Topic #10

Markowitz Portfolio Theory

Reading: Luenberger Chapter 6, Sections 6 - 10

Primbs/Investment Science 1
 The Markowitz Model

Solving the Optimization

The Two Fund Theorem

Markowitz Portfolio Theory
Inclusion of a Risk Free Asset

The One Fund Theorem

Markowitz’s Message

Primbs/Investment Science 2
 Picture of Markowitz
For a given mean return, you would like to
minimize your risk or the variance.

r
x


Minimum variance point for a given mean return

Primbs/Investment Science 3
 The Markowitz Model
Markowitz formulated the problem of being on the
efficient frontier as an optimization.

Assume there are n risky assets with

Mean returns: r1 , r2 , , rn
Covariances:  ij for i,j=1,...,n

Primbs/Investment Science 4
 Markowitz Optimization
1 n
Minimize: 
2 i , j 1
wi w j ij = 1/2 (Variance)
n
Subject to: wr  r
i 1
i i p = Mean Return

w
i 1
i 1 = Weights sum to 1.

Note: (1) We are allowing short selling!

(2) We assume all assets are risky!

Primbs/Investment Science 5
 The Markowitz Model

Solving the Optimization

The Two Fund Theorem

Markowitz Portfolio Theory
Inclusion of a Risk Free Asset

The One Fund Theorem

Markowitz’s Message

Primbs/Investment Science 6
 Solving a General Optimization
Min: f ( x, u )
s.t.: g1 ( x, u )  c1
g 2 ( x , u )  c2

Step 1: Convert all constraints to zero on the right hand side.

Min: f ( x, u )
s.t.: g1 ( x, u )  c1  0
g 2 ( x, u )  c2  0

Primbs/Investment Science 7
 Solving a General Optimization
Min: f ( x, u )
s.t.: g1 ( x, u )  c1
g 2 ( x , u )  c2

Step 2: Associate a Lagrange multiplier with each constraint.

Min: f ( x, u )
s.t.: g1 ( x, u )  c1  0 1
g 2 ( x, u )  c2  0 2

Primbs/Investment Science 8
 Solving a General Optimization
Min: f ( x, u )
s.t.: g1 ( x, u )  c1
g 2 ( x , u )  c2

Step 3: Form the Lagrangian by subtracting from the objective

each constraint multiplied by its Lagrange multiplier.
Min: f ( x, u )
s.t.: g1 ( x, u )  c1  0 1
g 2 ( x, u )  c2  0 2

L( x, u, 1 , 2 )  f ( x, u )  1 ( g1 ( x, u )  c1 )  2 ( g 2 ( x, u )  c2 )

Primbs/Investment Science 9
 Solving a General Optimization
Min: f ( x, u )
s.t.: g1 ( x, u )  c1
g 2 ( x , u )  c2

Step 4: Compute the partial derivatives of the Lagrangian

with respect to all its variables and set equal to zero.
L( x, u, 1 , 2 )  f ( x, u )  1 ( g1 ( x, u )  c1 )  2 ( g 2 ( x, u )  c2 )

L f g g L
  1 1  2 2  0   g1 ( x, u )  c1  0
x x x x 1
L f g g L
  1 1  2 2  0   g 2 ( x, u )  c2  0
u u u u 2

Primbs/Investment Science 10
 Solving a General Optimization
Min: f ( x, u )
s.t.: g1 ( x, u )  c1
g 2 ( x , u )  c2

Step 5: Solve these equations (for x, u, ) to find the optimal

solution.

L f g g L
  1 1  2 2  0   g1 ( x, u )  c1  0
x x x x 1
L f g g L
  1 1  2 2  0   g 2 ( x, u )  c2  0
u u u u 2

Primbs/Investment Science 11
 Markowitz Optimization
1 n
Minimize: 
2 i , j 1
wi w j ij = 1/2 (Variance)
n
Subject to: wr  r
i 1
i i p = Mean Return

w
i 1
i 1 = Weights sum to 1.

Note: (1) We are allowing short selling!

(2) We assume all assets are risky!

Primbs/Investment Science 12
 Markowitz Optimization
Step 1
1 n
Minimize: 
2 i , j 1
wi w j ij
n
Subject to: wr r
i 1
i i p 0

 w 1  0
i 1
i

Primbs/Investment Science 13
 Markowitz Optimization
Step 2
1 n
Minimize: 
2 i , j 1
wi w j ij
n
Subject to: wr r
i 1
i i p 0 

 w 1  0
i 1
i


Primbs/Investment Science 14
 Markowitz Optimization
Step 3
1 n
Minimize: 
2 i , j 1
wi w j ij
n
Subject to: wr r
i 1
i i p 0 

 w 1  0
i 1
i


Form the Lagrangian:

1 n  n   n 
L( w,  ,  )   wi w j ij     wi ri  rp      wi  1
2 i , j 1  i 1   i 1 

Primbs/Investment Science 15
 Markowitz Optimization
Step 4
Form the Lagrangian:
1 n  n   n 
L( w,  ,  )   wi w j ij     wi ri  rp      wi  1
2 i , j 1  i 1   i 1 

Differentiate with respect to wi ,  , 

n
wi 
j 1
ij w j  ri    0 for i=1,...,n
n

 wr  r
i 1
i i p
These equations
characterize efficient
n
funds.
 w
i 1
i 1

Primbs/Investment Science 16
 Markowitz Optimization
Step 5

Solve to obtain optimal weights. (n+2 equations, n+2 unknowns)

Differentiate with respect to wi ,  , 

n
wi 
j 1
ij w j  ri    0 for i=1,...,n
n

 wr  r
i 1
i i p
These equations
characterize efficient
n
funds.
 w
i 1
i 1

Primbs/Investment Science 17
 The Markowitz Model

Solving the Optimization

The Two Fund Theorem

Markowitz Portfolio Theory
Inclusion of a Risk Free Asset

The One Fund Theorem

Markowitz’s Message

Primbs/Investment Science 18
 The Two-Fund Theorem
Let’s make the following assumptions:

(1) Short selling is allowed.

(2) All assets are risky.
(3) All investors have the same estimates of
means, variances, and covariances.

Theorem: Investors seeking minimum variance portfolios

need only invest in combinations of two minimum
variance funds.

Primbs/Investment Science 19
 Importance of the Two Fund
Theorem

Only two efficient funds need to exist, and everyone can

invest in them!

Another Efficient
x
r Fund
x Fund 2
x It is just a portfolio of Fund 1 and Fund 2.
Fund 1


Primbs/Investment Science 20
 The Two-Fund Theorem
Theorem: Investors seeking efficient portfolios need only
invest in combinations of two efficient funds.
Proof:
Let rP1 , 1 ,  1 w1  ( w11 , w12 , , w1n )
and rP2 , 2 ,  2
w 2
 ( w1
2
, w2
2
,  , w 2
n)

be efficient funds. Hence they satisfy the equations for an

efficient fund on a previous slide.
n n

1:   ij w   ri    0
1
j
1 1
2: 
j 1
ij w2j  2 ri   2  0
j 1
n n n

w
n

w r  r  wi1  1 w r  r 1
2 2 2
1 1
i i P i
i i P i 1
i 1 i 1 i 1

Primbs/Investment Science 21
 The Two-Fund Theorem
1: 2: 3:
n


n n

  w 3
 3
r   3
0
 ij       2 2 2
 w   ri 
1
j   0 1
ij w r1
i 0 ij j
j i
j 1 j 1
j 1 n n

 w
n n
n
1

n 3

w r  r
3 3
w r r
w r  r w 1 
2 2 2
1 1 1
i i i P
w 1i i i P i
i i P i 1 i 1
i 1 i 1 i 1
i 1

Now consider an efficient fund with mean return rP3

Solve rP1  (1   )rP2  rP3 for and set:

w3  w1  (1   ) w 2 Are these the optimal weights

3  1  (1   )2 and Lagrange multipliers
3
corresponding to rP ?
 3   1  (1   )  2
Yes!...We need to show

Primbs/Investment Science 22
 The Two-Fund Theorem
1: 2: 3:
n


n n

  w 3
 3
r   3
0
 ij       2 2 2
 w   ri 
1
j   0 1
ij w r
1
i 0 ij j
j i
j 1 j 1
j 1 n n

 w
n n
n
1

n 3

w r  r
3 3
w r r
w r  r w 1 
2 2 2
1 1 1
i i i P
w 1i i i P i
i i P i 1 i 1
i 1 i 1 i 1
i 1

w3  w1  (1   ) w2
n
3  1  (1   )2
 ij j i
 w
j 1
3
 3
r   3
 3   1  (1   )  2
n
   ij (w1j  (1   ) w2j )  (1  (1   )2 )ri  ( 1  (1   )  2 )
j 1
 n   n 
     ij w j   ri     (1   )   ij w j   ri     0
1 1 1 2 2 2

 j 1   j 1 

Primbs/Investment Science 23
 The Two-Fund Theorem
1: 2: 3:
n


n n

  w 3
 3
r   3
0
 ij       2 2 2
 w   ri
1
j   0 ij w
1
ri
1
0 ij j j i
j 1 j 1
j 1 n n

 w
n n
n
1

n 3

w r  r
3 3
w r r
w r  r w 1 
2 2 2
1 1 1
i i i P
w 1i i i P i
i i P i 1 i 1
i 1 i 1 i 1
i 1

w3  w1  (1   ) w2
n
3  1  (1   )2
 i ri
w 3

i 1
 3   1  (1   )  2
n
  (wi1  (1   ) wi2 )ri
i 1

 n 1   n 2 
    wi ri   (1   )  wi ri   rP  (1   )rP  rP
1 2 3

 i 1   i 1 
Primbs/Investment Science 24
 The Two-Fund Theorem
1: 2: 3:
n


n n

  w 3
 3
r   3
0
 ij       2 2 2
 w   ri
1
j   0 ij
1
w ri
1
0 ij j j i
j 1 j 1
j 1 n n

 w
n n
n
1

n 3

w r  r
3 3
w r r
w r  r w 1 
2 2 2
1 1 1
i i i P
w 1i i i P i
i i P i 1 i 1
i 1 i 1 i 1
i 1

w3  w1  (1   ) w2
n 3  1  (1   )2
 i
w 3

i 1
 3   1  (1   )  2
n
  wi1  (1   ) wi2
i 1
 n 1  n 2
    wi   (1   )  wi     (1   )  1
 i 1   i 1 
( w3 , 3 ,  3 ) is efficient!
Primbs/Investment Science 25
 The Two-Fund Theorem

Asset Return Std. Dev. Covariance Matrix 1 2 3 Lambda (sum

Mu to
(mean
one) return)
Min Variance
1 10% 0.1581139 1 0.025 0 0 1 10% 0
2 25% 0.244949 2 0 0.06 0 1 25% 0
3 20% 0.1581139 3 0 0 0.025 1 20% 0
Constraint (weights sum
1 to 1) 1 1 0 0 1
Constraint (mean return)
10% 25% 20% 0 0 0.25

Solutions
Min Variance Mean Return = 25%
w1 0.3
0.413793103 -0.244898
w2 0.172413793 0.5102041
w3 0.413793103 0.7346939
0.25
Lagrange -0.01034483 0.0306122
Mu -0.244898 Efficient
0.2
Mean Return0.167241379 0.25 Frontier
Mean Return

Variance 0.010344828 0.0306122 1 and 2

Std. Dev. 0.101709526 0.1749636
0.15 1 and 3

Efficient Frontier 2 and

1 and 2 3
Weight on0.1
Min
Weight
Variance
on mean
w1return w2 w3 Mean ReturnVariance Std. Dev. Mean Return Variance
-1 2 -0.903589 0.84799437 1.055595 0.33276 0.091414 0.302348 Asset
0.4 1 0.265
-0.9
0.05 1.9 -0.83772 0.81421534 1.023505 0.32448 0.08351 0.288981 0.385 0.23685
-0.8 1.8 -0.771851 0.78043631 0.991414 0.31621 0.076011 0.275701 Asset
0.37 2 0.2104
-0.7 1.7 -0.705982 0.74665728 0.959324 0.30793 0.068918 0.262522 0.355 0.18565
-0.6 0 1.6 -0.640113 0.71287825 0.927234 0.29966 0.062229 0.249458 Asset
0.34 3 0.1626
-0.5 1.5 -0.574243 0.67909923 0.895144 0.29138 0.055947 0.23653 0.325 0.14125
0 0.05 0.1 0.15 0.2 0.25 0.3
-0.4 1.4 -0.508374 0.6453202 0.863054 0.2831 0.050069 0.223761 0.31 0.1216
-0.3 1.3 -0.442505 Standard
0.61154117Deviation
0.830964 0.27483 0.044597 0.211179 0.295 0.10365
-0.2 1.2 -0.376636 0.57776214 0.798874 0.26655 0.03953 0.198821 0.28 0.0874
-0.1 1.1 -0.310767 0.54398311 0.766784 0.25828 0.034868 0.186731 0.265 0.07285

Primbs/Investment Science 26
 The Markowitz Model

Solving the Optimization

The Two Fund Theorem

Markowitz Portfolio Theory
Inclusion of a Risk Free Asset

The One Fund Theorem

Markowitz’s Message

Primbs/Investment Science 27
 Inclusion of a Risk-Free Asset
We have assumed that all the assets are risky.

Now assume there exists a risk-free asset with

return rf.

r
rf


Primbs/Investment Science 28
 Inclusion of a Risk-Free Asset
What happens when we combine the risk free asset
with a risky portfolio

risk free: (rf ,0)

risky asset: ( r ,  2
)

Let’s form a portfolio consisting of  of the risk free

asset and  of the risk asset:
mean: rf  (1   )r
variance: (1   ) 2  2
standard deviation: (1   )

Primbs/Investment Science 29
 Inclusion of a Risk-Free Asset
For this portfolio we have
(mean, standard deviation)= (rf  (1   )r , (1   ) )

As we vary , this maps out a straight line

r
x
(r ,  2 )
rf

Primbs/Investment Science  30
 Expanded Feasible Region

r F
x Tangent to the feasible region of risky funds.
rf

Primbs/Investment Science 31
 The Markowitz Model

Solving the Optimization

The Two Fund Theorem

Markowitz Portfolio Theory
Inclusion of a Risk Free Asset

The One Fund Theorem

Markowitz’s Message

Primbs/Investment Science 32
 The One-Fund Theorem
Let’s make the following assumptions:

(1) Short selling is allowed.

(2) There is a risk free asset.
(3) All investors have the same estimates of
means, variances, and covariances.

Theorem: 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.

Primbs/Investment Science 33
 The One-Fund Theorem
Theorem: 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.

r F
x Tangent to the feasible region of risky portfolios.
rf


Primbs/Investment Science 34
 Computation of the One-Fund
The one-fund is the fund of risky assets that results in the maximum slope
with the risk-free rate.

Maximize this slope

rFund  rf
slope 
 Fund

r F
x Tangent to the feasible region of risky funds.
rf


Primbs/Investment Science 35
 Computation of the One-Fund
The one-fund is the fund of risky assets that results in the maximum slope
with the risk-free rate.
n
Maximize this slope
rFund  rf
 wi (ri  rf )
slope   max i 1
1/ 2
 Fund wi
 n 
  wi w j ij 
 
 i , j 1 
Take derivative wrt. wk for k=1...n and set equal to zero:
1/ 2 1/ 2
 n   n
 n   n 
  wi w j ij  (rk  rf )    wi (ri  rf )   wi w j ij    w j kj 
     
 i , j 1   i 1  i , j 1   j 1  0
 n 
  wi w j ij 
 
 i , j 1 
Primbs/Investment Science 36
 Computation of the One-Fund
1/ 2 1 / 2
 n   n  n   n 
  wi w j ij  (rk  rf )    wi (ri  rf )   wi w j ij    w j kj 
     
 i , j 1   i 1  i , j 1   j 1  0
 n 
  wi w j ij 
 
 i , j 1 
 n  n 
  wi (ri  rf )   w j kj 

 i 1  j 1  0
(rk  rf ) 
 n 
  wi w j ij 
 
 i , j 1 
 n 
  wi (ri  rf )   n 
Let
 i 1   (rk  rf )     w j kj   0

 n   j 1 
  wi w j ij 
 
 i , j 1  for k=1,...,n
Primbs/Investment Science 37
 Computation of the One-Fund
 n 
(rk  rf )     w j kj   0
 for k=1,...,n
 j 1 
n

 w 
j 1
j kj  (rk  rf ) for k=1,...,n

Let: w j  v j v 
j 1
j kj  (rk  rf ) for k=1,...,n

vj
Solve for j and normalize wj  n

v
i 1
i

These are the weights of the One-Fund!

Primbs/Investment Science 38
 The Markowitz Model

Solving the Optimization

The Two Fund Theorem

Markowitz Portfolio Theory
Inclusion of a Risk Free Asset

The One Fund Theorem

Markowitz’s Message

Primbs/Investment Science 39
 The Message of Markowitz
The Two-Fund and One-Fund theorems are important
consequences of Markowitz.

Beyond these, Markowitz says that we can form optimal

portfolios which take advantage of the correlations
between assets.

A serious difficulty with this theory is that among n

assets, there are n(n-1)/2 covariances. This is a lot! Just
consider trying to compute this for the market, which has
thousands of assets!
Assets are valuable as members of portfolios!!!

Primbs/Investment Science 40
 Problems

Primbs/Investment Science 41
The two-fund theorem:

Consider a market with 3 risky assets.

Two efficient funds are

A – (0.25,0.25,0.50) with expected return 10%
B – (0.10,0.70,0.20) with expected return 20%

What are the weights on an efficient portfolio with mean

return
a) 15%
b) 30%

Primbs/Investment Science 42
The one-fund theorem:

Consider a market with risky assets and a risk free asset.

The risk free return is 5%

The one-fund has mean 10% and standard deviation 15%

What is the standard deviation on an efficient portfolio

with mean return

a) 30%

b) 10%

Primbs/Investment Science 43
 Optimizations:
r1  0.1  11  0.3
 12  0.01
r2  0.2  22  0.4
There are two assets in the market with means and covariances given
above. Write the optimization problem for a portfolio with minimum
variance subject to a mean return constraint of 18%. Write the necessary
conditions for the solution. Solve them.

If the risk free rate is 5%, compute the one-fund.

Primbs/Investment Science 44
 Appendix 1:
Markowitz Theory using
Linear Algebra

Primbs/Investment Science 45
 The Markowitz Model

Constrained Optimization

Derivatives with Lin. Alg.

Markowitz Portfolio Theory Quadratic Opt. w/ Lin. Alg.

(The Structure of Optimal
The Two Fund Theorem
Portfolios)
Inclusion of a Risk Free Asset

The One Fund Theorem

Markowitz’s Message

Primbs/Investment Science 46
 Picture of Markowitz
For a given mean return, you would like to
minimize your risk or the variance.

r
x


Minimum variance point for a given mean return

Primbs/Investment Science 47
 The Markowitz Model
Markowitz formulated the problem of being on the minimum
variance set as an optimization.
Assume there are n risky assets with  r1 
 r1 
r  r 
r   2 r   2
   
Returns: r1 , r2 , , rn    
rn  rn 
Mean returns: r1 , r2 ,  , rn
 11  12   1n 
Covariances:  ij for i,j=1,...,n   22   2 n 
  21

    
 
 n1  n 2   nn 
Primbs/Investment Science 48
 Markowitz Optimization
1 n
Minimize: 
2 i , j 1
wi w j ij = 1/2 (Variance)
n
Subject to: wr  r
i 1
i i p = Mean Return

w
i 1
i 1 = Weights sum to 1.

Note: (1) We are allowing short selling!

(2) We assume all assets are risky!

Primbs/Investment Science 49
 Markowitz Optimization
Linear Algebra
Minimize: 1 T
w w = 1/2 (Variance)
2

Subject to: wT r  rp = Mean Return

wT 1  1 = Weights sum to 1.

Note: (1) We are allowing short selling!

(2) We assume all assets are risky!

Primbs/Investment Science 50
 The Markowitz Model

Constrained Optimization

Derivatives with Lin. Alg.

Markowitz Portfolio Theory Quadratic Opt. w/ Lin. Alg.

(The Structure of Optimal
The Two Fund Theorem
Portfolios)
Inclusion of a Risk Free Asset

The One Fund Theorem

Markowitz’s Message

Primbs/Investment Science 51
 Constrained Optimization
Min: f ( x, u )
s.t.: g ( x, u )  c

f f
At a minimum, variations in df  dx  du  0
f(x,u) are equal to zero. x u

However, only variations that g g

dg  dx  du  0
preserve the constraint are allowed. x u

1
 g  g
Solve for du in terms of dx. du    dx
 u  x
Primbs/Investment Science 52
 Constrained Optimization
f f
At a minimum, variations in df  dx  du  0
f(x,u) are equal to zero. x u
1
However, only variations that  g  g
preserve the constraint are allowed.
du    dx
 u  x
 f f  g  1 g 
Substitute into df df      dx
 x u  u  x 
 
1
f f  g  g
We are at a minimum if:    0
(and constraint is satisfied) x u  u  x

Primbs/Investment Science 53
 Constrained Optimization
1
f f  g  g
We are at a minimum if:    0
(and constraint is satisfied) x u  u  x
1
Let’s rewrite this f  g  f g
condition    and  0
u  u  x x

These equations along f g f g

with the constraint are  0 and  0
the optimality conditions u u x x

A convenient way to get to these equations is through the Lagrangian...

Primbs/Investment Science 54
 The Lagrangian
Min: f ( x, u )
s.t.: g ( x, u )  c

Define the Lagrangian L ( x, u ,  )  f ( x, u )   ( g ( x, u )  c )

L f g
Setting the partial of   0
the Lagrangian equal x x x
to zero gives the L f g Optimality
correct optimality   0 Conditions
conditions u u u
L
 g ( x, u )  c  0

Primbs/Investment Science 55
 The Markowitz Model

Constrained Optimization

Derivatives with Lin. Alg.

Markowitz Portfolio Theory Quadratic Opt. w/ Lin. Alg.

(The Structure of Optimal
The Two Fund Theorem
Portfolios)
Inclusion of a Risk Free Asset

The One Fund Theorem

Markowitz’s Message

Primbs/Investment Science 56
 Derivatives using Linear Algebra

x
( xT A)  AT

x ( Ax)  A

x (a T x)  a T


x ( xT Ax)  x T A  x T AT

x
( x T Ax)  2 x T A if A is symmetric (i.e. A=AT).
Note: I use the convention that derivatives (i.e. gradients) are row vectors.
This means the chain rule works from left to right:
Let: y  Bx then
y

x ( y Ax)  y A  x A
T T
 ( Bx)T A  xT AT B
T T
 xT B T A  xT AT B
x
Primbs/Investment Science 57
 The Markowitz Model

Constrained Optimization

Derivatives with Lin. Alg.

Markowitz Portfolio Theory Quadratic Opt. w/ Lin. Alg.

(The Structure of Optimal
The Two Fund Theorem
Portfolios)
Inclusion of a Risk Free Asset

The One Fund Theorem

Markowitz’s Message

Primbs/Investment Science 58
 A Constrained Quadratic
Optimization
Min: 1
2 xT Ax Assume A is symmetric

s.t.: bT x  c

Define the Lagrangian: L( x,  )  12 x T Ax   (bT x  c)

Take partials:
L Ax  b  0
( x ,  )  x T A  b T  0 transpose
x
L bT x  c
( x,  )  x T b  c  0

Solve to find optimum.

Primbs/Investment Science 59
 The Markowitz Model

Constrained Optimization

Derivatives with Lin. Alg.

Markowitz Portfolio Theory Quadratic Opt. w/ Lin. Alg.

(The Structure of Optimal
The Two Fund Theorem
Portfolios)
Inclusion of a Risk Free Asset

The One Fund Theorem

Markowitz’s Message

Primbs/Investment Science 60
 Markowitz Optimization
Linear Algebra

Minimize: 1 T
w w = 1/2 (Variance)
2

Subject to: wT r  rp = Mean Return

wT 1  1 = Weights sum to 1.

Primbs/Investment Science 61
 Markowitz Optimization
1 T
Minimize: w w Lagrange Mult.
2
Subject to: wT r  rp  0 
wT 1  1  0 
1 T
Lagrangian: L( w,  ,  )  w w   ( wT r  rp )   ( wT 1  1)
2
T
 L 
   w  r  1  0
 w 
Optimality  L 
T

   r w  rp  0
T
Conditions   
T
 L 
   1T w  1  0
  
Primbs/Investment Science 62
 The Structure of Optimality
w  r  1  0
Optimality
Conditions r T w  rp  0
1 w 1  0
T

 r 1  w  0 
Rewrite as: r T    
0 0     rP  

1T 0 0     1 

Primbs/Investment Science 63
 Solving the Markowitz Problem
 r 1  w  0 
Rewrite as: r T    
0 0     rP  

1T 0 0     1 

Given rp
1
Solve for optimal w  w   r 1 0
(and )by:     r T 0 0 r 
    P
   1T 0 0  1 
Primbs/Investment Science 64
 The Two Fund Theorem
Let (w1,1,1) and (w2,2,2) be Markowitz solutions
1
r r 2
corresponding to P and P , respectively. Then the solution to
3
the Markowitz problem for rP is:
(w1,1,1)= (w1,1,1)+(1-)(w1,1,1)
where  solves: r 3  r 1  (1   ) r 2
P P P
Proof: 1
1
 w3    r 1 0  r 1  0  0 
  1 
    r T 0 0 r 3   r T 0 0  2
 rP   (1   ) rP  
 3   P  1 
  3  1T 0 0  1  1T 0 0     1
 
1 1
 r 1 0  r 1 0  w1   w2 
  r T 0 0 r 1   (1   ) r T
 P  0 0 r 2        (1   )   
P  1  2
1T 0 0  1  1T 0 0  1   1    2 
Primbs/Investment Science 65
 Importance of Two Fund Theorem
Theorem: Investors seeking efficient portfolios need only
invest in combinations of two efficient funds.
Another Efficient
x
r Fund
x Fund 2
x It is just a portfolio of Fund 1 and Fund 2.
Fund 1


Only two efficient funds need to exist, and everyone can
invest in them!
Primbs/Investment Science 66
 The Markowitz Model

Constrained Optimization

Derivatives with Lin. Alg.

Markowitz Portfolio Theory Quadratic Opt. w/ Lin. Alg.

(The Structure of Optimal
The Two Fund Theorem
Portfolios)
Inclusion of a Risk Free Asset

The One Fund Theorem

Markowitz’s Message

Primbs/Investment Science 67
 Inclusion of a Risk-Free Asset
We have assumed that all the assets are risky.

Now assume there exists a risk-free asset with

return rf.

r
rf


Primbs/Investment Science 68
 Inclusion of a Risk-Free Asset
What happens when we combine the risk free asset
with a risky portfolio

risk free: (rf ,0)

risky asset: ( r ,  2
)

Let’s form a portfolio consisting of  of the risk-free

asset and  of the risky asset:
mean: rf  (1   )r
variance: (1   ) 2  2
standard deviation (1   )

Primbs/Investment Science 69
 Inclusion of a Risk-Free Asset
For this portfolio we have
(mean, standard deviation)= (rf  (1   )r , (1   ) )

As we vary , this maps out a straight line

r
x
(r ,  2 )
rf

Primbs/Investment Science  70
 Expanded Feasible Region

r F
x Tangent to the feasible region of risky funds.
rf

Primbs/Investment Science 71
 The Markowitz Model

Constrained Optimization

Derivatives with Lin. Alg.

Markowitz Portfolio Theory Quadratic Opt. w/ Lin. Alg.

(The Structure of Optimal
The Two Fund Theorem
Portfolios)
Inclusion of a Risk Free Asset

The One Fund Theorem

Markowitz’s Message

Primbs/Investment Science 72
 The One-Fund Theorem
Theorem: 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.

r F
x Tangent to the feasible region of risky portfolios.
rf


Primbs/Investment Science 73
 Derivation of the One-Fund
Theorem
Let w be the weights on risky assets, and w0 the weight on the risk free asset.

Minimize: 1 T
w w = 1/2 (Variance)
2
Subject to: w0 rf  wT r  rp = Mean Return

w0  wT 1  1 = Weights sum to 1.

w0  1  wT 1

Primbs/Investment Science 74
 Derivation of the One-Fund
Theorem
Minimize: 1 T
w w
2 Lagrange Mult.
Subject to: rf  w (r  rf 1)  rp 
T

1 T
Lagrangian: L( w,  )  w w   (rf  wT (r  rf 1)  rp )
2
T
 L 
Optimality    w   (r  rf 1)  0
 w 
Conditions T
 L 
   rf  (r  rf 1) w  rp  0
T

  

Primbs/Investment Science 75
 Derivation of the One-Fund
Theorem
T
 L 
Optimality    w   (r  rf 1)  0
 w 
Conditions T
 L 
   rf  (r  rf 1) w  rp  0
T

  
w   1 (rf 1  r )
But these weights won’t sum to 1. So normalize to sum to 1.
1 1

w  T (  1
( r f 1  r ))  (  1
(rf 1  r ))
1 ( (rf 1  r ))
1 1
1 ( (rf 1  r ))
T

This does not depend on rP . This is the one-fund of risky assets.

Primbs/Investment Science 76
 Appendix 2:
When the 1-Fund and 2-Fund
Theorems Hold

Primbs/Investment Science 77
 When the two fund theorem holds
The Two Fund Theorem

Short selling allowed Yes!

No short selling allowed No!

Primbs/Investment Science 78
 No Short Selling

Minimize: 1 T
w w = 1/2 (Variance)
2

Subject to: wT r  rp = Mean Return

wT 1  1 = Weights sum to 1

wi  0 = No short selling

Primbs/Investment Science 79
 No Short Selling
1 T
L( w,  ,  )  w w   ( wT r  rp )   ( wT 1  1)    i wi
2 i

With the inequality constraint, we must use the Kuhn-Tucker conditions.

T
 L 
   w  r  1    i  0
 w  i
r w  rp
T

1T w  1
wi  0
i  0 This condition is not linear in
 w 0 ( i , wi )
i i

Primbs/Investment Science 80
 When the One-Fund Theorem Holds
Risk Free Asset:
Short Sell No Short Sell
Risky Assets
Short Sell Yes No

No Short Sell Yes No

Primbs/Investment Science 81
 Pictures: Short Selling of Risk Free
Allowed
Efficient Frontier

r F
x Tangent to the feasible region of risky portfolios.
(Doesn’t matter if this is under short selling or no
rf short selling of risky assets!)

Primbs/Investment Science 82
 When the One-Fund Theorem Holds
Risk Free Asset:
Short Sell No Short Sell
Risky Assets
Short Sell Yes No

No Short Sell Yes No

Primbs/Investment Science 83
 Pictures: Short Selling of Risk Free
Not Allowed
Efficient Frontier

r F
x Tangent to the feasible region of risky portfolios.
(Doesn’t matter if this is under short selling or no
rf short selling of risky assets!)

Primbs/Investment Science 84
 General Optimization

1 T
Minimize: w w = 1/2 (Variance)
2
r
Subject to: f  wT
(r  rf 1)  rp = Mean Return
wi  0 = No Short Risky

1  1T w  0 = No Short Risk-free

Primbs/Investment Science 85
 First Order Optimality Conditions

L
 w   (r  rf 1)    1  0
w
rf  wT (r  rf 1)  rp  0

wi  0 i  0 wi i  0

11 w  0
T
0  (1  1T w)  0

Primbs/Investment Science 86
 Allow Shorting of the Risk-Free

L
 w   (r  rf 1)    1  0
w
rf  wT (r  rf 1)  rp  0

wi  0 i  0 wi i  0

11 w  0
T
0  (1  1T w)  0

Remove the restriction on no shorting of the risk free asset.

Keep restriction on shorting of risky assets.
Primbs/Investment Science 87
 One-Fund Theorem Holds!

L
 w   (r  rf 1)    0
w
rf  wT (r  rf 1)  rp  0

wi  0 i  0 wi i  0

Note that if (w,rP-rf) is a solution, then so is a(w,rP-rf) for any a  0.

Hence, a one-fund theorem holds in this case!

(Think about the picture and this should be clear!)

Primbs/Investment Science 88
 Allow Shorting of Risky

L
 w   (r  rf 1)    1  0
w
rf  wT (r  rf 1)  rp  0

wi  0 i  0 wi i  0

11 w  0
T
0  (1  1T w)  0

Remove the restriction on no shorting of the risky assets.

Keep the restriction on shorting of risk free asset.
Primbs/Investment Science 89
 One-Fund Theorem Doesn’t Hold!

L
 w   (r  rf 1)  1  0
w
rf  wT (r  rf 1)  rp  0

1  1T w  0 0  (1  1T w)  0

A one-fund theorem cannot hold because of the (1-1Tw) term!

Again, this should be clear from the picture.
Not allowing shorting of risky assets and the
risk free asset suffers from the same problem.

Primbs/Investment Science 90
 Appendix 3:
Lagrange Multipliers and the
Objective Function

Primbs/Investment Science 91
 Constrained Optimization
Why is  the derivative of the objective with respect to the constraint?

Min: f ( x, u ) f ( x, u )
s.t.: g ( x, u )  c g ( x, u )  c  0
f f
df  dx  du  0
Optimality Condition x u
g g
dg  dx  du  dc  0
x u
Solve for du
1
g g  g   g 
du  dc  dx du     dc  dx 
u x  u   x 

Primbs/Investment Science 92
 Constrained Optimization
1
f f  g   g 
So: df  dx     dc  dx 
x u  u   x 
 f f  g  1 g  f  g 
1

df     dx    dc
 x u  u  x  u  u 
 
1
f g f  g 
But:  0   
u u u  u 
 f g 
df     dx  dc
 x x 
df

dc
Primbs/Investment Science 93