Chapter I General Theory Economics of Risk and Time

Toulouse School of Economics Catherine Bobtche [email protected]

September 2013

Introduction

In this theory, only consequences matter (the process does not count, the framing is irrelevant...). The choices made by the agent may aect consequences. Most often, consequences will take the form of a summary variable (for instance wealth).

Outline

The Expected Utility Model

Risk Aversion

Changes in Risk

Critiques of Expected Utility Theory

The Expected Utility Model - Motivation

Necessity to build a preference functional that evaluates the level of satisfaction of the decision maker (DM) who bears the risk. Modern Theory of risk bearing has been founded by von Neumann and Morgenstern in 1944 (Theory of Games and Economic Behavior): - a rst axiomatic approach to the ordering of probability distributions introduced by Ramsey (1931), - stated in clearer and simpler terms by von Neumann and Morgenstern (1947). A nice axiomatic approach has been proposed by Luce and Raia (1957).

The Expected Utility Model - Lotteries

Two types of information are necessary to describe a risky environment: - enumeration of all possible outcomes * C : set of all possible outcomes that aect the well-being of the DM (it can be consumption bundles, monetary outcomes...), * C is assumed to be nite, - characterization of the risky environment * probability vector of the possible outcomes, * ps =probability of outcome s occurring, with A Lottery L is a vector (p1 , ..., pS ), the set of all lotteries on C is S L = {(p1 , ..., pS ) [0, 1] | s ps = 1}. A compound lottery is a lottery whose outcomes are lotteries. Only the probabilities of the potential outcomes matter (no matter whether they come from a simple or a compound lottery).
5

S s =1

ps = 1.

The Expected Utility Model - Lotteries

Denition The utility function U : L R has an expected utility form if there is an assignment of numbers (u1 , . . . , uS ) to the S outcomes such that for every simple lottery L = (p1 , . . . , pS ) L we have U (L) = u1 p1 + . . . + uS pS . A utility function U : L R with the expected utility form is called a von Neumann-Morgenstern expected utility function.

The Expected Utility Model - Axioms

Which lotteries should individuals prefer? : rationale preference relation over the set of lotteries L. is complete and transitive: - completeness (La , Lb ) L2 , either La

Lb or Lb

La (or both),

- transitivity if La Lb and Lb Lc , then La Lc . indierence relation: La Lb i La Lb and Lb La .

The Expected Utility Model - Axioms

Axiom 1: continuity is such that La , Lb , Lc L3 such that La Lb La + (1 )Lc . Lb Lc , [0, 1] such that

This implies that there exists a functional U : L R such that U (La ) U (Lb ) La Lb . With the two assumptions and Axiom 1 only, the theory of choice under uncertainty would not dier from the standard theory of consumer choice under certainty.

The Expected Utility Model - Axioms

Axiom 2: independence is such that La , Lb , Lc L, [0, 1] La Lb La + (1 )Lc Lb + (1 )Lc .

No parallel in the consumer theory under certainty. The independence axiom implies that the preference functional U must be linear in the probabilities:
S

U (L) =
s =1

us ps .

This is the most controversial point of this theory (see below).

9

The Expected Utility Model - The Theorem

Expected Utility Theorem Suppose that the rationale preference relation on the space of simple lotteries L satises the continuity and the independence axioms. Then can be represented by a preference functional that is linear in probabilities. That is, there exists a scalar us associated to each outcome for s = 1...S in such a manner that for any two lotteries a , p a , ..., p a ) and Lb = (p b , p b , ..., p b ), we have La = (p1 2 1 2 S S
S S a ps us s =1 s =1 b ps us .

La

Lb

10

The Expected Utility Model - The Theorem

Illustration of the theorem in the case where outcomes are monetary. A lottery on wealth can be expressed by a continuous random variable whose realization is an amount of money w : w1 w2 Eu (w1 ) Eu (w2 ) u (w )dF1 (w ) u (w )dF2 (w )

U (w1 ) U (w2 ) , where Fi is the cdf of the rv wi . Note that expected utility is cardinal whereas the utility function is ordinal. Last, this theory can be extended to subjective probabilities (Savage, 1954).
11

Outline

The Expected Utility Model

Risk Aversion

Changes in Risk

Critiques of Expected Utility Theory

12

Risk Aversion - Characterization

In this section, we study how dierent agents react in front of a given risk, and how to compare agents.

13

Risk Aversion - Characterization

Introduction of an additional restriction: DMs are risk averse. Due to John Pratt and based on the Jensens inequality. Denition An agent is risk averse if he dislikes all zero mean risks (or all pure risks) at all wealth levels. Equivalently, an agent is risk averse i Eu (z ) u (E (z )) = u (w ) where z = w + x with x a pure risk, namely E (x ) = 0. An agent is risk averse if replacing an uncertain nal wealth by its expected value makes him better o.

14

Risk Aversion - Characterization

Proposition An agent with vNM utility function u is risk averse i u is concave. Similarly, an agent is risk-neutral (risk-lover) i his vNM utility function is linear (convex). The observation of the human behavior strongly favors the assumption that human beings are risk averse.

15

Risk Aversion - Risk Premium

Denition A risk premium is the maximum amount of money that one is ready to pay to escape a pure risk (a zero mean risk). for a risk averse agent, is positive. is such that Eu (w0 + x ) = u (w0 ) where E x = 0. Notation: (w0 , u , x ). Note: an increasing linear transformation of u has no eect on the decision makers choice, i.e. on the risk premium.

16

Risk Aversion - Certainty Equivalent

Consider a lottery whose net gain is described by y = + x . Denition The certainty equivalent is the sure amount of money that makes the agent indierent between playing the lottery or not: Eu (w0 + y ) = u (w0 + C e ). Notation: C e (w0 , u , y ). C e (w0 , u , + x ) = (w0 + , u , x ).

17

Risk Aversion - The Arrow Pratt Approximation

Analysis of the characteristics of the risk premium for a small risk y , with y = + x where x is a pure risk and w0 : 1 1 2 (w0 + , u , x ) = E x 2 A(w0 + ) = x A(w0 + ), 2 2
(w ) where A(w ) = u u (w ) is the coecient of absolute risk aversion (of

dimension

1 e ).

The risk premium for a small pure risk is approximatively proportional to its variance.

18

Risk Aversion - Comparative Risk Aversion

But is it possible to have a measure of the intensity of risk aversion for any risk? No! It does not exist. But... Denition Let agents u and v have the same initial wealth. Agent u is more risk-averse than agent v if u dislikes all lotteries that v dislikes, independent of their initial wealth. The risk premium of any risk is larger for agent u than for agent v . Proposition (Pratt, 1964) The following three propositions are equivalent: 1. Agent u is more risk-averse than agent v , 2. w , Au (w ) Av (w ), 3. Function u is a concave transformation of function v : (.) with > 0 and 0 such that u (w ) = (v (w )), w .

19

Risk Aversion - Comparative Risk Aversion

The coecient of absolute risk aversion is economically and mathematically intuitive. For instance, roughly speaking, A(w0 ) measures the maximal amount that an agent with wealth w0 and utility u is ready to pay to get rid of a small risk with a variance of 2. But it is fairly tractable!

20

Risk Aversion - Decreasing Absolute Risk Aversion

Question: How is the risk premium for a zero mean risk aected by a change in initial wealth? Intuition: Wealthier people are generally less willing to pay for the elimination of xed risks. Denition: Preferences exhibit decreasing absolute risk aversion (DARA) if the risk premium associated to any risk is a decreasing function of wealth: (w0 , u , x )/ w0 0 for any w0 , x . Proposition: The following conditions are equivalent: 1. The risk premium is a decreasing function of wealth, 2. The absolute risk aversion is decreasing with wealth, meaning that u (w )/u (w ) is decreasing, 3. u is a concave transformation of u : u (w )/u (w ) u (w )/u (w ), for all w .
21

Risk Aversion - Some Utility Functions

- Harmonic Absolute Risk Aversion utility functions (HARA): the inverse of A(w ) (i.e. the prudence T (w )) is linear in wealth w w , with + > 0. - Constant Relative Risk Aversion utility functions (CRRA): HARA with = 0, and R (w ) = = constant (it also presents DARA) u (w ) = + w 1 /(1 ) ln w if = 1, if = 1.
1

u (w ) =

- Constant Absolute Risk Aversion utility functions (CARA): they exhibit increasing relative risk aversion and A(w ) = A exp (Aw ) . A - Quadratic utility functions: HARA with = 1, they exhibit increasing absolute risk aversion. The usual form is u (w ) = u (w ) = w
22

w2 . 2

Outline

The Expected Utility Model

Risk Aversion

Changes in Risk

Critiques of Expected Utility Theory

23

Changes in Risk - Motivation

In this part, we study how a given individual compares dierent risks.

24

Changes in Risk - Motivation

Previous section: changes in risk aversion (utility functions). Objective of this section: looking at changes in risk that have an unambiguous eect on the risk premium independently on the utility function. Some restrictions on the utility functions are imposed: - risk-averse agents (u 0), - usually, prudent agents (u More precisely, as (0, u , x1 ) (0, u , x2 ) Eu (x1 ) Eu (x2 ), the objective is to determine conditions under which this inequality is true for any utility function. 0, see below).

25

Changes in Risk - Motivation

Example: Which one of these two lotteries do you prefer?

x1 = (0.25, 300; 0.25, 100; 0.25, 0; 0.25, 200) x2 = (0.5, 200; 0.5, 100) Show that any risk-averse EU maximizer prefers x2 to x1 .

26

Changes in Risk - Increases in Risk

Study of changes in risk that preserve the expected outcome (mean preserving changes in risks). There are dierent approaches: 1. adding noise: from w to w + with E = 0. Remember that all risk averse agents dislike adding zero mean noises to the possible outcomes of their wealth. 2. single mean preserving spread (SMPS): w1 is a SMPS of w2 if
i. E w1 = E w2 , ii. there exists an interval I such that f1 (w ) f2 (w ), w I , and f1 (w ) f2 (w ), w outside I , OR ii. F1 (w ) F2 (w ) is positive and then negative as w increases.

Adding noise or constructing a series of SMPS are two equivalent ways to increase risk.

27

Changes in Risk - Second Order Stochastic Dominance

Denition Let F1 and F2 denote two cumulative distribution functions of w1 and w2 over the support [a, b ]. w2 dominates w1 in the sense of second order stochastic dominance (denoted w2 SSD w1 ) i

S ( ) =
a

(F1 (s ) F2 (s )) ds 0, [a, b ] and E (w1 ) = E (w2 ).

Remark: E (wi ) = b
a

Fi (s )ds .

28

Changes in Risk - Second Order Stochastic Dominance

Theorem (Rothschild and Stiglitz, 1970) An increase in risk from w2 to w1 with E (w1 ) = E (w2 ) can be dened by any of four statements that are equivalent: 1. The means are the same and risk-averse agents unanimously prefer w2 to w1 : u concave, Eu (w1 ) Eu (w2 ). 2. S (w ) 0, w [a, b ] and S (b ) = 0 (or w2 SSD w1 ). 3. w1 is obtained from w2 by a sequence of mean preserving spreads. 4. w1 is obtained from w2 by adding a white noise to it, with E (|w2 = w ) = 0 for all w .

29

Changes in Risk - Diversication

(well-known in nance)

Consider n iid lotteries x1 ,..., xn . A feasible strategy is characterized by a vector A = (1 , ...n ) with n i =1 i = 1. Proposition The distribution of nal wealth generated by the perfect diversication strategy (1/n, ..., 1/n) SSD dominates the distribution of nal wealth generated by any other feasible strategy.

30

Changes in Risk - Changes in Variance

A decrease in risk implies less variance.... HOWEVER: less variance does not imply a decrease in risk!!!

Example
x 0.3649 0.6065 1.6487 2.7183 u (x ) = ln(x ) -1 -0.5 0.5 1 p1 0.5 0.4 0 0.1 p2 0.742 0 0.258 0

E x1 V (x1 ) Eu (x1 )

0.6984 0.4660 -0.6000

E x2 V (x2 ) Eu (x2 )

0.6984 0.3140 -0.6130

31

Changes in Risk - First Order Stochastic Dominance

SSD is the condition under which every agent prefers one risk to the other in the case where the utility function is increasing and concave. FSD is the condition under which every agent prefers one risk to the other in the case where the utility function is increasing. Intuition: any change in risk that is generated by a transfer of probability mass form high wealth states to low wealth states is said to be FSD-deteriorating. Denition w2 is dominated by w1 in the sense of the rst degree stochastic dominance (FSD) if F2 (w ) F1 (w ), w . All consumers in the real world dislike FSD-dominated shifts in the distribution of nal wealth.
32

Changes in Risk - First Order Stochastic Dominance

Proposition The following propositions are equivalent: 1. All agents with a nondecreasing utility function prefer w1 to w2 : Eu (w2 ) Eu (w1 ) for all nondecreasing functions u . 2. w2 is dominated by w1 in the sense of FSD: F2 (w ) F1 (w ), w . 3. w1 is obtained from w2 by adding a nonnegative noise terms to the possible outcomes of w2 : w1 =d w2 + , where 0 with probability 1.
33

Changes in Risk - Main limits of FSD/SSD

- Incomplete orderings (just a subset of random variables can be ordered). - Not able to characterize the level of riskiness of a single random variable. - Need for an index of risk that would parallel the Arrow Pratts index of risk aversion. - A good index should both increase with more dispersion (SSD), and a lower location (FSD).

34

Changes in Risk - The Jewitt ordering

(Jewitt, 1989, Management Science)

Jewitt proposes an ordering on risks and not on agents. Denition Consider two lotteries w1 and w2 . w2 dominates w1 in the Jewitt sense i u , v more risk averse than u ; if u is indierent between w1 and w2 , v weakly prefers w2 to w1 .

35

Changes in Risk - The Jewitt ordering

Proposition w1 dominates w1 in the Jewitt sense i F1 (w ) F2 (w ) is positive and then negative when the wealth w increases.

36

Changes in Risk - Index of Riskiness

An index of riskiness that measures riskiness objectively - independently of the person or entity taking the risk has been developed by Aumann and Serrano (2008). Denition The index of riskiness R (Z ) of a lottery Z is dened by Ee

Z R (Z )

= 1.

R (Z ) is the reciprocal of the risk aversion index of a CARA agent who is indierent between accepting the lottery or not.

37

Changes in Risk - Index of Riskiness

Aumann and Serrano prove the dierent properties: - The riskiness of the lottery only depends on the lottery, and not on the wealth of the utility function of the DM for instance, - The index increases with more dispersion (SSD) and with a lower location (FSD), - R (nZ ) = nR (Z ), for all positive number n, - If two lotteries have the same index of riskiness, any compound lottery has this index as well, and the sum of the lotteries also, - If Z has a normal distribution, then R (Z ) =
1 Var (Z ) 2 E (Z ) .

38

Changes in Risk - Aversion to Downside Risk

Denition An agent is averse to downside risk if and only if he always prefers that a pure risk is contingent to a good outcome rather than to a bad outcome. Theorem An EU individual is averse to downside risk if and only if he is prudent (u is convex).

39

Outline

The Expected Utility Model

Risk Aversion

Changes in Risk

Critiques of Expected Utility Theory

40

The Expected Utility Model - Critics

Diculties in justifying the independence axiom: - the Allais Paradox shows that the independence axiom may be violated, - however, the EU proponents sustain that the paradox comes from a misconception by the players. They underline four important assumptions:
* * * * consequentialism, time consistency, context independence, reduction,

that imply the independence axiom. Other theories have been developed: rank dependent expected utility, ambiguity aversion, prospect theory and loss aversion, ...

41

Some Critiques of the Expected Utility Theory

Besides these critiques on the independence axiom for instance, some underlying assumptions do not seem reasonable. There exist both experimental and empirical evidences that seem to contradict the expected utility theory.

For instance, the objective function depends only on nal positions. Psychologists argue that peoples perception and evaluation of outcomes are aected by a reference position. There is an endowment eect. This leads to the notions of loss aversion and rst order risk aversion.

42

Some Critiques of the Expected Utility Theory

Moreover, there exist some paradoxes that contradict the theory. Besides the Allais paradox, the Ellsberg paradox has been the starting point of numerous academic works. Consider two urns: urn X is lled with 50 green balls and 50 red balls, whereas urn Y is lled up with some green balls and some red balls (100 balls in total). Four lotteries are considered: - lottery A: a ball is drawn from X . You get 10e if the ball is red, 0 otherwise, - lottery B : a ball is drawn from Y . You get 10e if the ball is red, 0 otherwise, - lottery C : a ball is drawn from X . You get 10e if the ball is green, 0 otherwise, - lottery D : a ball is drawn from Y . You get 10e if the ball is green, 0 otherwise.

43

Some Critiques of the Expected Utility Theory

Many people faced with the choice between A and B prefer lottery A, and many people prefer lottery C when they have to choose between C and D . Moreover, when oered both choices in dierent questions, say Question 1 for choice between A and B , and Question 2 for choice between C and D , many people choose A in Question 1 and C in Question 2. This choice is inconsistent with any subjective beliefs about the composition of res and green balls in Y . There is ambiguity aversion: people dislike the added uncertainty about the risk.

44

Some Alternatives - Prospect Theory

1979) The objective function is of the form: (pi ) v (xi )
i

(Kahneman and Tversky,

where - the reference point (relative to gains when x > 0 and loss when x < 0), - the valuation function dened on deviations from the reference point (and not on the nal outcomes), - and the weighting function need to be determined. The objective function is not linear in probabilities. This is the best supported alternative to expected utility theory.
45

Some Alternatives - Regret Theory

(Loomes and Sudgen)

A consequence function that is a function of the action a and the state of the world s expresses an outcome C (s ) = max F (a, s )
aA

leading to a utility u (C (s )). But the individual has to choose his action before knowing the state. Thus, the regret from having taken action a and then seen state s materializing, reads: R (a, s ) = u (C (s )) u (F (a, s )) . The expected regret of action a thus reads P S = s R (a , s ) =
s S s S

P S = s u (C (s ))
s S

P S = s u (F (a, s ))

Minimized expected regret can be viewed from a dierent perspective, as the additional value that would be attainable if one were able to postpone the decision until after the realization of the outcome. In other words, it is an option value. One may just want to minimize the maximum possible regret min max R (a, s ) .
aA s S 46

Some Alternatives - Rank-Dependent Expected Utility

This criterion is non linear in probabilities: 1) order the consequences C1 , C2 , . . . , Cn in increasing order of preferences, 2) choose a utility function u over the preferences 3) and a probability weighting function f satisfying f (0) = 0, f (1) = 1 and f (p ) < p , p (0, 1). The criterion function is
f (1) u (C1 )+ f (1 p1 ) (u (C2 ) u (C1 ))+ f (1 p1 p2 ) (u (C3 ) u (C2 ))+ . . .

See some examples.

47

Some Alternatives - Induced preferences

This theory recognizes that choice among random prospects is not the only thing people do; after they have made the choice, they make some other decisions that aect their eventual outcomes. When we focus solely on choice over random prospects (lotteries), these other choices are behind the scenes. Suppose such other action is labeled z , and the set of feasible actions is Z . That choice will aect the utility of a consequence, so write that utility as u (C , z ). The expected utility for a given action z Z is EU (L, z ) =
i

pi u (Ci , z ) .

The decision maker is allowed to choose this action after choosing the lottery (but before the realization of the actual outcome of the uncertainty). EU (L) = max EU (L, z ) = max
z Z z Z i

pi u (Ci , z ) .

z , EU (L) is a linear function of the probabilities. By choosing z optimally, we are taking the upper envelope of all these separate functions corresponding to dierent z . And the upper envelope of a family of linear functions is a convex function.
48

Some Alternatives

Among the active researchers in the area, you will nd some are totally convinced of the truth of one of the theories, but dierent people are committed to dierent approaches, and many others have not come to any conclusion at all.

Expected utility theory has provided the most detailed and richest body of applications so far, but other formulations are gradually catching up at the research level, especially in areas like behavioral nance.

49