Chapter 2
Chapter 2
1.2 Lotteries
We will define a simple lottery L as a list L = (p1 , ..., pN ) with pn ≥ 0, ∀n and
P
pn = 1, where pn is interpreted as the probability that outcome n occurs.
n
The consumer has a rational preference relation < over all possible lotteries.
As in the case of choice under certainty, there is some utility function U : £ → R,
1
from the space of lotteries to the set of real numbers, that summarizes such
preference relation. This utility function has an expected utility form if there
is an assignment of numbers (u1 , ..., uN ) to the N outcomes such that for every
simple lottery L, we have
U (L) = u1 p1 + ... + uN pN
Example 1 Allais’ Paradox. Assume there are three possible monetary prizes:
$2.5m, $0.5m, and $0. The individual must choose between the following two
lotteries:
L1 = (0, 1, 0) and L01 = (.10, .89, .01)
and typically people will prefer lottery L1 . On the other hand, if the individual
chooses between
then he will choose L02 . These choices are not consistent with expected utility.
If the individual prefers L1 , this means that
and thus the individual should prefer L2 to L02 , if he had a von Neumann-
Morgenstern utility function.
2
This way, the vNM utility function can be written as
Z
U (F ) = u(x)dF (x),
and the individual is risk-neutral if it holds with equality. Notice that this is a
direct application of Jensen’s inequality, since u(·) is a concave function in the
case of a risk-averse person. Because of concavity, the utility gain from an extra
dollar is smaller than the utility loss of losing one dollar.
The certainty equivalent of a lottery F is the amount of money for which the
individual is indifferent between the gamble F and the certain amount c(F, u),
i.e. Z
u(c(F, u)) = u(x)dF (x)
There are two relevant measures for risk aversion. First, the Arrow-Pratt
coefficient of absolute risk aversion at x is defined as
u00 (x)
a(x) = − ,
u0 (x)
and it informs us about the concavity of the Bernoulli utility function. Further-
more, the risk premium may be approximated by the expression
1
a(x)var(x).
2
Alternatively, the coefficient of relative risk aversion is
u00 (x)
r(x) = −x .
u0 (x)
Example 2 The demand for insurance. Suppose a consumer initially has mon-
etary wealth W. There is some probability p that he will lose an amount L. He
can purchase insurance that will pay him q dollars in the event of a loss. The
cost is π per dollar of coverage. The consumer wishes to determine how much
coverage to purchase. His problem is
u0 (W − L + (1 − π)q ∗ ) 1−p π
= .
u0 (W − πq ∗ ) p 1−π
3
On the other hand, the expected profit of the insurance company is
u0 (W − L + (1 − π)q ∗ ) = u0 (W − πq ∗ ),
which implies that if the consumer is strictly risk-averse, i.e. the second deriva-
tive is negative, then he will fully insure, i.e.
W − L + (1 − π)q ∗ = W − πq ∗ ⇒ q ∗ = L.
We do not have to worry about the case x > w, since, if 1 − π > 0, then
the expected loss would be infinite. The two relevant cases are x = 0 and x > 0.
The first-order condition is
π 1−π
− ≤ 0.
w+x w−x
Then, if x > 0,
π 1−π
= ⇒ x = (2π − 1),
w+x w−x
and if x = 0,
π 1−π 1
< ⇒π< .
w w 2