4.

Portfolio theory

Suppose there are N risky assets, whose rates of returns are given
by the random variables R1, · · · , RN , where
Sn(1) − Sn(0)
Rn = , n = 1, 2, · · · , N.
Sn(0)
Let w = (w1 · · · wN )T , wn denotes the proportion of wealth invested
N
X
in asset i, with wn = 1. The rate of return of the portfolio is
n=1
N
X
RP = wnRn.
n=1
Assumptions

1. There does not exist any asset that is a combination of other

assets in the portfolio.
2. µ = (R1 R2 · · · RN ) and 1 = (1 1 · · · 1) are linearly indepen-
dent.
The first two moments of RP are
N
X N
X
µP = E[RP ] = E[wnRn] = wnµn, where µn = Rn,
n=1 n=1
and
X N
N X N X
X N
2 = var(R ) =
σP wiwj cov(Ri, Rj ) = wiσij wj .
P
i=1 j=1 i=1 j=1
Let Ω denote the covariance matrix so that
2 = w T Ωw .
σP

Remark

2 . In mean-variance
The portfolio risk of return is quantified by σP
analysis, only the first two moments are considered in the port-
folio model. Investment theory prior to Markowitz considered the
maximization of µP but without σP .
Two-asset portfolio

Consider two assets with known means R1 and R2, variances σ12 and
σ22, of the expected rates of returns R1 and R2, together with the
correlation coefficient ρ.

Let 1 − α and α be the weights of assets 1 and 2 in this two-asset

portfolio.

Portfolio mean: RP = (1 − α)R1 + αR2, 0 ≤ α ≤ 1

2 = (1 − α)2σ 2 + 2ρα(1 − α)σ σ + α2 σ 2.

Portfolio variance: σP 1 1 2 2
We represent the two assets in a mean-standard deviation diagram
√
(recall: standard deviation = variance)

As α varies, (σP , RP ) traces out a conic curve in the σ − R plane.

With ρ = −1, it is possible to have σ = 0 for some suitable choice
of weight.
In particular, when ρ = 1,
q
σP (α; ρ = 1) = (1 − α)2σ12 + 2α(1 − α)σ1σ2 + α2σ22
= (1 − α)σ1 + ασ2.
This is the straight line joining P1(σ1, R1) and P2(σ2, R2).

When ρ = −1, we have

q
σP (α; ρ = −1) = [(1 − α)σ1 − ασ2]2 = |(1 − α)σ1 − ασ2|.
When α is small (close to zero), the corresponding point is close to
P1(σ1, R1). The line AP1 corresponds to

σP (α; ρ = −1) = (1 − α)σ1 − ασ2.

σ1
The point A corresponds to α = .
σ1 + σ2

σ1
The quantity (1 − α)σ1 − ασ2 remains positive until α = .
σ1 + σ2
σ1
When α > , the locus traces out the upper line AP2.
σ1 + σ2
Suppose −1 < ρ < 1, the minimum variance point on the curve that
represents various portfolio combinations is determined by
2
∂σP
= −2(1 − α)σ12 + 2ασ22 + 2(1 − 2α)ρσ1σ2 = 0
∂α
↑
set
giving
σ12 − ρσ1σ2
α= 2 .
σ1 − 2ρσ1σ2 + σ22
Mathematical formulation of Markowitz’s mean-variance analysis
N N
1 XX
minimize wi wj σij
2 i=1 j=1
N
X N
X
subject to wi Ri = µP and wi = 1.
i=1 i=1

Solution

We form the Lagrangian

N X
N
à N
! ÃN !
1 X X X
L= wi wj σij − λ1 wi − 1 − λ2 w i R i − µP
2 i=1 j=1 i=1 i=1

where λ1 and λ2 are Lagrangian multipliers.

We then differentiate L with respect to wi and set the derivative to zero.

N
X
∂L
= σij wj − λ1 − λ2 Ri = 0, i = 1, 2, · · · , N. (1)
∂wi j=1
N
X
∂L
= wi − 1 = 0; (2)
∂λ1 i=1
XN
∂L
= wi Ri − µP = 0. (3)
∂λ2 i=1
From Eq. (1), the portfolio weight admits solution of the form

w∗ = Ω−1(λ11 + λ2µ) (4)

where 1 = (1 1 · · · 1)T and µ = (R1 R2 · · · RN )T .

To determine λ1 and λ2, we apply the two constraints

1 =1T Ω−1Ωw∗ = λ11T Ω−11 + λ21T Ω−1µ. (5)

µP = µT Ω−1Ωw∗ = λ1µT Ω−11 + λ2µT Ω−1µ. (6)
T T
Write a = 1 Ω−11, b = 1 Ω−1µ and c = µT Ω−1µ, we have

1 = λ1a + λ2b and µP = λ1b + λ2c.

c − bµP aµP − b
Solving for λ1 and λ2 : λ1 = and λ2 = , where
∆ ∆
2
∆ = ac − b .

Note that λ1 and λ2 have dependence on µP , which is the target

mean prescribed in the variance minimization problem.
Assume µ 6= h1, and Ω−1 exists. Since Ω is positive definite, so
a > 0, c > 0. By virtue of the Cauchy-Schwarz inequality, ∆ > 0.
The minimum portfolio variance for a given value of µP is given by
T T
2
σP = w Ωw∗
∗ = w Ω(λ1Ω−1
∗ 1 + λ2Ω−1µ)
aµ2 P − 2bµP +c
= λ1 + λ2µP = .
∆
The set of minimum variance portfolios is represented by a parabolic
curve in the σP2 − µ plane. The parabolic curve is generated by
P
varying the value of the parameter µP .
 dµP
How about the asymptotic values of lim ?
µ→±∞ dσP

dµP dµP dσP2

= 2 dσ
dσP dσP P
∆
= 2σP
2aµ
√P − q 2b
∆
= aµ2P − 2bµP + c
aµP − b
so that
s
dµP ∆
lim =± .
µ→±∞ dσP a
Summary
c − bµP aµP − b
Given µP , we obtain λ1 = and λ2 = , and the optimal weight
∆ ∆
w∗ = Ω−1(λ1 1 + λ2µ).
To find the global minimum variance portfolio, we set
dσP2 2aµP − 2b
= =0
dµP ∆
so that µP = b/a and σP2 = 1/a. Also, λ1 = 1/a and λ2 = 0. We obtain

wg =
Ω−1 1= 1 .
Ω−1
T
a
1 Ω−11
Another portfolio that corresponds to λ1 = 0 is obtained when µP is taken to be
c
. The value of the other Lagrangian multiplier is given by
b
¡ ¢
a cb − b 1
λ2 = = .
∆ b
The optimal weight of this particular portfolio is
Ω−1 µ Ω−1µ
w∗d = = T
.
b
1 Ω−1µ
¡ c ¢2 ¡c¢
a − 2b +c c
Also, σd2 = b b
= .
∆ b2
Feasible set

Given N risky assets, we form various portfolios from these N assets.

We plot the point (σP , RP ) representing the portfolios in the σ − R
diagram. The collection of these points constitutes the feasible set
or feasible region.
Consider a 3-asset portfolio, the various combinations of assets 2
and 3 sweep out a curve between them (the particular curve taken
depends on the correlation coefficient ρ12).

A combination of assets 2 and 3 (labelled 4) can be combined with

asset 1 to form a curve joining 1 and 4. As 4 moves between 2 and
3, the curve joining 1 and 4 traces out a solid region.
Properties of feasible regions

1. If there are at least 3 risky assets (not perfectly correlated

and with different means), then the feasible set is a solid two-
dimensional region.

2. The feasible region is convex to the left. That is, given any two
points in the region, the straight line connecting them does not
cross the left boundary of the feasible region.
The left boundary of a feasible region is called the minimum variance
set. The most left point on the minimum variance set is called the
minimum variance point. The portfolios in the minimum variance
set are called frontier funds.

For a given level of risk, only those portfolios on the upper half
of the efficient frontier are desired by investors. They are called
efficient funds.

A portfolio w∗ is said to be mean-variance efficient if there exists

no portfolio w with µP ≥ µ∗P and σP2 ≤ σ ∗2 . That is, you cannot find
P
a portfolio that has a higher return and lower risk than those for an
efficient portfolio.
Two-fund theorem

Two frontier funds (portfolios) can be established so that any fron-

tier portfolio can be duplicated, in terms of mean and variance, as
a combination of these two. In other words, all investors seeking
frontier portfolios need only invest in combinations of these two
funds.

Remark

Any convex combination (that is, weights are non-negative) of ef-

ficient portfolios is an efficient portfolio. Let αi ≥ 0 be the weight
h i b
i I
of Fund i whose rate of return is Rf . Since E Rf ≥ for all i, we
a
have
n
X h i Xn
i b b
αiE Rf ≥ αi = .
i=1 i=1 a a
Proof

Let w1 = (w11 · · · wn
1 ), λ1, λ1 and w 2 = (w 2 · · · w 2)T , λ2, µ2 are two
1 2 1 n
known solutions to the minimum variance formulation with expected
rates of return µ1 2
P and µP , respectively.

n
X
σij wj − λ1 − λ2Ri = 0, i = 1, 2, · · · , n (1)
j=1
Xn
wiri = µP (2)
i=1
n
X
wi = 1. (3)
i=1

It suffices to show that αw1 + (1 − α)w2 is a solution corresponds

to the expected rate of return αµ1
P + (1 − α)µ2.
P
1. αw1 + (1 − α)w2 is a legitimate portfolio with weights that sum
to one.

2. Eq. (1) is satisfied by αw1 + (1 − α)w2 since the system of

equations is linear.

3. Note that
n h
X i
αwi1 + (1 − α)wi2 Ri
i=1
Xn n
X
= α wi1Ri + (1 − α) wi2Ri
i=1 i=1
= αµ1
P + (1 − α)µ 2.
P
Proposition

Any minimum variance portfolio with target mean µP can be uniquely

decomposed into the sum of two portfolios

w∗P = Awg + (1 − A)wd

c − bµP
where A = λ1a = a.
∆

Proof

For a minimum-variance portfolio whose solution of the Lagrangian

multipliers are λ1 and λ2, the optimal weight is

w∗P = λ1(Ω−11 + Ω−1µ) = λ1(awg ) + λ2(bwd).

Observe that the sum of weights is
c − µP b µP a − b ac − b2
λ1a + λ2b = a +b = = 1.
∆ ∆ ∆
We set λ1a = A and λ2b = 1 − A.
Indeed, any two minimum-variance portfolios can be used to substi-
tute for wg and wd. Suppose

wu = (1 − u)wg + uwd
wv = (1 − v)wg + v wd
we then solve for wg and wd in terms of wu and wv . Then

w∗P = λ1awg + (1 − λ1a)wd

λ1a + v − 1 1 − u − λ1a
= wu + wv ,
v−u v−u
where sum of coefficients = 1.
Example

Mean, variances, and covariances of the rates of return of 5 risky

assets are listed:

Security covariance Ri
1 2.30 0.93 0.62 0.74 −0.23 15.1
2 0.93 1.40 0.22 0.56 0.26 12.5
3 0.62 0.22 1.80 0.78 −0.27 14.7
4 0.74 0.56 0.78 3.40 −0.56 9.02
5 −0.23 0.26 −0.27 −0.56 2.60 17.68
Solution procedure to find the two funds in the minimum variance
set:

1. Set λ1 = 1 and λ2 = 0; solve the system of equations

5
X
σij vj1 = 1, i = 1, 2, · · · , 5.
j=1

Normalize vk1’s so that they sum to one

vi1
wi1 = Pn 1
.
j=1 vj
1
After normalization, this gives the solution to wg , where λ1 =
a
and λ2 = 0.
2. Set λ1 = 0 and λ2 = 1; solve the system of equations:
5
X
σij vj2 = Ri, i = 1, 2, · · · , 5.
j=1

Normalize vi2’s to obtain wi2.

After normalization, this gives the solution to wd, where λ1 = 0

1
and λ2 = .
b
The above procedure avoids the computation of a = 1T Ω−11
T
and b = 1 Ω−1µ.
 security v1 v2 w1 w2
1 0.141 3.652 0.088 0.158
2 0.401 3.583 0.251 0.155
3 0.452 7.284 0.282 0.314
4 0.166 0.874 0.104 0.038
5 0.440 7.706 0.275 0.334
mean 14.413 15.202
variance 0.625 0.659
standard deviation 0.791 0.812

* Note that w1 corresponds to the global minimum variance point.

We know that µg = b/a; how about µd?

Ω−1 µ c
T
µd = µ wd = µT = .
b b

c b ∆
Difference in expected returns = µd − µg = − = > 0.
b a ab

c 1 ∆
Also, difference in variances = σd2 − σg2 = 2 − = 2 > 0.
b a ab
How about the covariance of portfolio returns for any two minimum
variance portfolios?

Write
u = w T R and Rv = w T R
RP u P v
where R = (R1 · · · RN )T . Recall that

 
Ω−11 Ω−1 µ
σgd = cov  R, R
a b
 T Ã !
= 
Ω−1 1 Ω
Ω−1µ
a b

=
1Ω−1µ = 1 since b = 1Ω−1µ.
ab a
cov(RPu , Rv ) = (1 − u)(1 − v)σ 2 + uvσ 2 + [u(1 − v) + v(1 − u)]σ
P g d gd
(1 − u)(1 − v) uvc u + v − 2uv
= + 2 +
a b a
1 uv∆
= + 2
.
a ab
In particular,

cov(Rg , RP ) = wTg ΩwP =

1 Ω−1ΩwP 1
= = var(Rg )
a a
for any portfolio wP .

For any Portfolio u, we can find another Portfolio v such that these
two portfolios are uncorrelated. This can be done by setting
1 uv∆
+ 2
= 0.
a ab
Inclusion of a riskfree asset

Consider a portfolio with weight α for a risk free asset and 1 − α for
a risky asset. The mean of the portfolio is

RP = αRf + (1 − α)Rj (note that Rf = Rf ).

The covariance σf j between the risk free asset and any risky asset
is zero since
E[(Rj − Rj ) (Rf − Rf ) = 0.
| {z }
zero
2 is
Therefore, the variance of portfolio σP
2 = α2 σ 2 +(1 − α)2 σ 2 + 2α(1 − α) σ
σP f j fj
|{z} |{z}
zero zero
so that σP = |1 − α|σj .
The points representing (σP , RP ) for varying values of α lie on a
straight line joining (0, Rf ) and (σj , Rj ).

If borrowing of risk free asset is allowed, then α can be negative. In

this case, the line extends beyond the right side of (σj , Rj ) (possibly
up to infinity).
Consider a portfolio with N risky assets originally, what is the impact
of the inclusion of a risk free asset on the feasible region?

Lending and borrowing of risk free asset is allowed

For each original portfolio formed using the N risky assets, the new
combinations with the inclusion of the risk free asset trace out the
infinite straight line originating from the risk free point and passing
through the point representing the original portfolio.

The totality of these lines forms an infinite triangular feasible region

bounded by the two tangent lines through the risk free point to the
original feasible region.
No shorting of risk free asset

The line originating from the risk free point cannot be extended
beyond points in the original feasible region (otherwise entail bor-
rowing of the risk free asset). The new feasible region has straight
line front edges.
The new efficient set is the single straight line on the top of the
new triangular feasible region. This tangent line touches the original
feasible region at a point F , where F lies on the efficient frontier of
the original feasible set.

b
Here, Rf < .
a
One fund theorem

Any efficient portfolio (any point on the upper tangent line) can be
expressed as a combination of the risk free asset and the portfolio
(or fund) represented by F .

“There is a single fund F of risky assets such that any efficient

portfolio can be constructed as a combination of the fund F and
the risk free asset.”

Remark Under the assumptions that

• every investor is a mean-variance optimizer

• they all agree on the probabilistic structure of assets

• unique risk free asset

Then everyone will purchase a single fund, market portfolio.

New Lagrangian formulation
2
σP 1
minimize = wT Ωw
2 2

subject to wT µ + (1 − wT 1)r = µP .
1 T
Define L = w Ωw + λ[µP − r − (µ − r1)T w]
2
N
X
∂L
= σij wj − λ(µ − r1) = 0, i = 1, 2, · · · , N (1)
∂wi j=1
∂L
=0 giving (µ − r1)T w = µP − r. (2)
∂λ
Solving (1): w∗ = λΩ−1(µ − r1). Substituting into (2)

µP − r = λ(µ − r1)T Ω−1(µ − r1) = λ(c − 2rb + r2a).

Lastly, the relation between µP and σP is given by the following pair
of half lines
T T T
2 = w ∗ Ωw ∗ = λ(w ∗ µ − r w ∗
σP 1)
= λ(µP − r) = (µP − r)2/(c − 2rb + r2a).
With the inclusion of the riskfree asset, the set of minimum variance
portfolios are represented by portfolios on the two half lines
q
Lup : µP − r = σP ar2 − 2br + c (1a)
q
Llow : µP − r = −σP ar2 − 2br + c. (1b)

Recall that ar2 − 2br + c > 0 for all values of r since ∆ = ac − b2 > 0.
The minimum variance portfolios without the riskfree asset lie on
the hyperbola
2 aµ2
P − 2bµP + c .
σP =
∆
 b
When r < µg = , the upper half line is a tangent to the hyperbola.
a
The tangency portfolio is the tangent point to the efficient frontier
(upper part of the hyperbolic curve) through the point (0, r).
The tangency portfolio M is represented by the point (σP,M , µMP ),
and the solution to σP,M and µM
P are obtained by solving simultane-
ously

2 aµ2
P − 2bµP + c
σP =
∆
q
µP = r + σP c − 2rb + r2a.
Once µP is obtained, we solve for λ and w∗ from
µP − r ∗ = λΩ−1(µ − r 1).
λ= and w
c − 2rb + r2a
The tangency portfolio M is shown to be
Ω−1(µ − r1) c − br c − 2rb + r2a
w∗M = , µM
P = and 2
σP,M = 2
.
b − ar b − ar (b − ar)
 b
When r < , it can be shown that µM
P > r. Note that
a
µ ¶µ ¶ µ ¶
b b c − br b b − ar
µM
P − −r
= −
a a b − ar a a
c − br b2 br
= − 2+
a a a
ca − b2 ∆
= 2
= 2
> 0,
a a
M b b
so we deduce that µP > > r, where µg = . Indeed, we can
a a
b
deduce (σP,M , µM
P ) does not lie on the upper half line if r ≥ .
a
 b
When r < , we have the following properties on the minimum
a
variance portfolios.

1. Efficient portfolios

Any portfolio on the upper half line

q
µP = r + σP ar2 − 2br + c
within the segment F M joining the two points (0, r) and M
involves long holding of the market portfolio and riskfree asset,
while those outside F M involves short selling of the riskfree asset
and long holding of the market portfolio.
2. Any portfolio on the lower half line
q
µP = r − σP ar2 − 2br + c
involves short selling of the market portfolio and investing the
proceeds in the riskfree asset. This represents non-optimal in-
vestment strategy since the investor faces risk but gains no extra
expected return above r.
What happens when r = b/a? The half lines become
s s
µ ¶
b b2 ∆
µP = r ± σP c − 2 b± = r ± σP ,
a a a
which correspond to the asymptotes of the feasible region with risky
assets only.

b
When r = , µM
P does not exist. Recall that
a
w∗ = λΩ−1(µ − r1) so that
T
1 w = λ(1Ω−1µ − r1Ω−11) = λ(b − ra).
T
When r = b/a, 1 w = 0 as λ is finite. Any minimum variance port-
folio involves investing everything in the riskfree asset and holding
a portfolio of risky assets whose weights sum to zero.
 b
When r > , only the lower half line touches the feasible region with
a
risky assets only.

Any portfolio on the upper half line involves short selling of the
tangency portfolio and investing the proceeds in the riskfree asset.
Alternative approach

Given a point (σP , µP ) in the feasible region, we draw a line joining

this point and the risk free asset. Let θ be the angle of inclination
µP − rf
of this line, where tan θ = . The tangency portfolio is the
σP
feasible point that maximizes θ or tan θ.
n
X
Write RP = wiRi, where wi is the weight associated with the
i=1
n
X n
X
risky asset i. Since µP = wiRi and rf = wirf , we have
i=1 i=1
n
X
wi(Ri − rf )
tan θ =  i=1 1/2 .
n X
X n
 σij wiwj 
i=1 j=1
Set the derivative of tan θ with respect to each wk equal to zero:-
∂
tan θ
∂wi
 1/2 " # n −1/2
n
X n
X X n
X
¡ ¢
Ri − rf  σij wi wj  − wi (Ri − rf )  σi,j wi wj  σij wj
i,j=1 i=1 i,j=1 j=1
= n
X
σij wi wj
i,j=1
= 0.

This leads to the following system of equations:-

n
X
σij wj = λ(Ri − rf ), λ is some constant, i = 1, 2, · · · , n.
j=1
Hint Use the following relation
 1/2  −1/2
n
∂ X X X
wiσij wj  =  wiσij wj  σij wj .
∂wi i,j i,j j=1
We write λvj = wj for each j, the above system becomes
n
X
σij vj = ri − rf , i = 1, 2, · · · , n.
j=1
We then solve for vj by a linear system solver.

vj
Finally, we normalize wj ’s by wj = Pn , j = 1, · · · , n.
j=1 vk
Example (5 risky assets and one risk free asset)

Data of the 5 risky assets are given in the earlier example, and
rf = 10%.

The system of linear equations to be solved is

5
X
σij vj = Ri − rf = 1 × Ri − rf × 1, i = 1, 2, · · · , 5.
j=1

Recall that v 1 and v 2 in the earlier example are solutions to

5
X 5
X
σij vj1 =1 and σij vj2 = Ri, respectively.
j=1 j=1

Hence, vj = vj2 − rf vj1, j = 1, 2, · · · , 5 (numerically, we take rf =

10%).
Addition of risk tolerance factor

2 , with τ ≥ 0, where τ is the risk tolerance.

Maximize 2τ µP − σP

2 subject to
Optimization problem: max 2τ µP − σP 1 · w = 1.
w∈RN

Remark

τ is closely related to the relative risk aversion coefficient. Given an

initial wealth W0 and under a portfolio choice w, the end-of-period
wealth is W0(1 + RP ). Let µP = E[RP ] and σP 2 = var(R ).
P

Consider the Taylor expansion

W 2
u[W0(1 + RP )] ≈ u(W0) + W0u0(W0)RP + 0 u00(W0)RP
2 + ··· .
2
2] =
Neglecting third and higher order moments and noting E[RP
2 + µ2 .
σP P
 W 2
E[u(W0(1 + µP ))] ≈ u(W0) + W0u0(W0)µP + 0 u00(W0)(σP 2 + µ2 ) + · · ·
P
" 2 #
2
W0 00 0
2u (W0)
= u(W0) − u (W0) − µ − (σ 2 + µ2 )
P P P
2 W0u00(W0)
+···
W u00 (W )
0 0
Neglecting µ2
P compared to σP
2 and letting R = −
R , we
u0(W0)
2 2.
have the objective function: µP − σP
RR

Note that the expected utility can be expressed solely in terms of

2 when
mean µP and variance σP

(i) u is a quadratic function, or

(ii) RP is normal.
Quadratic optimization problem

max [2τ µT w − wT Ωw] subject to 1 · w = 1.

w∈RN
The Lagrangian formulation becomes:

L(w; λ) = 2τ µT w − wT Ωw + λ(wT 1 − 1).

The first order conditions are
(
2τ µ − 2Ωw∗ + λ1 = 0
.
1·w =1∗
Express the optimal solution w∗ as wg + τ z ∗, τ ≥ 0.

1. When τ = 0, 2Ωw = λ01 and 1T w g = 1

λ0 −1 T λ T
wg = Ω 1 and 1 = 1 wg = 0 1 Ω−11
2 2
hence
Ω−11
wg = T
(independent of µ).
1 Ω−11
 λ
2. When τ ≥ 0, w = τ Ω−1µ + Ω−11.
2
T T λ T −1 λ 1 − τ 1Ω−1µ
1 = 1 w = τ1 Ω−1µ + 1 Ω 1 so that = T .
2 2 1 Ω−11
T  T 

w= τ Ω−1µ+
1−τ 1 Ω−1µ −1
Ω 1 = τ Ω−1µ −
1 Ω−1µ Ω−11+w .
T T −1 g
1 Ω−11 1 Ω 1
We obtain
T −1
∗ −1 1
z =Ω µ− T
Ω µ −1 T
Ω 1 and 1 z ∗ = 0.
1 Ω−11
Observe that cov(Rwg , Rz∗) = z ∗T Ωwg = 0.
Financial interpretation

wg leads to a minimum risk position. This position is modified

by investing in the self-financing portfolio z ∗ so as to maximize
2τ µT w − wT Ω−1w.

Efficient frontier

Consider

µP = µT (wg + τ z ∗) = µg + τ µP,z ∗
2 = σ 2 + 2τ cov(R
σP , R ) + τ 2σ 2 .
g |
w g z
{z
∗
} z∗
z ∗T Ωwg =0
à !2
2 2 µP − µg 2 . Hence, the frontier is
By eliminating τ, σP = σg + σz ∗
µP,z ∗
2 )-diagram and hyperbolic in the (µ , σ )-
parabolic in the (µP , σP P P
diagram.
Inclusion of riskfree asset (deterministic rate of return R0 = r)

N
X
Let w = (w0 · · · wN )T and wi = 1.
i=0

Lagrangian formulation becomes

c
bTw
L = 2τ µ cT Ωw
c−w cT 1 − 1) + 2τ w0r + λw0
c + λ(w
T
b = (µ1 · · · µN )T ∈ RN , 1 = (1 · · · 1)T ∈
c = (w1 · · · wn)T ∈ RN , µ
where w
RN .

The optimality conditions become

2τ r + λ = 0 (i)
2τ µ c∗ + λ1 = 0
b − 2Ωw (ii)
N
X
wi∗ = 1 (iii)
i=0
Estimation of risk tolerance (inverse problem)

Reverse optimization: given an efficient portfolio w∗, it is possible

∗ ) and µ∗ . Taking the inner product
to express τ in terms of var(RP P
of w ∗
c with (ii), we obtain
T
2τ (rw0∗ + µ
bTw
c∗) − 2w
c∗ Ωw
c + λ = 0.

By eliminating λ using (i), we obtain the implied risk tolerance as

follows:
var(RP∗)
τ = ∗ .
µP − r
Marginal utilities

c
With w0 = 1 − 1 · w
c, we obtain the objective function

c
F (w b − r 1)T w
c) = 2τ [r + (µ cT Ωw
c] − w c
c
b − r 1) − 2Ωw
∇F = 2τ (µ c

so that

(∇F )i = 2τ [µi − r] − 2cov(RP , Ri), i = 1, 2, · · · , N.

An increase of amount dwi in the weight of asset i and a corre-
sponding reduction of the riskless asset leads to a marginal change
(∇F )i dwi of the objective function. By increasing (decreasing) the
positions with high (low) marginal utilities, an efficient portfolio w
can be considerably improved.
Summary

1. The objective function 2τ µT w − wT Ωw represents a balance of

maximizing return 2τ µT w against risk wT Ωw, where the weigh-
ing factor τ is related to the reciprocal of relative risk aversion
coefficient RR .

2. The optimal solution takes the form

w∗ = wg + τ z ∗
where wg is the portfolio weight of the global minimum variance
portfolio and the weights in z ∗ are summed to zero.
3. The additional variance above σg2 is given by

τ 2σz
2 .
∗

This is obvious since cov(Rwg , Rz ∗) = 0, that is, Rwg and Rz ∗

are uncorrelated

4. With the inclusion of riskfree asset, the marginal utility (∇F )i

of the ith asset can be increased by

(i) higher value of τ, τ ≥ 0,

(ii) higher positive value of µi − r
(iii) higher negative correlation between portfolio’s rate of return
RP and asset’s rate of return Ri.
Asset-Liability Model

Liabilities of a pension fund = future benefits − future contributions

Market value can hardly be determined since liabilities are not readily
marketable. Assume that some specific accounting rules are used to
calculate an initial value L0. If the same rule is applied one period
later, a value L1 results.

L1 − L0
Growth rate of the liabilities = RL = , where RL is expected
L0
to depend on the changes of interest rate structure, mortality and
other stochastic factors. Let A0 be the initial value of assets. The
investment strategy of the pension fund is given by the portfolio
choice w.
Surplus optimization

Depending on the portfolio choice w, the surplus gain after one

period

S1 − S0 = A0Rw − L0RL, where S0 = A0 − L0.

The return on surplus is defined by
S1 − S0 1
RS = = Rw − RL
A0 f0
where f0 = A0/L0 is the initial funding ratio.

Maximization formulation:-
( " # Ã !)
1 1
max 2τ E Rw − RL − var Rw − RL
w ∈ RN f0 f0
N
X 1
subject to wi = 1. Note that RL and var(RL) are independent
i=1 f0
of w so that they do not enter into the objective function.
 ( )
2
max 2τ E[Rw ] − var(Rw ) + cov(Rw , RL)
w ∈ RN f0
N
X
subject to wi = 1. Recall that
i=1
 
N
X N
X
cov(Rw , RL) = cov  wiRi, RL = wicov(Ri, RL).
i=1 i=1

max {2τ µT w + 2γ T w − wT Ωw} subject to 1T w = 1,

w ∈ RN

1
where γ T = (γ1 · · · γN ) with γi = cov(Ri, RL),
f0

µT = (µ1 · · · µN ) with µi = E[Ri], σij = cov(Ri, Rj ).

Remarks

1. The additional term 2γ T w in the objective function arises from

the correlation cov(Ri, RL) between return of risky asset i and
return of liability multiplied by the factor L0/A0.

2. Compared to the earlier model, we just need to replace µT by

1
µT + γ T . The efficient portfolios are of the form
τ
w∗ = wg + z L + τ z ∗, τ ≥ 0,
T −1
1
where z L = Ω−1γ − T
Ω γ
Ω−11 with
N
X
ziL = 0.
1 Ω−11 i=1
The occurrence of liabilities leads only to parallel shifts of the
set of efficient portfolios.
The mean-variance criterion can be reconciled with the expected
utility approach in either of two ways: (1) using a quadratic utility
function, or (2) making the assumption that the random returns are
normal variables.

Quadratic utility

b 2
The quadratic utility function can be defined as U (x) = ax − x ,
2
where a > 0 and b > 0. This utility function is really meaningful
only in the range x ≤ a/b, for it is in this range that the function is
increasing. Note also that for b > 0 the function is strictly concave
everywhere and thus exhibits risk aversion.
 mean-variance analysis ⇔ maximum expected utility criterion
based on quadratic utility

Suppose that a portfolio has a random wealth value of y. Using the

expected utility criterion, we evaluate the portfolio using
· ¸
b 2
E[U (y)] = E ay − y
2
b
= aE[y] − E[y 2]
2
b 2 b
= aE[y] − (E[y]) − var(y).
2 2

The optimal portfolio is the one that maximizes this value with
respect to all feasible choices of the random wealth variable y.
Normal Returns

When all returns are normal random variables, the mean-variance

criterion is also equivalent to the expected utility approach for any
risk-averse utility function.

To deduce this, select a utility function U . Consider a random

wealth variable y that is a normal random variable with mean value
M and standard deviation σ. Since the probability distribution is
completely defined by M and σ, it follows that the expected utility
is a function of M and σ. If U is risk averse, then
∂f ∂f
E[U (y)] = f (M, σ), with >0 and < 0.
∂M ∂σ
• Now suppose that the returns of all assets are normal random
variables. Then any linear combination of these asset is a normal
random variable. Hence any portfolio problem is therefore equiv-
alent to the selection of combination of assets that maximizes
the function f (M, σ) with respect to all feasible combinations.

• For a risky-averse utility, this again implies that the variance

should be minimized for any given value of the mean. In other
words, the solution must be mean-variance efficient.

• Portfolio problem is to find w∗ such that f (M, σ) is maximized

with respect to all feasible combinations.
 Two fund monetary separation

Consider a financial market with the riskfree asset and several risky
assets, suppose the utility function satisfies
u0(z)
− 00 = a + bz, valid for all z,
u (z)
then the optimal portfolio at different wealth levels is given by the
combination of the riskfree asset and market fund consisting of the
risky assets. The relative proportions of risky assets in the market
fund remain the same, irrespective of W0.

Remark

The class of utility functions include

(i) quadratic utility
(ii) log utility: a = 0
(iii) exponential utility: b = 0
(iv) power utility: a = 0.
Let W0 be the initial wealth, then the wealth amount a∗j (W0) of
risky asset j in the optimal portfolio satisfies

a∗j (W0) = αj h(W0) j = 1, 2, · · · , n, and αj is independent of W0,

so that the relative proportion bj is given by
a∗j (W0) αj
bj = n = n , independent of W0.
X X
a∗k (W0) αk
k=1 k=1
Lemma

1. Suppose the utility function satisfies

u0(W1)
− 00 = a + bW1, for all W1,
u (W1)
then the optimal portfolio is given by

a∗j (W0) = αj (a + bRW0), j = 1, 2, · · · , n, (A)

e = 1 + re , j = 1, 2, · · · , n.
where R = 1 + rf and R j j

2. Define V (W0) = max{aj }n f )],

E[u(W
j=1
1
 
m
X n
X n
X
f = W −
where W aj  R + e = RW +
aj R e −R),
aj (R
1 0 j 0 j
j=1 j=1 j=1
then
V 0(W0) a
− 00 = + bW0, for all initial wealth W0. (B)
V (W0) R
Proof

Assume that
a∗j (W0) = αj (W0)(a + bRW0)
where αj (W0), j = 1, · · · , n, is a differentiable function.

For any value of W0, from the optimality property of a∗j (W0), we
deduce that
f )]
∂E[u(W 1
∂ak
   
   
  N
 
  X  
 0  e − R)
e − R)α (W )(a + bRW ) (R
= E u RW0 + (Rj j 0 0 k  = 0,
   
  j=1  
 | {z } 
f1
W
(1)
k = 1, 2, · · · , N.
Next, we differentiate eq (1) with respect to W0. First, we observe
that
 
f N
X
dW 1 e − R)α (W )b
= R 1 + (R j j 0
dW0 j=1
N dα (W )
X j 0 e − R)(a + bRW ).
+ (Rj 0
j=1 dW 0

Hence, for the kth component, we obtain

N
X dαj (W0)
E[u00(W
f )(R
1
e
j
e
− R)(Rk − R)(a + bRW0)]
j=1 dW0
  
N
X
 00 f 
= −E u (W1)(Rk − R)R 1 + e − R)α (W )b ,
(R j j 0
j=1
k = 1, 2, · · · , N.
In matrix form, we have
   dα1 (W0 )

e1 − R)
(R 2 · · · (R e1 − R)(R eN − R) dW0
 (R e2 − R)(R e1 − R) · · · (R e2 − R)(R eN − R)  00 
dα2 (W0 ) 

E 
 ... ... ... 
 u ( f
W 1 )(a + bRW 0 )

dW0
... 
 
eN − R)(R
(R e1 − R) · · · (ReN − R)2 dαN (W0 )
dW0
 n h PN io 
E u (W00 f1)(R e1 − R)R 1 + e
j=1 (Rj − R)αj (W0 )b
 n h PN io 
 00 f1)(R e2 − R)R 1 + ej − R)αj (W0)b 
 E u (W j=1 ( R 
= − 
 ... 
 n h PN io 
00
E u (W f1)(R eN − R)R 1 + e
j=1 (Rj − R)αj (W0 )b
(2)
From the assumption
u0(W1)
− 00 = a + bW1,
u (W1)
we obtain
u0 (W
f )
1
u00(W
f )
1 = − h PN i
e
a + b RW0 + j=1(Rj − R)αj (W0)(a + bRW0)
u0 (W
f )
1
= − h PN i. (3)
e
(a + bRW0) 1 + j=1(Rj − R)αj (W0)b
observe that
 
N
X
−u00(W
f )(R
1
e − R)R 1 +
k
e − R)α (W )b
(R j j 0
j=1
R
= u0 (W
f )(R
1
e
k − R) , k = 1, 2, · · · N. (4)
a + bRW0
Recall the first order condition:

E[u0(W
f )(R
1
e − R)] = 0
k k = 1, 2, · · · , N.
Combining eqs (1) and (4), and knowing that the column vector on
the right hand side of eq (2) is a zero vector, we deduce that
dαj
(W0) = 0, j = 1, 2, · · · , n,
dW0
provided that the matrix in eq (2) is non-singular. We then have

αj (W0) = αj , independent of W0.

Now, a∗j = αj (a + bW0), a > 0. When b = 0, a∗j is independent of the
initial wealth W0.

The portfolio (a∗1(W0) · · · a∗N (W0)) is said to be partially separated if

a∗j (W0)/a∗j 0 (W0) is independent of W0, and it is said to be completely
separated if a∗j is independent of W0.
To show eq (B), we start from the optimality condition on

a∗j (W0) = αj (a + bRW0)

to obtain
  
n
X
V (W0) = E u RW0 + e − R)α (a + bRW )
(R j j 0
j=1
   
n
X N
X
= E u 1 + (Re − R)αj b RW0 + e − R)α a .
(R
j j j
j=1 j=1
Differentiate V (W0) twice with respect to W0
  
n
X
V 0(W 0) = E u0(W
f )R 1 +
1
e − R)α b
(R j j
j=1
  2 
n
X
 e − R)α b 
V 00(W0) = E u00(W
f )R2 1 +
1 ( Rj j .
j=1

Relating u00(W
f ) with u0(W
1
f ) using eq (3), we obtain
1
h ³ Pn ´i
RE u0 (W
f )R
1
e
1 + j=1(Rj − R)αj b R
V 00(W 0) = − =− V 0(W0).
a + bRW0 a + bRW0
Combining the results
V 0(W0) a
− 00 = + bW0.
V (W0) R
Formulation for finding the optimal portfolio

Let α be the weight of the riskfree asset so that the wealth invested in risky
assets is W0 (1 − α). Let bj be the weight of risky asset j within W0(1 − α) so that
Xn
bj = 1. The random wealth W f at the end of the investment period is
j=1

f )]
max E[u(W
{α,bj }

where
n
X
f
W = W0 α(1 + rf ) + W0 (1 − α)bj (1 + rej )
j=1
 
n
X
= W0 1 + αrf + (1 − α) bj rej 
j=1

subject to
n
X
bj = 1.
j=1
Lagrangian formulation
 
n
X
f )] + λ 1 −
max E[u(W bj  .
{α,bj ,λ} j=1
First order conditions give
  
n
X
E u0(W
f )W r −
0 f bj rej  = 0 (1)
j=1
h i
E 0 f
u (W )W0(1 − α)rej = λ, j = 1, 2, · · · , n, (2)
n
X
bj = 1. (3)
j=1
 
n
X
From eq. (1), E[u0(W
f )r ]
f = E u0(W
f) bj rej ,
j=1

and from eqs (2) and (3), we have

 
n
X
λ= E u0(W
f )W
0(1 − α) bj rej  .
j=1
Substituting into eq (2)
 
n
X
f )re ] = E u0(W
E[u0(W j
f) bj rej  , j = 1, 2, · · · , n,
j=1
and using eq (1), we obtain

E[u0(W
f )(re − r )] = 0
j f

or equivalently,
    
n
X
E u0 W0 1 + rf + (1 − α) b`(re` − rf ) (rej − rf ) = 0,
`=1
j = 1, 2, · · · , n. (4)
Exponential utility

Consider u0(z) = Ae−az , a > 0, substituting into eq. (4)

      
 n
X 
E A exp −a W0 (1 + rf ) + (1 − α) b`(re` − rf )  (rej − rf ) = 0
 
`=1
and since A exp(−aW0(1 + rf )) is non-random, we have
h Pn i
−a e`−rf ) e
`=1 W0 (1−α)b` (r
E e (r j − rf ) = 0, j = 1, 2, · · · , n. (5a)
For another initial wealth W00 , we have similar result
· Pn ¸
0 0 0 e −r )
E e−a `=1 W0 (1−α )b` (r` f (r
ej − rf ) = 0, j = 1, 2, · · · , n. (5b)
Suppose we postulate that the solution to the system of equations
h Pn i
−a e`−rf ) e
`=1 β` (r
E e (rj − rf ) = 0, j = 1, 2, · · · , n,
is unique, then by comparing eqs (5a,b), we obtain

W0(1 − α)b` = W00 (1 − α0)b0`.

Summing ` from 1 to n, we obtain

W0(1 − α) = W00 (1 − α0),

hence
b` = b0`, ` = 1, 2, · · · , n.
The total wealth amount W0(1 − α) invested in risky assets and the
wealth amount in each asset are independent of W0.