CH Pres 01

1.1: Preferences 1.2: Risk Premia 1.3: Portfolio Choice 1.
4: Conclusions
Expected Utility and Risk Aversion
George Pennacchi
University of Illinois
George Pennacchi University of Illinois

Expected utility and risk aversion 1/ 60
1.1: Preferences 1.2: Risk Premia 1.3: Portfolio Choice 1.4: Conclusions
Introduction
Expected utility is the standard framework for modeling investor

choices. The following topics will be covered:
1 Analyze conditions on individual preferences that lead to an

expected utility function.
2 Consider the link between utility, risk aversion, and risk premia
for particular assets.
3 Examine how risk aversion a¤ects an individual’s portfolio
choice between a risky and riskfree asset.

Preferences when Returns are Uncertain
Economists typically analyze the price of a good using supply

and demand. We can do the same for assets.
The main distinction between assets is their future payo¤s:

Risky assets have uncertain payo¤s, so a theory of asset
demands must specify investor preferences over di¤erent,
uncertain payo¤s.
Consider relevant criteria for ranking preferences. One

possible measure is the asset’s average payo¤.

Criterion: Expected Payo¤
Suppose an asset o¤ers a single random payo¤ at a particular

future date, and this payo¤ has a discrete distribution with n
possible outcomes P (x1 ; :::; xn ) and corresponding probabilities
(p1 ; :::; pn ), where ni=1 pi = 1 and pi 0.
Then the expected value of the payo¤

P (or, more simply, the
expected payo¤) is x E [e x ] = ni=1 pi xi .
Is an asset’s expected value a suitable criterion for

determining an individual’s demand for the asset?
Consider how much Paul would pay Peter to play the

following coin ‡ipping game.

St. Petersburg Paradox, Nicholas Bernoulli, 1713
Peter continues to toss a coin until it lands “heads.” He

agrees to give Paul one ducat if he gets heads on the very …rst
throw, two ducats if he gets it on the second, four if on the
third, eight if on the fourth, and so on.
If the number of coin ‡ips taken to …rst obtain heads is i, then

i
pi = 21 and xi = 2i 1 : Thus, Paul’s expected payo¤ equals
P1 1 1 1 1
x = i =1 pi xi = 2 1 + 4 2 + 8 4 + 16 8 + ::: (1)
1
= 2 (1 + 12 2 + 14 4 + 18 8 + :::
1
= 2 (1 + 1 + 1 + 1 + ::: = 1

St. Petersburg Paradox
What is the paradox?
Daniel Bernoulli (1738) explained it using expected utility.
His insight was that an individual’s utility from receiving a

payo¤ di¤ered from the size of the payo¤.
P
Instead of valuing an asset as x = ni=1 pi xi , its value, V ,
would be Xn
V E [U (e x )] = p i Ui
i =1
where Ui is the utility associated with payo¤ xi .
He hypothesized that Ui is diminishingly increasing in wealth.

Criterion: Expected Utility
Von Neumann and Morgenstern (1944) derived conditions on

an individual’s preferences that, if satis…ed, would make them
consistent with an expected utility function.
De…ne a lottery as an asset that has a risky payo¤ and
consider an individual’s optimal choice of a lottery from a
given set of di¤erent lotteries. The possible payo¤s of all
lotteries are contained in the set fx1 ; :::; xn g.
A lottery is characterized by an ordered set of probabilities
P
n
P = fp1 ; :::; pn g, where of course, pi = 1 and pi 0. Let a
i =1
di¤erent lottery be P = fp1 ; :::; pn g. Let , , and
denote preference and indi¤erence between lotteries.

Preferences Over Di¤erent Random Payo¤s
Speci…cally, if an individual prefers lottery P to lottery P,

this can be denoted as P P or P P .
When the individual is indi¤erent between the two lotteries,

this is written as P P.
If an individual prefers lottery P to lottery P or she is

indi¤erent between lotteries P and P, this is written as
P P or P P .
N.B.: all lotteries have the same payo¤ set fx1 ; :::; xn g, so we
focus on the (di¤erent) probability sets P and P .

Expected Utility Axioms 1-3
Theorem: There exists an expected utility function

V (p1 ; :::; pn ) if the following axioms hold:
Axioms:
1) Completeness
For any two lotteries P and P, either P P, or P P, or
P P.
2) Transitivity
If P P and P P, then P P.
3) Continuity
If P P P, there exists some 2 [0; 1] such that
P P + (1 )P, where P + (1 )P denotes a
“compound lottery”; namely, with probability one receives the
lottery P and with probability (1 ) one receives the lottery P.
Expected Utility Axioms 4-5

4) Independence
For any two lotteries P and P , P P if and only if for all 2
(0,1] and all P :
P + (1 )P P + (1 )P
Moreover, for any two lotteries P and P y , P P y if and only if for

all 2(0,1] and all P :
P + (1 )P P y + (1 )P
5) Dominance
Let P 1 be the compound lottery 1 P z + (1 y
1 )P and P
2 be the
compound lottery 2 P z + (1 y
2 )P . If P
z P y , then P 1 P 2 if
and only if 1 > 2 .
Discussion: Machina (1987)
The …rst three axioms are analogous to those used to establish

a real-valued utility function in consumer choice theory.
Axiom 4 (Independence) is novel, but its linearity property is

critical for preferences to be consistent with expected utility.
To understand its meaning, suppose an individual chooses P

P. By Axiom 4, the choice between P + (1 )P and
P + (1 )P is equivalent to tossing a coin that with
probability (1 ) lands “tails,” in which both lotteries pay
P , and with probability lands “heads,” in which case the
individual should prefer P to P.

Allais Paradox
But, there is some experimental evidence counter to this

axiom.
Consider lotteries over fx1 ; x2 ; x3 g = f$0; $1m; $5mg and two

lottery choices:
C1: P 1 = f0; 1; 0g vs P 2 = f:01; :89; :1g
C2: P 3 = f:9; 0; :1g vs P 4 = f:89; :11; 0g
Which do you choose in C1? In C2?

Allais Paradox
Experimental evidence suggests most people prefer P 1 P2

and P 3 P4.
But this violates Axiom 4. Why?
De…ne P 5 = f1=11; 0; 10=11g and let = 0:11. Note that P 2

is equivalent to the compound lottery:
P2 P 5 + (1 ) P1
0:11f1=11; 0; 10=11g + 0:89f0; 1; 0g
f:01; :89; :1g

Allais Paradox
Note also that P 1 is trivially the compound lottery
P 1 + (1 ) P 1 . Hence, if P 1 P 2 , the independence
axiom implies P 1 P 5 .
Now also de…ne P 6 = f1; 0; 0g, and note that P 3 equals the
following compound lottery:
P3 P 5 + (1 ) P6
0:11f1=11; 0; 10=11g + 0:89f1; 0; 0g
f:9; 0; :1g
while P4 is equivalent to the compound lottery
P4 P 1 + (1 ) P6
0:11f0; 1; 0g + 0:89f1; 0; 0g
f:89; 0:11; 0g
Allais Paradox
But if P 3 P 4 , the independence axiom implies P 5 P 1 ,

which contradicts the choice of P 1 P 2 that implies
P1 P5.
Despite the sometimes contradictory experimental evidence,

expected utility is still the dominant paradigm.
However, we will consider di¤erent models of utility at a later

date, including those that re‡ect psychological biases.

Deriving Expected Utility: Axiom 1
We now prove the theorem by showing that if an individual’s

preferences over lotteries satisfy the preceding axioms, these
preferences can be ranked by the individual’s expected utility
of the lotteries.
De…ne an “elementary” or “primitive” lottery, ei , which

returns outcome xi with probability 1 and all other outcomes
with probability zero, that is, ei = fp1 ; :::pi 1 ; pi ; pi +1 :::; pn g
= f0; :::0; 1; 0; :::0g where pi = 1 and pj = 0 8j 6= i.
Without loss of generality, assume that the outcomes are

ordered such that en en 1 ::: e1 . This follows from the
completeness axiom for this case of n elementary lotteries

Deriving Expected Utility: Axiom 3, Axiom 4
From the continuity axiom, for each ei , there exists a

Ui 2 [0; 1] such that
ei Ui en + (1 Ui )e1 (2)
and for i = 1, this implies U1 = 0 and for i = n, this implies

Un = 1.
Now a given arbitrary lottery, P = fp1 ; :::; pn g, can be viewed
as a compound lottery over the n elementary lotteries, where
elementary lottery ei is obtained with probability pi .
P p1 e1 + ::: + pn en

By the independence axiom, and equation (2), the individual

is indi¤erent between lottery, P, and the following lottery:
p1 e1 + ::: + pn en p1 e1 + ::: + pi 1 ei 1 + pi [Ui en + (1 Ui )e1 ]

+pi +1 ei +1 + ::: + pn en (3)
where the indi¤erence relation in equation (2) substitutes for

ei on the right-hand side of (3).
By repeating this substitution for all i, i = 1; :::; n, the
individual will be indi¤erent between P and
n
! n
!
X X
p1 e1 + ::: + pn en p i U i en + 1 pi Ui e1 (4)
i =1 i =1


P
n
Now de…ne pi Ui . Thus, P en + (1 )e1
i =1
Similarly, we can show that any other arbitrary lottery
Pn
P = fp1 ; :::; pn g en + (1 )e1 , where p i Ui .
i =1
We know from the dominance axiom that P P i¤ > ,
P
n P
n
implying p i Ui > p i Ui .
i =1 i =1
So we can de…ne the function
n
X
V (p1 ; :::; pn ) = p i Ui (5)
i =1
which implies that P P i¤ V (p1 ; :::; pn ) > V (p1 ; :::; pn ).

Deriving Expected Utility: The End
The function in (5) is known as von Neumann-Morgenstern

expected utility. It is linear in the probabilities and is unique
up to a linear monotonic transformation.
The intuition for why expected utility is unique up to a linear
transformation comes from equation (2). Here we express
elementary lottery i in terms of the least and most preferred
elementary lotteries. However, other bases for ranking a given
lottery are possible.
For Ui = U(xi ), an individual’s choice over lotteries is the
same under the transformation aU(xi ) + b, but not a
nonlinear transformation that changes the “shape” of U(xi ).

St. Petersburg Paradox Revisited

p
Suppose Ui = U(xi ) = xi . Then the expected utility of the
St. Petersburg payo¤ is
n
X 1
X 1
X 1
X
1p 1
(i +1) i
V = p i Ui = 2i 1 = 2 2 = 2 2
2i
i =1 i =1 i =1 i =2
2 3
= 2 + 2 + :::2 2
X 1
1 i 1 1 1
= p 1 p = 1 p
i =0
2 2 1 p12 2
1
= p = 1:707
2 2
A certain payment of 1:7072 = 2:914 ducats has the same
expected utility as playing the St. Petersburg game.
Super St. Petersburg

The St. Petersburg game has in…nite expected payo¤ because
the probability of winning declines at rate 2i , while the
winning payo¤ increases at rate 2i .
In a “super” St. Petersburg paradox, we can make the
winning payo¤ increase at a rate xi = U 1 (2i 1 ) to cause
expected utility to increase at 2i . For square-root utility,
xi = (2i 2)2 = 22i 2 ; that is, x1 = 1, x2 = 4, x3 = 16, and so
on. The expected utility of “super” St. Petersburg is
Xn 1
X 1
1 p 2i 2 X 1 i 1
V = p i Ui = 2 = 2 =1 (6)
2i 2i
i =1 i =1 i =1
Should we be concerned that if prizes grow quickly enough,
we can get in…nite expected utility (and valuations) for any
chosen form of expected utility function?
Von Neumann-Morgenstern Utility
The von Neumann-Morgenstern expected utility can be

generalized to a continuum of outcomes and lotteries with
continuous probability distributions. Analogous to equation
(5) is
Z Z
V (F ) = E [U (e
x )] = U (x) dF (x) = U (x) f (x) dx (7)
where F (x) is the lottery’s cumulative distribution function

over the payo¤s, x. V can be written in terms of the
probability density, f (x), when F (x) is absolutely continuous.
This is analogous to our previous lottery represented by the
discrete probabilities P = fp1 ; :::; pn g.

Risk Aversion
Diminishing marginal utility results in risk aversion: being

unwilling to accept a “fair” lottery. Why?
Let there be a lottery that has a random payo¤, e ", where
"1 with probability p
e
"= (8)
"2 with probability 1 p
The requirement that it be a “fair” lottery restricts its
expected value to equal zero:
E [e
"] = p"1 + (1 p)"2 = 0 (9)
which implies "1 ="2 = (1 p) =p, or solving for p,

p = "2 = ("1 "2 ). Since 0 < p < 1, "1 and "2 are of
opposite signs.
Risk Aversion and Concave Utility

Suppose a vN-M maximizer with current wealth W is o¤ered
a fair lottery. Would he accept it?
With the lottery, expected utility is E [U (W + e")]. Without
it, expected utility is E [U (W )] = U (W ). Rejecting it implies
U (W ) > E [U (W + e
")] = pU (W + "1 ) + (1 p)U (W + "2 )
(10)
U (W ) can be written as
U(W ) = U (W + p"1 + (1 p)"2 ) (11)
Substituting into (10), we have
U (W + p"1 + (1 p)"2 ) > pU (W + "1 )+(1 p)U (W + "2 )
(12)
which is the de…nition of U being a concave function.
Risk Aversion , Concavity

A function is concave if a line joining any two points lies
entirely below the function. When U(W ) is a continuous,
second di¤erentiable function, concavity implies U 00 (W ) < 0.

Risk Aversion , Concavity
To show that concave utility implies rejecting a fair lottery, we

can use Jensen’s inequality which says that for concave U( )
E [U(~
x )] < U(E [~
x ]) (13)
Therefore, substituting x~ = W + e
" with E [e
"] = 0, we have
E [U(W + e
")] < U (E [W + e
"]) = U(W ) (14)
which is the desired result.

Risk Aversion and Risk Premium

How might aversion to risk be quanti…ed? One way is to
de…ne a risk premium as the amount that an individual is
willing to pay to avoid a risk.
Let denote the individual’s risk premium for a lottery, e
".
is the maximum insurance payment an individual would pay to
avoid the lottery risk:
U(W ) = E [U(W + e
")] (15)
W is de…ned as the certainty equivalent level of wealth
associated with the lottery, e
".
For concave utility, Jensen’s inequality implies > 0 when e " is
fair: the individual would accept wealth lower than her
expected wealth following the lottery, E [W + e "], to avoid the
lottery.
Risk Premium
For small e
" we can take a Taylor approximation of equation
(15) around e" = 0 and = 0.
Expanding the left-hand side about = 0 gives
U(W ) = U(W ) U 0 (W ) (16)
and expanding the right-hand side about e

" gives
E [U(W + e
")] = E U(W ) + e "2 U 00 (W )
"U 0 (W ) + 12 e (17)
2
= U(W ) + 0 + 1
2 U 00 (W )
where 2 "2 is the lottery’s variance.
E e

Risk Premium cont’d
Equating the results in (16) and (17) gives

00 (W )
1 2U 1 2
= 2 2 R(W ) (18)
U 0 (W )
where R(W ) U 00 (W )=U 0 (W ) is the Pratt (1964)-Arrow

(1971) measure of absolute risk aversion.
Since 2 > 0, U 0 (W ) > 0, and U 00 (W ) < 0, concavity of the
utility function ensures that must be positive
An individual may be very risk averse ( U 00 (W ) is large), but
may be unwilling to pay a large risk premium if he is poor
since his marginal utility U 0 (W ) is high.

U 00 (W ) and U 0 (W )
Consider the following negative exponential utility function:
bW
U(W ) = e ;b > 0 (19)
Note that U 0 (W ) = be bW > 0 and

U 00 (W ) = b 2 e bW < 0.
Consider the behavior of a very wealthy individual whose
wealth approaches in…nity
lim U 0 (W ) = lim U 00 (W ) = 0 (20)

W !1 W !1
There’s no concavity, so is there no risk aversion?

b2 e bW
R(W ) = bW
=b (21)
be
Absolute Risk Aversion: Dollar Payment for Risk
We see that negative exponential utility, U(W ) = e bW ,
has constant absolute risk aversion.
If, instead, we want absolute risk aversion to decline in wealth,

a necessary condition is that the utility function must have a
positive third derivative:
00
@R(W ) @ UU 0 (W
(W )
) U 000 (W )U 0 (W ) [U 00 (W )]2
= =
@W @W [U 0 (W )]2
(22)

R(W ) ) U(W )
The coe¢ cient of risk aversion contains all relevant

information about the individual’s risk preferences. Note that
U 00 (W ) @ (ln [U 0 (W )])
R(W ) = = (23)
U 0 (W ) @W
Integrating both sides of (23), we have

Z
R(W )dW = ln[U 0 (W )] + c1 (24)
where c1 is an arbitrary constant. Taking the exponential

function of (24) gives
R
R (W )dW
e = U 0 (W )e c1 (25)

R(W ) ) U(W ) cont’d
Integrating once again, we obtain

Z R
e R (W )dW dW = e c1 U(W ) + c2 (26)
where c2 is another arbitrary constant.
Because vN-M expected utility functions are unique up to a

linear transformation, e c1 U(W ) + c2 re‡ects the same risk
preferences as U(W ).

Relative Risk Aversion
Relative risk aversion is another frequently used measure

de…ned as
Rr (W ) = WR(W ) (27)
Consider risk aversion for some utility functions often used in
models of portfolio choice and asset pricing. Power utility can
be written as
U(W ) = 1 W ; < 1 (28)
( 1)W 2 (1 )
implying that R(W ) = W 1
= W and, therefore,
Rr (W ) = 1 .
Hence, it displays constant relative risk aversion.

Logarithmic Utility: Constant Relative Risk Aversion

Logarithmic utility is a limiting case of power utility. Since
utility functions are unique up to a linear transformation, write
the power utility function as
1 1 W 1
W =
Next take its limit as ! 0. Do so by rewriting the

numerator and applying L’Hôpital’s rule:
W 1 e ln(W ) 1 ln(W )W
lim = lim = lim = ln(W )
!0 !0 !0 1
(29)
Thus, logarithmic utility is power utility with coe¢ cient of
2
relative risk aversion (1 ) = 1 since R(W ) = W W 1
= W1
and Rr (W ) = 1.
HARA: Power, Log, Quadratic
Hyperbolic absolute-risk-aversion (HARA) utility generalizes

all of the previous utility functions:
1 W
U(W ) = + (30)
1
W
s:t: 6= 1, > 0, 1 + > 0, and = 1 if = 1.
1
W
Thus, R(W ) = 1 + . Since R(W ) must be > 0, it
1
W
implies > 0 when > 1. Rr (W ) = W 1 + .
HARA utility nests constant absolute risk aversion ( = 1,
= 1), constant relative risk aversion ( < 1, = 0), and
quadratic ( = 2) utility functions.
Another Look at the Risk Premium
A premium to avoid risk is …ne for insurance, but we may also

be interested in a premium to bear risk.
This alternative concept of a risk premium was used by Arrow
(1971), identical to the earlier one by Pratt (1964).
Suppose that a fair lottery e
", has the following payo¤s and
probabilities:
1
+ with probability 2
e
"= 1 (31)
with probability 2
How much do we need to deviate from “fairness” to make a

risk-averse individual indi¤erent to this lottery?

Risk Premium v2
Let’s de…ne a risk premium, , in terms of the probability of

winning, p, minus the probability of losing, 1 p:
= Prob(win) Prob(lose) = p (1 p) = 2p 1 (32)
Therefore, from (32) we have
Prob(win) p = 12 (1 + )
Prob(lose) = 1 p = 12 (1 )
We want that equalizes the utilities of taking and not taking

the lottery:
1 1
U(W ) = (1 + )U(W + ) + (1 )U(W ) (33)
2 2
Risk Aversion (again)
Let’s again take a Taylor approximation of the right side,

around = 0
1
U(W ) = (1 + ) U(W ) + U 0 (W ) + 12 2 U 00 (W ) (34)
2
1
+ (1 ) U(W ) U 0 (W ) + 12 2 U 00 (W )
2
= U(W ) + U 0 (W ) + 12 2 U 00 (W )
Rearranging (34) implies

1
= 2 R(W ) (35)
which, as before, is a function of the coe¢ cient of absolute

risk aversion.
Risk Aversion (again)
Note that the Arrow premium, , is in terms of a probability,

while the Pratt measure, , is in units of a monetary payment.
If we multiply by the monetary payment received, , then
equation (35) becomes
1 2
= 2 R(W ) (36)
Since 2 is the variance of the random payo¤, e ", equation (36)

shows that the Pratt and Arrow risk premia are equivalent.
Both were obtained as a linearization of the true function
around e
" = 0.

A Simple Portfolio Choice Problem
Let’s consider an individual’s single-period portfolio choice

problem.
Assume there is a riskless security that pays a rate of return
equal to rf and just one risky security that pays a random rate
of return equal to e
r.
Also, let W0 be the individual’s initial wealth, and let A be the
monetary amount that the individual invests in the risky asset
at the beginning of the period. Thus, W0 A is the initial
investment in the riskless security.
The individual’s end-of-period wealth, W ~ , is given by:
~
W = (W0 A)(1 + rf ) + A(1 + ~r ) (37)
= W0 (1 + rf ) + A(~r rf )

Single Period Utility Maximization
The individual chooses A to maximize expected utility that is

increasing and concave in end-of-period wealth:
~ )] = maxE [U (W0 (1 + rf ) + A(~r
maxE [U(W rf ))] (38)
A A
The …rst-order condition with respect to A is:

h i
E U0 W ~ (~r rf ) = 0 (39)
Note that the second order condition

h i
E U 00 W~ (~r rf )2 0 (40)
~
is satis…ed because U 00 W 0 from concavity.

Obtaining A from FOC
Suppose E [~r ] = rf . Then we can show A=0 is the solution.

If A=0, then W ~ = W0 (1 + rf ) so that
U W0 ~ = U (W0 (1 + rf )) is nonstochastic. Hence,
0
h i
E U0 W ~ (~r rf ) = U 0 (W0 (1 + rf )) E [~r rf ] = 0.
Next, suppose E [~r ] > rf .

h i
~ (~r
A = 0 is not a solution because E U 0 W rf ) =
U 0 (W0 (1 + rf )) E [~r rf ] > 0 when A = 0.
Thus, when E [~r ] rf > 0, let’s show that A > 0.

Why must A > 0?
Let r h denote a realization of ~r > rf , and let W h be the

corresponding level of W ~
Also, let r l denote a realization of ~r < rf , and let W l be the
corresponding level of W ~.
Then U 0 (W h )(r h rf ) > 0 and U 0 (W l )(r l rf ) < 0.
~ (~r
For U 0 W rf ) to average to zero for all realizations of
~r , it must be that W h > W l so that U 0 W h < U 0 W l due
to the concavity of the utility function.
Why? Since E [~r ] > rf , the average r h is farther above rf than
the average r l is below rf . To preserve (39), the multipliers
must satisfy U 0 W h < U 0 W l to compensate, which
occurs when W h > W l and which requires that A > 0.
Distribution of Returns
r ] > rf , there is more probability mass for r h than r l .

When E [e

Concave Utility
When 0 h 0 l
h E [e
r ] > rf , need
i U W < U W for
f (e
E U0 W r rf ) = 0.

How does optimal A change with initial wealth, W0 ?
dA(W 0 )
Let us use implicit di¤erentiation to obtain dW 0 .
h i
De…ne f (A; W0 ) E U W f , and let maximized expected
utility when A is optimally chosen be v (W0 ) = max f (A; W0 ).
A
Also de…ne A (W0 ) as the value of A that maximizes f for a
given initial wealth, W0 .
Using the chain rule, the total derivative of v (W0 ) with
respect to W0 is dvdW
(W 0 )
0
= @f (A;W
@A
0 ) dA(W 0 )
dW 0 +
@f (A(W 0 );W 0 )
@W 0 .
@f (A;W 0 )
= 0 since it is the …rst-order condition for a
@A
maximum.

How does A change wrt W0 cont’d
The total derivative simpli…es to dvdW

(W 0 )
0
= @f (A(W 0 );W 0 )
@W 0 ,
implying that the maximized objective function with respect
to a parameter is just the partial derivative with respect to
that parameter.
Second, consider how the optimal value of the control

variable, A (W0 ), changes when the parameter W0 changes.
We do so by taking the total derivative of the F.O.C. (39),

@f (A (W0 ) ; W0 ) =@A = 0, with respect to W0 :
@(@f (A(W 0 );W 0 )=@A) 2f (A(W 0 );W 0 ) dA(W 0 ) @ 2 f (A(W 0 );W 0 )
@W 0 = 0 =@ @A 2 dW 0 + @A@W 0

How does A change wrt W0 cont’d
Rearranging the above gives us
dA (W0 ) @ 2 f (A (W0 ) ; W0 ) @ 2 f (A (W0 ) ; W0 )

= = (41)
dW0 @A@W0 @A2
We can then evaluate it to obtain
h i
(1 + r f )E U ~
00 (W )(~
r rf )
dA
= h i (42)
dW0 E U 00 (W ~ )(~r rf )2
The denominator of (42) is positive because of concavity, so

dA
the sign of dW 0
depends on the numerator.

dA
Implications for dW0 with DARA
Consider an individual with absolute risk aversion that is

decreasing in wealth. Assuming E [~r ] > rf so that A > 0:
R W h < R (W0 (1 + rf )) (43)
where, as before, R(W ) = U 00 (W )=U 0 (W ).

Multiplying both sides of (43) by U 0 (W h )(r h rf ), which is
a negative quantity, the inequality sign reverses:
U 00 (W h )(r h rf ) > U 0 (W h )(r h rf )R (W0 (1 + rf )) (44)
Then for A > 0, we have W l < W0 (1 + rf ). If absolute risk
aversion is decreasing in wealth, this implies
R(W l ) > R (W0 (1 + rf )) (45)
dA
Implications for dW0 with DARA
Multiplying (45) by U 0 (W l )(r l rf ), which is positive,

inequality (45) becomes
U 00 (W l )(r l rf ) > U 0 (W l )(r l rf )R (W0 (1 + rf )) (46)
Inequalities (44) and (46) are the same whether the

realization is ~r = r h or ~r = r l , so taking expectations over all
realizations of ~r implies
h i h i
~ )(~r rf ) > E U 0 (W
E U 00 (W ~ )(~r rf ) R (W0 (1 + rf ))
(47)
The …rst term on the right-hand side is just the FOC.

Implications for risk-taking with ARA/RRA
Therefore, inequality (47) reduces to

h i
~ )(~r rf ) > 0
E U 00 (W (48)
Thus, DARA ) dA=dW0 > 0: A increases with initial wealth.
What about the proportion of initial wealth? To analyze this,

de…ne
dA
dW 0 dA W0
A
= (49)
W
dW 0 A
0
which is the elasticity measuring the proportional increase in

the risky asset for an increase in initial wealth.

Implications for risk-taking with RRA
A
Adding 1 A to the right-hand side of (49) gives
(dA=dW0 )W0 A
=1+ (50)
A
Substituting dA=dW0 from equation (42), we have
h i h i
~ )(~r rf ) + AE U 00 (W
W0 (1 + rf )E U 00 (W ~ )(~r rf )2
= 1+ h i
~ )(~r rf )2
AE U 00 (W
(51)
Collecting terms in ~ )(~r
U 00 (W rf ), this can be rewritten as

Implications for risk-taking with RRA
h i
~ )(~r
E U 00 (W
rf )fW0 (1 + rf ) + A(~r rf )g
= 1+ h i (52)
AE U 00 (W ~ )(~r rf )2
h i
~ )(~r rf )W
E U 00 (W ~
= 1+ h i (53)
AE U 00 (W~ )(~r rf )2
The denominator h in (53) is positive

i for A > 0 by concavity.
~ )(~r rf )W
Therefore, if E U 00 (W ~ > 0 then > 1 and the
individual invests proportionally more in the risky asset with
an increase in wealth.
Can we relate this to the individual’s risk aversion?
Implications for risk-taking with DRRA
Consider an individual whose relative risk aversion is

decreasing in wealth.
Then for A > 0, we again have W h > W0 (1 + rf ). When

Rr (W ) WR(W ) is decreasing in wealth, this implies
W h R(W h ) < W0 (1 + rf )R (W0 (1 + rf )) (54)
Multiplying both sides of (54) by U 0 (W h )(r h rf ), which is

a negative quantity, the inequality sign reverses:
W h U 00 (W h )(r h rf ) > U 0 (W h )(r h rf )W0 (1+rf )R (W0 (1 + rf ))

(55)

For A > 0, we have W l < W0 (1 + rf ). If relative risk aversion

is decreasing in wealth, this implies
W l R(W l ) > W0 (1 + rf )R (W0 (1 + rf )) (56)
Multiplying (56) by U 0 (W l )(r l rf ), which is positive, it

equals
W l U 00 (W l )(r l rf ) > U 0 (W l )(r l rf )W0 (1+rf )R (W0 (1 + rf ))

(57)
Inequalities (55) and (57) are the same whether the
realization is ~r = r h or ~r = r l .
Next, take expectations over all realizations of ~r to obtain

h i h i
E W~ U 00 (W
~ )(~r rf ) > ~ )(~r
E U 0 (W rf ) W0 (1+rf )R(W0 (1+rf ))
(58)
Since the …rst term on the right-hand side of inequality (58) is
the FOC, the inequality simpli…es to
h i
E W ~ U 00 (W
~ )(~r rf ) > 0 (59)
Hence, decreasing relative risk aversion implies > 1 so an

individual invests proportionally more in the risky asset as
wealth increases.
The opposite is true for increasing relative risk aversion: < 1
so that this individual invests proportionally less in the risky
asset as wealth increases.
Risk-taking with ARA/RRA
The main results of this section can be summarized as:
Risk Aversion Investment Behavior

@A
Decreasing Absolute @W 0 > 0
@A
Constant Absolute @W 0 = 0
@A
Increasing Absolute @W 0 < 0
@A A
Decreasing Relative @W 0 > W 0
@A A
Constant Relative @W 0 = W 0
@A A
Increasing Relative @W 0 < W 0

Conclusions
We have shown:
– Why expected utility, rather than expected value, is a better

criterion for choosing and valuing assets.
– What conditions preferences can satisfy to be represented by
an expected utility function.
– The relationship between a utility function, U(W ), and risk
aversion.
– How ARA/RRA a¤ects the choice between risky and risk-free
assets.


CH Pres 01

Uploaded by

Copyright:

Available Formats

CH Pres 01

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

CH Pres 01

Uploaded by

Copyright:

Available Formats

1.1: Preferences 1.2: Risk Premia 1.3: Portfolio Choice 1.

Expected Utility and Risk Aversion

George Pennacchi University of Illinois

Expected utility is the standard framework for modeling investor

1 Analyze conditions on individual preferences that lead to an

George Pennacchi University of Illinois

Preferences when Returns are Uncertain

Economists typically analyze the price of a good using supply

The main distinction between assets is their future payo¤s:

Consider relevant criteria for ranking preferences. One

George Pennacchi University of Illinois

Criterion: Expected Payo¤

Suppose an asset o¤ers a single random payo¤ at a particular

Then the expected value of the payo¤

Is an asset’s expected value a suitable criterion for

Consider how much Paul would pay Peter to play the

George Pennacchi University of Illinois

St. Petersburg Paradox, Nicholas Bernoulli, 1713

Peter continues to toss a coin until it lands “heads.” He

If the number of coin ‡ips taken to …rst obtain heads is i, then

George Pennacchi University of Illinois

St. Petersburg Paradox

What is the paradox?

Daniel Bernoulli (1738) explained it using expected utility.

His insight was that an individual’s utility from receiving a

He hypothesized that Ui is diminishingly increasing in wealth.

George Pennacchi University of Illinois

Criterion: Expected Utility

Von Neumann and Morgenstern (1944) derived conditions on

George Pennacchi University of Illinois

Preferences Over Di¤erent Random Payo¤s

Speci…cally, if an individual prefers lottery P to lottery P,

When the individual is indi¤erent between the two lotteries,

If an individual prefers lottery P to lottery P or she is

George Pennacchi University of Illinois

Expected Utility Axioms 1-3

Theorem: There exists an expected utility function

Expected Utility Axioms 4-5

Moreover, for any two lotteries P and P y , P P y if and only if for

Discussion: Machina (1987)

The …rst three axioms are analogous to those used to establish

Axiom 4 (Independence) is novel, but its linearity property is

To understand its meaning, suppose an individual chooses P

George Pennacchi University of Illinois

But, there is some experimental evidence counter to this

Consider lotteries over fx1 ; x2 ; x3 g = f$0; $1m; $5mg and two

Which do you choose in C1? In C2?

George Pennacchi University of Illinois

Experimental evidence suggests most people prefer P 1 P2

But this violates Axiom 4. Why?

De…ne P 5 = f1=11; 0; 10=11g and let = 0:11. Note that P 2

George Pennacchi University of Illinois

But if P 3 P 4 , the independence axiom implies P 5 P 1 ,

Despite the sometimes contradictory experimental evidence,

However, we will consider di¤erent models of utility at a later

George Pennacchi University of Illinois

Deriving Expected Utility: Axiom 1

We now prove the theorem by showing that if an individual’s