Business 35904 John H.

Cochrane
Problem Set 1
Due Thursday, Week 2
Note: There are a lot of problems, but theyre all easy, and Im only looking for very
short answers.
Part I.
Do problems 1,4,8,9 in Ch 1. Asset Pricing p. 31. In 4, the suggestion is that we can
get the Sharpe ratio indicated by the following graph:
) (R
) (R E
In 9, nd the mean log return of portfolios d1,1 = cd1
n
+ (1 c):
)
dt, use reasonable
numbers to plot the mean log return as a function of c, and nd the composition of the
growth-optimal portfolio, where 1
n
denotes the market return.
Additional problems:
1. Prove that power utility reduces to log if 1. (Hint: Youre always allowed to
add constants to utility functions, so write the power function as
c
1
1
1
. Hint 2:
d (c
1
) ,dc = (1 )c

but d (c
1
) ,d 6= (1 )c

)
2. Risk aversion coecients
(a) Suppose you only care about consumption tomorrow, so the utility function is
simply 1 [n(c)] . You have enough income to support consumption c for sure.
i. Suppose you are forced to gamble a small fraction of of future consump-
tion, thus the gamble is c with 50% chance of each and your new
consumption level will be c = c + c or c = c c with equal probability.
How much utility do you lose? Use a second order Taylor approximation
to show that the answer is
1n(c) n( c) =
1
2
n
00
( c) c
2

2
1
ii. Similarly, nd out how much better o you are if you are given a gift o c.
(You only need a rst order expansion here.)
iii. Now, how much gift o c to I have to give you in order to compensate you
for taking on the risk ? Equate your answer in i and ii, solve for o in terms
of . Optional: express your answer in terms of the mean and variance of
the risk.
iv. What if you are asked to evaluate a dollar risk c = c + or c = c and
a dollar bribe o?
3. An investor has utility function
n(c
t+1
) =
c
1
t+1
1
(a) Suppose the investor will consume $50,000 next year. How much would he be
willing to give up at t+1 to avoid an even bet of gaining or losing $5,$50,$500,$5000
if = 0. 1. 2. 10. 50?
Hints: To answer this question, equate utility of $50,000 minus x for sure to
the expected utility of winning and losing the bet. Its prettier if you set the
problem up as, what fraction of consumption r would you give up for sure to
avoid a bet on a fraction of consumption , i.e. how much would you give up
for sure rc to avoid the 50/50 chance of gaining or losing c.
(b) How much would you give up in this situation? Think of t + 1 as the rest of
your life right now you expect to consume $50,000 per year for the rest of
your life. How much are you willing to give up on average not to take the
bet that will raise/lower your consumption by the $5-$5000 amounts? What
is your risk aversion? (Dont be surprised if your answers are not consistent
across the amounts of money involved.)
4. Suppose an investor has leverage, he must pay back an amount A. Equivalently,
suppose A represents a backstop level of consumption that the investor is simply
not willing to risk no matter what. (Id rather die than y commercial, honey.)
Now the utility function is
n(c) =
(c A)
1
1
(a) Plot this utility function. (Use = 2 if you want to use a computer. Freehand
sketch is ne too.)
(b) What is the risk aversion coecient
c&
00
(c)
&
0
(c)
for this investor? Try to make your
formula pretty, showing how is modied by the ratio c,A.
(c) If this investor has a loss, so that it is likely c will be much closer to A, does
this make him more or less risk averse?
2
This is an important problem. I think it captures a lot of what happened
in the fall of 2008. Many investors have leverage or backstop commitments
(mortgages) or even an accustomed level of consumption A. As they lose
money, they become more risk averse and sell, making markets go down further.
Theres nothing irrational about it if youre leveraged, you have to sell after
a loss.
Part II. Some continuous-time problems to get you a bit more familiar with d. and
dt and a cool paradox of long term returns.
1. Whats wrong with this:
dr
r
= od.
Z
dr
r
= o
Z
d.
Look up any integral table and
Z
a
2
a
1
dr
r
= ln r
2
ln r
1
Thus,
lnr
T
ln r
0
= o(. .
0
).
In the notes I came up with an additional
1
2
o
2
term. Where is the mistake?
2. Comparing geometric and arithmetic returns. Consider a stock with zero dividend
yield and price process
d1 = j1dt + o1d.
where j, o, are constants.
(a) Find the instantaneous arithmetic 1 (d1,1) and geometric1 (d log 1) mean
stock return. Explain their dierence.
(b) Does the dierence between arithmetic and geometric mean matter much? Use
values for o corresponding to i) the market return which varies about 1% per
day ii) a small growth stock which can vary 7% per day. (Transform to annual
units. Yes, there are stocks that vary 7% per day, dont be scared by the
annualized volatility. )
3. Suppose that log prices j follow a diusion with stochastic volatility,
dj
t
= jdt + o
t
d.
t
do
t
= d.
t
3
(Yes, this o can be both positive and negative, but the variance o
2
will always be
positive. o
t
is also unpleasantly nonstationary. A more reasonable specication
would have a square root process in the second equation, and a second d., and
start with dj,j rather than log prices, but I want to keep this problem simple.
) Solve this system nd j
t
in terms of j
0
, .
t
, .
0
, o
0
. Is j
t
still normally or
lognormally distributed? Hint: First solve o
c
in terms of .
c
.
0
, and plug that back
in the j equation. To evaluate
R
t
c=0
.
c
d.
c
apply Itos Lemma to d(.
2
). Of course
R
t
c=0
d(.
2
) = .
2
t
.
2
0
. You will get an answer with .
2
t
.
2
0
as well as .
t
.
0
on the
right hand side.
4. Solve the continuous-time AR(1),
dr
t
= cr
t
dt + od.
t
to do this, rst nd
d

c
t
r
t

from Itos lemma. Then you can just integrate. (solve means express it with r
t
on the left, and some integral against d.
t
on the right.) Does the answer remind you
of a discrete-time result? In a later problem set well come back to this question and
see how could you solve this if you werent so clever as to try d

c
t
r
t

.
5. A useful trick. Prove that if r is normal, 1(c
a
) = 1
1(a)+
1
2
o
2
(a)
. a) Write the
denition of 1(c
a
) =
1

2o
R

c
a
c

()
2
2
2
dr. Merge the two exponentials, pull a
c
j+
1
2
o
2
out of the integral, leaving an integral of a new normal distribution. The
integral of a normal distribution is one, so youre done.
4