topic 2 decision theory 2015 (1)
TRANSCRIPT
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
1/51
Decision Theory under
Uncertainty (Parts I & II)
Economics of Information
(ECON3016)
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
2/51
Objectives of this Part
Review how economists model
individuals or firms choices under
uncertainty Review the standard assumptions
about attitudes towards risk
Understand concepts of Paretoefficiency in situations with uncertainty
Application: Optimal Risk Sharing
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
3/51
Uncertainty
Consumers and firms are frequentlyuncertain about the payoffs from their
choices. Example 1: A farmer chooses to
cultivate either apples or cherries
Uncertainty about profits
Not clear what will (turn out to be) thebetter choice
Profits depend on rain conditions,
plagues, market prices,
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
4/51
Uncertainty
Example 2: Rolling a fair die
Win 2 if 1, 2, or 3,
Neither win nor lose if 4, or 5 Lose 6 if 6
Example 3: Johns monthlyconsumption:
3600 if he does not get ill
400 if he gets ill (so he cannot work)
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
5/51
Modelling Uncertainty
Element 1: States of the world 1,2, , ill/healthy, number on the die, weather,
Element 2: Prizes in each state of the world Consumption level, monetary payoff, profit,
Element 3: Probabilities of the states of
the world
Assumption: Probabilities are known!
Assumption:Actors only care about prizes
and their likelihoods, not about states!
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
6/51
Lotteries
A lottery consists of
outcomes or prizes
, ,
Probability of outcomes : 0 One outcome will be realized:
probabilities add up to one
= ++ 1
Note that we are hiding the states of the
world in this description of uncertainty
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
7/51
Probability
The probability of a repetitive event is the
relative frequency with which it occurs
Probability of heads when flipping a coin is 0.5(If the coin is fair) More generally, probabilities can be
estimated from data and experience
If available!
Warning: Our conclusions may not carry
over to situations where there is no or
limited information about probabilities.
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
8/51
Expected Value
Expected value of a lottery with monetary outcomes 1, , and probabilities 1, 2,
++
=
Note: We will only consider lotteries with
monetary prizes
(Or other one-dimensional variables)
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
9/51
Expected Value of Monthly
Consumption (Example 3)
Example 3: Johns monthly consumption:
1 3600 if he does not get ill 2 400 if he gets ill (so he cannot work) Probability of illness 0.25 Probability of no illness 1 0.75 The expected value is:
+ 0.75 3600 + 0.25 400 2800
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
10/51
Compare Lotteries with the
same Expected Value
Lottery A: (lottery without risk)
get 2800 for sure (ill or healthy)
() 2800 Lottery B:
get 3600 if healthy ( 0.75) get 400 if ill ( 0.25) expected value also 2800
Which Lottery is better?
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
11/51
Risk Neutrality
A person is called risk-neutral if she alwayschooses the lottery with the highestexpected value
For a risk-neutral person we have in Ex. 3:
~ The person is indifferent between Lotteriesand because E ()
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
12/51
Indifference Curves
Lets draw indifference curves for a
risk-neutral person
2 monetary prizes , Fixed probabilities , Indifference Curve
Combinations of , that have thesame expected value
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
13/51
Risk-Neutral Indifference Curves: combinations
(1, 2)with constant expected value
1 1 2 2 1 1 1 2
1
2 11 1
1 2 2 1
1 1
2 1
1 1
(1 )
1 1
0, ; 0,1
(slope)1
E p x p x E p x p x
pE
x xp p
E Eif x then x if x then x
p p
dx p
dx p
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
14/51
Risk-Neutral Indifference Curves: combinations
(1, 2)with constant expected value
1
2
/(11)
/1
Decreasing line with slope:
21
11 1
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
15/51
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
16/51
Risk-Neutral Indifference Curves: combinations
(1, 2)with constant expected value
(Ex. 3)
2800/0.75
45-degree line:lotteries without risk
2800
2800
3600
400
Lottery A
Lottery B
Decreasing line with slope:
21
11 1
0.750.25 3
2
1
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
17/51
Risk Aversion
Does the expected value criterion (risk-neutrality) provide reasonable predictions?
Would you prefer Lottery or Lottery ? Most individuals would avoid the risk in and choose the safe option
The expected value does not capture the
uncertainty involved in a lottery! BUT: Will see that in some cases, risk-
neutrality is a reasonable assumption.
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
18/51
Expected Utility
Standard model for prefs. over lotteries
Individuals evaluate outcomes/prizes using
a strictly increasing utility function () Lottery is preferred over lottery if theexpected utility for is higher than for :
=
> =
is called the Bernoulli utility function
(or von Neumann-Morgenstern utility function)
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
19/51
Expected Utility
Note that Expected Utility is computed
similar to Expected Value:
The mathematical expectation is taken overutility values rather than monetary prizes.
=
++
Remark: EU criterion can also be applied for
non-monetary lotteries.
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
20/51
Expected Utility Example 3
Suppose John has Bernoulli utility function Lottery: 1 (2800) 1 2800 53 Lottery
:
0.75 3600 + 0.25 (400) 0.75 3600 + 0.25 400 50 John strictly prefers
over
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
21/51
Indifference Curves
2 monetary prizes , Fixed probabilities
, Indifference Curve Combinations of , that yield the
same expected utility
for given Bernoulli utility function (.)
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
22/51
Slope of Indifference Curves
+
+ (1 ) Total differentiation:+ 1 0Rearrange:
1
1 1
< 0
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
23/51
Indifference Curves
Features of indifference curves:
Decreasing
Slope on 45-degree line is= 11 1
11 1
Same slope as the constant expected value curve!
Individuals are approx. risk-neutral on the 45-degree line.
If is concave, then indifference curves areconvex (see first problem set).
Will see that concavity of the Bernoulli utility function
means that a person is risk-averse.
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
24/51
Concavity and Convexity
A (twice differentiable) function () is: concave if 0, convex if 0, strictly concave if < 0, strictly convex if > 0.
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
25/51
Indifference Curves (Ex. 3)
Lotteries without Risk:45-degree line Constant Expected Value:
21 11 1 0.750.25 3
2
1
Indifference Curves:
11 1
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
26/51
Indifference Curves and Risk Aversion
Convex Indifference Curve:
Among the lotteries with
constant expected value
(red line) the lottery on the45-degree line is most
preferred.
Person chooses certain
outcome ( ) if shedoes not have to reduceexpected value to do this.
2
1
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
27/51
Indifference Curves and Risk Aversion
2800/0.752800
2800
3600
400
Lottery A
Lottery B
2
1
Convex Indifference Curve:
Lottery is preferred toLottery . (no uncertainty) (uncertainty)
We have Indifference
between B and C even
though 2800Lottery C
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
28/51
Risk Aversion: Formal Definition
Consider decision maker (DM) with Bernoulli utility
function (). Definition: The Certainty Equivalent
CE(X)of a
lottery is the amount of money for which the DMis indifferent between the lottery and gettingCE(X) for sure:
(CE(X))EU(X) Definition: DM is risk averse (risk loving) if forevery lottery that involves risk, the certaintyequivalent is lower (higher) than exp. value of
CE(X))
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
29/51
Risk Aversion: Formal Definition
Related to the certainty equivalent:
Definition: The risk-premiumR for a lotteryis the difference between the expected value of
and its certainty equivalent:R E(X)CE(X) Easy to see from definitions:
DM is risk averse/risk loving if her risk-premium ispositive/negative for all lotteries that involve some
risk.
One can show:
Risk aversion is equivalent to concavity of ().
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
30/51
Indifference Curves and Risk Aversion
2800/0.752800
2800
3600
400 Lottery B
2
1
Suppose 2.800
EU 50 (see last week) CE EU CE 50 2500 ( )R CE 300
Lottery C
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
31/51
Examples of commonly used Utility
functions for risk averse individuals
( ) ln( )
( )
( ) 0 1
( ) exp( * ) 0
a
U x x
U x x
U x x where a
U x a x where a
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
32/51
Risk Neutrality
DM is risk-neutral if Equivalently: For all lotteries:
and 0 When does risk-neutrality arise? Average of many independent draws of the
same lottery equals expected value.
This is similar to diversification. Large companies (e.g. insurance companies)
can diversify risks and can therefore behave
risk-neutral.
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
33/51
Classification
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
34/51
Measuring Risk Aversion
Intuitively, risk-aversion becomes stronger
(CE smaller) if is more concave orindifference curves are more convex
Definition:Arrow-Pratt measure of absolute
risk aversion:
For risk averse individuals, () < 0 () is positive for risk averse individuals
"( )
( ) '( )
U Xr X
U X
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
35/51
Comparing Risk Aversion of different
Decision Makers/Utility Functions
Consider and such that for alllevels of wealth:
> is more risk-averse than :
For every lottery that involves risk: CE < CE and R > R
More risk-averse DM is willing to pay a
higher premium to avoid risk.
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
36/51
Risk Aversion at different wealth
levels for the same utility function
Consider a lottery with prizes , , . How do R(X) and CE() change if DM gets
wealthier?
Endow DM with additional certain wealth . New lottery + with outcomes+ , , +
Result: If is is decreasing(increasing/const.) in , then CE + is increasing (decr./const.) in
R
+ is decreasing (incr./const.) in
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
37/51
Risk Aversion and Wealth
Example 1: () ln()where > 0 Arrow-Pratt measure:
1
Risk aversion decreases as wealth
increases
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
38/51
Risk Aversion and Wealth
Example 2:
exp()where > 0. Arrow-Pratt measure:
Risk aversion is constant as wealth
increases
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
39/51
Willingness to Pay for Insurance
Consider a person with a current wealth of
100,000 who faces a 25% chance of losing
his automobile worth 20,000 Lottery Suppose the Bernoulli utility function is
() ln ()
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
40/51
Willingness to Pay for
Insurance
The expected utility isEU 0.75 100 + 0.25 80
EU 0.75 ln 100 + 0.25 ln (80) 11.46 Certainty equivalent:
ln CE EU CE . 94,574 Risk-Premium:R CE
95 000 94 574 426
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
41/51
Willingness to Pay for Full
Insurance
Pay premium for full insurance: Certain final wealth of 100,000 (Warning:
is not the risk premium)
DM is willing to for full insurance if: < 100,000 < 100,000
< 100,000 < 100,000 + () 5,426 Result: Willingness to pay is equal to
Expected Loss plus Risk Premium.
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
42/51
Actuarially Fair Premium
Definition: Insurance policy with
actuarially fair premium yields zero
expected profits for insurancecompany
(Note: marketing and administration
expenses are not included in thecomputation of the actuarially fair
premium)
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
43/51
Actuarially Fair Premium
Previous example: Expected profit0.75 + 0.25 20,000 0 Result:Actuarially fair premium is:
5,000 Notice: the actuarially fair premium is smaller thanthe maximum premium that the individual is willingto pay (5426). There is room for the insurance company and the individual to
trade and improve their profits/welfare. (Even with administrativecosts.)
This happens if (and only if) an individual is risk averse.
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
44/51
Summary (so far)
The expected value is an adequate criterion to choose among
lotteries if DM is risk neutral, but not if DM dislikes risk.
If DM prefers to receive B rather than a lottery with expected
value greater than B, then we say the DM is risk averse. If U(x) is the utility function, we always assume U(x)>0
If an individual is risk averse, then U(x)
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
45/51
Efficiency and Risk Sharing
Will analyse situations in which two
parties can allocate risk
Allocation: Specifies outcome for eachperson in each state of the world.
Definition: Ex-ante (Pareto) efficiency
An allocation is efficient if it is notpossible to increase the expected
utility of one person without decreasing
EU of someone else.
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
46/51
Efficiency and Risk Sharing: Example
Two states of the world 1,2 Two people:
, Allocation: (, , , ) Feasibility constraint:+ < and + <
where is the total amount of money availablein state . The allocation defines lotteries and.
Bernoulli Utilities:
and
.
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
47/51
Example: Finding PE allocations
Fix EU of one individual: EU EU Maximise EU of the other one:
max EU s.t. EU EUand feasibility Insert everything:
max + (1 ) s.t. + (1 ) EU
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
48/51
Efficiency and Risk Sharing: Example
-- Finding PE allocations
Set up Lagrangean:
Frist-order Conditions yield:
Make sure you can compute this condition.
PE allocations: Indifference curves touch!
Careful: Sometimes the optimal solution is at the
boundary! Then the we have inequalities from Kuhn-
Tucker conditions and indifference curves may intersect.
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
49/51
Efficiency and Risk Sharing: Example
Suppose (no aggregate uncertainty)and individuals are risk-averse.
> implies < Risk-aversion (concavity of ) implies:
< 1
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
50/51
Efficiency and Risk Sharing: Example
Suppose > (aggregate uncertainty) A is risk-averse and B is risk-neutral.
B risk-neutral implies 1
Hence: PE-efficient allocations are on the 45-linethrough the origin for A. (Plus corner solutions!)
Risk averse A is fully insured and risk-neutral B
takes all the risk.
-
8/9/2019 Topic 2 Decision Theory 2015 (1)
51/51
Summary (Risk Sharing)
Efficient risk-sharing: Ratio of marginal
utilities are equal for different DMs (except
at corner solutions)
No aggregate risk & risk aversion:
PE requires full insurance for both parties
Aggregate risk & one risk-neutral DM
PE implies full insurance for risk-averse DM
More generally: Less risk averse DM bears
more risk.