I.C.

Asset Pricing and International Risk Sharing

In the previous sections we used the Euler equation to derive optimal consumption and investment decisions. Now note that the Euler equation defines a relation between consumption (or other real variables) and prices.

Path of prices Path of real variables

Previously we took the prices as given and derived the optimal path of consumption. We may, however, use the Euler relation in the other direction. Take the path of real variables, e.g., consumption as given and derive what the prices have to be. A straightforward way to do this is to assume that output is exogenous, like manna from heaven, and cannot be stored. In that environment we may introduce markets for capital and production facilities. This is the setup in the seminal Lucas (Econometrica, 1978) paper. We will later relax the assumption about exogenous and non-storable output. This will give us the basic stochastic growth (or RBC) model. 1. CAPM and the Lucas Tree Model and other

(1.100)

1) Large number of identical agents. 2) Equal number of trees with stochastic crop dt The distribution of dt is Markov. Distribution is F d t d t 1 = F d t d t 1, d t 2 , . The process known by all agents.

7 2

K7

3) Purpose: find pt the price of a tree as a function of the state of the economy (dt). 4) The gross return on a tree is 5) No safe asset. 6) Perfect market in ownership of trees. All equal so no trade in equilibrium. 7) No storage or foreign trade so consumption ct = dt

pt +1 + d t +1 (per capita) pt

zt +1 from the Euler equation gives Substitute for ~

U ct = Et (1 + ) 1U ct +1 or

1 6

pt = Et

 U 1 c 6 1 p ! U 1 c 6
t +1 t

 !

1 61p

t +1 + d t +1

pt

t +1 + d t +1

6"# $

6 "# $

(1.101)

By 7)

pt = Et

 U 1 d 6 1 p ! U 1 d 6
t +1 t

t +1 + d t +1

6"# $

(1.102)

For any type of expectations in (1.102) based on dt we can compute a price today as a function of dt Let the individuals subjective expectations of dt+1 be described by by the function ps(dt+1). We then have

F s d t +1 d t and the expectations about the relation between pt+1 and dt+1 be given

pt =

p dt .

1 6
2

U d t +1 U dt

1 6 4 p (d 1 6
s

t +1 ) + d t +1

9dF 2d
s

t +1 d t

(1.103)

Now Lucas defined the very powerful concept of rational expectations. Lets require that p p and F d t +1 d t F
s

with (1.103) defines a unique and constant pricing function p . Use recursions on (1.102)
pt = E t

05

05

2d

t +1 d t . Lucas proves that this together

05

= Et

 !  !

U ( d t +1 ) U ( dt

U ( d t +1 U ( dt
j

= Et

j =1

U dt + j dt + j U (d t

  ( )   ( ) (K )"# )  ! ( ) ! ( ) $ ( ( )  )  ) "#  "# (K ) ) ! ( ) ! ( ) $ $

Et +1 U dt + 2 U dt + 1 Et + 2 U dt +3 U dt +2 +d t + 3 U dt +2 U dt +1 U dt +3 U dt + 1 +d t + 3 + dt +2

+ dt + 2

+ dt +1

+ lim
j

U d t + j pt + j U ( dt

 "#  "# $ $  "# $

+ dt +1

(1.104)

A discounted sum of dividends. Stochastic discount rates unless marginal utility is constant. A simple example with log utility

pt = Et

j =1

U dt + j dt + j
t

3 8 U 1 d 6

= Et

j =1

dt dt + j dt + j

= dt

j = dt 1
j =1

(1.105)

With dt i.i.d. so EtU d t + j d t + j = Et U d t + s d t + s j, s > 0

3 8

1 6

pt = Et

j =1

U dt + j dt + j j
t

3 8 U 1 d 6

= EU d t +1 d t +1 1 U dt

644744 8 0 5 1 6 1 6
constant

(1.106)

Note that the price increases in dt (if U is concave). Assume an autocorrelation in dt then Et

jU 3d t + j 8d t + j
j =1

depends on dt Both

income and substitution effect, with log utility they cancel.

The Consumption and market CAPM Consider a slight addition to the problem (1.4), namely that individuals can invest a freely chosen share of their assets in a risky asset and the remainder in the risk free asset. We assume away other income. The asset accumulation becomes

~ ) At +1 = At ct (1 + rt +1 )(1 t ) + (1 + r t +1 t ~ r ) = A c 1 + r + (r
t t t +1 t +1 t +1 t

0 0

50 50

(1.107)

and the Bellman equation

Vt At =
ct , t

1 ~ r ) max E t U ct + (1 + ) Vt + 1 At ct 1 + rt + 1 + ( r t +1 t +1 t

0 5

05

00

50

55

(1.108)

Using the envelope theorem to substitute for V, the first order conditions are

~ r ) ct ; U ct = E t (1 + ) 1U ct +1 1 + rt +1 + ( r t +1 t +1 t ~ r t ; E t (1 + ) 1U ct +1 ) r t +1 t +1 At ct ~ r Et U ct +1 ) r t +1 t +1 = 0
The second equation in (1.109) implies

05

0 50

0 50

0 510

50

5 =0
.

56
(1.109)

~ = E U c ) r E t U ct +1 r t +1 t t +1 t +1
Substituting into the first equation in (1.109) yields

0 5

0 5

(1.110)

 1 + r U 0 c ! 1+ ~  1+ r U 0 c U 0 c 5 = E + 1 !
U ct = E t
t t

05

t +1

t +1

t +1

t +1

5"#$ 5"#$

(1.111)

Consider a similar problem but with n risky assets. FOC for the portfolio share in each risk asset i yields

~ rt +1 Et U ct +1 = Et U ct +1 r i , t +1 = Et
So

0 5 U 0 c 5 E
t +1

0 5

~ ~ r i ,t +1 + cov U ct +1 , ri ,t +1 .

0 5

(1.112)

~ cov U ct +1 , r i , t +1 ~ E t ri , t +1 = rt +1 Et U ct +1
This is the Consumption CAPM, yielding the equilibrium expected return that must result on a competitive market. So a risky asset can have an expected rate of return that is larger or smaller than the safe return. Note that a positive covariance between consumption and the risky return implies a negative covariance between marginal utility and ~ zi ,t +1 so that asset will have a risk premium. Note that

0 5 0 5

(1.113)

di ,t +1 + pi , t +1 ~ r = 1 i ,t +1 pi ,t
where d is dividends and p is the price of the assets. So CAPM yields a relation between todays asset price and the expected price tomorrow. CAPM is thus in itself not enough to pin down the asset prices. A no-bubble condition is also needed. If we have CRRA ( U = c

(1.114)

) then we have

~ Et r i , t +1 = rt +1

~ cov ct+ 1 , ri ,t + 1 Et ct+ 1 ~ cov ct+ 1 ct , ri ,t +1 Et ct+ 1

= rt +1

ct

(1.115)

rt +1 +
where I have used that

~ cov ct +1 ct , r i ,t +1
E t ct+ 1 ct

cov( f ( x ), y) f ( x ) cov( x, y ) .
and thus

Now assume there exists an asset m which return is perfectly negatively correlated with marginal utility. So

~ = cov r m, t + 1
Then from (1.113)

~ U (ct +1 ) = r m,t +1 ~ ,r .
i , t +1

~ cov U (ct +1 ), r m, t + 1

~ cov t U ct +1 , r m, t + 1 ~ E t rm,t +1 = rt +1 Et U ct +1
E t U ct +1
Substitute (1.117) into (1.113)

0 5 0 5

(1.116)

0 5

~ var r m,t +1 = ~ Et r m,t +1 rt +1

(1.117)

~ Et r i , t +1

~ ~ cov r m, t +1 , ri ,t +1 ~ = rt +1 + Et r m, t +1 rt +1 ~ var r ~ = rt +1 + i ,t +1 Et r m, t + 1 rt +1 .

m, t +1

(1.118)

This is the market or traditional CAPM. Note that i is the (true) regression coefficient ~ in a regression of asset i on m. The term Et r m, t +1 rt +1 can be interpreted as the price of aggregate or systematic risk.

Empirics

The Mehra Prescott Puzzle Consider a representative household that maximizes

c1 E1 1 t =1 s. t . ct = yt , yt = t yt 1

(1.119)

is a stochastic growth rate that can take n different values {1 , , n } all >0. The probability of a specific growth rate depends only on last periods growth rate.

Pr t = j t 1 = i = ij .
Assume there is a share that entitles the owner to the entire output the next period. From (1.104) we have that the price of this share is

=
j

(1.120)

pt = Et

j =1 j =1

U yt + j y t + j
t

= Et

3 8 U 1 y 6

= Et

j =1

yt+ j yt + j

( yt )

kj =1 t + k

yt

yt yt

(1.121)

= p( yt , t ) = w( t ) yt wi yt = wi ct
where i {1,...,n}. So the price is H(1) in c and y. We can also use (1.102) to get

 U 1 y 6 1wy + y ! U1 y 6   y  "# = E   1wy + y 6# y   ! $ = E 0 5 1 wy + y 6 w = 4 3w + 189

wi yt = Et
t t +1 t t +1 t t t t t 1 n t t i ij 1 j j j =1

t +1

6"# $
(1.122)

This is a linear equation system in n unknowns so we can solve it for the price of the share in all states of the world. Now we can calculate the net return on the asset 6

~ = p ( j y t , j ) + j y t p ( yt , i ) r ij p( y t , i ) w j j yt + j yt wi yt j (w j + 1) = = 1 wi yt wi
and expected return is

(1.123)

~ ri e =

ij rij
i =1
t +1 t

(1.124)

We can also compute the price of a safe asset in this economy.

f

  y  "#  U 1 y 6 " (y , ) = E =E ! U 1 y 6 #$ !  y  #$
t t t t t t

= Et =
with a return of ri = 1 pi 1 .
f f

ij j pif .
j =1

(1.125)

Now we want to find the unconditional (average) returns on the assets. First we need the unconditional probabilities of the states .

Assume ergodic growth rates

= lim T 0 0 , 01 = 1
s 1 11

4 9  p( ) "#  L "# M # = M O M # ! p( )$ ! L $
s n1 n T n 1n

(1.126)

Then

~ r e = ~ ri e r = ri
f f

(1.127)

The risk premium is then defined as ~ re rf .

Results Now simplify and assumed there is two states and a common transition probability.

1 = 1 + + , 2 =1+ , 1 0.5 T = = . 1 0.5

 !

"# $

 "# ! $

(1.128)

Now we calibrate the model using the mean growth rate of GDP, its variance and autocorrelation.

E ( ) = = 0.018 Var ( )0.5 = = 0.036 + (1 )( ) Corr ( t , t 1 ) = = 2 1 = 014 . 2 = 0.43.

(1.129)

Using these value and some reasonable values for and <10 we find that the risk premium should be something in the order of 0 to 0.4%. But on average the US stock market has yielded 6% more on average than government. Consumption versus market (Mankiw & Shapiro 1986) Mankiw and Shapiro first estimates (1.118) from 464 different stocks. They first estimate the market , i.e., mi

m,t +1 time. Note that the coefficient on should be equal to the equity premium (around 6%) and constant for all periods and assets. Then they use this variable to predict the return on the corresponding asset.

~ ~ cov r m,t +1 , ri ,t +1 ~ var r

assuming it is constant over

~ zi ,t = 0 + 1 mi + t
The coefficient is significant and around 6 for most estimation methods. Now rewrite (1.115)

(1.130)

~ cov ct +1 ct , r i , t +1 ~ ri ,t +1 rt +1 + + t +1 Et ct+ c 1 t ~ ~ cov ct +1 ct , r m,t +1 cov ct +1 ct , ri , t +1 = rt +1 + + t +1 ~ c ,r cov c E c c

t t +1 t t +1 t m, t + 1
8 (1.131)

Assume that all relevant moments in (1.131) are constant over time. We then ci as sample moments

~ cov ct +1 ct , r i , t +1 ci ~ cov c c ,r
t +1 t
Then we can run the regression

(1.132)

m,t +1

~ zi ,t +1 = 0 + 1 ci + t +1
Now the estimate of 1 is insignificant and unstable. M&S also run a regression with both , then only the coefficient on mi comes out significant.

(1.133)

2.

International Risk Sharing

Consider a world with N countries indexed by n, inhabited with individuals with identical CRRA utility functions. Each country has a production sector that yields a stochastic output in each period, denoted Ytn. Shares in the production sector are traded on the world market where there is also a risk free on period bond traded with a safe return rt+1 between period t and t+1. . Buying a share in country ns production sector in period t, yields a stochastic return. The problem is

 " V 0 A 5 max E 01 + 5 U 0c 5# > C ! $  ~ s. t . A = 0 A c 5 1 + r + ( r  lim 01 + 5 A 0.

ts t c s , is t s=t s N t +1 t t t +1 i =1 i t T T T

i t +1 rt +1 )

, 
  

(1.134)

The Bellman Equation becomes

  V 0 A 5 max U 0c 5 + E V  0 A c 5 1 + r 1+  
1
t ct , t
i

i ~i t + 1 + t ( rt + 1 rt + 1 ) i =1

(1.135)

The first order condition for optimal portfolio shares is

1 ~i r ) = 0 EtV At +1 ( At ct )( r t +1 t +1 1+ ~i r ) = 0 E V A (r
t
t +1

0 5 0 5

(1.136)

t +1

t +1

Now assume log utility and use the result in (1.21) to guess that the value function is aln(A) + b. (Make sure you know how to verify the guess). Then we get

10

Et

a ~i ( rt +1 rt +1 ) = 0 At +1

Et

 0 A c 51 + r
t t

1 ~i r + it ( r t +1 t +1
i =1 N

t +1

 ) 

~i r ) = 0 (r t +1 t +1

(1.137)

Et

1 ~i r ) 1 + rt +1 + it ( r t +1 t +1
i =1 N

~ i r ) = 0. (r t +1 t +1

This is a complicated equation to solve, but we can see one thing immediately. The equation does not involve individual wealth. This means that all individuals, regardless of where they live, should choose the same portfolio shares. In other words, the individuals in each country should simply hold a portion of the world portfolio of assets. This results remains if we use CRRA functions with elasticities different from unity. The equal portfolio shares result is grossly at odds with the empirics. For example, the average US investor held 94% of her wealth in domestic assets. The predicted share is much lower. This is called the home-bias puzzle. The gain from diversifying may be large in 1987 an average return of 10% with a standard deviation of 17.3% on the S&P500 could be transformed into an average return of 13% with a standard deviation of less than 16% by using 14 non-US bond and equity markets. Attempted explanations for the home-bias puzzle 1. Non-traded goods. 2. Restrictions and frictions. 3. Irrational behavior. 4. Informational asymmetries.

11