6 Modern Portfolio Theory

A Generalizing the Portfolio Problem

B Justifying Mean-Variance Utility
C The Gains From Diversification
D The Efficient Frontier
E A Separation Theorem
F Strengths and Shortcomings of MPT
Generalizing the Portfolio Problem

We can elaborate on our previous portfolio problem

max E {u[Y0 (1 + rf ) + a(˜

r − rf )]}
a

by allowing the investor to allocate funds to N > 1 risky

assets.

r˜1 , r˜2 , . . . , r˜N risky (random) returns

a1 , a2 , . . . , aN amounts allocated at the risky assets
wi = ai /Y0 share of initial wealth allocated to each
risky asset (portfolio w eights)
Generalizing the Portfolio Problem

Ỹ1 = random terminal wealth

N
! N
X X
= (1 + rf ) Y0 − ai + ai (1 + r˜i )
i=1 i=1
N
X
= (1 + rf )Y0 + ri − rf )
ai (˜
i=1
N
X
= (1 + rf )Y0 + ri − rf )
wi Y0 (˜
i=1
Generalizing the Portfolio Problem

With
N
X
Ỹ1 = (1 + rf )Y0 + ri − rf ),
wi Y0 (˜
i=1

the generalized problem can be stated as

( " N
#)
X
max E u Y0 (1 + rf ) + ri − rf )
wi Y0 (˜
w1 ,w2 ,...,wN
i=1
Generalizing the Portfolio Problem

Modern Portfolio Theory examines the solution to this

extended problem assuming that investors have mean-variance
utility, that is, assuming that investors’ preferences can be
represented by a trade-off between the mean (expected value)
and variance (or standard deviation) of terminal wealth.

MPT was developed by Harry Markowitz (US, b.1927, Nobel

Prize 1990) in the early 1950s, the classic paper being his
article “Portfolio Selection,” Journal of Finance Vol.7 (March
1952): pp.77-91.
Generalizing the Portfolio Problem

The mean-variance utility hypothesis seemed natural at the

time the MPT first appeared, and it retains some intuitive
appeal today. But viewed in the context of more recent
developments in financial economics, particularly the
development of vN-M expected utility theory, it now looks a
bit peculiar.

A first question for us, therefore, is: Under what conditions

will investors have preferences over the means and variances of
asset returns?
Generalizing the Portfolio Problem

Under what conditions will investors have preferences over the

means and variances of asset returns?
1. Portfolio risks are small, so that a quadratic
approximation to a general Bernoulli utility function is
accurate.
2. The Bernoulli utility function is quadratic, so that the
approximation in (1) is always exact.
3. Asset returns are normally distributed, so that terminal
wealth is normally distributed as well.
Justifying Mean-Variance Utility
If we start by assuming an investor has preferences over
terminal wealth Ỹ1 described by a vN-M expected utility
function
E [u(Y˜1 )]
we can write

Ỹ1 = E (Ỹ1 ) + [Ỹ1 − E (Y˜1 )]

and interpret the portfolio problem as a trade-off between the

expected payoff
E (Y˜1 )
and the size of the “bet”

[Y˜1 − E (Y˜1 )]
Justifying Mean-Variance Utility

With this interpretation in mind, consider a second-order

Taylor approximation of the Bernoulli utility function u once
the outcome [Ỹ1 − E (Ỹ1 )] of the bet is known:

u(Ỹ1 ) ≈ u[E (Y˜1 )] + u 0 [E (Y˜1 )][Y˜1 − E (Y˜1 )]

1
+ u 00 [E (Y˜1 )][Y˜1 − E (Y˜1 )]2
2
Justifying Mean-Variance Utility

u(Y˜1 ) ≈ u[E (Y˜1 )] + u 0 [E (Y˜1 )][Y˜1 − E (Y˜1 )]

1
+ u 00 [E (Y˜1 )][Y˜1 − E (Y˜1 )]2
2
Now go back to the beginning of the period, before the
outcome of the bet is known, and take expected values to
obtain
1
E [u(Y˜1 )] ≈ u[E (Y˜1 )] + u 00 [E (Y˜1 )]σ 2 (Y˜1 )
2
since

E [Y˜1 − E (Y˜1 )] = 0 and E {[Y˜1 − E (Y˜1 )]2 } = σ 2 (Y˜1 )

Justifying Mean-Variance Utility

1
E [u(Y˜1 )] ≈ u[E (Y˜1 )] + u 00 [E (Y˜1 )]σ 2 (Y˜1 )
2
The right-hand side of this expression is in the desired form: if
u is increasing, it rewards higher mean returns and if u is
concave, it penalizes higher variance in returns.

So one possible justification for mean-variance utility is to

assume that the size of the portfolio bet Y˜1 − E (Y˜1 ) is small
enough to make this Taylor approximation a good one.

But is it safe to assume that portfolio bets are small?

Justifying Mean-Variance Utility

A second possibility is to assume that the Bernoulli utility

function is quadratic, with

u(Y ) = a + bY1 + cY12 ,

with b > 0 and c < 0. Then

u 0 (Y1 ) = b + 2cY1 and u 00 (Y1 ) = 2c.

In this case, the second-order (quadratic) Taylor approximation

holds exactly, even for large bets.
Justifying Mean-Variance Utility

Note, however, that for a quadratic utility function

u 00 (Y1 ) 2c
RA (Y1 ) = − = −
u 0 (Y1 ) b + 2cY1

which is increasing in Y1 .

Hence, quadratic utility has the undesirable implication that

the amount of wealth allocated to risky investments declines
when wealth increases.
Justifying Mean-Variance Utility

Fortunately, there is a result from probability theory: if Y˜1 is

normally distributed with mean µY = E (Y˜1 ) and standard
deviation σY1 = {E [Y˜1 − E (Y˜1 )]2 }1/2 then the expectation of
any function of Y˜1 can be written as a function of µY and σY .

Hence, in particular, there exists a function v such that

E [u(Ỹ1 )] = v (µY , σY )
Justifying Mean-Variance Utility

The result follows from a more basic property of the normal

distribution: its location and shape is described completely by
its mean and variance.
Justifying Mean-Variance Utility

If Ỹ1 is normally distributed, there exists a function v such that

E [u(Ỹ1 )] = v (µY , σY ).

Moreover, if Ỹ1 is normally distributed and

1. u is increasing, then v is increasing in µY
2. u is concave, then v is decreasing in σY
3. u is concave, then indifference curves defined over µY and
σY are convex
Justifying Mean-Variance Utility

Since µY is a “good” and σY is a “bad,” indifference curves

slope up. But if u is concave, these indifference curves will still
be convex.
Justifying Mean-Variance Utility

Problems with the normality assumption:

1. Returns on assets like options are highly non-normal.
2. Departures from normality, including skewness
(asymmetry) and excess kurtosis (“fat tails”) can be
detected in returns for both individual stocks and stock
indices.
Justifying Mean-Variance Utility

The mean-variance utility hypothesis is intuitively appealing

and can be justified with reference to vN-M expected utility
theory under various additional assumptions.

Still, it’s important to recognize its limitations: you probably

wouldn’t want to use it to design sophisticated investment
strategies that involve very large risks or make use of options
and you probably wouldn’t want to use it to study how
portfolio strategies or risk-taking behavior changes with wealth.
The Gains From Diversification

One of the most important lessons that we can take from

modern portfolio theory involves the gains from diversification.

To see where these gains come from, consider forming a

portfolio from two risky assets:
r˜1 , r˜2 = random returns
µ1 , µ2 = expected returns
σ1 , σ2 = standard deviations
Assume µ1 > µ2 and σ1 > σ2 to create a trade-off between
expected return and risk.
The Gains From Diversification

If w is the fraction of initial wealth allocated to asset 1 and

1 − w is the fraction of initial wealth allocated to asset 2, the
random return r˜P on the portfolio is

r˜P = w r˜1 + (1 − w )˜
r2

and the expected return µP on the portfolio is

µP = E [w r˜1 + (1 − w )˜
r2 ]
= wE (˜r1 ) + (1 − w )E (˜r2 )
= w µ1 + (1 − w )µ2
The Gains From Diversification

µP = w µ1 + (1 − w )µ2

The expected return on the portfolio is a weighted average of

the expected returns on the individual assets.

Since µ1 > µ2 , µP can range from µ2 up to µ1 as w increases

from zero to one. Even higher (or lower) expected returns are
possible if short selling is allowed.
The Gains From Diversification

But now let’s calculate the variance of the random portfolio

return
r˜P = w r˜1 + (1 − w )˜
r2

σP2 = rP − µP )2 ]
E [(˜
= E {[w r˜1 + (1 − w )˜r2 − w µ1 − (1 − w )µ2 ]2 }
= E {[w (˜r1 − µ1 ) + (1 − w )(˜r2 − µ2 )]2 }
= E [w 2 (˜
r1 − µ1 )2 + (1 − w )2 (˜
r2 − µ2 )2
+ 2w (1 − w )(˜ r1 − µ1 )(˜
r2 − µ2 )]
The Gains From Diversification

σP2 = E [w 2 (˜
r1 − µ1 )2 + (1 − w )2 (˜
r2 − µ2 )2
+ 2w (1 − w )(˜ r1 − µ1 )(˜
r2 − µ2 )]

σP2 = w 2 E [(˜
r1 − µ1 )2 ] + (1 − w )2 E [(˜
r2 − µ2 )2 ]
+ 2w (1 − w )E [(˜ r1 − µ1 )(˜
r2 − µ2 )]
The Gains From Diversification

In probability theory, the covariance between two random

variables X1 and X2 is defined as

σ(X1 , X2 ) = E {[X1 − E (X1 )][X2 − E (X2 )]}

and the correlation between X1 and X2 is defined as

σ(X1 , X2 )
ρ(X1 , X2 ) =
σ(X1 )σ(X2 )
The Gains From Diversification
The covariance

σ(X1 , X2 ) = E {[X1 − E (X1 )][X2 − E (X2 )]}

is positive if
X1 − E (X1 ) and X2 − E (X2 )
tend to have the same sign, negative

X1 − E (X1 ) and X2 − E (X2 )

tend to have opposite signs, and zero if

X1 − E (X1 ) and X2 − E (X2 )

show no tendency to have the same or opposite signs.

The Gains From Diversification

Mathematically, therefore, the covariance

σ(X1 , X2 ) = E {[X1 − E (X1 )][X2 − E (X2 )]}

measures the extent to which the two random variables tend

to move together.

Economically, buying two assets with returns that are

imperfectly, and especially, negatively correlated is like buying
insurance: one return will be high when the other is low and
vice versa, reducing the overall risk of the portfolio.
The Gains From Diversification

The correlation
σ(X1 , X2 )
ρ(X1 , X2 ) =
σ(X1 )σ(X2 )

has the same sign as the covariance, and is therefore also a

measure of co-movement.

But “scaling” the covariance by the two standard deviations

makes the correlation range between −1 and 1:

−1 ≤ ρ(X1 , X2 ) ≤ 1
The Gains From Diversification
Hence

σP2 = w 2 E [(˜
r1 − µ1 )2 ] + (1 − w )2 E [(˜
r2 − µ2 )2 ]
+ 2w (1 − w )E [(˜ r1 − µ1 )(˜
r2 − µ2 )]

implies

σP2 = w 2 σ12 + (1 − w )2 σ22 + 2w (1 − w )σ12

= w 2 σ12 + (1 − w )2 σ22 + 2w (1 − w )σ1 σ2 ρ12

where
σ12 = the covariance between r˜1 and r˜2
ρ12 = the correlation between r˜1 and r˜2
The Gains From Diversification
This is the source of the gains from diversification: the
expected portfolio return

µP = w µ1 + (1 − w )µ2

is a weighted average of the expected returns on the individual

asset returns, but the standard deviation of the portfolio return

σP = [w 2 σ12 + (1 − w )2 σ22 + 2w (1 − w )σ1 σ2 ρ12 ]1/2

is not a weighted average of the standard deviations of the

returns on the individual assets and can be reduced by
choosing a mix of assets (0 < w < 1) when ρ12 is less than
one and, especially, when ρ12 is negative.
The Gains From Diversification
To see more specifically how this works, start with the case
where ρ12 = 1 so that the individual asset returns are perfectly
correlated. This is the one case in which there are no gains
from diversification. With ρ12 = 1,

σP = [w 2 σ12 + (1 − w )2 σ22 + 2w (1 − w )σ1 σ2 ρ12 ]1/2

= [w 2 σ12 + (1 − w )2 σ22 + 2w (1 − w )σ1 σ2 ]1/2
= {[w σ1 + (1 − w )σ2 ]2 }1/2
= |w σ1 + (1 − w )σ2 |.

In this special case, the standard deviation of the return on the

portfolio is a weighted average of the standard deviations of
the returns on the individual assets.
The Gains From Diversification

When ρ12 = 1, so that individual asset returns are perfectly

correlated, there are no gains from diversification.
The Gains From Diversification
Next, let’s consider the opposite extreme, in which ρ12 = −1
so that the individual asset returns are perfectly, but
negatively, correlated:

σP = [w 2 σ12 + (1 − w )2 σ22 + 2w (1 − w )σ1 σ2 ρ12 ]1/2

= [w 2 σ12 + (1 − w )2 σ22 − 2w (1 − w )σ1 σ2 ]1/2
= {[w σ1 − (1 − w )σ2 ]2 }1/2
= |w σ1 − (1 − w )σ2 |.

In this special case, the setting

σ2
w=
σ1 + σ2
creates a “synthetic” risk free portfolio!
The Gains From Diversification

When ρ12 = −1, so that individual asset returns are perfectly,

but negatively correlated, risk can be eliminated via
diversification.
The Gains From Diversification

µP = w µ1 + (1 − w )µ2

σP = [w 2 σ12 + (1 − w )2 σ22 + 2w (1 − w )σ1 σ2 ρ12 ]1/2

In all intermediate cases, there will still be gains from

diversification. These gains will become stronger as ρ12
declines from 1 to −1.
The Gains From Diversification

As ρ12 decreases from 0.5 to 0 to -0.5 to -0.75, the gains from

diversification strengthen.
The Efficient Frontier

µP = w µ1 + (1 − w )µ2

σP = [w 2 σ12 + (1 − w )2 σ22 + 2w (1 − w )σ1 σ2 ρ12 ]1/2

In the case with two risky assets, the choice of w

simultaneously determines µP and σP . But with more than
two risky assets, the portfolio problem takes on an added
dimension, since then we can ask: how can we select
w1 , w2 , . . . , wN to minimize σP for any given choice of µP ?
The Efficient Frontier

Consider two portfolios, A and B, with expected returns µA

and µB and standard deviations σA and σB .

Recall that portfolio A is said to exhibit mean-variance

dominance over portfolio B if either

µA > µB and σA ≤ σB

or
µA ≥ µB and σA < σB
The Efficient Frontier

Hence, choosing portfolio shares to minimize variance for a

given mean will allow us to characterize the efficient frontier:
the set of all portfolios that are not mean-variance dominated
by any other portfolio.

This is a useful intermediate step in modern portfolio theory,

since investors with mean-variance utility will only choose
portfolios on the efficient frontier.
The Efficient Frontier

With three assets, for example, an investor can choose

w1 = share of initial wealth allocated to asset 1

w2 = share of initial wealth allocated to asset 2

1 − w1 − w2 = share of wealth allocated to asset 3
The Efficient Frontier

Given the choices of w1 and w2 :

r˜P = w1 r˜1 + w2 r˜2 + (1 − w1 − w2 )˜

r3

µP = w1 µ1 + w2 µ2 + (1 − w1 − w2 )µ3

σP2 = w12 σ12 + w22 σ22 + (1 − w1 − w2 )2 σ32

+ 2w1 w2 σ1 σ2 ρ12
+ 2w1 (1 − w1 − w2 )σ1 σ3 ρ13
+ 2w2 (1 − w1 − w2 )σ2 σ3 ρ23
The Efficient Frontier

Our problem is to solve

min σP2 subject to µP = µ̄

w1 ,w2

for a given value of µ̄.

But since we are more used to solving constrained

maximization problems, consider the reformulated, but
equivalent, problem:

max −σP2 subject to µP = µ̄

w1 ,w2
The Efficient Frontier

Set up the Lagrangian, using the expressions for σP and µP

derived previously:

L = −w12 σ12 − w22 σ22 − (1 − w1 − w2 )2 σ32

− 2w1 w2 σ1 σ2 ρ12
− 2w1 (1 − w1 − w2 )σ1 σ3 ρ13
− 2w2 (1 − w1 − w2 )σ2 σ3 ρ23
+ λ[w1 µ1 + w2 µ2 + (1 − w1 − w2 )µ3 − µ̄]
The Efficient Frontier

L = −w12 σ12 − w22 σ22 − (1 − w1 − w2 )2 σ32

− 2w1 w2 σ1 σ2 ρ12
− 2w1 (1 − w1 − w2 )σ1 σ3 ρ13
− 2w2 (1 − w1 − w2 )σ2 σ3 ρ23
+ λ[w1 µ1 + w2 µ2 + (1 − w1 − w2 )µ3 − µ̄]

First-order condition for w1 :

0 = −2w1∗ σ12 + 2(1 − w1∗ − w2∗ )σ32 − 2w2∗ σ1 σ2 ρ12

− 2(1 − w1∗ − w2∗ )σ1 σ3 ρ13 + 2w1∗ σ1 σ3 ρ13
+ 2w2∗ σ2 σ3 ρ23 + λ∗ µ1 − λ∗ µ3
The Efficient Frontier

L = −w12 σ12 − w22 σ22 − (1 − w1 − w2 )2 σ32

− 2w1 w2 σ1 σ2 ρ12
− 2w1 (1 − w1 − w2 )σ1 σ3 ρ13
− 2w2 (1 − w1 − w2 )σ2 σ3 ρ23
+ λ[w1 µ1 + w2 µ2 + (1 − w1 − w2 )µ3 − µ̄]

First-order condition for w2 :

0 = −2w2∗ σ22 + 2(1 − w1∗ − w2∗ )σ32 − 2w1∗ σ1 σ2 ρ12

+ 2w1∗ σ1 σ3 ρ13 − 2(1 − w1∗ − w2∗ )σ2 σ3 ρ23
+ 2w2∗ σ2 σ3 ρ23 + λ∗ µ2 − λ∗ µ3
The Efficient Frontier
The two first-order conditions and the constraint

0 = −2w1∗ σ12 + 2(1 − w1∗ − w2∗ )σ32 − 2w2∗ σ1 σ2 ρ12

− 2(1 − w1∗ − w2∗ )σ1 σ3 ρ13 + 2w1∗ σ1 σ3 ρ13
+ 2w2∗ σ2 σ3 ρ23 + λ∗ µ1 − λ∗ µ3

0 = −2w2∗ σ22 + 2(1 − w1∗ − w2∗ )σ32 − 2w1∗ σ1 σ2 ρ12

+ 2w1∗ σ1 σ3 ρ13 − 2(1 − w1∗ − w2∗ )σ2 σ3 ρ23
+ 2w2∗ σ2 σ3 ρ23 + λ∗ µ2 − λ∗ µ3

w1∗ µ1 + w2∗ µ2 + (1 − w1∗ − w2∗ )µ3 = µ̄

form a system of three equations in the three unknowns: w1∗ ,
w2∗ , and λ∗ .
The Efficient Frontier
Moreover, the equations are linear in the unknowns w1∗ , w2∗ ,
and λ∗ :

0 = −2w1∗ σ12 + 2(1 − w1∗ − w2∗ )σ32 − 2w2∗ σ1 σ2 ρ12

− 2(1 − w1∗ − w2∗ )σ1 σ3 ρ13 + 2w1∗ σ1 σ3 ρ13
+ 2w2∗ σ2 σ3 ρ23 + λ∗ µ1 − λ∗ µ3

0 = −2w2∗ σ22 + 2(1 − w1∗ − w2∗ )σ32 − 2w1∗ σ1 σ2 ρ12

+ 2w1∗ σ1 σ3 ρ13 − 2(1 − w1∗ − w2∗ )σ2 σ3 ρ23
+ 2w2∗ σ2 σ3 ρ23 + λ∗ µ2 − λ∗ µ3

w1∗ µ1 + w2∗ µ2 + (1 − w1∗ − w2∗ )µ3 = µ̄

Given specific values for µ1 , µ2 , µ3 , σ1 , σ2 , σ3 , ρ12 , ρ13 , ρ23 ,
and µ̄ they can be solved quite easily.
The Efficient Frontier
In linear algebra, a vector is just a column of numbers. With
N ≥ 3 assets, you can organize the portfolio shares and
expected returns into a vectors:
   
w1 µ1
 w2   µ2 
w =  ..  and µ =  .. 
   
 .   . 
wN µN
where
w1 + w2 + . . . + wN = 1
Also in linear algebra, the transpose of a vector just
reorganizes the column as a row; for example:
w 0 = w1 w2 . . . wN
 
The Efficient Frontier

Meanwhile, the variances and covariances can be organized

into a matrix – a collection of rows and columns:
 
σ12 σ1 σ2 ρ12 . . . σ1 σN ρ1N
 σ1 σ2 ρ12 σ22 . . . σ2 σN ρ2N 
Σ=
 
.. .. .. 
 . . ... . 
σ1 σN ρ1N σ2 σN ρ2N . . . σN2
The Efficient Frontier
Using the rules from linear algebra for multiplying vectors and
matrices, the expected return on any portfolio with shares in
the vector w is
µ0 w
and the variance of the random return on the portfolio is

w 0 Σw .

Hence, the problem of minimizing the variance for a given

mean can be written compactly as

max −w 0 Σw subject to µ0 w = µ̄ and `0 w = 1

w

where ` is a vector of N ones.

The Efficient Frontier

max −w 0 Σw subject to µ0 w = µ̄ and `0 w = 1

w

Problems of this form are called quadratic programming

problems and can be solved very quickly on a computer even
when the number of assets N is large.

We can also add more constraints, such as wi ≥ 0, ruling out

short sales.
The Efficient Frontier

Going back to the case with three assets, once the optimal
shares w1∗ and w2∗ have been found, the minimized standard
deviation can be computed using the general formula

σP2 = w12 σ12 + w22 σ22 + (1 − w1 − w2 )2 σ32

+ 2w1 w2 σ1 σ2 ρ12
+ 2w1 (1 − w1 − w2 )σ1 σ3 ρ13
+ 2w2 (1 − w1 − w2 )σ2 σ3 ρ23

Doing this for various values of µ̄ allows us to trace out the

minimum variance frontier.
The Efficient Frontier

Adding assets shifts the minimum variance frontier to the left,

as opportunities for diversification are enhanced.
The Efficient Frontier

However, the minimum variance frontier retains its sideways

parabolic shape.
The Efficient Frontier

The minimum variance frontier traces out the minimized

variance or standard deviation for each required mean return.
The Efficient Frontier

But portfolio A exhibits mean-variance dominance over

portfolio B, since it offers a higher expected return with the
same standard deviation.
The Efficient Frontier

Hence, the efficient frontier extends only along the top arm of
the minimum variance frontier.
The Efficient Frontier
Recall that any of the following assumptions imply that
indifference curves in this σ − µ diagram slope upward and are
convex:
1. Risks are small enough to justify a second-order Taylor
approximation to any increasing and concave Bernoulli
utility function within the vN-M expected utility
framework
2. Investors have vN-M expected utility with quadratic
Bernoulli utility functions
3. Asset returns are normally distributed and investors have
vN-M expected utility with increasing and concave
Bernoulli utility functions
The Efficient Frontier

Portfolios along U 1 are suboptimal. Portfolios along U 3 are

infeasible. Portfolio P ∗ , located where U 2 is tangent to the
efficient frontier, is optimal.
The Efficient Frontier

Investor B is less risk averse than investor A. But both choose

portfolios along the efficient frontier.
The Efficient Frontier

Thus, the mean-variance utility hypothesis built into Modern

Portfolio Theory implies that all investors choose optimal
portfolios along the efficient frontier.
A Separation Theorem

So far, however, our analysis has assumed that there are only
risky assets. An additional, quite striking, result emerges when
we add a risk free asset to the mix.

This implication was first noted by James Tobin (US,

1918-2002, Nobel Prize 1981) in his paper “Liquidity
Preference as Behavior Towards Risk,” Review of Economic
Studies Vol.25 (February 1958): pp.65-86.
A Separation Theorem

Consider, therefore, the larger portfolio formed when an

investor allocates the fraction w of his or her initial wealth to
a risky asset or to a smaller portfolio of risky assets and the
remaining fraction 1 − w to a risk free asset with return rf .
A Separation Theorem
If the risky part of this portfolio has random return r˜, expected
return µr = E (˜ r ), and variance σr2 = E [(˜
r − µr )2 ] then the
larger portfolio has random return r˜P = w r˜ + (1 − w )rf with
expected return

µP = E [w r˜ + (1 − w )rf ] = w µr + (1 − w )rf

and variance

σP2 = E [(r˜P − µP )2 ]
= E {[w r˜ + (1 − w )rf − w µr − (1 − w )rf ]2 }
= E {[w (˜r − µr )]2 } = w 2 σr2 .
A Separation Theorem

The expression for the portfolio’s variance

σP2 = w 2 σr2

implies
σP = w σr
and hence
σP
w= .
σr
Hence, with σr given, a larger share of wealth w allocated to
risky assets is associated with a higher standard deviation σP
for the larger portfolio.
A Separation Theorem

But the expression for the portfolio’s expected return

µP = w µr + (1 − w )rf

indicates that so long as µr > rf , a higher value of w will yield

a higher expected return as well.

What is the trade-off between risk σP and expected return µP

of the mix of risky and riskless assets?
A Separation Theorem
To see, substitute
σP
w=
σr
into
µP = w µr + (1 − w )rf
to obtain
   
σP σP
µP = µr + 1 − rf
σr σr
 
µr − rf
= rf + σP
σr
A Separation Theorem
The expression
 
µr − rf
µP = rf + σP
σr

shows that for portfolios of risky and riskless assets:

1. The relationship between σP and µP is linear.
2. The slope of the linear relationship is given by the Sharpe
ratio, defined here as the “expected excess return” offered
by the risky components of the portfolio divided by the
standard deviation of the return on that risky component:
µr − rf
.
σr
A Separation Theorem

Hence, any investor can combine the risk free asset with risky
portfolio A to achieve a combination of expected return and
standard deviation along the red line.
A Separation Theorem

However, any investor with mean-variance utility will prefer

some combination of the risk free asset and risky portfolio B
to all combinations of the risk free asset and risky portfolio A.
A Separation Theorem

And all investors with mean-variance utility will prefer some

combination of the risk free asset and risky portfolio T to any
other portfolio.
A Separation Theorem

Investor B is less risk averse than investor A. But both choose

some combination of the “tangency portfolio” T and the risk
free asset.
A Separation Theorem

Note that the tangency portfolio T can be identified as the

portfolio along the efficient frontier of risky assets that has the
highest Sharpe ratio.
A Separation Theorem

This is the two-fund theorem or separation theorem implied by

Modern Portfolio Theory.

Equity mutual fund managers can all focus on building the

unique portfolio that lies along the efficient frontier of risky
assets and has the highest Sharpe ratio.

Each individual investor can then tailor his or her own portfolio
by choosing the combination of the riskless assets and the risky
mutual fund that best suits his or her own aversion to risk.
Strengths and Shortcomings of MPT

We’ve already considered one shortcoming of the MPT: its

mean-variance utility hypothesis must rest on one of two more
basic assumptions.

Either utility must be quadratic or asset returns must be

normal.
Strengths and Shortcomings of MPT

A second problem involves the estimation or “calibration” of

the model’s parameters.

With N risky assets, the vector µ of expected returns contains

N elements and the matrix Σ of variances and covariances
contains N(N + 1)/2 unique elements. When N = 100, for
example, there are 100 + (100 × 101)/2 = 5150 parameters to
estimate!

And to use data from the past to estimate these parameters,

one has to assume that past averages and correlations are a
reliable guide to the future.
Strengths and Shortcomings of MPT

On the other hand, the MPT teaches us a very important

lesson about how individual assets with imperfectly, and
especially negatively, correlated returns can be combined into
a diversified portfolio to reduce risk.

And the MPT’s separation theorem suggests that a retirement

savings plan that allows participants to choose between a
money market mutual fund and a well-diversified equity fund is
fully optimal under certain circumstances and perhaps close
enough to optimal more generally.
Strengths and Shortcomings of MPT

Finally, our first equilibrium model of asset pricing, the Capital

Asset Pricing Model, builds directly on the foundations
provided by Modern Portfolio Theory.