Chapter 2

The Efficient Frontier

2.1

The Efficient Front ier

A portfolio consists of various amounts held in different assets. The

number of possible assets can be quite large. Standard and Poor's lists
500 assets ( S&P 500) and in Canada, the Toronto Stock Exchange lists
300 assets ( TSX 300) . The basic portfolio optimization problem is to
decide how much of an investor's wealth should be optimally invested
in each asset.
Consider a universe of n assets. Let [Link] denote the expected return 1
on asset i, i = 1 , 2, ... , n and aij denote the covariance between the
returns of assets i and j, 1 :::;: i, j :::;: n. Let

, [Link] ) ' and E = [aij]

E is called the covariance matrix2 for the assets and is symmetric
and positive semidefinite. Throughout this chapter we will make the
J.L

( [Link] , J.L2 ,

1 Here, return is defined as the ratio between the price of the i-th asset at any

future time and the current price of that i-th asset.

2In mathematics, the symbol :E is used to denote summation. In the finance
literature, :E is used to denote the covariance matrix for risky assets and we shall
use that convention throughout this chapter.

21

Chap. 2

The Efficient Frontier

22

stronger assumption that I: is positive definite. Let Xi denote the pro

portion of wealth to be invested in asset i and let x = (x 1 , x 2 , . . . , Xn ) '.
In terms of x, the expected return of the portfolio /-lp and the variance
of the portfolio ; are given by
/-lp = 1-l x and 2 = x' "".wx.

CJ

CJP

Let l = ( 1 , 1 , . . . , 1)'; i.e. , l is an n-vector of ones. Since the components

of x are proportions, they must sum to one; i.e. , l'x = 1. The constraint
l' x = 1 is usually called the budget constraint.
The goal is to choose a value for x which gives a large value for /-lp
and a small value for ; These two goals tend to be in conflict. Suppose
we have two portfolios, both having the same expected return but the
first having a small variance and the second having a large variance. The
first portfolio is obviously more attractive because it bears less risk for
the same expected return. This is the key idea behind H. Markowitz's
definition of an efficient portfolio.

CJ .

Definition 2.1 A

portfolio is variance-efficient if for a fixed /-lp ,

there is no other portfolio which has a smaller variance CJ; .

Definition 2. 1 implies that a portfolio is efficient if for some fixed

/-lp , ; is minimized. Thus the efficient portfolios are solutions of the
optimization problem

CJ

min{x' I:x I 1-l'x

/-lp , l'x

1}.

(2. 1)

In (2. 1 ) , /-lp is to vary over all possible values. For each value of /-lp ,
we will in general get a different efficient portfolio. The mathematical
structure of (2. 1 ) is that of minimizing a quadratic function subject
to two linear equality constraints, the first of which is parametric (the
parameter being /-lp ) . Thus (2. 1 ) is a parametric quadratic programming
problem.
There is an alternative (and equivalent) definition of an efficient
portfolio. Suppose we have two portfolios both having the same vari
ance but the first having a large expected return and the second having

Sec. 2.1

The Efficient Frontier

23

a small expected return. The first portfolio is more attractive because

it gives a higher expected return for the same risk as the second port
folio. This prompts the following alternative definition of an efficient
portfolio.
Definition 2.2 A
a; ,

portfolio is expected return-efficient if for fixed

there is no other portfolio with a larger [Link]

Definition 2.2 implies that a portfolio is efficient if for some fixed

is maximized. Using Definition 2.2, the efficient portfolios are
the optimal solutions for
a; , [Link]

max{J.t1X I x'Ex

= a; ,

l'x =

1}.

(2.2)

Note that (2.2) has a linear objective function, a quadratic equality

constraint and a linear equality constraint.
There is a third optimization problem which also produces efficient
portfolios. It is a somewhat more convenient formulation than (2. 1) or
(2.2) and we shall use it to develop the theoretical aspects of portfolio
optimization. Let t be a scalar parameter and consider the problem
1
.
mm{
tJ.t'x + 2x'Ex I l'x = 1 } .
(2.3)
-

The intuition behind (2.3) is as follows. For t 0, the parameter t bal

ances how much weight is placed on the maximization of J.t1x ( equiv
alently, the minimization of -J.t'x) and the minimization of x' Ex. If
t = 0, (2.3) will find the minimum variance portfolio. As t becomes
very large, the linear term in (2.3) will dominate and portfolios will be
found with higher expected returns at the expense of variance.
Definition 2.3 A

portfolio is parametric-efficient if it is an optimal

solution for {2. 3} for some nonnegative parameter t .

Each of (2. 1 ) , (2.2) and (2.3) generates a family of optimal solutions

as their respective parameters ( [Link] , a; and t, respectively) are allowed

24

The Efficient Frontier

Chap. 2

to vary. A key result is that the families of optimal solutions for the
efficient portfolios for each of these problems are identical, provided 1-l
is not a multiple of l. ( See Exercise 2.??) . Thus we can continue our
analysis using any one of the three problems. In what follows, it will be
convenient to use (2.3) . We next derive a relationship between /-lp and
u; for efficient portfolios. In doing so, we shall assume that E is positive
definite and consequently nonsingular ( see Exercise 1.7 ) . The approach
we use is to obtain the optimal solution for (2.3) as an explicit function
of t, use that to obtain expressions for /-lp and u; in terms of t, and
then eliminate t.
From Definition 1.1 ( see Section 1 . 1 ) , the optimality conditions for
(2.3) are:

ti-l - Ex

ul, and, l ' x

1.

(2.4)

From Theorem 1.2 ( see Section 1 . 1 ) these conditions are both necessary
and sufficient for optimality. Solving the first for x gives
(2.5)
Applying the budget constraint gives a single equation for the multiplier
u:
l ' x = 1 = -ul ' E - 1 l + tl ' E - 1 /-l,
which has solution

Substituting u into (2.5) gives the efficient portfolios as explicit linear

functions of the parameter t:

Notice that of the quantities in (2.6) , l'E- 1 1 and l'E- 1 1-l are both scalars
whereas E- 1 z and E- 1 1-l are both vectors. For ease of notation, let

The Efficient Frontier

Sec. 2.1

25

Then the efficient portfolios are

x(t)

ho + th1 .

We next use (2.8) to find /-Lp and 0'; in terms of t:

/-Lp = J.L1 X(t) = J.L1 ho + tJ.L' hb

(2.8)
(2.9)

and

CT;

(ho + th l ) ' E (ho + th 1 )

hEh0 + 2thEh0 + t2 h Eh 1

(2. 10)

and we have used the fact that hEh1 = h Eh0 because E is symmetric
and the transpose of a scalar is just the scalar itself. Let
(2. 1 1 )
and
=

hEho and fJ2 = h Eh 1 .

(2. 12)
Note that {32 > 0 if and only if J.L is not a multiple of l (Exercise 2 . 1 ) .
We will assume throughout that J.L is not a multiple o f l and this implies
f3o

hEho , fJ1

(2. 13)
!32 > 0.
The a's and {3's are just constants defined in terms of the data for
the model problem (2.3) . From (2.9) , (2. 10) , (2. 1 1 ) and (2. 12) , we now
have
/-Lp = ao + a1t and CT; = f3o + 2 f31 t + fJ2 t 2 ,
(2. 14)
which shows /-Lp is a linear function of t and CT; is a quadratic function
of t. In addition, it can be shown (Exercise 2.2) that
(2.15)
Equation (2. 14) gives a parametric relationship between /-Lp and CT; .
We can eliminate the parameter t and obtain an explicit relationship
between /-Lp and 0'; . Solving (2. 14) for t and t 2 and using (2. 15) gives
( CTp2 - f3o )
([Link] - ao )
2
,
and
t
=
(2. 16)
t =

a1

The Efficient Frontier

26

Chap. 2

It can be shown (Exercise 2.3) that

fJ2 = a1 .
Using (2. 17) and eliminating t in (2. 16) gives
a; - f3o = (P,p - ao ) 2 /a 1 .

(2. 17)
(2. 18)

Equation (2. 18) shows the relationship between the variance ( a; )

and expected return (p,p ) for efficient portfolios and is called the effi
cient frontier. Two of the constants, a0 and {30, in (2. 18) can be in
terpreted as follows. When t = 0 in (2.3) , the problem becomes that
of minimizing the variance subject to the budget constraint. The ex
pected return plays no role. The resulting portfolio is called the global
minimum variance portfolio or simply the minimum variance portfolio.
From (2.8) we see that h0 is precisely the minimum variance portfolio
and from (2. 1 1 ) , (2. 12) and (2. 14) , a0 and {30 are the expected return
and variance of the minimum variance portfolio, respectively.
llp

CX0

+-- Minimum Variance Portfolio

'
'

.....
cl

Figure

2. 1

The Efficient Frontier.

The Efficient Frontier

Sec. 2.1

27

From ( 2.18) , the algebraic relationship between a ( as opposed to

aFigure
)
is that of a parabola. The efficient frontier is illustrated in
P and2 . 1 . The
"nose" of the efficient frontier corresponds to the min
/-lp

imum variance portfolio ( t = 0) where the investor wants the smallest

risk and is not interested in expected return. As t ( in ( 2.16)) is in
creased from 0, the investor becomes less risk averse and trades off an
increase in expected return with increased risk. Thus t is a measure of
the investor's aversion to risk.
Points on the efficient frontier below the minimum variance point,
correspond to negative values of t. This would correspond to portfolios
which are not efficient and therefore only the top half of the efficient
frontier is used.
llp

Inefficient Portfolio

c?p

0
Figure

2.2

Inefficient Portfolio.

Part of the value of the efficient frontier is that it condenses in

formation concerning a problem having n assets into a 2-dimensional
graphical representation. Consider the portfolio shown inside the ef
ficient frontier in Figure 2.2. It is inefficient because moving up and

28

The Efficient Frontier

Chap. 2

in a vertical direction would produce a portfolio with higher expected

return at the same level of variance. Similarly, moving to the left in
a horizontal manner would produce a portfolio with decreased risk at
the same level of expected return. Thus, all portfolios strictly within
the efficient frontier are inefficient. The set of all efficient portfolios
corresponds precisely to points on the top half of the efficient frontier.
So far, we have chosen to think of the efficient frontier in ( a; , [Link] )
space; i.e. , mean-variance space. Sometimes it is helpful to think of it as
a curve in ( ap , [Link] ) space; i.e. , mean-standard deviation space. Rewriting
(2. 18) as
a; - ([Link] - ao ) 2 / f32 = f3o
shows that the efficient frontier depends on the difference of the squares
of the two variables [Link] and ap Thus in ( ap, [Link] ) space, the graph of the
efficient frontier is a hyperbola. This is illustrated in Figure 2.3.
Jlp

a.o

. . . . . . . . . .
:

f- Minimum Variance Portfolio

..... .....

112
0

Figure

2.3

Efficient Frontier in Mean-Standard Deviation Space.

Sec. 2.1

The Efficient Frontier

29

We complete this section by illustrating the key concepts with a

small numerical example.
Example 2.1

Consider a universe with n = 3 assets, J1 = ( 1 . 1 , 1.2, 1.3 ' and =

diag ( 1 X 10 - 2 ' 5 X 10- 2 ' 7 X 10 - 2 . Find the efficient portfolios, find
and sketch the efficient frontier, and computationally verify ( 2. 15 and
(2. 17 for this problem.

The efficient portfolios are given by ( 2.8 where h0 and h 1 are ob

tained from ( 2.7 . Recall the l = ( 1 , 1 , 1 ' for this problem. In order to
compute h0, we first must compute - 1 l :
0
1
102
0
1
2
1
2
o
1 = 10 1 ,
z=
o i x 10
f7
0
0 l7 X 102
1

from which we compute

[' - 1 [ = 102 ( 1 + +

Now we can compute

h0

- 1 z
l '- 1 l

35
47

10 - 2 x 102

][ ] [ ]
X

102 = 134.285714.

[il [] - [
y

- 35
47

0.7446808
0. 1489362
0.1063830

Note that x(O) = h0 + 0 h 1 = h0 is an efficient portfolio and as

such must satisfy the budget constraint; i.e. , l'ho = 1 . As a check on
our computations, we compute l'h0 for the above h0:

as expected.

l[ l [ l [

We now proceed with the computation of h 1 . First we need - 1 f-t:

E - 1 J. =

1 02
0
0

1 02

0
1 02

1.1

1.2

1.3

= 1 02

1.1

0.24

\3

1 10
24
1 8.57143

30

The Efficient Frontier

Chap. 2

We also need l'L,- 1 f-t = -t'L.- 1 l :

l'L-- 1 -t = 102 ( 1 . 1 + 0.24 +

\3 )

152.5714286.

We can now speed up the computations for h 1 by noting that

and so

hi

1 10
24
18.57143

- 152.5714286

0.7446808
0. 1489362
0.1063830

] [
=

-3.6170135
1 .2765912
2.3404223

We have previously noted that l'ho = 1 . Because x(t) = h0+th 1 is an

efficient portfolio for all t, x(t) must also satisfy the budget constraint;
i.e. , l'x(t) = l'ho + tl'h 1 . But this implies l'h 1 = 0. We can use this to
check our computations:

l ' hi
( to

(1 , 1 , 1)

-3.6170135
1 .2765912
2.3404223

0,

within hand calculator accuracy ) as expected.

] [

Thus, the efficient portfolios for this example are

x(t)

0. 7446808
0. 1489362
0.1063830

+t

-3.6170135
1 .2765912
2.3404223

We next evaluate the efficient set constants eto , a1 , f3o , {31 and {32 .
From (2. 1 1 ) , we obtain

a0 = -t' ho = ( 1 . 1 , 1.2, 1 .3)

0.7446808
0.1489362
0.1063830

1 . 136170212766,

The Efficient Frontier

Sec. 2.1
and
a1

11'h 1

( 1 . 1 , 1.2, 1 .3)

- 3.6170135
1 .2765912
2.3404223

31

0.595744680851 .

From (2. 12) we obtain

f3o
X

{31

{32
X

h'L,ho
(0.7446808, 0. 1489362, 0.1063830)
10- 2
o
o.7446808
o
0
0. 1489362
0 5 X 10- 2
2
o
o 7 x 100.1063830
0.00744680851 1 ,

l[

h'L,h 1
(0.7446808, 0. 1489362, 0.1063830)
10- 2
o
o
-3.6170135
0
1 .2765912
5 X 10- 2
o 7 x 10- 2
2.3404223
0.0,

h 'L,h 1
( -3.6170135, 1.2765912, 2.3404223)
0
-3.6170135
10- 2
o
1 .2765912
0 5 X 10- 2
0
2.3404223
0
0 7 x 10- 2

l[

l[

0.595744680851 .
Note from these results that {31 = 0 and {32
verifies (2. 15) and (2. 17) , respectively.

a 1 which computationally

From (2. 18) the equation of the efficient frontier is

0"; - 0.00744680851 1 = ([Link] - 1 . 136170212766) 2 / 0.595744680851.

32

The Efficient Frontier

Chap. 2

This can be used to construct a table of values for this efficient frontier
as shown in Table 2 . 1 . Note that because the efficient frontier is a
parabola opening up to the right, each specified value of u; will result
in two values of /-lp , one on the upper part and one on the lower part.
These values are then used to plot the efficient frontier as shown in
Figure 2.4.
TABLE

2. 1

Table of Values for Efficient Frontier

of Example

0.00744680851 1
0.008150764218
0.010262631340
0.013782409876
0.018710099826
0.025045701 192
0.032789213971
0.041940638165
0.052499973774

1 . 136170212766
1 . 156648930210
1 . 177127647654
1 . 197606365097
1 .218085082541
1 .238563799985
1 .259042517429
1 .279521234872
1 .299999952316

2.1

1 . 136170212766
1 . 1 15691495322
1.095212777878
1 .074 734060435
1 .054255342991
1 .03377662554 7
1 .013297908103
0.992819190659
0.972340473216

In all of the numerical examples in this text, we take the components

of J-l to be "one plus rate of return." The actual rate of return could
equally well be used since the efficient portfolios would be identical
in either case. Indeed, if J-l were replaced with J-l + Ol, where 0 is any
constant, our model problem
1
(2.19 )
min{ - tJ-t' x + -x' Ex I l ' x = 1 }
2
would be changed to
min{ - t(J-t + Ol) ' x +

x' Ex

I l'x

1}.

(2.20 )

However, the budget constraint, l'x = 1 , implies that the linear part of
the objective function for (2.20 ) can be written
tJ-t1 X
tOl ' X
- t(J-t + Ol) ' X =
=
- tJ.L1 X - tO.
-

Computer Programs

Sec. 2.2

33

'

0=0.0074

Figure

2.4

fl

Efficient Frontier for Example 2 . 1 .

Since this differs from the objective function for (2. 19) by the constant
-tO, the optimal solutions for the two problems will be identical.

2.2

Computer Programs

Figure 2.5 shows the m-file Example2pl.m which solves Example 2.1 .
Lines 2 and 3 define the data for that problem. Line 4 invokes checkdata
(see Figure 1 .5) to check if the given covariance matrix is symmetric and
positive definite. Line 5 calls the function EFMV coeff (Efficient Frontier
Mean Variance coefficients, see Figure 2.6) with input arguments J-t and

The Efficient Frontier

34

Chap. 2

. It returns with values for alphaO, alphal , betaO, beta2, hO and hl

which represent a0 , a l l {30, {32 , h0 and h1 , respectively. Line 6 calls the
function EFMVplot (Efficient Frontier Mean Variance plot, see Figure
2. 7) which uses the efficient frontier coefficients just computed to plot
the efficient frontier.
1
2
3
4
5

% E xamp l e 2 p 1 . m
mu = [ 1 . 1 1 . 2 1 . 3 ] '
S i gma = [ 1 . e-2 0 0 ; 0 S . O e-2 0 ; 0 0 7 . 0 e-2 ]
che c k dat a ( S i gma , 1 . e - 6 ) ;
[ a lp ha O , a lph a 1 , bet a O , be t a 2 , h O , h 1 ] = EFMVc o e f f ( mu , S i gma )
EFMVp l ot ( 0 . 0 , 0 . 5 , 0 . 0 1 , a l pha O , a lpha 1 , be t a O , bet a 2 ) ;

Figure

2.5

Program for Example 2 . 1 : Example2pl .m.

The file EFMVcoeff.m shown in Figure 2.6 calculates the coefficients

of the efficient frontier from the input arguments mu and Sigma. Line 4
constructs n from the length of mu and sets ell to be an n-vector of 1 's.
Line 5 computes templ = - l z and temp2 = l' - 1 l. From this, Line
6 constructs hO = (l' - l l) - l - 1 l. Line 7 computes temp3 = - l J.L and
% E FMVc o e f f . m
f u n c t i o n [ a lp h a O , a lpha 1 , bet a O , b e t a 2 , h O , h 1 ]
3
EFMVc o e f f ( mu , S i gma )
4
n = l e ng t h ( mu ) ;
e l l = ones ( 1 , n ) ' ;
5
t emp 1 = S i gma - 1 * e l l ;
t emp 2 = t emp 1 ' * e l l ;
6 h O = t emp 2 - 1 * t emp 1
1
t emp 3 = S i gma - 1 * mu ;
t emp 4
e l l ' * t emp 3 ;
s
h 1 = S i gma - 1 * mu - t emp 4 * h O
g
a l pha O = mu ' * h O
1 0 a l p h a 1 = mu ' * h 1
h O ' * S i gma * h O
1 1 bet a O
12 b e t a 2 = h 1 ' * S i gma * h 1
1

Figure

2.6

Calculation of Efficient Frontier Coefficients: EFMVcoeff.m.

Sec. 2.2

Computer Programs

35

temp4 = l'L,- 1 J.L. Line 8 computes h1 = r:, - l J.L - l'L,- 1 [Link]. Finally, lines
9 through 12 compute alphaO = J.L'h0, alpha1 = J.L1 h 1 , betaO = hL,h0
and beta2 =h 'L,h1 .
Figure 2.7 shows the function EFMVplot. The first three arguments
are tlow, thigh and tine. They specify that the parameter t should start
at tlow and increase to thigh in increments of tine. The last arguments
are alphaO, alpha1 , betaO and beta2 which represent the efficient fron
tier coefficients a0, a1 , {30 and {32 , respectively. Lines 5, 6 and 7 compute
mup = a0 + ta 1 , muplow = a0 - ta 1 (for the lower or inefficient part
of the efficient frontier) and the variance sigma2p = {30 + t2 {32 . Line
8 plots the points, lines 9 and 10 label the axes and line 1 1 provides
a title. Line 12 inserts text showing the minimum variance portfolio.
Finally, lines 14 through 18 set up the axes.
1

2
4
5
6
1
8
9

10
11
12
13
14
15
16
11
18
19

20

% EFMVp l ot . m
funct i o n EFMVp l ot ( t l o w , t h i g h , t i n c , a lph a O , a lph a l ,
b e t a 0 , be t a 2 )
t
t l ow : t i n c : t h i g h ;
mup
a l p h a O + t * a l pha l ;
mup l o w
alphaO - t * alpha l ;
s i gma 2 p
bet a O + t . " 2 * be t a 2 ;
p l o t ( s i gma 2 p , mup , ' -k ' , s i gma 2 p , mup l ow, ' --k ' )
x l ab e l ( ' P o rt f o l i o Va r i a n c e \ s i gm a _p " 2 ' )
y l ab e l ( ' P o rt f o l i o Me a n \ mu _p ' )
t i t l e ( ' E f f i c i e n t F r o nt i e r : Mean-Va r i a n c e S p a c e ' ,
' F o nt s i z e ' , l 2 )
t e xt ( be t a O , a l ph a O , ' \ l e f t a r r o w { \ rm m i n imum v a r i a n c e
porfol io} ' )
% s et a x e s
xmi n
0.;
xmax
bet a O + t h i gh . " 2 * bet a 2 ;
ymi n
a l p h a O - t h i gh * a lpha l ;
a l p h a O + t h i gh * a lpha l ;
ymax
a x i s ( [ xm i n xmax ymi n yma x ] )
=

Figure

2.7

Plot the Efficient Frontier: EFMVplot .m.

The Efficient Frontier

36

Chap. 2

Efficient Frontier: Mean-Variance Space

1 .4
1 .3

-:::.0.
.
c:

.Q

0
c..

1 .2
minimum variance porfolio

1 .1

0.9
0.05

0.1
2

Portfolio Variance a

0. 1 5

Figure 2.8

2.3

Plot of the Efficient Frontier: EfficFrontPlot .m.

Exercises

2.1 Show {32

0 if and only if f-l is a multiple of l.

2.2 Verify equation (2.15) ({31

2.3 Verify equation (2. 17) ({32

0) .

a1 ) .

2.4 The purpose of this exercise is to establish relationships between

the optimal solutions for (2. 1 ) , (2.2) and (2.3) and their respective
parameters f-lp, a; and t.
(a) Write out the optimality conditions for (2. 1) and solve them
in terms of I;- 1 . Show that these efficient portfolios are iden
tical to those of (2.7)-(2.8) . Assume f-l is not a multiple of

Exercises

Sec. 2.3

37

l. Hints: (i) (2. 1) has two linear constraints so that the op

timality conditions for it will include two multipliers, (ii) it
may be helpful to recall a formula for the inverse of a (2, 2)
matrix, (iii) /lp is a parameter in (2. 1 ) and t is a parame
ter in (2.3) . Both parameters can vary over an appropriate
range.
(b) Show that the optimal solutions for (2.2) are identical to
those for (2. 1) and (2.3) for the top part of the efficient
frontier; i.e. , for t 0 in the case of (2.3) . What variation of
(2.2) will produce the efficient portfolios for the bottom (in
efficient) part of the efficient frontier? Hint: The optimality
conditions for (2.2) can be shown to be
(1) x' Ex = a 2 l ' x = 1
(2)

11 =

P'

'

2lhEx + fhl, 0 1 0.

2.5 For each . of the following problems, find, by hand calculation,

h0 , h 1 , x(t) , ao , a l l {30, {31 and {32 (as in Example 2 . 1 ) , and sketch
the efficient frontier.
(a) n = 2 , 11 = ( 1 . 1 , 1 .2)' and E = diag(l0- 2 , w- 1 ) ,
(b) n = 3, 11 = ( 1 . 1 , 1 . 15, 1 .2)' and E = diag(1o-4, 10- 3 , 10- 2 ) .
In each case, verify your results by using the computer programs
of Section 2.2.

2.6 Suppose E = diag( ai ) Show that the efficient portfolios have

components

where
and

02 = 01 ( 1-tda1 + 1-t2 /a2 + + 1-Ln f an )

Also show the multiplier for the budget constraint is u = -0 1 +
t (}2

38

The Efficient Frontier

2.7 Let

JLi+ l

(1 +

n-z

-.

2
) and O"i+l

Chap. 2
10 - 2 ( n - z) 2 ,

for i = 0, 1 , . . , n - 1 and let = diag(0" 1 , 0"2 , . . , O"n) Write a

MATLAB program to implement the solution for the diagonal
covariance case and for each n = 10 and n = 20 use it to solve
the problem with the given data. For each such value of n, also
solve the problem using a variation of the program EFMVcoeff.m
of Figure 2.6. Compare the results.
.

2.8 Show the slope of the efficient frontier in (O"; , J1,p ) space is for
all t > 0.
2.9 Determine the efficient frontier for each of (2. 1 ) , (2.2) and (2.3)
when JL is a multiple of l; i.e. , J-1, = (}l for some scalar B . Illustrate
graphically.
2.10 Consider the problem
1
min{ - tjJ,1X + 2x'x I l'x

0} ,

where the data is identical to (2.3) and the problem differs from
(2.3) only in that the budget constraint l 'x = 1 has been replaced
by l'x = 0. This is usually called a portfolio rebalance problem.
Obtain a closed form solution for this problem and show that the
efficient frontier (in mean-standard deviation space) is a pair of
straight lines.
2 . 1 1 Suppose n = 4,

JL = ( 1 .2, 1 .3, 1 .4, 1.5)' and = diag(10-4, 10- 3 , 10- 2 , 10- 1 ) .

(a) Obtain ho and h 1 .

(b) Find the equation of the efficient frontier. Sketch it.
(c) As t is increased, some asset is eventually reduced to zero.
Which one and for what value of t? Assuming that the
model problem is now modified by adding nonnegativity
constraints x 2: 0, and that this asset remains at 0 as t is in
creased further, find the new optimal portfolio. What is the

Exercises

Sec. 2.3

39

next asset to be reduced to zero and for what value of t does

this occur? Continue until a third asset is reduced to zero.
Sketch the resulting four "pieces" of the efficient frontier.
2.12 Is it possible for an efficient portfolio to have one or more com
ponents with negative values? These correspond to short selling;
i.e. , the investor borrows money to buy additional amounts of an
attractive asset. Sometimes the portfolio optimization problem is
formulated to preclude short selling as we will see in subsequent
chapters. The purpose of this exercise is to show that if short
selling does occur, the efficient frontier is divided into three seg
ments: the leftmost has portfolios with short selling, the middle
has no short selling and the rightmost has portfolios with short
selling. See Figure 2.9.

( a) Use (2.8) to show that either there are no nonnegative ef

ficient portfolios or there are numbers t1 and th such that
x(t) 0 for all t with t1 ::; t ::; th and x(t) has mixed sign
outside of the interval (th th)
( b ) Use the results of part ( a) to show that the nonnegative
portfolios all lie on a segment of the efficient frontier, as
shown in Figure 2.9.

nonnegative
portfolio

Figure

2.9

Nonnegative Portfolios.