176

CHAPTER

:L'

CHOICE UNDER

6:

+ :L > L. This is equivalent

UNCERTAINTY

to saying that

But since L I,', the independence axiom implies that we must have iL' + : L
: L + i L (see Exercise 6.B.1). This contradicts (6.B.3), and so we must conclude that
indifference curves are straight lines.
Figure h.B.S(c) depicts two straight but nonparallel indifference lines. A violation
of the independence axiom can be constructed in this case, as indicated in the figure.
There we have L 2 L' (in fact, L -- L'), but i L + :L" 2 iL' + $L" does not hold for
the lottery I," shown in the figure. Thus, indifference curves must be parallel, straight
lines if preferences satisfy the independence axiom.
In Proposition 6.R.3, we formally state and prove the expected utility theorem.
Proposition 6.6.3: (Expected Utility Theorem) Suppose that the rational preference
relation 2 on the space of lotteries .Y satisfies the continuity and independence
axioms. Then 2 admits a utility representation of the expected utility form. That
is, we can assign a number u, to each outcome n = 1, . . . , N in such a manner
that for any two lotteries L = ( p , ,. . . , p N ) and L' = ( p ; ,. . . , p h ) , we have

2 L'

if and only if

n=l

u,pn 2

C
n-1

u,p;.

Proof: We organiie the proof in a succession of steps. For simplicity, we assume that
there
are best and worst lotteries in Y , L and L (so, E 2 L 2 L for any L E 27.' If
I,
L, then all lotteries in V are indifferent and the conclusion of the proposition

holds trivially. Hence, from now on, we assume that

Sicp 1.

L > L.

/ ~ ' I , > L ' u n l l a ~ ( 0 , 1then

) , L > c c L + ( I - a ) L ' > L'.

This claim makes sense. A nondegenerate mixture of two lotteries will hold a
preference position strictly intermediate between the positions of the two lotteries.
Formally, the claim follows from the independence axiom. In particular, since L > L',
the independence axiom implies that (recall Exercise 6.B.1)

L
S i ~ 2.
p

al, + ( I

a ) L > a L + ( 1 - cc)L' > EL' + ( 1 - a)L' = L'.

Let a,/j E [O, I ] . Then pL

> cc.

+ (1

j?)k

> ccL + ( 1

a)L

if and only

if

Suppose that /j > a. Note first that we can write

+(I

/?)L=

+ ( I - y)[aL + ( 1 - a ) L ] ,

+
+

where y = [ ( P - a ) / ( l - a ) ] E (0, I ] . By Step 1 , we know that ?I > a 1 ( 1 - a)&.

Applying Step 1 again, this implies that
(1 - y)(aE ( 1 - a)&) > a 1 ( 1 - a)L,
and so we conclude that PL + ( 1 - P)L, > ccL + ( I - a)&.
For the converse, suppose that fi I
cc. If P = a, we must have
+ (1 - P)L,
a L + ( I - cc)L. So suppose that P < a. By the argument proved in the previous

7. In fact, with our assumption of a finite set of outcomes, this can be established as a
conscqucncc o f the independence axiom (see Exercise 6.B.3).

SECTION

paragraph (reversing the roles of a and

/l)L.

/It +(I

P),

6.8:

EXPECTED UTILITY

we must then have a L

+ (1 - a ) L >.

Step 3.

For uny L E Y , there is u unique a, such that

[a,L

+ (1

cr,)L]

- L.

Existence of such an a , is implied by the continuity of 2 and the fact that L and L
are, respectivcly, the best and the worst lottery. Uniqueness follows from the result
of Step 2.
The cxistcncc of a,, is established in a manner similar to that used in the proof of Proposition
3.C.1. Specifically, define thc sets

[ a ~ [ O , l ] : a l ( 1 - a ) L k L } and

Hy the continuity and completeness of

{ c t ~ [ O , l ] :L k a L + (1 -a)l.).

2,both sets are closed, and

any a 6 [O,1] belongs to

;it least onc of thc two scts. Sincc both sets arc nonempty and [0, I] is connected, it follows
that thcrc is some a belonging to both. This establishes the existence of an a, such that

+ (I

u,,L

--

Stcp 4.

a,,)/, I..
7'/1(>/ u t ~ i * t ~Uo n: Y' 4 R that u.ssiqn.s U ( L )= a, fbr ull L E .Y' represents
t hc prclfcrcncc~rolution 2.

Observe that, by Step 3, for any two lotteries L, I,' E 2', we have

LkI,'

ifandonlyif

Thus, by Step 2, L
Step .5.

2 L' if and

a,L+(l -a,)L,ka,.L+(l-a,.)L.

only if a , 2 a,.,.

Thr utility function U ( . ) (hut us.siyns U ( L )= a, for ull L

rrnd thrrc~forchu.5 the expected utility form.

Wc want to show that for any L, L' E Y ,and fi E [0,1],we have U ( B L

/lU(L)+ ( I - /I)U(IJ1).
By definition, we have

E:

+ (1

Y is linear

P)L1)=

and
Therefore, by thc indcpcndcnce axiom (applied twice),

- P[u(L)L+

( 1 - U(L))L] ( 1

P)[u(L')L+ ( 1 - U ( L ' ) ) L ] .

Rearranging terms, we see that the last lottery is algebraically identical to the
lottery
In other words, the compound lottery that gives lottery [ U ( L ) L+ (1 - U ( L ) ) L ]with
has the
probability /I and lottery [ u ( L ' ) L ( I - U ( L 1 ) ) L ]with probability ( I - /I)
same reduced lottery as the compound lottery that gives lottery L with probability
[/IU(L)+ ( I - P ) U ( L f ) ]and lottery L with probability [I - P U ( L ) - ( 1 - B)U(Lf)].
Thus
PL + ( 1 - /I)Lt
[ / I U ( L ) ( 1 - P ) u ( L ' ) ] L [I - /?U(L)- ( 1 - /?)U(L1)]L.

THEORY

177

178

CHAPTER

6:

C H O I C E UNDER

UNCERTAINTY

By the construction of U(.) in Step 4, we therefore have

U(/jL + (1 - /I)L') = PU(L)

+ (1

P)U(L1),

as we wanted.
Together. Steps I to 5 establish the existence of a utility function representing
that has the expected utility form.

Di.scu.ssion of the Theory of Expected Utility

A first advantage of the expected utility theorem is technical: It is extremely
convenient analytically. This, more than anything else, probably accounts for its
pcrvasive use in economics. It is very easy to work with expected utility and very
dillicult to do without it. As we have already noted, the rest of the book attests to
the importance of the result. Later in this chapter, we will explore some of the
analytical uses of expected utility.
A second advantage of the theorem is normative: Expected utility may provide
a valuable guide to action. People often find it hard to think systematically about
risky alternatives. But if an individual believes that his choices should satisfy the
axioms on which the theorem is based (notably, the independence axiom), then the
theorem can be used as a guide in his decision process. This point is illustrated in
Example 6.B. 1.

Example 6.H.1: Expc.ctc~d Utility as u Guide to Introspection. A decision maker may

not be able to assess his preference ordering between the lotteries L and L' depicted
in Figure 6.8.6. The lotteries are too close together, and the differences in the
probabilities involved are too small to be understood. Yet, if the decision maker
belicves that his preferences should satisfy the assumptions of the expected utility
theorem, then he may consider L" instead, which is on the straight line spanned by
I, and L' but at a significant distance from L. The lottery L" may not be a feasible
choice, but if hc detcrmines that L" > L, then he can conclude that L' > L. Indeed,
if L" >- L, then there is an indifference curve separating these two lotteries, as shown
in the figurc, and it follows from the fact that indifference curves are a family of
parallel straight lines that there is also an indifference curve separating L' and L, so
that L' >- I,. Note that this type of inference is not possible using only the general
Figure 6.B.6

Expected utility as a
guide to introspection.

SECTION

6.8:

EXPECTED UTILITY THEORY

179

(2 500000 dollars)

Figure 6.B.7

(0 dollars)

(500 000 dollars)

choice theory of Chapter I because, without the hypotheses of the expected utility
theorem, the indilkrence curves need not be straight lines (with a general indifference
map, we could perfectly well have L" > L and L > L').
A concrete example of this use of the expected utility theorem is developed in
Excrcise h.B.4. rn
As a descriptive thcory, however, the expected utility theorem (and, by implication,
its central assumption, the independencc axiom), is not without difficulties. Examples
6.B.2 and 6.B.3 arc designed to test its plausibility.

Example 6.R.2: The Alluis P~irudox.This example, known as the Allais paradox [from
Allais (1953)], constitutes the oldest and most famous challenge to the expected utility
theorem. It is a thought experiment. There are three possible monetary prizes (so the
number of outcomes is N = 3):
First Prize

Second Prize

Third Prize

2 500 000 dollars

500 000 dollars

0 dollars

The decision maker is subjected to two choice tests. The first consists of a choice
betwecn the lotteries L , and L',:
L,=(O,l,O)

L;=(.10,.89,.01).

The second consists of a choice between the lotteries L, and L;:

L2=(0,.11,.9)

L;=(.10,0,.90).

The four lotteries involved are represented in the simplex diagram of Figure 6.B.7.
I t is common for individuals to express the preferences L, > L', and L; > L,.'
8. In our classroom experience, roughly half the students choose this way.

Depiction of the Allais

paradox in the simplex.