected return for
value of the return of an asset,” we simply use the term asset i, denoted var(Ri ), is
“expected return.” Mathematically, the expected return of
asset i is expressed as: var(Ri ) = p1 [r1 − E(Ri )]2 + p2 [r2 − E(Ri )]2
+ · · · + p N [r N − E(Ri )]2
E(Ri ) = p1 R1 + p2 R2 + · · · + p N RN (1.4)
or,
where,

N
Rn = the nth possible rate of return for asset i var(Ri ) = pn [rn − E(Ri )]2 (1.5)
pn = the probability of attaining the rate of return n for i=1
asset i
Using the probability distribution of the return for stock
N = the number of possible outcomes for the rate of
XYZ, we can illustrate the calculation of the variance:
return
var(RXY Z ) = 0.18(12% − 7%)2 + 0.24(10% − 7%)2
How do we specify the probability distribution of re-
turns for an asset? We shall see later on in this chapter + 0.29(8% − 7%)2 + 0.16(4% − 7%)2
that in most cases the probability distribution of returns is + 0.13(−4% − 7%)2
based on historical returns. Probabilities assigned to dif-
= 24.1%
ferent return outcomes that are based on the past perfor-
mance of an uncertain investment act as a good estimate The variance associated with a distribution of returns
of the probability distribution. However, for purpose of measures the compactness with which the distribution is
illustration, assume that an investor is considering an in- clustered around the mean or expected return. Markowitz
vestment, stock XYZ, which has a probability distribution argued that this compactness or variance is equivalent to
for the rate of return for some time period as given in the uncertainty or riskiness of the investment. If an asset
Table 1.1. The stock has five possible rates of return and is riskless, it has an expected return dispersion of zero. In
the probability distribution specifies the likelihood of oc- other words, the return (which is also the expected return
currence (in a probabilistic sense) for each of the possible in this case) is certain, or guaranteed.
outcomes.
Substituting into equation (1.4) we get
Standard Deviation
E(RXY Z ) = 0.18(12%) + 0.24(10%) + 0.29(8%)
Since the variance is squared units, it is common to see the
+ 0.16(4%) + 0.13(−4%) variance converted to the standard deviation by taking the
= 7% positive square root of the variance:

Thus, 7% is the expected return or mean of the probability SD(Ri ) = var(Ri )
distribution for the rate of return on stock XYZ.
For stock XYZ, then, the standard deviation is
√
SD(RXYZ ) = 24.1% = 4.9%
MEASURING PORTFOLIO RISK The variance and standard deviation are conceptually
The dictionary defines risk as “hazard, peril, exposure to equivalent; that is, the larger the variance or standard de-
loss or injury.” With respect to investments, investors have viation, the greater the investment risk.
JWPR026-Fabozzi c01 June 22, 2008 6:54

INVESTMENT MANAGEMENT 7

There are two criticisms of the use of the variance as a formula is:
measure of risk. The first criticism is that since the variance
measures the dispersion of an asset’s return around its var(R p ) = var(Ri ) + var(R j ) + 2wi w j cov(Ri , R j ) (1.6)
expected return, it treats both the returns above and below where
the expected return identically. There has been research in
the area of behavioral finance to suggest that investors do cov(Ri , R j ) = covariance between the return for assets
not view return outcomes above the expected return in iand j
the same way as they view returns below the expected In words, equation (1.6) states that the variance of the
return. Whereas returns above the expected return are portfolio return is the sum of the squared weighted vari-
considered favorable, outcomes below the expected return ances of the two assets plus two times the weighted covari-
are disliked. Because of this, some researchers have argued ance between the two assets. We will see that this equation
that measures of risk should not consider the possible can be generalized to the case where there are more than
return outcomes above the expected return. two assets in the portfolio.
Markowitz recognized this limitation and, in fact, sug-
gested a measure of downside risk—the risk of realizing
an outcome below the expected return—called the semi- Covariance
variance. The semivariance is similar to the variance ex- Like the variance, the covariance has a precise mathemati-
cept that in the calculation no consideration is given to cal translation. Its practical meaning is the degree to which
returns above the expected return. However, because of the returns on two assets co-vary or change together. In
the computational problems with using the semivariance fact, the covariance is just a generalized concept of the
and the limited resources available to him at the time, he variance applied to multiple assets. A positive covariance
used the variance in developing portfolio theory. between two assets means that the returns on two assets
Today, practitioners use various measures of downside tend to move or change in the same direction, while a
risk. However, regardless of the measure used, the basic negative covariance means the returns tend to move in
principles of portfolio theory developed by Markowitz opposite directions. The covariance between any two as-
and set forth in this chapter are still applicable. That is, sets i and j is computed using the following formula:
the choice of the measure of risk may affect the calculation
but doesn’t invalidate the theory. cov(Ri , R j ) = p1 [ri1 − E(Ri )][r j1 − E(R j )]
The second criticism is that the variance is only one + p2 [ri2 − E(Ri )][r j2 − E(R j )]
measure of how the returns vary around the expected
+ · · · + p N [ri N − E(Ri )][r j N − E(R j )]
return. When a probability distribution is not symmetrical
around its expected return, then a statistical measure of the (1.7)
skewness of a distribution should be used in addition to
the variance (see Ortobelli et al., 2005). where
Because expected return and variance are the only two rin = the nth possible rate of return for asset i
parameters that investors are assumed to consider in rjn = the nth possible rate of return for asset j
making investment decisions, the Markowitz formula- pn = the probability of attaining the rate of return n for
tion of portfolio theory is often referred to as a “two- assets i and j
parameter model.” It is also referred to as “mean-variance N = the number of possible outcomes for the rate of
analysis.” return.
The covariance between asset i and i is just the variance
Measuring the Portfolio Risk of a Two-Asset Portfolio of asset i.
Equation (1.5) gives the variance for an individual asset’s To illustrate the calculation of the covariance between
return. The variance of a portfolio consisting of two assets two assets, we use the two stocks in Table 1.2. The first is
is a little more difficult to calculate. It depends not only on stock XYZ from Table 1.1 that we used earlier to illustrate
the variance of the two assets, but also upon how closely the calculation of the expected return and the standard de-
the returns of one asset track those of the other asset. The viation. The other hypothetical stock is ABC whose data

Table 1.2 Probability Distribution for the Rate of Return for Asset XYZ and Asset ABC

n Rate of Return for Asset XYZ Rate of Return for Asset ABC Probability of Occurrence
1 12% 21% 0.18
2 10 14 0.24
3 8 9 0.29
4 4 4 0.16
5 −4 −3 0.13
Total 1.00
Expected return 7.0% 10.0%
Variance 24.1% 53.6%
Standard deviation 4.9% 7.3%
JWPR026-Fabozzi c01 June 22, 2008 6:54

8 Portfolio Selection

are shown in Table 1.2. Substituting the data for the two In (1.10), the terms for which h = g results in the variances
stocks from Table 1.2 in equation (1.7), the covariance be- of the G assets, and the terms for which h = g results in
tween stocks XYZ and ABC is calculated as follows: all possible pair-wise covariances amongst the G assets.
cov(RXY Z , RABC ) = 0.18(12% − 7%)(21% − 10%) Therefore, (1.10) is a shorthand notation for the sum of all
G variances and the possible covariances amongst the G
+ 0.24(10% − 7%)(14% − 10%) assets.
+ 0.29(8% − 7%)(9% − 10%)
+ 0.16(4% − 7%)(4% − 10%)
PORTFOLIO DIVERSIFICATION
+ 0.13(−4% − 7%)(−3% − 10%)
Often, one hears investors talking about diversifying their
= 34 portfolio. By this an investor means constructing a portfo-
lio in such a way as to reduce portfolio risk without sacri-
Relationship between Covariance and Correlation ficing return. This is certainly a goal that investors should
The correlation is analogous to the covariance between the seek. However, the question is how to do this in practice.
expected returns for two assets. Specifically, the correla- Some investors would say that including assets across
tion between the returns for assets i and j is defined as all asset classes could diversify a portfolio. For example, a
the covariance of the two assets divided by the product of investor might argue that a portfolio should be diversified
their standard deviations: by investing in stocks, bonds, and real estate. While that
might be reasonable, two questions must be addressed
cor(Ri , R j ) = cov(Ri , R j )/[SD(Ri )SD(R j )] (1.8)
in order to construct a diversified portfolio. First, how
The correlation and the covariance are conceptually much should be invested in each asset class? Should 40%
equivalent terms. Dividing the covariance between the of the portfolio be in stocks, 50% in bonds, and 10% in
returns of two assets by the product of their standard de- real estate, or is some other allocation more appropriate?
viations results in the correlation between the returns of Second, given the allocation, which specific stocks, bonds,
the two assets. Because the correlation is a standardized and real estate should the investor select?
number (that is, it has been corrected for differences in the Some investors who focus only on one asset class such
standard deviation of the returns), the correlation is com- as common stock argue that such portfolios should also be
parable across different assets. The correlation between diversified. By this they mean that an investor should not
the returns for stock XYZ and stock ABC is place all funds in the stock of one corporation, but rather
should include stocks of many corporations. Here, too,
cor(RXYZ , RABC ) = 34/(4.9 × 7.3) = 0.94 several questions must be answered in order to construct
The correlation coefficient can have values ranging from a diversified portfolio. First, which corporations should
+1.0, denoting perfect co-movement in the same direction, be represented in the portfolio? Second, how much of
to –1.0, denoting perfect co-movement in the opposite di- the portfolio should be allocated to the stocks of each
rection. Also note that because the standard deviations corporation?
are always positive, the correlation can be negative only if Prior to the development of portfolio theory, while in-
the covariance is a negative number. A correlation of zero vestors often talked about diversification in these general
implies that the returns are uncorrelated. Finally, though terms, they did not possess the analytical tools by which to
causality implies correlation, correlation does not imply answer the questions posed above. For example, in 1945,
causality. D. H. Leavens (1945) wrote:
An examination of some fifty books and articles on in-
vestment that have appeared during the last quarter of
Measuring the Risk of a Portfolio a century shows that most of them refer to the desirabil-
Comprised of More than Two Assets ity of diversification. The majority, however, discuss it
So far, we have defined the risk of a portfolio consisting in general terms and do not clearly indicate why it is
of two assets. The extension to three assets—i, j, and k—is desirable.
as follows: Leavens illustrated the benefits of diversification on the
var(R p ) = wi2 var(Ri ) + w 2j var(R j ) + wk2 var(Rk ) assumption that risks are independent. However, in the
last paragraph of his article, he cautioned:
+ 2wi w j cov(Ri , R j ) + 2wi wk cov(Ri , Rk )
The assumption, mentioned earlier, that each security
+ 2w j wk cov(R j , Rk )
(1.9) is acted upon by independent causes, is important, al-
though it cannot always be fully met in practice. Di-
In words, equation (1.9) states that the variance of the
versification among companies in one industry cannot
portfolio return is the sum of the squared weighted vari- protect against unfavorable factors that may affect the
ances of the individual assets plus two times the sum of whole industry; additional diversification among in-
the weighted pair-wise covariances of the assets. In gen- dustries is needed for that purpose. Nor can diversifi-
eral, for a portfolio with G assets, the portfolio variance is cation among industries protect against cyclical factors
given by, that may depress all industries at the same time.

G

G A major contribution of the theory of portfolio selection
var(R p ) = wg wh cov(Rg , Rh ) (1.10) is that using the concepts discussed above, a quantitative
g=1 h=1 measure of the diversification of a portfolio is possible,
JWPR026-Fabozzi c01 June 22, 2008 6:54

INVESTMENT MANAGEMENT 9

and it is this measure that can be used to achieve the The Effect of the Correlation of Asset
maximum diversification benefits. Returns on Portfolio Risk
The Markowitz diversification strategy is primarily
concerned with the degree of covariance between as- How would the risk change for our two-asset portfolio
set returns in a portfolio. Indeed a key contribution of with different correlations between the returns of the com-
Markowitz diversification is the formulation of an asset’s ponent stocks? Let’s consider the following three cases for
risk in terms of a portfolio of assets, rather than in isola- cor(RC ,RD ): +1.0, 0, and –1.0. Substituting into equation
tion. Markowitz diversification seeks to combine assets (1.11) for these three cases of cor(RC ,RD ), we get
in a portfolio with returns that are less than perfectly
positively correlated, in an effort to lower portfolio risk cor E SD
(variance) without sacrificing return. It is the concern for
maintaining return while lowering risk through an analy- +1.0 15% 35%
sis of the covariance between asset returns, that separates 0.0 15 25
−1.0 15 5
Markowitz diversification from a naive approach to diver-
sification and makes it more effective.
Markowitz diversification and the importance of asset As the correlation between the expected returns on
correlations can be illustrated with a simple two-asset stocks C and D decreases from +1.0 to 0.0 to –1.0, the
portfolio example. To do this, we will first show the gen- standard deviation of the expected portfolio return also
eral relationship between the risk of a two-asset portfolio decreases from 35% to 5%. However, the expected portfo-
and the correlation of returns of the component assets. lio return remains 15% for each case.
Then we will look at the effects on portfolio risk of com- This example clearly illustrates the effect of Markowitz
bining assets with different correlations. diversification. The principle of Markowitz diversification
states that as the correlation (covariance) between the re-
Portfolio Risk and Correlation turns for assets that are combined in a portfolio decreases,
In our two-asset portfolio, assume that asset C and as- so does the variance (hence the standard deviation) of the
set D are available with expected returns and standard return for the portfolio. This is due to the degree of corre-
deviations as shown: lation between the expected asset returns.
The good news is that investors can maintain expected
Asset E(R) SD(R) portfolio return and lower portfolio risk by combining
Asset C 12% 30%
assets with lower (and preferably negative) correlations.
Asset D 18% 40% However, the bad news is that very few assets have small
to negative correlations with other assets! The problem,
then, becomes one of searching among large numbers of
If an equal 50% weighting is assigned to both stocks C
assets in an effort to discover the portfolio with the min-
and D, the expected portfolio return can be calculated as
imum risk at a given level of expected return or, equiva-
shown:
lently, the highest expected return at a given level of risk.
E(R p ) = 0.50(12%) + 0.50(18%) = 15% The stage is now set for a discussion of efficient portfo-
The variance of the return on the two-stock portfolio from lios and their construction.
equation (1.6) is:
var(R p ) = wC2 var(RC ) + w 2D w D 2 var(RD )
+2wC w D cov(RC , RD )
= (0.5)2 (30%)2 + (0.5)2 (40%)2 CHOOSING A PORTFOLIO
+ 2(0.5)(0.5) cov(RC , RD ) OF RISKY ASSETS
From equation (1.8), Diversification in the manner suggested by Professor
Markowitz leads to the construction of portfolios that have
cor(RC , RD ) = cov(RC , RD )/[SD(RC )SD(RD )] the highest expected return at a given level of risk. Such
so portfolios are called efficient portfolios. In order to con-
cov(RC , RD ) = SD (RC ) SD(RD ) cor(RC , RD ) struct efficient portfolios, the theory makes some basic as-
Since SD(RC ) = 30% and SD(RD ) = 40%, then sumptions about asset selection behavior by the entities.
The assumptions are as follows:
cov(RC , RD ) = (30%)(40%) cor(RC , RD )
Substituting into the expression for var(Rp ), we get Assumption 1: The only two parameters that affect an in-
vestor’s decision are the expected return and the vari-
var(R p ) = (0.5)2 (30%)2 + (0.5)2 (40%)2 ance. (That is, investors make decisions using the two-
+ 2(0.5)(0.5)(30%)(40%) cor(RC , RD ) parameter model formulated by Markowitz.)
Assumption 2: Investors are risk averse. (That is, when
Taking the square root of the variance gives faced with two investments with the same expected
SD(R
 p)
return but two different risks, investors will prefer the
= (0.5)2 (30%)2 + (0.5)2 (40%)2 + 2(0.5)(0.5)(30%)(40%) cor(RC , RD ) one with the lower risk.)
= 625 + (600) cor(RC , RD ) Assumption 3: All investors seek to achieve the highest
(1.11) expected return at a given level of risk.
JWPR026-Fabozzi c01 June 22, 2008 6:54

10 Portfolio Selection

Table 1.3 Portfolio Expected Returns and Standard Deviations 20%

for Five Mixes of Assets C and D 5
4
Proportion Proportion 15% 3
Portfolio of Asset C of Asset D E(Rp ) SD(Rp )
2
1 100% 0% 12.0% 30.0% 1

ER(Rp)
2 75 25 13.5% 19.5% 10% Feasible set represented by curve 1–5
3 50 50 15.0% 18.0% Markowitz efficient set: portion of curve 3–5
4 25 75 16.5% 27.0%
5 0 100 18.0% 40.0% 5%
Asset C: E(RC ) = 12%, SD(RC ) = 30%
Asset D: E(RD ) = 18%, and SD(RD ) = 40%
Correlation between Asset C. and D = cor(RC ,RD ) = −0.5 0%
0% 10% 20% 30% 40% 50%
SD(Rp)
Assumption 4: All investors have the same expectations re-
garding expected return, variance, and covariances for Figure 1.2 Feasible and Efficient Portfolios for Assets C
all risky assets. (This is referred to as the homogeneous and D
expectations assumption.)
Assumption 5: All investors have a common one-period
investment horizon. possible combinations of C and D lie between portfolios 1
and 5, or on the curve labeled 1–5. In the case of two assets,
any risk/return combination not lying on this curve is not
Constructing Efficient Portfolios attainable, since there is no mix of assets C and D that will
result in that risk/return combination. Consequently, the
The technique of constructing efficient portfolios from curve 1–5 can also be thought of as the feasible set.
large groups of stocks requires a massive number of calcu- In contrast to a feasible portfolio, an efficient portfolio is
lations. In a portfolio of G securities, there are (G2 − G)/2 one that gives the highest expected return of all feasible
unique covariances to calculate. Hence, for a portfolio of portfolios with the same risk. An efficient portfolio is also
just 50 securities, there are 1,224 covariances that must said to be a mean-variance efficient portfolio. Thus, for
be calculated. For 100 securities, there are 4,950. Further- each level of risk there is an efficient portfolio. The collec-
more, in order to solve for the portfolio that minimizes risk tion of all efficient portfolios is called the efficient set.
for each level of return, a mathematical technique called The efficient set for the feasible set presented in Figure
quadratic programming must be used. A discussion of this 1.2 is differentiated by the bold curve section 3–5. Efficient
technique is beyond the scope of this chapter. However, portfolios are the combinations of assets C and D that re-
it is possible to illustrate the general idea of the construc- sult in the risk/return combinations on the bold section of
tion of efficient portfolios by referring again to the simple the curve. These portfolios offer the highest expected re-
two-asset portfolio consisting of assets C and D. turn at a given level of risk. Notice that two of our five port-
Recall that for two assets, C and D, E(RC ) = 12%, folio mixes—portfolio 1 with E(Rp ) = 12% and SD(Rp ) =
SD(RC ) = 30%, E(RD ) = 18%, and SD(RD ) = 40%. We now 20% and portfolio 2 with E(Rp ) = 13.5% and SD(Rp ) =
further assume that cor(RC ,RD ) = –0.5. Table 1.3 presents 19.5%—are not included in the efficient set. This is be-
the expected portfolio return and standard deviation for cause there is at least one portfolio in the efficient set (for
five different portfolios made up of varying proportions example, portfolio 3) that has a higher expected return
of C and D. and lower risk than both of them. We can also see that
portfolio 4 has a higher expected return and lower risk
than portfolio 1. In fact, the whole curve section 1–3 is not
Feasible and Efficient Portfolios efficient. For any given risk/return combination on this
A feasible portfolio is any portfolio that an investor can con- curve section, there is a combination (on the curve section
struct given the assets available. The five portfolios pre- 3–5) that has the same risk and a higher return, or the same
sented in Table 1.3 are all feasible portfolios. The collection return and a lower risk, or both. In other words, for any
of all feasible portfolios is called the feasible set of portfo- portfolio that results in the return/risk combination on
lios. With only two assets, the feasible set of portfolios is the curve section 1–3 (excluding portfolio 3), there exists a
graphed as a curve that represents those combinations of portfolio that dominates it by having the same return and
risk and expected return that are attainable by construct- lower risk, or the same risk and a higher return, or a lower
ing portfolios from all possible combinations of the two risk and a higher return. For example, portfolio 4 domi-
assets. nates portfolio 1, and portfolio 3 dominates both portfolios
Figure 1.2 presents the feasible set of portfolios for all 1 and 2.
combinations of assets C and D. As mentioned earlier, the Figure 1.3 shows the feasible and efficient sets when
portfolio mixes listed in Table 1.3 belong to this set and there are more than two assets. In this case, the feasible
are shown by the points 1 through 5, respectively. Starting set is not a line, but an area. This is because, unlike the
from 1 and proceeding to 5, asset C goes from 100% to two-asset case, it is possible to create asset portfolios that
0%, while asset D goes from 0% to 100%—therefore, all result in risk/return combinations that not only result in
JWPR026-Fabozzi c01 June 22, 2008 6:54

INVESTMENT MANAGEMENT 11

Feasible set: All portfolios on and bounded u3

by curve I–II–III
Markowitz efficient set: All portfolios on III u2
curve II–III u1

Markowitz
efficient
ER(Rp)

frontier

ER(Rp)
u3
P*MEF
P*MEF
II u2
u1

I
Risk [SD(Rp)]
SD(Rp)
Figure 1.3 Feasible and Efficient Portfolios with More
u1, u2, u3 = indifference curves with u1 < u2 < u3
Than Two Assets*
*
* = Optimal portfolio on Markowitz efficient frontier
PMEF
The picture is for illustrative purposes only. The actual shape of
the feasible region depends on the returns and risks of the assets Figure 1.4 Selection of the Optimal Portfolio
chosen and the correlation among them.

In Figure 1.4, three indifference curves representing a

utility function and the efficient frontier are drawn on the
combinations that lie on the curve I–II–III, but all combi-
same diagram. An indifference curve indicates the com-
nations that lie in the shaded area. However, the efficient
binations of risk and expected return that give the same
set is given by the curve II–III. It is easily seen that all the
level of utility. Moreover, the farther the indifference curve
portfolios on the efficient set dominate the portfolios in
from the horizontal axis, the higher the utility.
the shaded area.
From Figure 1.4, it is possible to determine the opti-
The efficient set of portfolios is sometimes called the
mal portfolio for the investor with the indifference curves
efficient frontier, because graphically all the efficient port-
shown. Remember that the investor wants to get to the
folios lie on the boundary of the set of feasible portfolios
highest indifference curve achievable given the efficient
that have the maximum return for a given level of risk.
frontier. Given that requirement, the optimal portfolio
Any risk/return combination above the efficient frontier
is represented by the point where an indifference curve
cannot be achieved, while risk/return combinations of
is tangent to the efficient frontier. In Figure 1.4, that is
the portfolios that make up the efficient frontier dominate ∗ ∗
the portfolio PMEF . For example, suppose that PMEF cor-
those that lie below the efficient frontier.
responds to portfolio 4 in Figure 1.2. We know from
Table 1.3 that this portfolio is made up of 25% of asset
C and 75% of asset D, with an E(Rp ) = 16.5% and SD(Rp ) =
27.0%.
Choosing the Optimal Portfolio
Consequently, for the investor’s preferences over risk
in the Efficient Set and return as determined by the shape of the indifference
Now that we have constructed the efficient set of portfo- curves represented in Figure 1.4, and expectations for as-
lios, the next step is to determine the optimal portfolio. set C and D inputs (returns and variance-covariance) rep-
Since all portfolios on the efficient frontier provide the resented in Table 1.2, portfolio 4 is the optimal portfolio
greatest possible return at their level of risk, an investor because it maximizes the investor’s utility. If this investor
or entity will want to hold one of the portfolios on the had a different preference for expected risk and return,
efficient frontier. Notice that the portfolios on the efficient there would have been a different optimal portfolio. For
frontier represent trade-offs in terms of risk and return. example, Figure 1.5 shows the same efficient frontier but
Moving from left to right on the efficient frontier, the risk three other indifference curves. In this case, the optimal
∗∗
increases, but so does the expected return. The question is portfolio is PMEF , which has a lower expected return and
∗
which one of those portfolios should an investor hold? The risk than PMEF in Figure 1.4. Similarly, if the investor had
best portfolio to hold of all those on the efficient frontier a different set of input expectations, the optimal portfolio
is the optimal portfolio. would be different.
Intuitively, the optimal portfolio should depend on the At this point in our discussion, a natural question is how
investor’s preference over different risk/return trade-offs. to estimate an investor’s utility function so that the indif-
As explained earlier, this preference can be expressed in ference curves can be determined. Unfortunately, there
terms of a utility function. is little guidance about how to construct one. In general,
JWPR026-Fabozzi c01 June 22, 2008 6:54

12 Portfolio Selection

u3 u2 u1 follows (Markowitz, 1959, pp. 96–101):

ri = αi + βi F + ui (1.12)
Markowitz
efficient
frontier where
P**
MEF
ri = the return on security i
F = value of some index
ER(Rp)

ui = error term

u3
(Note that Markowitz (1959) used the notation I in propos-
ing the model given by equation (1.12).) The expected
u2 value of ui is zero and ui is uncorrelated with F and every
u1 other uj .
The parameters α i and β i are parameters to be estimated.
When measured using regression analysis, β i is the ratio
of the covariance of asset i’s return and F to the variance
SD(Rp) of F.
Markowitz further suggested that the relationship need
u1, u2, u3 = indifference curves with u1 < u2 < u3 not be linear and that there could be several underlying
P**
MEF
= Optimal portfolio on Markowitz efficient frontier factors.
Figure 1.5 Selection of Optimal Portfolio with Different
Indifference Curves (Utility Function)
Single-Index Market Model
Sharpe (1963) tested equation (1.12) as an explanation of
economists have not been successful in estimating utility how security returns tend to go up and down together
functions. with general market index, F. For the index in the mar-
The inability to estimate utility functions does not mean ket model he used a market index for F. Specifically,
that the theory is flawed. What it does mean is that once Sharpe estimated using regression analysis the following
an investor constructs the efficient frontier, the investor model:
will subjectively determine which efficient portfolio is ap-
rit = αi + βi rmt + uit (1.13)
propriate given his or her tolerance to risk.

where
rit = return on asset i over the period t
INDEX MODEL’S rmt = return on the market index over the period t
APPROXIMATIONS TO THE α i = a term that represents the nonmarket component
of the return on asset i
COVARIANCE STRUCTURE β i = the ratio of the covariance of the return of asset i
The inputs to mean-variance analysis include expected and the return of the market index to the variance
returns, variance of returns, and either covariance or cor- of the return of the market index
relation of returns between each pair of securities. For uit = a zero mean random error term
example, an analysis that allows 200 securities as possible
candidates for portfolio selection requires 200 expected The model given by equation (1.13) is called the single
returns, 200 variances of return, and 19,900 correlations or index market model or simply the market model. It is im-
covariances. An investment team tracking 200 securities portant to note that when Markowitz discussed the pos-
may reasonably be expected to summarize their analyses sibility of using equation (1.12) to estimate the covariance
in terms of 200 means and variances, but it is clearly unrea- structure, the index he suggested was not required to be a
sonable for them to produce 19,900 carefully considered market index.
correlations or covariances. Suppose that the Dow Jones Wilshire 5000 is used to
It was clear to Markowitz that some kind of model of represent the market index, then for a portfolio of G assets
covariance structure was needed for the practical appli- regression analysis is used to estimate the values of the βs
cation of normative analysis to large portfolios. He did and αs. The beta of the portfolio (β p ), is simply a weighted
little more than point out the problem and suggest some average of the computed betas of the individual assets
possible models of covariance for research. (β i ), where the weights are the percentage of the market
One model Markowitz proposed to explain the correla- value of the individual assets relative to the total market
tion structure among security returns assumed that the re- value of the portfolio. That is,
turn on the ith security depends on an “underlying factor,

G
the general prosperity of the market as expressed by some βp = wi βi
index.” Mathematically, the relationship is expressed as i=1
JWPR026-Fabozzi c01 June 22, 2008 6:54

INVESTMENT MANAGEMENT 13

So, for example, the beta for a portfolio comprised of the of the portfolio. The variance or the standard deviation
following: of an asset’s returns measures its risk. Unlike the port-
folio’s expected return, a portfolio’s risk is not a simple
Company Weight β weighting of the standard deviation of the individual as-
sets in the portfolio. Rather, the covariance or correlation
General Electric 20% 1.24 between the assets that comprise the portfolio affects the
McGraw-Hill 25% 0.86 portfolio risk. The lower the correlation, the smaller the
IBM 15% 1.22
risk of the portfolio.
General Motors 10% 1.11
Xerox 30% 1.27 Markowitz has set forth the theory for the construction
of an efficient portfolio, which has come to be called a ef-
ficient portfolio—a portfolio that has the highest expected
would have the following beta: return of all feasible portfolios with the same level of risk.
Portfolio beta = 20%(1.24) + 25%(0.86) + 15%(1.22) The collection of all efficient portfolios is called the effi-
+ 10%(1.11) + 30%(1.27) cient set of portfolios or the efficient frontier.
= 1.14 The optimal portfolio is the one that maximizes an in-
vestor’s preferences with respect to return and risk. An
investor’s preference is described by a utility function,
Multi-Index Market Models which can be represented graphically by a set of indif-
Sharpe concluded that equation (1.12) was as complex a ference curves. The utility function shows how much an
covariance as seemed to be needed. This conclusion was investor is willing to trade off between expected return
supported by research of Cohen and Pogue (1967). King and risk. The optimal portfolio is the one where an indif-
(1966) found strong evidence for industry factors in ad- ference curve is tangent to the efficient frontier.
dition to the marketwide factor. Rosenberg (1974) found
other sources of risk beyond a marketwide factor and in-
dustry factor.
One alternative approach to full mean-variance analysis REFERENCES
is the use of these multi-index or factor models to obtain Rosenberg, B. (1974). Extra-market components of covari-
the covariance structure. ance in security returns. Journal of Financial and Quanti-
tative Analysis 19, 2: 23–274.
Cohen, K. J., and Pogue, G. A. (1967). An empirical eval-
uation of alternative portfolio selection models. Journal
SUMMARY of Business 40, 2: 166–193.
In this chapter we have introduced portfolio theory. De- King, B. F. (1966). Market and industry factors in stock
veloped by Markowitz, this theory explains how investors price behavior. Journal of Business 39, 1 (Part 2: Supple-
should construct efficient portfolios and select the best ment on Security Prices): 139–190.
or optimal portfolio from among all efficient portfolios. Leavens, D. H. (1945). Diversification of investments.
The theory differs from previous approaches to portfolio Trusts and Estates 80: 469–473.
selection in that he demonstrated how the key parame- Markowitz, H. M. (1952). Portfolio selection. Journal of
ters should be measured. These parameters include the Finance 7, 1: 77–91
risk and the expected return for an individual asset and a Markowitz, H. M. (1959). Portfolio Selection: Efficient Diver-
portfolio of assets. Moreover, the concept of diversifying sification of Investments. Cowles Foundation Monograph
a portfolio, the goal of which is to reduce a portfolio’s risk 16, New York: John Wiley & Sons.
without sacrificing expected return, can be cast in terms Ortobelli, S., Rachev, S. T., Stoyanov, S., Fabozzi, F. J., and
of these key parameters plus the covariance or correlation Biglova, A. (2005). The proper use of risk measures in
between assets. All these parameters are estimated from portfolio theory. International Journal of Theoretical and
historical data and other sources of information and draw Applied Finance 8, 8: 1–27.
from concepts in probability and statistical theory. Sharpe, W. F. (1963). A simplified model for portfolio anal-
A portfolio’s expected return is simply a weighted av- ysis. Management Science 9, 2: 277–293.
erage of the expected return of each asset in the portfolio. von Neumann, J., and Morgenstern, O. (1944). Theory of
The weight assigned to each asset is the market value of Games and Economic Behavior. Princeton, NJ: Princeton
the asset in the portfolio relative to the total market value University Press.
JWPR026-Fabozzi c01 June 22, 2008 6:54

14