FM1 Notes, Spring 2019

Two Fund Separation, CAPM and APT

Two Fund Separation

We know that all portfolios on the portfolio frontier can be generated by any two
distinct frontier portfolios. Thus, if individuals prefer frontier portfolios, they
can simply hold a linear combination of two frontier portfolios.

In this case, given any feasible portfolio, for an individual investor, there exists a
portfolio of two mutual funds (also called separating portfolios) such that the
investor prefers at least as much as the original portfolio. This is called two
(mutual) fund separation.

When two fund separation obtains, any two distinct frontier portfolios can serve

1
as separating portfolios. In particular, we can pick pmvp and zc(p) to be the
separating portfolios.

One fund separation: A vector of asset returns r exhibits one fund separation if
there exists a feasible portfolio  (also called the separating portfolio) such that
every risk-averse individual prefers  to any other feasible portfolio.

The Market Portfolio

Let W0i  0 be individual i’s initial wealth (a scalar). Let wij be the proportion
of the initial wealth invested in the jth security by individual i.

I
The total wealth of the economy Wmo  W0i , where I is the number of
i 1

individuals in the economy. In equilibrium, the total wealth Wmo is equal to the

2
total value of securities.

Let wmj – proportion of total wealth contributed by the total value of the jth
security. This is called the portfolio weights of the market portfolio.

For markets to clear, we must have

I

 ij 0  wmjWmo
w
i 1
W i

Dividing both sides by Wmo

I
 W0i 
   wij  wmj
i 1  Wmo 

i.e., the market portfolio weights are a convex combination of the portfolio
weights for individuals.

3
Claim: When two fund separation obtains, the market portfolio is a frontier
portfolio.

Proof:

When two fund separation holds, the separating portfolios must be frontier
portfolios and any two distinct frontier portfolios can be separating portfolios.

In this case, each individual will hold a linear combination of the two separating
portfolios.

Any linear combination of frontier portfolios is also a frontier portfolio  each

individual holds a frontier portfolio.

In equilibrium, the market portfolio is a convex combination of individuals’

4
portfolios. Thus, the market portfolio is a linear combination of frontier
portfolios and is itself on the frontier.

QED.

Recall from the last class that if p is a frontier portfolio, with p  mvp and let q
be any feasible portfolio, we have the following linear relationship

E[rq ]  (1   qp ) E[rzc ( p ) ]   qp E[rp ]

When 2-fund separation obtains, the market portfolio is a frontier portfolio. If m

is not the mvp, then we can have

E[rq ]  (1   qm ) E[rzc ( m ) ]   qm E[rm ]

5
 N Cov(rq , rm )
where the market portfolio’s return is rm   wmj rj and  qm 
j 1 Var (rm )

Since any risky asset j is itself a feasible portfolio, we have

E[rj ]  (1   jm ) E (rzc ( m ) )   jm E (rm ) for j = 1, 2, …, N.

The security market line can be expressed as:

E[rj ]  E[rzc ( m ) ]   jm [ E (rm )  E (rzc ( m ) )]

6
 E (~
rj )

E (~
rm )  E (~
rzc (m ) )

E (~
rzc (m ) )

0 βjm

7
Asset pricing with a risk-free asset
A
Assume that rf   E (rmvp ) , and let e be the tangency portfolio. We know that
C
for any feasible portfolio q, rq  (1   qe )rf   qe re  qe , with
cov(re , qe )  E (qe ) =0.

A/C e

rf

8
CAPM
Assume that two fund separation holds.
A
When rf  and risky assets are in strictly positive supply, the tangency
C
portfolio e must be the market portfolio of risky assets in equilibrium.

We have E[rq ]  rf   qm [ E (rm )  rf ] for any portfolio q in market equilibrium.

This is the traditional CAPM.

A
When rf  , if there is an equilibrium, then we claim that the riskless asset is
C
in strictly positive supply and the risky assets are in zero net supply.

To see this, when 2-fund separation holds, individuals hold frontier portfolios.

9
 A
When rf  , an individual puts all his wealth into the riskless asset and holds a
C
self-financing portfolio of risky assets. That is, an individual invests everything
in the riskless asset and holds an arbitrage portfolio of risky assets I T W p  0,
where W p is the portfolio weights of all risky assets.

For markets to clear, it is necessary that the riskless asset be in strictly positive
supply and the risky assets be in zero net supply.

10
 E[~
rp ] CML

A/C

rf

0  (~
rp )

Remark 1: when there exists a riskless asset, an investor will never choose to
hold a portfolio with return < rf . The reason is we can always invest all money
in rf. So the conclusion holds even when borrowing is not allowed.

Remark 2: When riskless borrowing is strictly higher than riskless lending rate,
rb  rL , we have E (rq )  E ( rzc ( m ) )   qm [ E ( rm )  E (rzc ( m ) )] for any feasible
portfolio and E (rm )  E (rzc ( m ) )  0, and rb  E (rzc ( m ) )  rL .

11
 Alternative derivation of the CAPM
Relating market risk premium to investors’ optimal portfolio position

Assume that: there is a risk free asset, with return rf , and N risky assets, with
returns {ri }iN1 which are multivariate normally distributed (required to apply the
Stein’s Lemma). Let W be the optimally invested random time-1 wealth for
i

person i.

From the FOC E[ui' (Wi )( rj  rf )]  0, i, j

N N
where Wi  W [1   wij rj  (1   wij ) rf ]
0
i

j 1 j 1

N
 W [(1  rf )   wij (rj  rf )] `
0
i

j 1

12
Using the definition of covariance, the FOC can be written as

E[ui' (Wi )]E[ rj  rf ]   cov[ui' (Wi ), rj ]

Note: Wi and rj are bi-variate normal.

[Stein’s Lemma:
If x and y are bi-variate normal, then Cov( g ( x), y )  E[ g '( x)]Cov( x, y ) ,
provided that g is differentiable and satisfies some regularity conditions.]

Assume that ui is twice differentiable and apply Stein’s Lemma

E[u '(Wi )]E[rj  rf ]   E[ui'' (Wi )]Cov(Wi , rj )

13
Define the ith investor’s global absolute risk aversion
 E[ui'' (Wi )]
i 
E[u ' (W )] i i

Dividing both sides by E[ui'' (Wi )] , and summing across i (assume I individuals)

I I I
( ) E (rj  rf )   Cov(Wi , rj )  Cov(Wi , rj )
i
1

i 1 i 1 i 1
I
Now W  W
i 1
i mo (1  rm ) is total wealth in the world in period 1.

I
Therefore substituting to the above yields E[ rj  rf ]  Wmo (i1 ) 1 Cov( rm , rj ) ,
i 1
I
where I ( i1 ) 1 is the harmonic mean of individuals’ global absolute risk
i 1
I
aversion. Wmo ( i1 ) 1 can be interpreted as the aggregate relative risk
i 1

14
aversion of the economy.

We therefore have
I
E[ rm  rf ]  Wmo (i1 )  Var (rm )
i 1

i.e., the risk premium on the market portfolio is proportional to the aggregate
relative risk aversion of the economy.

E[rm  rf ]  0 when u(.) is increasing and strictly concave.

Cov(rm , rj )
Direct substitution yields E[rj  rf ]  E (rm  rf ) . We therefore obtain
Var (rm )
the CAPM.

15
 The Arbitrage Pricing Theory (APT)

Potential problems with the CAPM

1. Existence of transactions cost and other trading frictions

2. Existence of non-tradable assets, e.g., human capital
3. Market portfolio hard to identify

Motivations for multifactor APT model

Assume that asset return is driven by multiple risk factors and by an

idiosyncratic component.

APT is a relative pricing model.

16
No assumption about investor preferences is made. It uses arbitrage pricing to
restrict an asset’s risk premium.

Assume the following return generating process:

k
ri  ai   biz f z  i (1)
z 1

where ai is the expected return of asset i,

E ( )  E ( f )  E ( f )  0
i z i z

E (ij )  0, i  j
E ( f x f y )  0, x  y
E ( f 2 )  1
z

E ( )  si2  S 2  
i
2

i rj )   ij
2
E (r

17
Definition: Let a portfolio containing n assets be described by the vector of
W1n 
 n
W
investment amounts in each of the n assets, W n   2  .
 ... 
 n
Wn 
Consider a sequence of these portfolios where n is increasing. An asymptotic
arbitrage exists if the following conditions hold:

(A) The portfolio requires zero net investment

n

 i 0
W n

i 1

(B) The portfolio return becomes certain as n gets large

n n
lim Wi nW jn ij  0
n
i 1 j 1

18
(C) The portfolio’s expected return is positive
n

W
i 1
i
n
ai    0 .

Claim (APT): If no asymptotic arbitrage opportunities exist, then the expected

return of asset i is described by the following linear relation

k
ai  0   z biz   i
z 1

where 0 is a constant, z is the risk premium for risk factor f z and

n

 i 1
i 0 (i)

b i 1
iz i  0, z  1,2,..., k (ii)

19
 1 n 2
lim  i  0 (iii)
n n
i 1

Note that if the economy contains a risk-free rate (implying biz  0 ), the
risk-free return will be approximated by 0 .

Proof: For a given number of assets, n>k, consider the following cross-sectional
regression
k
ai  0   z biz   i
z 1

We know that (i) and (ii) can be satisfied by the properties of OLS regression.

To show that (iii) is also satisfied, consider a zero-net-investment arbitrage

portfolio with the following investment amounts:

20
 i
Wi 
n
n  i2
i 1

The total arbitrage portfolio return is given by

21
 rp
n
 Wi ri
i 1

1 n

n
 ( r )i i

n i
2 i 1

i 1

1 n k

n
[ (a   b
i i iz
f   )]
z i

n i2
i 1 z 1

i 1

1 n

n
[ (a   )]
i i i

n i2
i 1

i 1

n
where we have used b 
i 1
iz i  0 in the last equality.

22
We can calculate the mean and variance of the portfolio as follows
1 n
E (rp ) 
n
 ( a )
i i

n i
2 i 1

i 1

since E (i )  0 .

k n
Substituting ai  0   biz z   i , and using the fact that b  iz i  0 and
z 1 i 1
n


i 1
i  0 , we have

23
 E (rp )
1 n k n n
 [0  i   (z  i biz )   i2 ]
n
n i2
i 1 z 1 i 1 i 1

i 1

1 n

n
i
 2

n i2
i 1

i 1

1 n 2
 
n i 1
i

To calculate the variance

rp  E (rp )
1 n

n
 i i
n i2
i 1

i 1

24
Using the fact that E (ij )  0, i  j and E (i 2 )  si2  S 2   , we have

E[ rp  E ( rp )]2

n

 2 2
i i s
 i 1
n
n i2
i 1
n
S 2
 i
2

 i 1
n
n i2
i 1

S2
  0, as n  
n

In other words, the portfolio return becomes certain. This implies that in the
limit, the actual return equals the expected return

25
 1 n 2
lim rp  E (rp ) 
n 

n i 1
i .

If there are no asymptotic arbitrage opportunities, this certain return on the

portfolio must equal zero. Therefore, we have

1 n 2
lim  i  0
n n
i 1

as required. QED.

26