0% found this document useful (0 votes)
41 viewsChapter 5 The Expected Utility Approach To The Problem of Choice Under Uncertainty
Uploaded by
20203111Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
0% found this document useful (0 votes)
41 viewsChapter 5 The Expected Utility Approach To The Problem of Choice Under Uncertainty
Uploaded by
20203111Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
/ 26
THE EXPECTED UTILITY APPROACH TO THE PROBLEM
OF CHOICE UNDER UNCERTAINTY!
1 INTRODUCTION
In this chapter we begin to develop a model for con-
sumption-investment decisions by individuals under
conditions of uncertainty. The general problem can be
deseribed as follows: Consider an individual who must
make @ consumption-investment decision at each of +
discrete points in his lifetime. At the first decision point
he has a quantity of wealth iy that represents the maxi-
mum possible level of consumption during time period 1.
At the beginning of period 1, tw, must be split between
current consumption cand investment fy = 121 ~ c.
At the beginning of period 2, the individual's wealth
level is
= (+ Rs) = (1 - a) (+ &),
‘It is acgumed here that the reader is familiar with elementary
statistical concepts, such as expected value, probability distribu-
tion, and variance. A brief review of these concepts is provided in
the Appendix to this chapter. 190 Uncertainty Models
where Rs is the one-period or percentage return at the beginning of period
2 per dollar of investment at the beginning of period 1.2 The return Ry is
assumed to be a random variable, that is, the observed value of fis drawn
from some probability distribution. Thus the wealth level ds is also a
random variable?
At the beginning of period 2, , must in turn be allocated to consumption
‘and investment, and the consumption-investment decision problem is
faced at the beginning of each subsequent period until period r, the last
period of the individual's life, at which time the entire available wealth 3,
is consumed and a bequest is considered consumption. The individual is
assumed to derive satisfaction only from consumption, and his problem is
to map out a consumption-investment strategy that maximizes the level of
satisfaction provided by anticipated consuinption over his lifetime.
‘Under uncertainty the decision problem is of course complicated by the
fact that the actual lifetime consumption sequence is to some extent
unpredictable, because, as indicated above, the wealth levels produced
through time by any given investment strategy are usually random vari-
ables. Thus in order to solve the individual’s sequential consumption-
investment problem, we need a theory of choice under uncertainty that
defines the criteria that the individual uses in choosing among different,
probability distributions of lifetime consumption. Developing such a
theory of choice is the purpose of this chapter.
W, THE EXPECTED UTILITY MODEL: GENERAL AXIOMATIC TREATMENT
The theory of choice under uncertainty that we apply to the consumption-
investment problem is the “expected utility hypothesis.” In general terms,
the expected utility hypothesis states that when faced with a set of mutually
exclusive actions, each involving its own probability distribution of “out-
‘comes,” the individual behaves as if he attaches numbers called, purely for
convenience, utilities to each outeome and then chooses that action whose
assotiated probability distribution of outeomes provides maximum ex-
pected utility.«
‘In general, the one-period or percentage return at {+1 on the investment Ay under-
taken at fis
mt
* Tildes (~) are used throughout this and following chapters to denote random variables.
‘When we talk about 1 specific observed value of auch a variable, however, the tilde
i dropped.
“The expected or average value of utility of any probebility distribution of outcomes
is computed like any other expected value; that is, the utility of each possible outcome Ch. 5, See. 1] Approach to Problem of Choice under Uncertainty 191
In the + period consumption-investment problem, an outcome is a com-
plete sequence of lifetime consumptions C, = (c1cz,...¢r), and an action.
is a r period consumption-investment strategy that produces a probability
distribution for different possible lifetime consumption sequences. But
because the consumption-investment problem is just one possible applica~
tion of the expected utility model, we initially present the model in the
‘most general terms and then turn to its applications that are of major
interest in finance
Thus we envisage @ decision maker faced with a set S of prospects,
whose characteristics are at this point purposely left unspecified, from
which a choice must be made. 5 includes all the prospects that are relevant,
for the decision at hand. Moreover, S ean include both “elementary
prospects,”” for each of which there is only one possible outcome, and
“random prospects,” which are probability distributions of mutually
exclusive elementary prospects.
At this point we could simply assume that the decision maker's behavior
conforms to the expected utility model; that is, he behaves as if he assigned
utilities to elementary prospects and then ranked random prospects on the
basis of expected utility. Alternatively, we can show that behavior in
conformity with the expected utility model is implied by a more basic set of
axioms concerning how the individual ranks outcomes and probability
distributions of outcomes, just as behavior in conformity with the ordinary
utility model in Chapter 1 can be shown to follow from a more basic set of
axioms concerning consumer choice under conditions of certainty. Because
an axiomatic treatment can help to produce a better understanding of the
model, this approach is taken here.
is weighted (multiplied) by the probability of the outcome, and the sum of these
products over all pomible outcomes is the expected or average vale of utility for this
probability distribution of outcomes.
‘Note that we aay that the individual behaves a if he wore an expected utility
maximises. Av always, we donot prerume that he formally goes through th optimisation
‘process prescribed by the theory. Rather his obwervable Behavior is agumed to be ax
I his decision proctas conformed to the model (compare Chap. 1, Sec. 1). As usual
however, we wwe words a ite lotely and talk about an individual maximizing his
tity, But such statements are alwaya meant tobe interpreted in an “an if” sense
"The reader who finds the general treatment in this ection overly abytact is nevertheles
encouraged to continue. Later sections provide concrete applications of the expected
utility model, and these give some perspective on the initial development of the model
presented bere. (Indeed the reader may find it helpful to reconsider the material
in this section after reading through the chapter.) Our mode of presentation is
‘designed to provide » more general understanding of the expected utility model than
srould be obtained only from its specie application to the consumption investment
problem. 192 Uncertainty Models
WA. The Axiom System
‘The set of axioms we use is as follows:
Axiom 1 (Comparability). The individual can define a complete
Preference ordering over the set of prospects in S; that is, for any two
Prospects z and y in 8, he can say that z > yory > zorz~y*
Axiom # (Transitivity), The ordering of prospects assumed in Axiom 1
is also completely transitive. For example, 2 > y and y > z imply z > 2;
or 2-~y and y~e imply 2~2; or 2~y and y > z imply z > z; and
50 on.
Aviom 8 (Strong Independence). If z ~ y, then for any third prospect
zin 8, G(2,2:a) ~ G(yt:a). Here G(z,2: a) represents a gamble, that is,
a random prospect, in which the individual gets either z, with probability
a, or z, with probsbility 1 — a, and G(y,t: a) likewise represents a gamble
that produces either y ors, with probabilities « and 1 — a. We also assume
that if 2 > y, then G(z2:a) > G(ye:a); or if = > y, then G(ze:a) =
G(y#:a). In short, the rankings of two prospects are not changed when
each is combined in the same way into a gamble or probability distribution
involving a common third prospect.
Axiom 4. If the prospects z, y, and z are such that either z > y > 2 or
2 2 y > 2, there is a unique a such that
y~G(zz:a).
Axiom 6.1f z>y2e and zeuzz, and y~G(zz:a) and
u~ G(z,: a3), then a1 > az implies y > u and a = as implies y ~ u.
Axioms 1 and 2 are analogous to the axioms of comparability and tran-
sitivity assumed in the theory of choice under certainty in Chapter 1. And
as in the certainty model, these two axioms are sufficient to define a con-
sistent preference ordering for all prospects in S. But if in addition we wish
to say that the ordering of random prospects in S is according to expected
utility, additional behavioral postulates are required, Axioms 3 to 5 provide
one possible set of such additional behavioral restrictions.
"Tho notation 2> y is read ‘xis strictly preferred to y,” and 2 ~ y is read “z and y
sre regarded as equivalent.” Likewise = > y is read “‘z is at least equivalent to ¥.”
Note that in the statements of the atioms there is no restriction on whether the
Prospects are elementary or random. For example, in Axiom 1, 2 and y can be elementary
or random prospects. {Ch. 5, See. 1} Approach to Problem of Choice under Uncertainty 193
Axiom 3 is the ‘strong independence” axiom. Intuitively, the importance
of the axiom to the expected utility model is ensy to see. If the expected
utility rule is to be applicable—that is, if the utility of a random prospect
is to be just the weighted sum of the utilities of its component elementary
prospects, with weights equal to the probabilities of obtaining each of the
elementary prospects—then necessarily the decision maker's attitudes
toward particular (mutually exclusive) prospects cannot be affected when
these are combined in various ways into random prospects. And this is the
direct assumption of Axiom 3.
In judging the reasonableness of the strong independence axiom as a
description of behavior, one must keep in mind that a random prospect is a
probability distribution of mutually exclusive elementary prospects;
ultimately one obtains only one of the elementary prospects in the proba-
bility mixture. Thus although in general one may not be willing to say that
an ordering of objects is unchanged when the objects are combined into
some mixture, in the case of a probability mixture of mutually exclusive
outcomes this assumption may seem more reasonable.
For example, the elementary prospects under consideration may be
bundles of consumption goods with all the goods in a given bundle to be
consumed by the decision maker. The decision maker's rankings of different
elementary prospects (bundles) is usually affected by the degrees of
complementarity and substitutability among the goods in a particular
bundle, so that the separate ranks of each good may not be a good indication
of the rank of the bundle. (Or in utility terms, it is not usually possible to
rank a consumption bundle by assigning utilities to each good, without
regard to the quantities of other goods in the bundle, and then obtain the
utility of the bundle as the sum of the separate utilities of each good.) But
in ranking probability distributions of such bundles, because a particular
distribution ultimately yields only one of its component bundles, one may
well find it reasonable to assume the strong independence axiom for such
probability mixtures. Thus, for example, one may be willing to say that if
x and y are two commodity bundles such that z~ y, then for any third
bundle z, the individual is indifferent between (1) a gamble in which either
+ is obtained with probability « or z with probability 1 — a and (2) the
corresponding gamble in which either y is obtained with probability « or 2
with probability 1 — a.
Intuitively it is clear that ranking random prospects, which are just
probability distributions of elementary prospects, according to expected
utility requires @ utility function in which the differences between the
utility levels assigned to different elementary prospects have some meaning;
that is, if the utility of a random prospect is to be just the expected or
average value of the separate utilities of each of its component elementary 196 Uncertainty Models
prospects, differences in utility levels must have some meaning.” In later
discussions we shall see that Axioms 4 and 5 play a critical role in defining
utility functions for which this is the case.
Before moving on, however, we should note that the axiom system that
we have chosen is far from the least restrictive set of behavioral postulates
that could be shown to lead to the expected utility rule. We chose this
Particular set of axioms to simplify the derivation of the expected utility
rule and to make this derivation contribute as much as possible to a fuller
understanding of the model. The reader who is interested in a derivation
from less restrictive assumptions is encouraged to seek out the references
at the end of the chapter, especially Herstein and Milnor [2].
NB. Derivetion of the Expected Usiity Rule
To show that the expected utility rule follows from the axioms, we must
show that the axioms imply two things:
1. There exists an order-preserving utility function; that is, if U(-) is the
function, U(z) > U(y) implies z > y and U(z) = Uy) implies
rT yt
The ordering of random prospects given by the function is according to
expected utility; that is, U(G(zy:a)) = aU(z) + (1 — a)U(y).
For simplicity let us suppose that the set S of prospects is bounded by
‘two extreme prospects a and 6, such that a > b and for any prospect
in Seither :
a>z>b
or azz>b.
Axioms 4 and 5 can then be used to rank all prospects in S in terms of the
two extreme prospects @ and 6. Thus let us define the function a(z) as the
probability such that
2~G(a,b: a(2));
that is, a(z) is the probability value for which the individual is indifferent
between (1) obtaining the prospect z for certain and (2) engaging in a
gamble, or equivalently, obtaining a probability distribution, that yields
1 This is of course in contrast with a purely ordinal function, such as those in Chap.
41, or those which would be implied by the first two axioms of the expected utility model,
in which only the ordering of prospects provided by the function is meaningful and,
xcept for sign, differences in assigned levels of utility are completely arbitrary.
* We use the notation U/(-) when we talk about the function in general terms, that is,
without reference to any specifi value of ita argument. ICh. §, See, IT] Approach to Problem of Choice under Uncertainty 193
either prospect @ with probability a(z) or prospect b with probability
1 — a(z). The existence and uniqueness of a(2) for any + in $ is guaranteed
by Axiom 4, and existence and uniqueness mean that «(z) is a function
defined for all prospeets in S. Thus for the prospect y, a(y) is the probability
value such that
y~G(ap:a(y)).
Indeed from Axiom 5 we ean see immediately that (2) is an order
preserving utility function; thet is, a(z) > a(y) implies 2 > y, and
(2) = a(y) implies z ~y. Thus to show that the expected utility rule
follows from the axioms, it remains only to how that the function a(z)
ranks random prospects according to expected utility; that i, consider a
random prospect G(z,y: 8) in which the individual obtains either 2, with
probability 8, or y, with probability 1 ~ f. From Axiom 4 we know that
there is always a unique probability «(@(z,y:8)) such that
G(z,y: 8) ~ G(a,b: «(G(z,y: 8)));
that is, there is always a probability a(G(z,y:6)) such that the individual
is indifferent between (1) engaging in the gamble G(z,y:8) and (2)
engaging in the gamble G(a,b: a(G(2,y:8))) in which he obtains either a,
with probability a(G(z,y: 8), or 6, with probability 1 — a(@(2y:6)).
We already know that @(G(z,y:8)) ranks G(zy:8) relative to other
prospects in S. It remains only to show that
a(G(z,y:8)) = Ba(z) + (1 — B)a(y);
that is, the ranking is according to expected utility, where a{-) is the
utility function. And we see that at this stage in the analysis, Axiom 3
(strong independence) begins to play a eritical role.
Because
2~ G(a,b:a(z)),
from Axiom 3 we can conclude that
G(z,y:8) ~ G(G(a,b: a(z)),y: 8). (5.1)
Here G(G(a,b: a(2)),y:8) represents a gamble in which (1) with proba-
bility ¢ the individual obtains the prospect G(a,b: «(z)), whieh is of course
itself a random prospect, or (2) with probability 1— 8 he obtains the
prospect y. Likewise, because
y~G(ab:a(y)),
from Axiom 3 we can again conclude that
G(G(a,b: a(z)),y:8) ~ G(G(a,b: a(2)),G(a,b: a(y)):8). (5.2) 196 Uncertainty Models
Applying Axiom 2 (transitivity) to expressions (5.1) and (5.2), we obtain
G(z,y: 8) ~ G(G(a,b: a(z))),G(a,b: «(y)) 8). (5.3)
Now G(G(a,b: a(z)),G(a,b: a(y)):8) is a double gamble in which either
(1) with probability 6 the individual engages in the gamble G(a,b: a(z))
or (2) with probability 1— 6 he engages in the gamble G(a,b: a(y)).
But both of these component gambles involve only the extreme prospects
a and b. Indeed the reader can easily determine that G(G(a,b: a(z)),
G(a,b: a(y)):8) is identical with G(a,b: Baz) + (1 — 8)a(y)), a gamble
in which the individual obtains either the prospect a, with probability
Ba(z) + (1 — B)a(y), or the prospect b, with probability 1 — [Ba(z) +
(1 = 8)a(y) J. Thus from Equation (5.3) and Axioin 2
G(2,y: 8) ~ G(a,b: Ba(z) + (1 — B)@(y))
But from Axiom 4
Gay: 8B) ~ (a,b: a(G(z,y:8)).
‘And because Axiom 4 also tells us that a(G(z,y: 8)) is unique, we must have
a(G(z,y:8)) = Ba(z) + (1 — B)a(y).
‘Thus the ranking of random prospects provided by the function a(+) is
indeed according to expected utility.
‘To review, Axioms 4 and 5 allowed us to define a utility function that
ranked prospects in S in terms of the two extreme prospects a and b. The
strong independence axiom then played a critical role in showing that this
utility function a(-) ranks random prospects according to expected utility.
N.C. Some Properties of the Unity Functions Implied
by the Expected Utility Med
It is useful to examine the utility function a(+) in a little more detail.
Note first that we must have
a~Glab:1) and bb ~G(a,b:0),
0 that a(a) = 1 and a(b) = 0. Thus the funetion ranges from 0 to 1,
‘and the utility value assigned to any particular prospect depends, from
Axiom 4, on how the probsbilities of obtaining gamble involving a and 5
must be balanced to make the individual indifferent between the prospect:
and the gamble.
‘We must emphasize, though, that a(-) is not the only utility function
consistent with the expected utility rule. Once we have shown that the
axioms imply the expected utility rule, any function that provides the same
rankings of prospect as a(+) and also ranks random prospects according to
expected utility is equivalent to a(+) as a representation of the individual's [Ch 5, See. Approach to Problem of Choice under Uncertainty 197
tastes. In fact, we now see that any positive linear transformation of a(+)
is equivalent to a(-).
‘Thus consider the function
U)=ntne), n>0
‘We want to examine the rankings of two arbitrary gambles, G(2,y:) and
@(u,z: 4), provided by the functions a(-) and U(+). We know that the
rankings provided by a(+) are according to expected utility; that is,
a(G(z,y:8)) = Ba(z) + (1 — B)a(y), (5.4)
a(G(u,z:¢)) = da(u) + (1 — ¢)a(2). (5.5)
On the other hand, if we use the function U(+) to compute the expected
utilities £(0) of the two gambles, we get?
For G(z,y:6):£(0) = aU(z) + (1 — 8)U(y)
= Bln + ye0(2)] + (1-8) + v0(y)]
= n+ vffa(z) + (1 - aly) J; (6.6)
For G(u2:4):£(0) = n+ rfea(u) + (1 - ¢)a(z)). (6.7)
Comparing Equations (5.6) and (5.7) with Equations (5.4) and (5.5), we
see that the expected utility rankings of the two gambles given by the
function U(-) is the same as that given by the function a(-).
This result is easy to explain. An expected utility is just linear com-
bination of the utilities of elementary prospects, that is, the expected
utility of a given probability distribution of elementary prospects is ob-
tained by first multiplying the probability of each elementary prospect by
the utility of the prospect and then summing the resulting products. Thus
when we take a positive linear transformation of the function a(), the
result is the same positive linear transformation of the expected utilities of
all random prospects, that is, utilities and expected utilities are all first
multiplied by a positive constant 7:, and then another constant 7 is added
to each. But these operations leave the rankings of the prospects completely
unchanged. And the same kind of reasoning can be used to conclude that in
general a nonlinear transformation of the utility function a(+) does not
provide the same expected utility rankings of random prospects as the
function a(-). We summarize these results with the statement that the
utility functions implied by the expected utility hypothesis are unique up to
Positive linear transformations.
"Note that the level of utility tobe obtained from a gamble is indeed a random variable:
thus the tilde over the Yin B(0). 198 Uncertainty Models
In sum, Axioms 1 and 2 would provide an ordinal utility function, say,
V(-), for prospects in the set S. As always, the function would be ordinal
in the sense that it would provide only an ordering of prospects in S; any
transformation of V(-), say, Z(V(-)), that is an increasing function of
values of V would provide a representation of the individual's tastes that is
equivalent to V(-). Adding Axioms 3 to 5, however, has allowed us to
imply a finer or more exact calibration of utilities: finer in the sense that
random prospects can be ranked on the basis of expected utilities. But the
utility functions implied by the expanded axiom set are themselves unique
only up to positive linear traitsformations, and so in a mathematical sense
they are not strictly cardinal, that is, unique. Following the eommon
usage of the utility literature, however, we henceforth refer to these
utility functions as cardinal
Tt is clear that with utility functions that are unique only up to positive
linear transformations, levels of utility do not have hedonic meaning, so
that, for example, interpersonal comparisons of utility are not possible.
‘The latter require utility functions that are strictly cardinal, that is,
unique.
But with the utility functions of the expected utility model, the change in
the level of utility from one prospect to another is unique up to a propor-
tionality factor; that is, if U(-) and (+) are two utility functions that
provide the same expected utility rankings of all prospects, there must be a
Positive constant y such that for any two prospects z and y in S,
U() — U(y) = yla(z) — a(y)], This means that with the utility
functions implied by the expected utility rule, one can sensibly talk about
increasing and decreasing marginal utility, which, as we soon see, is impor-
tant in developing meaningful notions of risk aversion and risk preference.
By way of contrast, recall that with a purely ordinal utility function only
the signs of marginal utilities have meaning.
ML, THE TIMELESS EXPECTED UTILITY OF WEALTH MODEL
‘The initial application of the expected utility model is to the problem of
choice from among various available “timeless” gambles. Specifically, we
assume that at the beginning of period 1 the individual has the opportunity
to.use his initial wealth w to engage in gambles whose outcomes, in this case,
levels of wealth, are known before the consumption-investment decision for
period 1 is made. The gambles are timeless in the sense that no consumption
takes place between the time when a gamble is undertaken and the time
its outcome is realized. We assume that the individual’s behavior in
this decision is in conformity with the axioms of the expected utility model,
‘0 that he chooses the gamble or the “portfolio” of gambles that maximizes
his expected utility. We want to show first how a utility of wealth funetion {Ch 5 Se. 11 Approach to Problem of Choice under Uncertainty 199
can be obtained from the axioms and then introduce the notions of risk
aversion, risk preference, and risk neutrality.
MLA, Obteining the Usity of Wealth Function from
Aslon 4
In the timeless expected utility of wealth model, elementary prospects
are just levels of wealth; random prospects are probability distributions of
wealth levels. Moreover, a nonsatiation axiom is usually added to the five
axioms presented in the preceding section; that is, itis assumed that more
wealth for certain is preferred to les.
‘To say that the utility of wealth functions implied by the axioms of the
expected utility model are unique only up to positive linear transformations
is equivalent to saying that two points on a utility function can be assigned
itrarily, as long as they are assigned in accordance with the nonsatiation
axiom, Then, as we now see, the remaining points on the utility function
can be determined from Axiom 4.
‘Thus suppose that we assign
($0) =0 and ($100) = 100.
According to Axiom 4, for any specific wealth level w between $0 and
$100, there is a unique probability « such that
w~ G(100,0; a).
‘Thus, given that we now know that the axiom system implies ranking of
random prospects according to expected utility, the utility of w is just the
expected utility of the gamble
U(w) = al (100) + (1 — a)U(0) = 1008.
Thus one can determine U(w) for # particular individual by asking the
following question: What is the probability a that would make you in
different between w dollars of wealth for certain and a gamble that could
result in $100 of wealth with probability a and $0 with probability (1 — «)?
Because we know U(0) and U(100), the answer to this question defines
uw).
‘The utility of a wealth level w greater than $100 can be determined by
noting that secording to Axiom 4 the individual can always define a proba~
bility « such that
100 ~ G(w,0: 2),
from which, applying the expected utility rule to G, we can infer that
(100) = als(w) + (1 @)U(0), for a specifie w > $100.
% And the reader may well find it useful to restate these five axioms in terms of the
current problem. 200 Uncertainty Models
‘Thus one asks the individual the following question: What is the probability
a that would make you indifferent between a wealth level of $100 for certain
and s gamble involving a probability a that wealth will be $w and (1 — a)
that it will be $0? The answer to this question defines U(w) as
U(100) — (1 = «) U(0)
Uw) =
=10
Thus once two points on an individuals utility of wealth funetion have
been assigned arbitrarily, the remaining points on the function ean, in
principle, be determined by a series of questions of the type presented above,
LB, Uwel Types of Unity of Wealth Functions: Risk
‘Aversion, Risk Preference, and Risk Neutrality
‘The nonsatiation axiom says that all utility functions must be monotone-
increasing functions of wealth. Thus to indicate that marginal utility is
always positive, a graph of utility against wealth must have a positive slope
at all levels of grealth. Given this basic restriction, there are three general
types of utility functions: (1) linear, (2) concave, and (3) convex, which
apply respectively to individuals who (1) have neither risk aversion nor
risk preference, (2) have risk aversion, and (8) have risk preference.
‘We can best elaborate on the implications of the individual’s attitudes
toward risk for the shape of his utility of wealth function by reference to
Figure 5.1, which shows the three general types. Let us begin by noting some
of the general properties of the three types of functions. A linear utility
function (Figure 5.1a) implies constant marginal utility of wealth at all
levels of wealth. Mathematically, the first derivative of the funetion is a
constant. A loss in wealth of ¢ decreases utility by exactly the same amount.
that an equivalent gain of ¢ would increase it. On the other hand, if the
individual's utility function is strictly concave (Figure 5.16), marginal
utility is a decreasing function of wealth. The graph of utility against
wealth is monotone-increasing, but utility increases with wealth at a slower
and slower rate. Mathematically, the first derivative of utility with respect
to wealth is positive, but the second derivative is negative. A loss in wealth
of ¢ decreases the individual's level of utility more than an equivalent gain
of ¢ would increase it. Finally, if the individual's utility of wealth function is
strictly convex (Figure 5.1¢), marginal utility is an increasing function of
wealth; utility increases with wealth at a faster and faster rate. Mathe-
matically, the second derivative of utility with respect to wealth is positive,
80 that the first derivative is an increasing function of wealth. An increase {Ch. 5, See. 111) “Approach to Problem of Choice under Uncertainty 201
Ut
U,(aw}
ULE ET)
Uso
u,t2w3)
£03)
stem)
Figure 5.1 Utility of Wealth Functions
in wealth of ¢ increases the individual's level of utility more than an equiva-
lent loss of « would decrease it.!*
Suppose now that at the beginning of period 1 the individual's wealth
level is w’. Suppose that if gamble A is chosen, his wealth w is 2u’ with
probability 0.5 and 0 with probability 0.5. The expected utility of A‘is
E(0) = 0.50 (0) + 0.5U(2w’),
which is just halfway between 0 and U(2u’) along the vertical axes of
‘each of the three graphs in Figure 5.1.
If we assume that the three graphs in Figure 5.1 represent the utility
of wealth functions of three different individuals, what can we say about
‘The statements concerning first and second derivatives hold only if these derivatives
‘exist, It ia eaay to show that the axioms of the expected utility hypothesis imply the
‘existence of a continuous utility of wealth function, but the function need not be
differentiable. 202 Uncertainty Models
the attitudes of these three individuals toward A? For individual i, ¢ =
1, 2,3, define the certainty equivalent level of wealth corresponding to the
probability distribution of wealth provided by the gamble A os the level of
wealth B,, i = 1, 2,3, such that:
UB) = EO);
that is, as far as individual ¢ with utility function Us is concerned, a proba-
bility distribution of wealth with expected utility E(0,) is exaetly equiva-
lent to B, of wealth obtained for certain. The individual would be indifferent,
in a choice between By for certain and the probability distribution of
‘wealth associated with action A.
‘The relationship between the expected value of wealth provided by &
given probability distribution and the certainty equivalent level of wealth
for this distribution gives us a way of defining whether the individual is a
risk averter, has risk preference, or is risk-neutral. For example, for the
individual in Figure 5.1b, the certainty equivalent level of wealth B: for the
probability distribution of A is less than w’ ="(«), the expected value of
wealth provided by A. Or in other words, for this individual the money
value of A is less than the expected wealth provided by A. (Or reading
along the vertical axis of the figure, the expected utility of A is less than the
utility of the expected wealth provided by A.) The individual regards any
level of wealth obtained for certain that is greater than B: as superior to the
probability distribution of wealth provided by the gamble A. Thus if A
were the only gamble available, he would be willing to pay an insurance
premium of w” — K, (K 2 Bz), in order to avoid A and obtaiti a level of
wealth of K for certain.!*
In general, for an individual with a strictly concave utility of wealth
function, the certainty equivalent level of wealth associated with a given
probability distribution is always less than the expected value of wealth
‘associated with the distribution. Or in other words, the money value of the
distribution to the individual is less than the expected value of its payoff.
‘Thus it is convenient to classify such a person as a risk averter.
By contrast, for the individual in Figure 5.Lc, whoee utility of wealth
funetion is strictly convex, the certainty equivalent level of wealth By for
the probability distribution of the gamble A is greater than w’ = E(2),
the expected level of wealth provided by A. For this individual the money
value of A is greater than the expected payoff from A. In essence for him
the chance of large gain is more than sufficient to compensate for an
equal chance of an equivalent large loss. Moreover, given a strictly convex
‘Thus this analyeis explains in part why people buy insurance, although insurance is
in general an unfair gamble; that is, the expected value of the payoff is less than the
premium that: must be paid. In essence they are willing to pay to avoid the probability
Aistributions of loasea that must be faced in the absence of insurance. [Ch 5, Bee. IV] ‘Approach to Problem of Choice under Uncertainty 203
utility function, these results hold with respect to all probability dis-
tributions of terminal wealth; the individual would always prefer to have
the distribution rather than its expected value for certain. Thus it is
meaningful to classify such a person as having risk prefer
For the individual in Figure 5.14, whose utility of wealth function is
linear, the certainty equivalent level of wealth Bi for the probability
distribution of A is exactly equal to 1’ = E(); the money value of the
distribution is exactly equal to its expected money payoff. Or in other
words, the expected utility of the distribution and the utility of its expected
value are equal. Moreover, these results hold with respect to all probability
distributions of wealth; if his utility function is linear, the individual
chooses among probability distributions of wealth solely in terms of their
expected values, always choosing the distribution with maximum expected
value. The “dispersion” or “riskiness” of the distributions has no effect on
his choice. Thus such a person is neutral with respect to risk; he has neither
risk aversion nor risk preference.
Finally, for convenience of exposition we have assumed that an individual
is s consistent risk averter, risk preferrer, or neutral with respect to risk at
all levels of wealth. This is, of course, not necessarily the case, and mixtures
of the three types of utility functions are possible. The only real restrictions
on the shape of the function are that it must becontinuousand itsslope must
be positive at all levels of wealth. Given these conditions, there is no reason
why the function cannot be concave over some regions of wealth, convex
over others, and linear over still others.
1V. EXPECTED UTILITY AND THE THEORY OF FINANCE
IV.A. The Multiperiod Expected Unity of Consumption
Model
‘The timeless expected utility of wealth model is useful for introducing
important concepts like risk aversion and risk preference. But the basic
problem in finance is the allocation of resources through time. Thus in the
+r period consumption-investment problem, the individual is concerned with
his lifetime consumption sequence
).
In this model the elementary prospects are the different possible lifetime
consumption sequences, that is, the different possible values of C,, and
random prospects are probability distributions of lifetime consumption
sequences, that is, probability distributions of C,. If we assume that the
individual's behavior in solving this problem conforms to the axioms of the
expected utility model, when these axioms are, of course, restated in terms
C= (easy 204 Uncertainty Models
of C,, then we can infer that the individual's tastes can be represented by
a utility function
U(C,) = Ulery.» 160)»
and the rankings of random prospects are according to expected utility.
If we assume the nonsatiation axiom—that is, holding consumption in
other periods constant, more consumption in any given period is preferred
to less—then the marginal utility of consumption in any period is positive.
‘Moreover, as in the timeless expected utility of wealth model, concavity,
convexity, and linesrity of the utility function U(C,) respectively imply
risk aversion, risk preference, and risk neutrality.
‘Thus, by definition, strict concavity of the function U(C,) says that for
any two consumption sequences Cy = (city. -s6r) and Cy = (éé1y.. sé),
and any a such that0 a (C100). + -s6r) + (1 = @) Ui (5.8)
‘or equivalently
Uae, + (1— aE.) > aU (C,) + (1- a) UC), O
