Ec2723, Fall 2013: Assignment 3

Due in class, Tuesday December 3

1. A representative agent has Epstein-Zin preferences, where is the discount factor,
risk aversion, the elasticity of intertemporal substitution, and = (1
)(1 1= ). Dene
= exp( r ); where r is the pure rate of time preference. Suppose the consumption growth
log(Ct+1 =Ct ) of the representative agent is i.i.d. Recall that, under the budget constraint
and regularity conditions, the Euler equation is given by
2(
3
)
1
1
Ct+1
1
1 = Et 4 exp( r )
(1 + Ri;t+1 )5 ;
Ct
1 + Rw;t+1

where 1 + Rw;t+1 is the gross simple return on the portfolio of all invested wealth.

a) Consider an asset paying a dividend stream Dt = Ct for some constant . Iterating

the Euler equation forward and imposing a standard no-bubble condition, show that the
present value formula for the asset price Pt is
Pt = Et

1
X
j=1

exp( r j)

Ct+j
Ct

1
1 + Rw;t+j

Ct+j ;

where (1 + Rw;t+j ) is the j-period gross simple return on the wealth portfolio.
b) Following a guess and verify approach, we suppose that the consumption-wealth
ratio C=W is a constant. Denote exp( ) = 1 + C=W . Express 1 + Rw;t+j only in terms of
consumption and the constant .
For the random variable X = log(Ct+j =Ct ), dene the cumulant-generating function
c(#) = log E exp(#X) ;
for all # for which the expectations are nite. (Note: Do not confuse the preference parameter
with the argument # of the cumulant-generating function.) Use c( ) when necessary in the
following steps.
c) Based on your answers in parts a) and b), derive the log dividend-price ratio d=p =
log(1 + Dt =Pt ) in terms of and model parameters r , , , , and .
d) Solve for the constant
the case with = 1.)

in terms of model parameters r , , , and . (Hint: consider

e) Based on your answers in part c) and d), derive the log riskfree rate rf , in terms of
model parameters r , , , and . (Hint: consider the case with = 0.)
f) Finally, derive the log of the expected gross return, er = log(1 + ERi;t+1 ), as well as
the log risk premium rp = er rf , in terms of model parameters r , , , , and . (Hint:
We know that the price-dividend ratio is a constant.)
g) Discuss which of your answers to previous parts are aected by the use of Epstein-Zin
preferences rather than power utility.

2. Consider the following model for the stochastic discount factor, Mt+1 :
mt+1

log(Mt+1 );

mt+1 = xt +
xt+1 = xt +

t+1 ;

t+1 ;

where t+1 is randomly drawn from one of two distributions. With probability , state 1
occurs at time t + 1 and t+1 = 1;t+1 , where 1;t+1 N (0; 21 ); with probability 1
, state
2 occurs at time t + 1 and t+1 = 2;t+1 , where 2;t+1 N (0; 22 ).
a) Is the stochastic discount factor Mt+1 lognormally distributed, conditional on information available at time t? Is it lognormally distributed, conditional on information available
at time t and knowledge of the state (1 or 2) that occurs at time t + 1?
b) Use the formula for the one-period zero-coupon bond yield,
1 + Y1t =

1
;
Et Mt+1

to solve for the one-period bond yield in this economy. Show that the log yield, y1t
log(1 + Y1t ), is linear in the state variable xt .
c) Use the recursive equation for bond prices,
Pnt = Et [Pn

1;t+1 Mt+1 ];

where Pnt is the price of an n-period zero-coupon bond at time t, to show that all log bond
yields are linear in xt . Derive an expression for the slope coe cient relating n times the log
bond yield of maturity n to the state variable xt . What is the standard name for models
with log bond yields linear in state variables? What makes this model dierent from other
models you have seen with this property?
d) Which of the following phenomena are displayed by this model? Explain.
(i) Time-varying risk premia in the term structure of interest rates.
(ii) Changing volatility of interest rates.
(iii) Excess kurtosis of interest rate movements.
(iv) Imperfect conditional correlation of bond returns.

3. Consider an innite-lived investor with Epstein-Zin preferences allocating a portfolio

between a risky asset, with log return rt+1 , and a short-term safe asset, with log return
rf;t+1 . The risky asset return is lognormally distributed, with a constant risk premium over
the short-term safe asset. If the asset returns and the investors consumption are jointly
conditionally lognormal, Epstein and Zin have shown that the rst-order conditions for the
investors optimization problem can be written as
Et rt+1

rf;t+1 +

2
t

Covt (rt+1; ct+1 )

+ (1

)Covt (rt+1; rp;t+1 );

where 2t is the conditional variance of the risky asset return, ct+1 is the change in the
investors log consumption, and rp;t+1 is the log return on the investors portfolio. The
parameters dening preferences are the coe cient of relative risk aversion , the elasticity
of intertemporal substitution in consumption , and = (1
)=(1 1= ):
Using a loglinear approximation to the investors intertemporal budget constraint, the
innovation in log consumption can be written as
ct+1

Et ct+1 = rp;t+1

Et rp;t+1 + (1

)(Et+1

Et )

1
X

rp;t+1+j ;

j=1

where

expfE(ct

wt )g is a parameter of loglinearization.

a) Give an intuitive explanation of this expression for the innovation in log consumption.
b) Use this expression to rewrite the investors rst-order condition so that consumption
does not appear. Show that the parameter drops out of the rst-order condition when it
is rewritten in this way.
c) Write t for the share of the risky asset in the portfolio. Derive an equation for t .
[Hint: Relate the conditional covariance Covt (rt+1; rp;t+1 ) to t and 2t .] Show that t = ,
a constant, and show that has two components that can be interpreted as myopic demand
and intertemporal hedging demand respectively.
d) A real perpetuity or consol bond pays one real dollar each period forever. Using a
loglinear approximation, the return on this bond can be written as
rc;t+1 = rf;t+1 +

(Et+1

Et )

1
X

j
c rf;t+1+j ;

j=1

where c captures a constant risk premium on the consol, c 1 expfE( pc;t )g, and pc;t is
the log price of the consol including its current coupon. Use this expression to show that if
the risky asset is a consol, then an investor who is innitely risk-averse and innitely averse
to intertemporal substitution will invest all her wealth in the consol. Explain the economic
intuition behind this result, and explain why it requires innite aversion to intertemporal
substitution.

4. Consider the risksharing problem of two innitely lived agents, who receive random
shares of a xed endowment e. Each period there are two states of the economy. In state
1, agent 1 receives e=2 + k and agent 2 receives e=2 k, while in state 2 agent 1 receives
e=2 k and agent 2 receives e=2 + k. The conditional probabilities of the two states are
constant and equal. Each agent maximizes a discounted sum of expected period utilities,
with time discount factor . Period utility is given by

where e
c

u(c) = e
c

e
c2 ;

e=2 and c is the level of consumption during the period.

Following Alvarez and Jermann (Econometrica 2000) assume that the only punishment
for default is permanent exclusion from the nancial market.
a) Calculate an agents expected utility under perfect risksharing.
b) Calculate an agents expected utility under autarchy, if the state is initially bad and
if the state is initially good.
c) Derive a condition under which no risksharing is possible.
d) Show that this condition is equivalent to high implied interest ratesin autarchy as
dened by Alvarez and Jermann.
e) Characterize the conditions under which partial risksharing, but not full risksharing,
is possible.

5. Following Athanasoulis and Shiller (Review of Financial Studies 2000), consider a

single-period model with two agents. Agent 1 has income x1 = c + v + ", while agent 2 has
income x2 = c + v ". The term c is deterministic, while v and " are normally distributed
shocks with mean zero. The variance of the common income shock v is 1, while the variance
of the relative income shock " is a. Agent 1 has constant absolute risk aversion coe cient
1 , while agent 2 has constant absolute risk aversion coe cient 2 .
A single futures contract exists in this economy. The payo on the contract is 1 v + 2 ",
and it costs p payable next period. Normalize the variance of the payo to 1. (Note: this
imposes restrictions on 1 and 2 .) The contract is in zero net supply.
a) Given the price p, solve for the quantities of the futures contract q1 and q2 that each
agent wishes to hold.
b) Solve for the price p that must hold in equilibrium given that the futures contract is
in zero net supply.
c) Write down an expression for the maximized utility of each agent. Add the maximized
utility of agent 1 to the maximized utility of agent 2 to get an expression for social welfare.
d) Now assume that 1 = 2 . Show that a futures contract designer who wishes to
maximize social welfare will set 1 = 0. Interpret this result.
e) Now assume that 1 = 0 while 2 > 0. Show that the welfare-maximizing contract
designer will set 2 =
1 . Interpret this result.
f) Now assume that 2 = 0 while 1 > 0. Show that the welfare-maximizing contract
designer will set 2 = 1 . Interpret this result.
g) What assumptions would be needed for the welfare-maximizing contract to have
0? Explain.

6. This question asks you to use Generalized Method of Moments (GMM) to analyze the
two-beta asset pricing model of Campbell and Vuolteenaho (2004). It is not necessary to
use a powerful programming language or statistics package to work with the data. Microsoft
Excel is su cient (but Matlab would make life easier). If you plan to use Excel, rst read
the help les on matrix functions (MMULT, MDETERM, MMULT, TRANSPOSE), how to
dene matrixes with CTRL+SHIFT+ENTER, etc.
Download the le problem set 3 data.xls from the course web site. The le has the
following columns: <date>, <excess return on the market over the risk-free asset, RM t
Rf t >, <expected log excess market return known at time t 1, Et 1 (reM t )>, <minus the
markets discount-rate news, NDR;t >, <the markets cash-ow news, NCF;t >, <net riskfree rate (for quarter t, known at t-1), Rf t >, and <net simple test-asset returns on 9 of the
25 Fama-French ME-BE/ME portfolios, Ri;t 1 >.
a) Estimate the parameters of a linear stochastic discount factor, specied as
Mt = a + b( NDR;t ) + cNCF;t
using GMM. Use ten moment conditions (one for the risk-free asset and nine for the stock
portfolios) of the form 0 = E[Mt (1 + Rit ) 1]. Use the identity weighting matrix W =
I. Clearly write down the expression that you are minimizing. What are the estimated
parameter values?
b) Next, generate and report the covariance matrix of the parameter estimates. Use
formulas in Cochrane, Asset Pricing, that take into account the fact that you used a prespecied (identity) weighting matrix. Clearly write out the formulas you use in your answer.
Also report the z-statistics and p-values for the hypothesis that the coe cient is zero for
each coe cient individually.
c) Suppose a slightly modied version of the rst-order condition in the paper Bad Beta,
Good Betaholds:
E(Rit

Rf t ) = Cov[(Rit

Rf t ); NCF;t ] + Cov[(Rit

Rf t ); NDR;t ]:

What restriction does this rst-order condition impose on the parameters of the linear SDF
specication? (Use the fact that the means of both NDR;t and NCF;t are zero.) Test this
restriction. Is it rejected?
d) Again, suppose that the rst-order condition in part c) holds. Based on the estimated
parameter vector [a; b; c]0 and its covariance matrix you produced above, give an estimate of
and the standard error of this estimate. (Use the delta method.)
e) Finally, use your S estimate to produce the asymptotically optimal weighting matrix
specied as W = S 1 . Use this weighting matrix to get optimal parameter estimates and
their covariance matrix. Compare the parameter estimates and covariance matrix to those
produced in part b). Perform the J-test of overidentifying restrictions. What is the p-value
of the J-test?