Você está na página 1de 15

Finals Microeconomics: MT11

Lectures on Risk and Expected utility:Lecture 3


Sujoy Mukerji
Oxford University
November 16, 2011
Sujoy Mukerji (Oxford University ) UGFnlsMicroRiskMT11 November 16, 2011 1 / 14
Comparing lotteries: return and risk
Fix a set of outcomes and consider dierent lotteries generated by
positing dierent probability distributions over these outcomes.
There are two natural ways to compare distribution over outcomes
level of returns higher vs. lower
dispersion of returns less risky vs more risky
Sujoy Mukerji (Oxford University ) UGFnlsMicroRiskMT11 November 16, 2011 2 / 14
Comparing lotteries: return and risk
Comparing levels: First order stochastic dominance
We want to attach meaning to the expression: "The distribution F(.)
yields unambiguously higher returns than the distribution G(.)."
(Note, we are considering CDFs here.)
At least two sensible criteria suggest themselves.
First, we could test whether every expected utility maximizer (for
whom "more is better") prefers F(.) over G(.).
Alternatively, we could verify whether, for every amount of money x,
the probability of getting at least x is higher under F(.) than under
G(.).
Fortunately, these two criteria lead to the same concept: they are
equivalent statements dening F(.) rst-order stochastically
dominates G(.)
Sujoy Mukerji (Oxford University ) UGFnlsMicroRiskMT11 November 16, 2011 3 / 14
Comparing lotteries: return and risk
Comparing levels: First order stochastic dominance
Recall, F(x) (a CDF) gives the probability that outcome is less than
x. So, F(.) rst-order stochastically dominates G(.) means
1 F(x) _ 1 G(x) at each x
= F(x) _ G(x) at each x.
(In pictures): The lottery `
1
rst order stochastically dominates the
lottery `
2
if and only if the CDF corresponding to `
1
lies below the
CDF corresponding to `
2
.
Example: `
1
=

1, 2, 3;
1
8
,
1
4
,
5
8

, `
2
=

1, 2, 3;
1
3
,
1
3
,
1
3

; . Check,
Pr (x _ 2 [ `
1
) =
7
8
>
2
3
= Pr (x _ 2 [ `
2
) ;
Pr (x _ 3 [ `
1
) =
3
8
>
1
3
= Pr (x _ 3 [ `
2
) .
Note if `
1
f.o.s.d. `
2
then the mean outcome under `
1
is greater thant
the mean outcome under `
2
but the converse is not generally true.
Sujoy Mukerji (Oxford University ) UGFnlsMicroRiskMT11 November 16, 2011 4 / 14
Comparing lotteries: return and risk
Comparing riskiness: Mean preserving spread
Restrict to comparisons between lotteries with same mean
Every risk averse expected utility maximizer prefers `
1
to `
2
if and
only if `
2
is a mean preserving spread of `
1
.
A sucient condition is that the CDF of `
1
crosses the CDF of `
2
from below, and the CDFs cross just once.
Sujoy Mukerji (Oxford University ) UGFnlsMicroRiskMT11 November 16, 2011 5 / 14
Comparing lotteries: return and risk
Mean preserving spread: a picutre
1
f
2
f
I
I
1
F
2
F
wealth wealth
density
cumulative probability
Example of a mean-preserving spread for a continuous distribution.
Sujoy Mukerji (Oxford University ) UGFnlsMicroRiskMT11 November 16, 2011 6 / 14
Comparing lotteries: return and risk
Why Mean preserving spread and not variance
Ranking lotteries in terms of variance does not guarantee all risk
averse agents preferences will rank lotteries the same way (as they
would be, if the ranking was in terms of mean preserving spreads).
Example
Lottery 1: pays 3 w.p. 1/2 and -3 w.p. 1/2. Mean: 0 Variance 9
Lottery 2: pays 4 w.p. 1/4 0 w.p. 1/2 and -4 w.p. 1/4. Mean: 0
Variance: 8
Consider a (CARA) Risk Averter i.e., an agent with vN-M utility given by
u = expx
Eu(Lottery 1) = approx -10
Eu(Lottery 2) = approx -14
This risk averse agent prefers the lottery with higher variance!
Sujoy Mukerji (Oxford University ) UGFnlsMicroRiskMT11 November 16, 2011 7 / 14
Mitigating risk
Risk spreading: the question
Consider risk averse agents with sure initial wealth w and vN-M
utility function u(.).
Suppose an agent is oered a risky project. The project yields a net
payo +200 (with probability 0.5) and -100 otherwise. Even though
the expected net payo is positive, the CE might well not be and a
single agent may, in that case, refuse taking on the project. However,
two agents together may well take it on; i.e., each getting net payos
of +100 and -50 in the respective states. Even if two do not, there
will be some number n such that a group of size n or larger will take
on the project.
The general proposition is, suppose the project give net payos x
1
and x
2
in states 1 and 2 respectively such that
E(w +X) = p
1
(w +x
1
) +p
2
(w x
2
) > w, while CE(w +X; u) < w.
Then, there will be an n > 0 such that for for all n > n,
CE(w +
X
n
; u) > w.
Spreading the risk over sucient number of agents will enable any
positive expected value project to be undertaken.
Sujoy Mukerji (Oxford University ) UGFnlsMicroRiskMT11 November 16, 2011 8 / 14
Mitigating risk
Risk spreading: the proof in a picture
On separate slide.
Sujoy Mukerji (Oxford University ) UGFnlsMicroRiskMT11 November 16, 2011 9 / 14
45

Payoffinstate1
Payoffinstate2
w
1
2
slope -
p
p
=
w
1
w x +
2
w x
Allpointstotherightofthis(thickgreen)line
hasexpectedvaluegreaterthanw
2
w x n
1
w x n +
Wholeriskyproject
Riskyprojectspreadovernagents
Mitigating risk
Risk pooling
Consider risk averse agents with sure initial wealth w and vN-M
utility function u(.).
Suppose each agent is oered a DIFFERENT risky project. Each
project yields a net payo +200 in case it is succesful (with probability
0.5) and -100 if it fails. Very importantly, the random variables
describing project net payos are independent random variables.
Again the expected net payo is positive but each agent may be
suciently risk averse to not want to take on a project.
However, agents could pool the risks. Under pooling each agent
draws the average income from the pool. That is, each draws the
income

n
i =1
X
i
n
.
Sujoy Mukerji (Oxford University ) UGFnlsMicroRiskMT11 November 16, 2011 10 / 14
Mitigating risk
Risk pooling
To see how this works, suppose there are 2 agents and 2 projects.
First consider payos to an agent from owning a project individually
(single ownership)
state F S
payo w 100 w + 200
probalility 0.5 0.5
Sujoy Mukerji (Oxford University ) UGFnlsMicroRiskMT11 November 16, 2011 11 / 14
Mitigating risk
Risk pooling
Now consider payos under the pooling contract.
state FF FS SF SS
avg py w +
[100100]
2
w +
[100+200]
2
w +
[+200100]
2
w +
[200+200]
2
prob. 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5
Sujoy Mukerji (Oxford University ) UGFnlsMicroRiskMT11 November 16, 2011 12 / 14
Mitigating risk
Risk pooling
payo to a single agent w 100 w + 50 w + 200
prob with individual projects 0.5 0 0.5
prob with pooled projects 0.25 0.5 0.25
Consider n independent random variables with identical probability
distributions: X
i
, i = 1, ..n. Let E(X
i
) = . The general result is as
n , Pr
h

n
i =1
X
i
n
i
=

1. This is a (rough) version of the


famous law of large numbers.
This is why insurance companies can behave as if they care only
about expected value...and are, eectively, risk neutral.
Sujoy Mukerji (Oxford University ) UGFnlsMicroRiskMT11 November 16, 2011 13 / 14
Ecient allocation of risk
Sujoy Mukerji (Oxford University ) UGFnlsMicroRiskMT11 November 16, 2011 14 / 14

Você também pode gostar