Portfolio Optimization

Definition: Portfolio
A portfolio is a collection of two or more assets say, 𝑎1 , 𝑎2 , . . 𝑎𝑛 ,
represented by an ordered 𝑛 − 𝑡𝑢𝑝𝑙𝑒 Θ = Θ(𝑥1 , 𝑥2 , . . 𝑥𝑛 ) , where
𝑥𝑖 ∈ 𝑅 , 𝑖 = 1 … 𝑛 is the number of units of the asset 𝑎𝑖 (𝑖 = 1, : . . , 𝑛)
owned by the investor.
We consider only a single period model, ie, in between the initial time taken
as 𝑡 = 0 and the final transaction time taken as 𝑡 = 𝑇, no transaction
ever takes place.

Let 𝑉𝑖 (0) and 𝑉𝑖 (𝑇) be the values of the ith asset at 𝑡 = 0 and 𝑡 = 𝑇.
respectively. Let 𝑉Θ (0) and 𝑉Θ (𝑇) denote the values of the portfolio
Θ = Θ(𝑥1 , 𝑥2 , . . 𝑥𝑛 ) at 𝑡 = 0 and 𝑡 = 𝑇, respectively. Then,
𝑛 𝑛

𝑉Θ (0) = ∑ 𝑥1 𝑉𝑖 (0) & 𝑉Θ (𝑇) = ∑ 𝑥𝑖 𝑉𝑖 (𝑇)

𝑖=1 𝑖=1

Then the quantity

𝑉Θ (𝑇) − 𝑉Θ (0)
𝑟Θ (𝑇) =
𝑉Θ (0)

is referred as the return of the portfolio Θ(𝑥1 , 𝑥2 , . . 𝑥𝑛 ).

Definition: Asset Weights

The weight 𝑤𝑖 of the asset 𝑎𝑖 is the proportion of the value of the asset in
the portfolio for (𝑖 = 1, : . . , 𝑛) at 𝑡 = 0, i.e.

𝑥𝑖 𝑉𝑖 (0) 𝑎𝑖 𝑉𝑖 (0)
𝑤𝑖 = = 𝑛 , 𝑖 = 1, : . . , 𝑛
𝑉Θ (0) ∑𝑖=1 𝑥𝑖 𝑉𝑖 (0)
It can be observed that 𝑤1 + 𝑤2 + ⋯ + 𝑤𝑛 = 1 .
Therefore, a portfolio can now be represented by the weights as
(𝑤1 , 𝑤2 , . . 𝑤𝑛 ).
§§ 𝑤𝑖 < 0 for some ‘𝑖’ is also possible in a portfolio, it indicates that the
investor has taken a short position on the 𝑖 − 𝑡ℎ asset 𝑎𝑖 .

Let 𝑟𝑖 be the return on the 𝑖 − 𝑡ℎ asset. Then

𝑉𝑖 (𝑇) − 𝑉𝑖 (0)
𝑟𝑖 = , 𝑖 = 1, . . 𝑛
𝑉𝑖 (0)

1 | Unit-4
Definition: Mean of the Portfolio Return
(referred as return of the portfolio)
Let (𝑤1 , 𝑤2 , . . 𝑤𝑛 ) be a portfolio of ‘n’ assets 𝑎1 , 𝑎2 , . . 𝑎𝑛 . Let 𝑟𝑖 , ( 𝑖 =
1, . . 𝑛) be the return on the 𝑖𝑡ℎ asset 𝑎𝑖 and 𝐸(𝑟𝑖 ) = 𝜇𝑖 , (𝑖 = 1, : . . , 𝑛), be its
expected value . Then the mean of the portfolio return is defined as
𝑛 𝑛 𝑛
𝜇 = 𝐸 (∑ 𝑤𝑖 𝑟𝑖 ) = ∑ 𝑤𝑖 𝐸(𝑟𝑖 ) = ∑ 𝑤𝑖 𝜇𝑖
𝑖=1 𝑖=1 𝑖=1

[ Since (total return of the portfolio) 𝑅𝑖 = ∑𝑛

𝑖=1 𝑤𝑖 𝑟𝑖 , hence 𝜇 = 𝐸(𝑅𝑖 ) ]

Definition: Variance of the Portfolio

(referred as risk of the portfolio)
Let (𝑤1 , 𝑤2 , . . 𝑤𝑛 ) be a portfolio of ‘n’ assets 𝑎1 , 𝑎2 , . . 𝑎𝑛 . Then the
variance of the portfolio defined as
𝑛 𝑛 𝑛
𝜎 = 𝑉𝑎𝑟 (∑ 𝑤𝑖 𝑟𝑖 ) = ∑ ∑ 𝑤𝑖 𝑤𝑗 𝜎𝑖𝑗 , 𝑤ℎ𝑒𝑟𝑒 𝜎𝑖𝑗 = 𝐶𝑜𝑣(𝑟𝑖 , 𝑟𝑗 )
2

𝑖=1 𝑖=1 𝑗=1

Also, 𝜎2 𝑖 = 𝑉𝑎𝑟(𝑟𝑖 ) & 𝜎2𝑗 = 𝑉𝑎𝑟(𝑟𝑗 )

If 𝜌𝑖𝑗 is correlation coefficient between 𝑟𝑖 & 𝑟𝑗 , then

𝐶𝑜𝑣(𝑟𝑖 , 𝑟𝑗 ) 𝜎𝑖𝑗
𝜌𝑖𝑗 = = ⇒ 𝜎𝑖𝑗 = 𝜌𝑖𝑗 𝜎𝑖 𝜎𝑗
𝜎𝑖 𝜎𝑗 𝜎𝑖 𝜎𝑗

Hence variance is also expressed as

𝑛 𝑛
𝜎 = ∑ ∑ 𝑤𝑖 𝑤𝑗 𝜌𝑖𝑗 𝜎𝑖 𝜎𝑗
2

𝑖=1 𝑗=1

Therefore given a portfolio

𝐴: (𝑤1 , 𝑤2 , . . 𝑤𝑛 ) , we can compute
its mean 𝜇𝐴 and standard
deviation 𝜎𝐴 and therefore get the
point A (𝜎𝐴 , 𝜇𝐴 ) in (𝜎, 𝜇) −
𝑝𝑙𝑎𝑛𝑒. Thus irrespective of the
number assets, a portfolio can
always be identified as a point in the
(𝜎, 𝜇) − 𝑝𝑙𝑎𝑛𝑒 .

2 | Unit-4
The portfolio optimization problem refers to the problem of determining
weights 𝑤𝑖 , ( 𝑖 = 1, . . 𝑛) such that the return of the portfolio is maximum
and the risk of the portfolio is minimum. Thus we aim to solve the following
optimization problem,

i. Minimize the risk, ie, 𝑚𝑖𝑛 ∑𝑛 𝑛

𝑖=1 ∑𝑗=1 𝑤𝑖 𝑤𝑗 𝜎𝑖𝑗
ii. Maximize the return, ie, 𝑚𝑎𝑥 ∑𝑛
𝑖=1 𝑤𝑖 𝜇𝑖

subject to, 𝑤1 + 𝑤2 + ⋯ + 𝑤𝑛 = 1
[ the search of combination of weights (𝑤1 , 𝑤2 , . . 𝑤𝑛 ) ]

Two Assets Portfolio Optimization

Consider a portfolio with two assets , say, 𝑎1 & 𝑎2 with weights 𝑤1 & 𝑤2
returns 𝑟1 & 𝑟2 and standard deviations 𝜎1 & 𝜎2 respectively. Then the
portfolio expected return 𝜇 and portfolio variance 𝜎 2 are given by

𝜇 = 𝐸(𝑤1 𝑟1 + 𝑤2 𝑟2 ) = 𝑤1 𝜇1 + 𝑤2 𝜇2 …..(1)

𝜎 2 = 𝑉𝑎𝑟(𝑤1 𝑟1 + 𝑤2 𝑟2 ) = 𝑤12 𝜎12 + 𝑤22 𝜎22 + 2𝑤1 𝑤2 𝜌 𝜎1 𝜎2 …..(2)

Here 𝜌 is the coefficient of correlation between 𝑟1 & 𝑟2 and lies in [−1, 1].
The value of 𝜌 provides a measure of the extent of diversification of the
portfolio so as to reduce risk. Larger the value of 𝜌 with negative sign,
smaller will be the value of 𝜎 2 .

Since , 𝑤1 + 𝑤2 = 1
Moreover, in case of short selling, the weights can be negative, hence

Let 𝑤2 = 𝑠 𝑡ℎ𝑒𝑛 𝑤1 = 1 − 𝑠 , 𝑠 ∈ 𝑅.
𝜇 = (1 − 𝑠)𝜇1 + 𝑠𝜇2 …..(3)
𝜎 2 = (1 − 𝑠)2 𝜎12 + 𝑠 2 𝜎22 + 2𝑠(1 − 𝑠) 𝜌 𝜎1 𝜎2 …..(4)
or, 𝜎 2 = (𝜎12 + 𝜎22 − 2 𝜌 𝜎1 𝜎2 )𝑠 2 − 2(𝜎1 − 𝜌𝜎2 )𝜎1 𝑠 + 𝜎21
Without loss of generality we assume that 0 < 𝜎1 ≤ 𝜎2 .
We discuss equ. (3 & 4) under two cases

(1) 𝜌 = ±1 (2) −1 < 𝜌 < 1

Case (1): for 𝜌 = ±1 , equ. (3 & 4) reduces to

𝜇 = (1 − 𝑠)𝜇1 + 𝑠𝜇2
𝜎 = |(1 − 𝑠)𝜎1 ± 𝑠𝜎2 |

3 | Unit-4
For 𝑠 ∈ [0,1] both weights are non-negative (no short selling). We have

𝜇 = (1 − 𝑠)𝜇1 + 𝑠𝜇2
𝜎̃ = (1 − 𝑠)𝜎1 + 𝑠𝜎2 (𝜌 = 1) 𝑜𝑟 𝜎̃ = (1 − 𝑠)𝜎1 − 𝑠𝜎2 (𝜌 = −1)
For, we plot the points (𝜎
̃ , 𝜇) (obtained by considering the different values of
𝑠) in (𝜎̃, 𝜇) − 𝑝𝑙𝑎𝑛𝑒 .

We get the following fig. (graphs are essentially straight lines). The bold
parts corresponding to 𝑠 ∈ [0,1].

Subsequently, we plot on (𝜎, 𝜇) − 𝑝𝑙𝑎𝑛𝑒 , and we the following graph

(i) 𝜌 = 1 𝑎𝑛𝑑 𝜎1 < 𝜎2 , we have

𝜎 2 = (1 − 𝑠)2 𝜎12 + 𝑠 2 𝜎22 + 2𝑠(1 − 𝑠)𝜎1 𝜎2 = {(1 − 𝑠)𝜎1 + 𝑠𝜎2 }2 …..(5)

𝑑𝜎 2
For extremum, = 2(𝜎1 − 𝜎2 ){(1 − 𝑠)𝜎1 + 𝑠𝜎2 } = 0 or
𝑑𝑠
𝜎1
𝑠𝑚𝑖𝑛 = − <0 … . . (6)
𝜎2 − 𝜎1
and

4 | Unit-4
 𝑑2 𝜎2
= 2(𝜎2 − 𝜎1 )2 > 0
𝑑𝑠 2
hence,
𝜎1 𝜎2
1 − 𝑠𝑚𝑖𝑛 = 1 + = >0
𝜎2 − 𝜎1 𝜎2 − 𝜎1
Let 𝜇𝑚𝑖𝑛 & 𝜎 2 𝑚𝑖𝑛 denote the expected return and variance of the portfolio,
and
𝜎2
𝑤1 = 1 − 𝑠𝑚𝑖𝑛 = >0 ,
𝜎2 − 𝜎1
𝜎1
& 𝑤2 = 𝑠𝑚𝑖𝑛 = − <0 … . . (7)
𝜎2 − 𝜎1
𝜎2 𝜇1 − 𝜎1 𝜇2
𝜇𝑚𝑖𝑛 = & 𝜎 2 𝑚𝑖𝑛 = 0 … . . (8)
𝜎2 − 𝜎1
Since 𝑠𝑚𝑖𝑛 < 0 ie, 𝑤2 < 0 an investor can eliminate risk in the portfolio
by taking a short position with respect to asset 𝑎2 .

(ii) 𝜌 = −1 & 𝜎1 ≤ 𝜎2 we have

𝜎 2 = (1 − 𝑠)2 𝜎12 + 𝑠 2 𝜎22 − 2𝑠(1 − 𝑠)𝜎1 𝜎2 = {(1 − 𝑠)𝜎1 − 𝑠𝜎2 }2

𝑑𝜎 2
= 2(𝜎1 + 𝜎2 ){(1 − 𝑠)𝜎1 − 𝑠𝜎2 } = 0 … . . (9)
𝑑𝑠
𝑑2 𝜎2
= 2(𝜎2 + 𝜎1 )2 > 0
𝑑𝑠 2
𝜎1
𝑤2 = 𝑠𝑚𝑖𝑛 = >0 ,
𝜎2 + 𝜎1
𝜎2
& 𝑤1 = 1 − 𝑠𝑚𝑖𝑛 = >0 … . . (10)
𝜎2 + 𝜎1
𝜎2 𝜇1 + 𝜎1 𝜇2
𝜇𝑚𝑖𝑛 = &
𝜎2 + 𝜎1
𝜎 2 𝑚𝑖𝑛 = 0 … . . (11)
Since 𝑤1 & 𝑤2 both are positive hence the investor can eliminate the risk
without selling.

Case (2): −1 < 𝜌 < 1 from equ. ( 3 & 4)

𝜇 = (1 − 𝑠)𝜇1 + 𝑠𝜇2

5 | Unit-4
 𝜎 2 = (𝜎12 + 𝜎22 − 2 𝜌 𝜎1 𝜎2 )𝑠 2 − 2(𝜎1 − 𝜌𝜎2 )𝜎1 𝑠 + 𝜎21
It represents the parametric equation of a parabola in (𝜎 2 , µ) − 𝑝𝑙𝑎𝑛𝑒.

𝑑𝜎 2 (𝜎1 − 𝜌𝜎2 )𝜎1

=0⇒𝑠= 2
𝑑𝑠 (𝜎1 + 𝜎22 − 2 𝜌 𝜎1 𝜎2 )
𝑑 2 𝜎2 2
2
= 2(𝜎21 + 𝜎22 − 2 𝜌 𝜎1 𝜎2 ) = 2 [(𝜎1 − 𝜌𝜎2 ) + (𝜎22 − 𝜌2 𝜎22 )] > 0 (∵ 𝜌
𝑑𝑠
< 1)
With the result we get

𝜎12 − 𝜌𝜎1 𝜎2 = (𝜎1 − 𝜌𝜎2 )𝜎1

𝑤2 = 𝑠𝑚𝑖𝑛 = &
(𝜎12 + 𝜎22 − 2 𝜌 𝜎1 𝜎2 )
𝑤1 = 1 − 𝑠𝑚𝑖𝑛
𝜎22 − 𝜌𝜎1 𝜎2 = (𝜎2 − 𝜌𝜎1 )𝜎2
= … . . (12)
(𝜎12 + 𝜎22 − 2 𝜌 𝜎1 𝜎2 )
𝜇𝑚𝑖𝑛 = (𝜇2 − 𝜇1 )𝑠𝑚𝑖𝑛 + 𝜇1 &

2
𝜎12 𝜎22 (1 − 𝜌2 )
𝜎 𝑚𝑖𝑛 = 2 … . . (13)
(𝜎1 + 𝜎22 − 2 𝜌 𝜎1 𝜎2 )
𝜎
i. If −1 ≤ 𝜌 < 𝜎1 then using equ. (12) we have 0 < 𝑠𝑚𝑖𝑛 < 1 then
2
minimum risk is attained without short selling. Also in this case

𝜎 2 𝑚𝑖𝑛 = 0 𝑓𝑜𝑟 𝜌 = −1
𝜎1
ii. If 𝜌 = ⟺ 𝑠𝑚𝑖𝑛 = 0 ⟺ 𝜎 2 𝑚𝑖𝑛 = 𝜎12 (∵ 𝑠𝑚𝑖𝑛 = 𝑤2 = 0)
𝜎2
𝜎1
iii. If < 𝜌 ≤ 1 ⟹ 𝑠𝑚𝑖𝑛 < 0 , ie, In this case the investor has taken a
𝜎2
short position on asset 𝑎2 in order to minimize the portfolio risk.
𝜎 2 𝑚𝑖𝑛 = 0 𝑓𝑜𝑟 𝜌 = 1

6 | Unit-4
 Some Results and Discussion
§§ The variance 𝜎 2 of a portfolio cannot exceed the greater of the

variances 𝜎12 & 𝜎22 of the components, ie,

𝜎 2 ≤ max{𝜎12 , 𝜎22 } if short selling is not allowed.
Proof:

Let us assume that 𝜎12 ≤ 𝜎22 . If short sales are not allowed, then
𝑤1 & 𝑤2 > 0 and
𝑤1 𝜎1 + 𝑤2 𝜎2 ≤ (𝑤1 + 𝑤2 )𝜎2 = 𝜎2
Since the correlation coefficient satisfies −1 ≤ 𝜌 ≤ 1, hence

𝜎 2 = 𝑤12 𝜎12 + 𝑤22 𝜎22 + 2𝑤1 𝑤2 𝜌 𝜎1 𝜎2

≤ (𝑤1 𝜎1 + 𝑤2 𝜎2 )2 ≤ 𝜎22
If 𝜎1 > 𝜎2 , the proof is analogous.

7 | Unit-4
§§ For −1 < 𝜌 < 1 the portfolio with minimum variance is attained at

𝜎12 − 𝜌𝜎1 𝜎2
𝑠𝑔 = 2
(𝜎1 + 𝜎22 − 2 𝜌 𝜎1 𝜎2 )
If short sales are not allowed, then the smallest variance is attained at
0 𝑖𝑓 𝑠𝑔 < 0
𝑠𝑚𝑖𝑛 = {𝑠𝑔 𝑖𝑓 0 ≤ 𝑠𝑔 ≤ 1
1 𝑖𝑓 𝑠𝑔 > 1
Proof: The relation for 𝑠𝑔 has already been obtained for global minimum
(since 𝜎 2 is a quadratic function of ) in equ. (12).
The subsequent result for no short sell can be visualized through the
following graph drawn in the (𝑠, 𝜎 2 ) − 𝑝𝑙𝑎𝑛𝑒.

Since the expression for σ2 is quadratic in s, hence the graph is parabola,

and bold parts is corresponding to the portfolio’s with no short selling,
ie, 0 ≤ 𝑠 ≤ 1, and we only obtain a segment of the curve. As 𝑠 increases
from 0 to 1, the corresponding point (𝜎, 𝜇) travels along the curve in the
direction from (𝜎1 , 𝜇1 ) 𝑡𝑜 (𝜎2 , 𝜇2 ) .

8 | Unit-4
§§ −1 ≤ 𝜌 ≤ 1 we have the following possibilities for 𝜎1 ≤ 𝜎2
𝜎
i. If −1 ≤ 𝜌 < 𝜎1 , then there is a portfolio without short selling such
2
that 𝜎 < 𝜎1 .
𝜎1
ii. If 𝜌 = , then 𝜎 ≥ 𝜎1 for each portfolio.
𝜎 2
𝜎1
iii. If < 𝜌 ≤ 1 , then there is a portfolio with short selling such that
𝜎2
𝜎 < 𝜎1 .
Proof:
𝜎
i. If −1 ≤ 𝜌 < 𝜎1 , then from expression of 𝑠𝑔
2
𝜎1 𝜎1
, ∵0 < 𝑠𝑔 < <1
𝜎1 + 𝜎2 𝜎1 + 𝜎2
Hence, 0 < 𝑠𝑔 < 1
That means no short selling and 𝜎 < 𝜎1 .
𝜎1
ii. If 𝜌= , then 𝑠𝑔 = 0 as a result we have 𝜎 ≥ 𝜎1 for every
𝜎2
portfolio because minimum variance is 𝜎12 .
𝜎
iii. If 1 ≥ 𝜌 > 𝜎1 then 𝑠𝑔 < 0 , that means the portfolio with minimum
2
variance that correspond to 𝑠𝑔 involves short selling of security 𝑎2
and satisfy 𝜎 < 𝜎1 . And if short selling is not allowed then minimum
variance will be 𝜎1 , hence 𝜎 > 𝜎1 for all other portfolio.

shows two typical examples of such curve, with 𝜌 close to but greater than
−1 (left) and with 𝜌 close to but smaller than +1 (right). Portfolios without
short selling are indicated by the bold line segments.

9 | Unit-4
Multi Asset Portfolio Optimization
The weights of the various assets 𝑎1 , 𝑎2 , . . 𝑎𝑛 in the portfolio are written in
the
vector form 𝑤 𝑇 = [𝑤1 , 𝑤2 , . . 𝑤𝑛 ] .
Let

𝑒 𝑇 = (1, . . . , 1) ∈ 𝑅𝑛 , then 𝑤1 + 𝑤2 +. . . 𝑤𝑛 = 1 can be expressed as

𝑇
𝑒 𝑤 = 1.
Let 𝑚𝑇 = (𝜇1 , 𝜇2 , . . 𝜇𝑛 ) be the expected return vector of the portfolio ,
where, 𝜇𝑖 = 𝐸(𝑟𝑖 ) and 𝐶 = [𝑐𝑖𝑗 ] denotes the 𝑛 𝑥 𝑛 variance-covariance
matrix, ie,

𝑐𝑖𝑗 = 𝐶𝑜𝑣(𝑟𝑖 , 𝑟𝑗 ) , 𝑖, 𝑗 = 1, . . 𝑛 .
Obviously C is a symmetric matrix. Now the expected return µ of the
portfolio is given by
𝑛 𝑛 𝑛
𝜇 = 𝐸 (∑ 𝑤𝑖 𝑟𝑖 ) = ∑ 𝑤𝑖 𝐸(𝑟𝑖 ) = ∑ 𝑤𝑖 𝜇𝑖 = 𝑚𝑇 𝑤
𝑖=1 𝑖=1 𝑖=1

and the variance of the portfolio is

𝑛 𝑛 𝑛
𝜎 2 = 𝑉𝑎𝑟 (∑ 𝑤𝑖 𝑟𝑖 ) = ∑ ∑ 𝑤𝑖 𝑤𝑗 𝜎𝑖𝑗 = 𝑤 𝑇 𝐶 𝑤
𝑖=1 𝑖=1 𝑗=1

Here 𝐶 is certainly positive semidefinite. In practice, it is also assumed to be

positive definite (and hence invertible) because the minimum risk of a general
n-asset portfolio is rarely zero.

The Feasible Region of a Portfolio Problem

Let 𝑊 = {𝑤 ∈ 𝑅 𝑛 : 𝑒 𝑇 𝑤 = 1} be the collection of all portfolios. Each
portfolio corresponds to a point in the (𝜎, 𝜇) − 𝑝𝑙𝑎𝑛𝑒 say (𝜎 𝑤 , 𝜇𝑤 ) . Then
the set {(𝜎𝑤 , 𝜇𝑤 ): 𝑤 ∈ 𝑊 } is called the feasible region.

Consider the 𝑛-dimensional hyperplane 𝑒 𝑇 𝑤 = 1 , in which the weight

vector 𝑤

resides. Let 𝑓 be the mapping that takes each weight vector in the
weight hyperplane to the corresponding portfolio point in the (𝜎, 𝜇) − 𝑝𝑙𝑎𝑛𝑒.
Our aim is to find the image of any straight line in the weight hyperplane
𝑒 𝑇 𝑤 = 1 under the mapping 𝑓.

10 | Unit-4
 The parametric equation of any line in the weight hyperplane is of the form
𝑙(𝜉) = (𝑠1 𝜉 + 𝑏1 , 𝑠2 𝜉 + 𝑏2 , … + 𝑠𝑛 𝜉 + 𝑏𝑛 )𝑇
= 𝑠𝜉 + 𝑏 , −∞<𝜉 <∞
Where , 𝑠 = (𝑠1 , 𝑠2 , . . 𝑠𝑛 )𝑇 & 𝑏 = (𝑏1 , 𝑏2 , . . 𝑏𝑛 )𝑇
Let 𝑤 be any point this line, Then,

𝜇 = 𝑚𝑇 𝑤
= 𝑚𝑇 (𝑠𝜉 + 𝑏)
= 𝜉(𝑚𝑇 𝑠) + (𝑚𝑇 𝑏)
Let , (𝑚𝑇 𝑠)−1 = 𝛼 & − (𝑚𝑇 𝑏)(𝑚𝑇 𝑠)−1 = 𝛽 , then
𝜉 = 𝛼𝜇 + 𝛽
Now, 𝜎 2 = 𝑤𝑇 𝐶 𝑤
= (𝑠𝜉 + 𝑏)𝑇 𝐶 (𝑠𝜉 + 𝑏)

= (𝑠 𝑇 𝐶𝑠)𝜉 2 + (𝑠 𝑇 𝐶𝑏 + 𝑏 𝑇 𝐶𝑠)𝜉 + 𝑏 𝑇 𝐶𝑏
2
= 𝛾 𝜉 + 𝛿𝜉 + 𝜂

Substituting the value of 𝜉 , we have

𝜎 2 = 𝛾(𝛼𝜇 + 𝛽)2 + 𝛿(𝛼𝜇 + 𝛽) + 𝜂 …..(A)

For −∞ < 𝜉 < ∞ , the ordered pair (𝜎 2 , 𝜇) traces a parabola given by
equ. (A) with axis parallel to 𝜎 -axis in (𝜎 2 , 𝜇) − 𝑝𝑙𝑎𝑛𝑒 .

For (𝜎, 𝜇) − 𝑝𝑙𝑎𝑛𝑒

𝜎 = √𝛾(𝛼𝜇 + 𝛽)2 + 𝛿(𝛼𝜇 + 𝛽) + 𝜂 ….(B)

The curve represented by equ.(B) in (𝜎, 𝜇) − 𝑝𝑙𝑎𝑛𝑒 is called Markowitz
Curve.

(The Markowitz curve given by equ.(B) is not a parabola in (𝜎, 𝜇) − 𝑝𝑙𝑎𝑛𝑒 .

The M.C. behaves almost as a straight line as 𝜇 → ∞, whereas it is always
possible to draw tangent on parabola for any value of 𝜇 .
As we cover all possible weight lines of the weight hyperplane, we trace a
family of Markowitz curves in the (𝜎, 𝜇) − 𝑝𝑙𝑎𝑛𝑒 , and this generates a solid
region in the (𝜎, 𝜇) − 𝑝𝑙𝑎𝑛𝑒 , and its shape will be like a bullet, which is
appropriately called the Markowitz bullet.)

11 | Unit-4
The solid region generated by all
possible weight vectors. Its left
boundary is called the minimum
variance set. The return corresponding
to points 𝑃0 , 𝑃1 , &𝑃2 are 𝜇0 , 𝜇1 , &𝜇2
respectively with known level of risk .
There is a point 𝑃𝑚𝑖𝑛 which has the
least variance. This point is called the
minimum variance point.

12 | Unit-4
To find the minimum variance point we need to solve the following risk
minimization problem

𝑚𝑖𝑛 𝜎2 = 𝑤𝑇𝐶 𝑤
} . . . . (1)
𝑠𝑢𝑏𝑗𝑒𝑐𝑡 𝑡𝑜 𝑒𝑇 𝑤 = 1
Theorem: Portfolio with minimum risk has weight given by

𝐶 −1 𝑒
𝑤 = 𝑇 −1
𝑒 𝐶 𝑒
Proof: Using the method of Lagrange Multiplier to solve equ. (1), we have

𝐿(𝑤, 𝜆) = 𝑤 𝑇 𝐶 𝑤 + 𝜆(1 − 𝑒 𝑇 𝑤)
𝑛 𝑛

= ∑ ∑ 𝑤𝑖 𝑤𝑗 𝜎𝑖𝑗 + 𝜆(1 − 𝑒 𝑇 𝑤)
𝑖=1 𝑗=1

differentiating w.r.t. ′𝑤′ and equating to zero, we get

𝜆
2𝑤 𝑇 𝐶 − 𝜆𝑒 𝑇 = 0 ⟹ 𝑤 = 𝐶 −1 𝑒 …..(2)
2

substituting in second equation of (1)

𝜆 𝑇 −1
𝑒 𝐶 𝑒=1
2
𝜆 1
= … . . (3)
2 𝑒 𝑇 𝐶 −1 𝑒
Putting in equ (2) from (3)

𝐶 −1 𝑒
𝑤 = 𝑇 −1
𝑒 𝐶 𝑒

13 | Unit-4
Markovitz Efficient Frontier
Looking at the minimum variance set in the (𝜎, 𝜇) − 𝑝𝑙𝑎𝑛𝑒 ,

we observe that for a given level of risk, (say 𝜎1 ), there are two values of

returns 𝜇0𝐿 < 𝜇0𝑈 . Since we want to maximize the return hence obvious

choice is 𝜇0𝑈 . Therefore in the minimum variance set, it is only the upper
half which is of importance for investment.

This upper half portion of the minimum variance set is called the Markowitz
efficient frontier.
Sometimes investor is more concern about the return. Hence for a given
return 𝜇 ,we are to find the value of the weights for minimum risk.

14 | Unit-4
Theorem: For a given expected return , the portfolio with minimum risk
has weights given by

𝜇 𝑚𝑇 𝐶 −1 𝑒 −1 𝑚𝑇 𝐶 −1 𝑚 𝜇 −1
𝑑𝑒𝑡 ( ) 𝐶 𝑚 + 𝑑𝑒𝑡 ( )𝐶 𝑚
𝑤= 1 𝑒 𝑇 𝐶 −1 𝑒 𝑒 𝑇 𝐶 −1 𝑚 1
𝑚𝑇 𝐶 −1 𝑚 𝑚𝑇 𝐶 −1 𝑒
𝑑𝑒𝑡 ( 𝑇 −1 )
𝑒 𝐶 𝑚 𝑒 𝑇 𝐶 −1 𝑒

Proof. We have to solve the following quadratic programming problem

1
𝑚𝑖𝑛 𝜎2 = 𝑤𝑇𝐶 𝑤 …..(1)
2

Subject to, 𝑚𝑇 𝑤 = 𝜇 .....(2)

𝑒𝑇𝑤 = 1 ....(3)
We define the Lagrangian with 2-parameters
1
𝐿(𝑤, 𝛼, 𝛽) = 𝑤 𝑇 𝐶 𝑤 + 𝛼(𝜇 − 𝑚𝑇 𝑤) + 𝛽(1 − 𝑒 𝑇 𝑤), 𝛼, 𝛽 ∈ 𝑅
2
𝜕𝐿
= 0 = 𝑤𝑇 𝐶 − 𝛼𝑚𝑇 − 𝛽𝑒𝑇
𝜕𝑤
or 𝑤 = 𝐶 −1 (𝛼𝑚 + 𝛽𝑒) …(4)
substituting in equ (2 & 3) from equ (4)

(𝑚𝑇 𝐶 −1 𝑚)𝛼 + (𝑚𝑇 𝐶 −1 𝑒)𝛽 = 𝜇 ...(5)

(𝑒 𝑇 𝐶 −1 𝑚)𝛼 + (𝑒 𝑇 𝐶 −1 𝑒)𝛽 = 1 …(6)

Solving for 𝛼 & 𝛽 we get

𝜇 𝑚𝑇 𝐶 −1 𝑒 𝑚𝑇 𝐶 −1 𝑚 𝜇
𝑑𝑒𝑡 ( 𝑇 −1
) 𝑑𝑒𝑡 (
𝑇 −1
)
𝛼= 1 𝑒 𝐶 𝑒 , & 𝛽= 𝑒 𝐶 𝑚 1
𝑚𝑇 𝐶 −1 𝑚 𝑚𝑇 𝐶 −1 𝑒 𝑚𝑇 𝐶 −1 𝑚 𝑚𝑇 𝐶 −1 𝑒
𝑑𝑒𝑡 ( 𝑇 −1 ) 𝑑𝑒𝑡 ( 𝑇 −1 )
𝑒 𝐶 𝑚 𝑒 𝑇 𝐶 −1 𝑒 𝑒 𝐶 𝑚 𝑒 𝑇 𝐶 −1 𝑒
Putting the value in equ. (4) we get the result
Karush-Kuhn-Tucker optimality conditions. The result is known as the two
fun d theorem.
Theorem 5.6.3 (Two Fund Theorem) Two efficient portfolios can be
established so that any other efficient portfolio can be duplicated, in terms
of mean and variance, as a linear combination of these two assets. In other
words, it says that, an investor seeking an efficient portfolio need to invest
only in the combination of these two assets. To be decided

15 | Unit-4
Capital Asset Pricing Model (CAPM)
So far we considered the portfolio consisting of risky assets. Now we
consider the portfolio containing risk free asset also.

Consider a portfolio with 𝑛 risky assets 𝑎1 , 𝑎2 , . . 𝑎𝑛 with weight 𝑤1 , 𝑤2 , . . 𝑤𝑛

and

one risk-free asset 𝑎𝑟𝑓 with weight 𝑤𝑟𝑓 · Then

𝑛

𝑤𝑟𝑖𝑠𝑘𝑦 + 𝑤𝑟𝑓 = ∑ 𝑤𝑖 + 𝑤𝑟𝑓 = 1 … (1)

𝑖=1

the expected return and the variance associated with this portfolio are given
by
𝑛

𝜇 = ∑ 𝑤𝑖 𝜇𝑖 + 𝑤𝑟𝑓 . 𝜇𝑟𝑓 … (2)

𝑖=1

and
𝑛 𝑛

𝜎 2 = 𝑉𝑎𝑟 (∑ 𝑤𝑖 𝑟𝑖 + 𝑤𝑟𝑓 𝜇𝑟𝑓 ) = 𝑉𝑎𝑟 (∑ 𝑤𝑖 𝑟𝑖 ) = 𝜎 2 𝑟𝑖𝑠𝑘𝑦 … (3)

𝑖=1 𝑖=1

respectively.
If we remove the risk-free asset from the portfolio and readjust the weight of
the risky assets so that their sum remain 1, the resultant portfolio so
obtained is referred to as the derived risky portfolio. We shall use
2
𝜇𝑑𝑒𝑟 & 𝜎𝑑𝑒𝑟 to represent the expected return and risk of the derived risky
portfolio. Then
𝑛
𝑤𝑖
𝜇 = 𝑤𝑟𝑖𝑠𝑘𝑦 ∑ 𝜇𝑖 + 𝑤𝑟𝑓 . 𝜇𝑟𝑓
𝑤𝑟𝑖𝑠𝑘𝑦
𝑖=1

= 𝑤𝑟𝑖𝑠𝑘𝑦 . 𝜇𝑑𝑒𝑟 + 𝑤𝑟𝑓 . 𝜇𝑟𝑓

= 𝑤𝑟𝑖𝑠𝑘𝑦 . 𝜇𝑑𝑒𝑟 + (1 − 𝑤𝑟𝑖𝑠𝑘𝑦 ). 𝜇𝑟𝑓

= 𝑤𝑟𝑖𝑠𝑘𝑦 (𝜇𝑑𝑒𝑟 − 𝜇𝑟𝑓 ) + 𝜇𝑟𝑓 … (4)

Also,
𝑛 𝑛
𝑤𝑖 2
𝑤𝑖
𝜎 2 = 𝑉𝑎𝑟 (𝑤𝑟𝑖𝑠𝑘𝑦 ∑ 𝑟𝑖 ) = 𝑤𝑟𝑖𝑠𝑘𝑦 𝑉𝑎𝑟 (∑ 𝑟𝑖 )
𝑤𝑟𝑖𝑠𝑘𝑦 𝑤𝑟𝑖𝑠𝑘𝑦
𝑖=1 𝑖=1

16 | Unit-4
 2 2
𝜎
= 𝑤𝑟𝑖𝑠𝑘𝑦 𝜎𝑑𝑒𝑟 ⟹ 𝑤𝑟𝑖𝑠𝑘𝑦 = … (5)
𝜎𝑑𝑒𝑟
Putting in equ. (4) we get

(𝜇𝑑𝑒𝑟 − 𝜇𝑟𝑓 )𝜎
𝜇 = 𝜇𝑟𝑓 + … (7)
𝜎𝑑𝑒𝑟
It is an equation of the line in (𝜎, 𝜇) − 𝑝𝑙𝑎𝑛𝑒 joining the points
(0, 𝜇𝑟𝑓 ) & (𝜎𝑑𝑒𝑟 , 𝜇𝑑𝑒𝑟 ) .

Now, for a given risk , if we choose various weight combinations of risk-

free
asset and risky assets for which equ. (1) holds, we generate different lines
represented by equ. (7) (𝜎, 𝜇) − 𝑝𝑙𝑎𝑛𝑒.
The line that produces the point with highest expected return for a given
risk is tangent to the upper portion of the Markowitz bullet.

Definition: Capital Market Line (CML)

Among all the lines satisfying equ. (7) for various weight combinations of
risk-free and risky assets, the line giving the highest return for a given risk
is called the capital market line, and point of contact on the Markowitz
bullet is said to represent Market Portfolio.

Theorem: One Fund Theorem

There exists a single portfolio, namely the market portfolio M, of risky
assets such that any efficient portfolio can be constructed as a linear
combination of the market portfolio M and the risk-free asset.

17 | Unit-4
Theorem: For any expected risk-free return 𝜇𝑟𝑓 , the weight vector 𝑤𝑀 of the
market portfolio is given by
𝐶 −1 (𝑚 − 𝜇𝑟𝑓 𝑒)
𝑤𝑀 = 𝑇 −1
𝑒 𝐶 (𝑚 − 𝜇𝑟𝑓 𝑒)
Proof:
For any (𝜎, 𝜇) in the Morkovitz bullet, the
slope of the line joining (0, 𝜇𝑟𝑓 ) and (𝜎, 𝜇)
is
𝜇 − 𝜇𝑟𝑓 ∑𝑛𝑖=1 𝜇𝑖 𝑤𝑖 − 𝜇𝑟𝑓
𝜇= = 1/2
𝜎 (∑𝑛𝑖=1 ∑𝑛𝑗=1 𝑤𝑖 𝑤𝑗 𝑐𝑖𝑗 )

For the line joining (0, 𝜇𝑟𝑓 ) and (𝜎, 𝜇) to be

a tangent line to the Markovitz bullet, we to solve,

𝑚𝑇 𝑤 − 𝜇𝑟𝑓
max 1/2
(𝑤𝑇 𝐶 𝑤) ⟩ … (1)
𝑠𝑢𝑏𝑗𝑒𝑐𝑡 𝑒𝑇 𝑤 = 1
The Lagrange’s function 𝐿: 𝑅 𝑛 × 𝑅 → 𝑅 for equ. (1) will be

𝑚𝑇 𝑤 − 𝜇𝑟𝑓
𝐿(𝑤, 𝜆) = + 𝜆(1 − 𝑒 𝑇 𝑤)
(𝑤 𝐶 𝑤)
𝑇 1/2

𝜕𝐿
considering =0
𝜕𝑤

1
[(𝑤 𝑇 𝐶 𝑤)1/2 𝑚 − (𝑚𝑇 𝑤 − 𝜇𝑟𝑓 )(𝑤 𝑇 𝐶 𝑤)−1/2 𝐶𝑤] = 𝜆𝑒
𝑤𝑇𝐶 𝑤
𝐶𝑤
or 𝜎 𝑚 − (𝜇 − 𝜇𝑟𝑓 ) = 𝜆𝜎 2 𝑒
𝜎

or 𝜎 2 𝑚 − (𝜇 − 𝜇𝑟𝑓 )𝐶𝑤 = 𝜆𝜎 3 𝑒
...(2)

or 𝜎 2 𝑤 𝑇 𝑚 − (𝜇 − 𝜇𝑟𝑓 )𝑤 𝑇 𝐶 𝑤 = 𝜆𝜎 3 𝑤 𝑇 𝑒
since 𝑤𝑇𝑚 = 𝜇 , 𝑒𝑇 𝑤 = 1 , & 𝜎2 = 𝑤𝑇𝐶 𝑤
𝜎 2 𝜇 − (𝜇 − 𝜇𝑟𝑓 )𝜎 2 = 𝜆𝜎 3
𝜇𝑟𝑓
𝜆= … (3)
𝜎
Hence the requisite value of weight vector 𝑤𝑀 can be obtained using
equ. (2 & 3)

18 | Unit-4
§§ If the market portfolio (𝜎𝑚 , 𝜇𝑚 ) is k:nown. Then the equation of
the capital market line is given by
(𝜇𝑀 − 𝜇𝑟𝑓 )
𝜇 = 𝜇𝑟𝑓 + 𝜎
𝜎𝑀
If the investor is willing to take a positive risk 𝜎 , it will in an
(𝜇𝑀 −𝜇𝑟𝑓 )
additional return 𝜎 over and above the risk-free return.
𝜎𝑀

§§ Now suppose an investor is willing to take risk 𝜎𝑃 , then for this risk
the expected return 𝜇𝑃 is maximum if the point (𝜎𝑃 , 𝜇𝑃 ) lies on
CML (Capital Market Line) , thus,
(𝜇𝑀 − 𝜇𝑟𝑓 )
𝜇𝑃 = 𝜇𝑟𝑓 + 𝜎𝑃
𝜎𝑀
𝜎𝑃
If we let 𝑤𝑃 = , then
𝜎𝑀
𝜇𝑃 = 𝜇𝑟𝑓 (1 − 𝑤𝑃 ) + 𝑤𝑃 𝜇𝑀
𝜎𝑃
That means investor should invest 𝑤𝑃 = proportion of investment
𝜎𝑀
in index and (1 − 𝑤𝑃 ) proportion in risk free.

Theorem: Suppose the market portfolio is (𝜎𝑀 , 𝜇𝑀 ) . The expected

return of an asset 𝑎𝑖 is given by

𝐶𝑜𝑣(𝑟𝑖 , 𝑟𝑀 )
𝜇𝑖 = 𝜇𝑟𝑓 + 𝛽𝑖 (𝜇𝑀 − 𝜇𝑟𝑓 ) , 𝑤ℎ𝑒𝑟𝑒 𝛽𝑖 =
𝜎𝑀2
Proof: Suppose an investor portfolio comprises of asset 𝑎𝑖 with weight 𝑤
and the market portfolio 𝑀 with weight 1 − 𝑤. Then the expected return
and risk of the portfolio will be

𝜇 = 𝑤𝜇𝑖 + (1 − 𝑤)𝜇𝑀
} ...(1)
𝜎 2 = 𝑤 2 𝜎𝑖2 + (1 − 𝑤)𝜎𝑀2 + 2𝜌𝑤(1 − 𝑤)𝜎𝑖 𝜎𝑗
Where 𝜌 is coefficient of correlation between the returns of asset 𝑎𝑖 and the
market portfolio 𝑀.

As 𝑤 varies, these values trace out

a curve in the (𝜎, 𝜇) − 𝑝𝑙𝑎𝑛𝑒. When
𝑤 =0 the capital market line (CML)
becomes tangent to the curve at 𝑀.
So we have the condition that slope
of the curve is slope of the CML at 𝑀,
differentiating (1)

19 | Unit-4
 𝑑𝜇 𝑑𝜇 𝑑𝑤 𝑑𝑤
| = | = (𝜇𝑖 − 𝜇𝑀 ) |
𝑑𝜎 𝑤=0 𝑑𝑤 𝑑𝜎 𝑤=0 𝑑𝜎 𝑤=0
𝑑𝜎 𝑤𝜎𝑖2 − (1 − 𝑤)𝜎𝑀2 + 𝜌𝜎𝑖 𝜎𝑀 (1 − 2𝑤)
𝑎𝑙𝑠𝑜 | = |
𝑑𝑤 𝑤=0 𝜎 𝑤=0

𝜎𝑖𝑀 − 𝜎𝑀2
= , ( 𝑠𝑖𝑛𝑐𝑒 𝜎𝑖𝑀 = 𝜌𝜎𝑖 𝜎𝑀 )
𝜎𝑀
𝑑𝜇 (𝜇𝑖 − 𝜇𝑀 )𝜎𝑀
𝑡ℎ𝑒𝑟𝑒𝑓𝑜𝑟𝑒 , | = … (2)
𝑑𝜎 𝑤=0 𝜎𝑖𝑀 − 𝜎𝑀2
Equating with slope of line, we get,
𝜇𝑖 − 𝜇𝑟𝑓 𝑑𝜇 (𝜇𝑖 − 𝜇𝑀 )𝜎𝑀
= | =
𝜎𝑀 𝑑𝜎 𝑤=0 𝜎𝑖𝑀 − 𝜎𝑀2
(𝜇𝑀 − 𝜇𝑟𝑓 )𝜎𝑀
𝑜𝑟 𝜇𝑖 = 𝜇𝑟𝑓 +
𝜎𝑀2
𝑜𝑟 𝜇𝑖 = 𝜇𝑟𝑓 + 𝛽𝑖 (𝜇𝑀 − 𝜇𝑟𝑓 ) … (3)
𝜎𝑖𝑀
Here 𝛽𝑖 = 2 is called the beta of an asset.
𝜎𝑀

If 𝜷𝒊 = 𝟎 ie, asset is completely uncorrelated with the market. Thus

CAPM gives 𝜇𝑖 = 𝜇𝑟𝑓 𝑢𝑠𝑖𝑛𝑔 (3)
Showing that however large is the risk 𝜎𝑖 , return will always be
limited to risk free, ie, no risk premium.
If 𝜷𝒊 < 𝟎 , then 𝜇𝑖 < 𝜇𝑟𝑓 , correlated negatively with the market. Hence
can be used to reduce the overall risk of the portfolio when other
assets are not doing well. For this reason it is called insurance.
(The above discussion suggests that though for a portfolio an appropriate
mea- sure of risk is 𝜎 but for an individual asset the proper measure of risk is
its beta.)

Example 5.7.2
Let the risk-free ·rate Pr! be 8% and the market has µM = 12% and aM =
15%. Let an asset a be given which has covariance of 0.045 with the market.
Determine the expected rate of return of the given asset.
Therefore the expected rate of return of the given asset is 16%.

20 | Unit-4
Definition: Beta of the Portfolio
The overall 𝛽 of the portfolio is defined as
𝑛

𝛽 = ∑ 𝑤𝑖 𝛽𝑖
𝑖=1

Ie, the weighted average of the betas of the individual assets in the portfolio
with the weights being those that define the portfolio.

Definition: Security Market Line

A linear equation
𝐶𝑜𝑣(𝑟𝑖 , 𝑟𝑀 )
𝜇 = 𝜇𝑟𝑓 + 𝛽(𝜇𝑀 − 𝜇𝑟𝑓 ) , 𝑤ℎ𝑒𝑟𝑒 𝛽𝑖 =
𝜎𝑀2
that describes the expected return
for all assets in the market is called
the security market line.

CAPM as a Pricing Formula

Let an asset be purchased at price 𝑃 and later sold at price 𝑄 , then rate
of return
𝑄−𝑃
𝑟=
𝑃
Here 𝑃 is known but 𝑄 is random. If we write 𝐸(𝑄) = ̃
𝑄 , then the
CAPM formula gives

𝑄̃ − 𝑃
= 𝜇𝑟𝑓 + 𝛽(𝜇𝑀 − 𝜇𝑟𝑓 )
𝑃
𝑄̃
𝑜𝑟 𝑃=
1 + 𝜇𝑟𝑓 + 𝛽(𝜇𝑀 − 𝜇𝑟𝑓 )
𝛽 is the beta of given asset.
𝑄
̃ is known then price 𝑃 can be determined , since 𝑃 =
If 𝑄 , hence
1+𝜇𝑟𝑓

𝜇𝑟𝑓 + 𝛽(𝜇𝑀 − 𝜇𝑟𝑓 ) can be interpreted as risk adjusted interest rate.

21 | Unit-4
Limitation of Markovitz Model
Markovitz model is known to be valid if returns 𝑟𝑖 are normally distributed
and the investor is ‘risk averse’ preferring lower standard deviation (S.D.).
The idea of standard deviation being a measure of risk may not be very
appealing to investor. This means the perception about risk of an investor is
symmetric about mean. There are several empirical studies, which reveal
that most 𝑟𝑖 are not normally are even symmetrically distributed.
If it is not symmetric about me then consideration of skewness and kurtosis
may also be considered. Thus the efficient Frontier in Markovitz model may
be generated in ( mean, variance, skewness, kurtosis )-space.

22 | Unit-4