Chapter 3 - Utility Theory: ST Petersburg Paradox
Chapter 3 - Utility Theory: ST Petersburg Paradox
Chapter 3 - Utility Theory: ST Petersburg Paradox
St Petersburg Paradox
• Suppose that a coin is to be tossed repeatedly until the head appears. The gambler
will be paid $2n if the head appears in the n-th toss.
Reference: Wayne L. Winston, Introduction to Probability Models, Operations Research v2, 2004 Thomson
AMA484 Decision Analysis 2
• A businessman rules out potential decision because it could bankrupt the firm,
even expected return is high. It is because the monetary value is not always a true
indicator of the overall value of the result of the decision.
• The overall worth of a particular outcome is called utility, and rational people
make decisions that maximize the expected utility.
AMA484 Decision Analysis 3
Representation of Lotteries
A lottery L
Event Consequence
Θ1 c1
Θ2 c2
Θ3 c3
Or
L = {(Θ1 , c1 ), (Θ2 , c2 ), (Θ3 , c3 )}.
AMA484 Decision Analysis 4
Lotteries
If a decision maker will
then, we can by pure deductive logic find the solution of any decision problem, which
is logically consistent with the decision maker’s own preferences and judgments.
Choosing among the available alternatives (different lotteries).
Lotteries: Possible course of action
• List of events: All the events on the list are mutually exclusive if no more than one
of them can occur;
• The events taken together are collectively exhaustive if one of them must occur.
AMA484 Decision Analysis 5
Utility Theory
(p1 , r1 ; p2 , r2 ; · · · ; pn , rn ).
• A lottery is often represented by a tree in which each branch stands for a possible
outcome of the lottery.
AMA484 Decision Analysis 6
• The number on each branch represents the probability that the outcome will occur.
• Our goal is to determine a method that a person can use to choose between
lotteries.
• The utility of the reward ri , written u(ri ), is the number qi such that the decision
maker is indifferent between the following lotteries:
(i.e., u(ri ) = qi )
qi
Most favorable outcome
1 ri and
1 − qi
Least favorable outcome
A decision maker wishes to choose among several acts (lotteries) when the
consequences of one or more of these acts depends on which one of a set of possible
events occurs.
AMA484 Decision Analysis 8
Preliminary Evaluations
If the decision maker wishes to solve his problem by formal analysis, then there are
two sets of preliminary evaluations which he must make:
(a) quantify his judgments about the possible events by assigning a probability P (Θi )
to each event Θi in a mutually exclusive and collectively exhaustive list,
{Θ1 , · · · , Θi , · · · , Θn }.
• We begin by assuming that the least favorable outcome (say, −$10, 000) has a
utility of 0 and that the most favorable outcome (say, $30, 000) has a utility of 1.
1
• Next we define a number x 12 having u(x 12 ) = 2
1
2
$30, 000 (Most favorable outcome)
1 x 12 and 1
2
−$10, 000 (Least favorable outcome)
0.8
0.6
u(x)
0.4
0.2
0
−10,000 0 10,000 20,000 30,000
x
AMA484 Decision Analysis 11
Interpretation of u(x)
If {(1, c)} ∼ {(α, c2 ), (1 − α, c1 )} or written as c ∼ {(α, c2 ), (1 − α, c1 )},
then
u(c) = αu(c2 ) + (1 − α)u(c1 ).
Thus
u(c) − u(c1 )
α= .
u(c2 ) − u(c1 )
Example: Determining u
Let C = {−$10, 000, $11, 000, $30, 000, $60, 000}.
Rank all monetary returns from the best to the worst and assign a utility of 1.0 to the
best and 0 to the worst:
Monetary value Utility
$60, 000 1.0
$30, 000 not yet known
$11, 000 not yet known
−$10, 000 0
Determine the utilities to the intermediate monetary values:
A series of questions are asked until the decision-maker is indifferent to the certain
money and the lottery.
AMA484 Decision Analysis 13
Probability-equivalence Approach
Question: Which of the following would you prefer?
(B) A lottery which will give a 70% chance of $60, 000 and a 30% chance of
−$10, 000?
Answer: A 30% chance of losing $10, 000 is too risky, I will take the certain money.
(B) A lottery which will give a 90% chance of $60, 000 and a 10% chance of
−$10, 000?
Answer: I now stand such a good chance of winning the lottery that I will buy the
lottery.
AMA484 Decision Analysis 14
(B) A lottery which will give a 85% chance of $60, 000 and a 15% chance of
−$10, 000?
Answer: I am indifferent between the certain money and the lottery ticket
$30, 000 ∼ {(0.85, 60, 000), (1−0.85, −10, 000)} ∼ {(0.85, 60, 000), (0.15, −10, 000)}.
(B) A lottery which will give a 60% chance of $60, 000 and a 40% chance of
−$10, 000?
Answer: I am indifferent between the certain money and the lottery ticket.
$11, 000 ∼ {(0.6, 60, 000), (1 − 0.6, −10, 000)} ∼ {(0.6, 60, 000), (0.4, −10, 000)}.
Thus
u($11, 000) = 0.6u($60, 000) + 0.4u(−$10, 000) = 0.6.
AMA484 Decision Analysis 16
Compound Lotteries
Suppose we ask a decision maker to rank the following lotteries:
0.5
$30, 000
1
L1 $10, 000 L2
0.5
$0
0.02
−$10, 000
1
L3 $0 L4
0.98
$500
AMA484 Decision Analysis 18
Suppose that for r1 = $10, 000, the decision maker is indifferent between
0.9
$30, 000
1
(1) $10, 000 and
0.1
−$10, 000
AMA484 Decision Analysis 19
Using (1)–(3), the decision maker can construct lotteries L′1 , L′2 , L′3 and L′4 such that
L′i iLi and each L′i involves only the best ($30, 000) and the worst (−$10, 000) possible
outcomes. Thus, from (1), we find that L1 iL′1 , where
0.9
$30, 000
L′1
0.1
−$10, 000
Returning to our discussion of L′′2 , we observe that L′′2 is a lottery that yields a
0.5 + 0.5 × 0.6 = 0.8 chance at $30, 000 and a 0.4 × 0.5 = 0.2 chance at −$10, 000.
Thus, L2 iL′′2 iL′2 , where
0.8
$30, 000
L′2
0.2
−$10, 000
Using (2), we find that the decision maker is indifferent between L4 and L′′4 , where
0.02
−$10, 000
L′′4
0.62
$30, 000
0.98
0.38
−$10, 000
In actuality, however, L′′4 yields a 0.98 × 0.62 = 0.6076 chance at $30, 000 and a
0.02 + 0.38 × 0.98 = 0.3924 chance at −$10, 000. Thus, L4 iL′′4 iL′4 , where
0.6076
$30, 000
L′4
0.3924
−$10, 000
AMA484 Decision Analysis 23
• The specification of u(ri ) for all rewards ri is called the decision maker’s utility
function.
• Given two lotteries L1 &L2 , we may choose between them via the expected utility
criteria
L1 r2 and L2
1−c
r3
AMA484 Decision Analysis 28
L1 and L2
1−c 1−c
r3 r3
and
0.5
$30, 000
L′′2
0.6
$30, 000
0.5
0.4
−$10, 000
AMA484 Decision Analysis 30
and
• Even if a decision maker’s utility function does not have these values, we can
transform his utility function into a utility function having u = 1.
AMA484 Decision Analysis 33
Lemma 1
Lemma 1 Given a utility function u(x), a real number a > 0 and any b, we define
v(x) = au(x) + b. Given any two lotteries L1 and L2 it will be the case that
1. A decision maker using u(x) as his utility function will have L1 pL2
a decision maker using v(x) as his utility function will have L1 pL2 .
2. A decision maker using u(x) as his utility function will have L1 iL2
a decision maker using v(x) as his utility functions will have L1 iL2 .
AMA484 Decision Analysis 34
Proof. Let
L1 = (p1 , r1 ; p2 , r2 ; · · · ; pn , rn ),
L2 = (p′1 , r1′ ; p′2 , r2′ ; · · · ; p′m , rm
′
).
Suppose the decision maker using u(x) prefers L1 to L2 . Then by the expected utility
criterion, we know that
n
X m
X
pi u(ri ) > p′j u(rj′ ). (1)
i=1 j=1
Since
n
X m
X
pi = p′j = 1,
i=1 j=1
Since a > 0, the inequality (3) follows from (1). Thus, if the u(x) decision maker has
L1 pL2 , the v(x) decision maker has L1 pL2 .
Similarly, if (3) holds, then (1) will hold. Thus, if v(x) decision maker will also have
L1 pL2 . A similar argument can be used to prove part (2) of Lemma 1.
AMA484 Decision Analysis 36
• Using Lemma 1, we can show that without changing how an individual ranks
lotteries, we can transform the decision maker’s utility function into one having u
(least favorable outcome) = 0 and u (most favorable outcome) = 1.
AMA484 Decision Analysis 37
Utility Function
Let C be a set of consequences (monetary values) and u : C → [0, +∞) be a function.
For example,
Utility of a Lottery
A lottery is L = {(pi , ri ) : i = 1, 2, · · · , n}, where ri is the value and pi is the
corresponding probability. Its utility value is given by
n
X
U (L) = pi u(ri ).
i=1
Let u : C → [0, +∞). Then u is a utility function if, for any two lotteries
L = {(pi , ri ) : i = 1, 2, · · · , n}, L = {(pj , rj ) : j = 1, 2, · · · , m},
′ ′ ′ ′′ ′′ ′′
n
X Xm
U (L′ ) = p′i u(ri′ ), U (L′′ ) = p′′j u(rj′′ ),
i=1 j=1
we have
′ ′′
L pL
>
L′ iL′′ ⇐⇒ U (L′ ) = U (L′′ ).
′′ ′
L pL <
AMA484 Decision Analysis 39
u is Monotone
The monotonicity of u means
c⋆ c⋆
AMA484 Decision Analysis 42
Risk neutral
Risk neutral (DM considers every lottery as its EMV):
Risk averse (DM considers no lottery more desirable than its EMV):
c⋆ c⋆
AMA484 Decision Analysis 44
c⋆ c⋆
AMA484 Decision Analysis 46
Increase in utility
0.8
through winning
0.6
0.4
Decrease in utility
through losing the 0.2
gamble
0 $
0 1000 2000
AMA484 Decision Analysis 47
By the EMV criterion, he should be indifferent between keeping his money and
gambling.
AMA484 Decision Analysis 48
When applying the utility function to the decision, it provides $1, 000 with a utility of
0.9.
If gambling, he has a 50% chance of increasing his assets so that the utility is increased
to 1 and a 50% chance of ending with assets with a utility of 0.
Hence the expected utility of the gamble is
which is less than 0.9. Thus, by the (maximum) expected utility criterion, the certain
money is more attractive than the risky option of gambling.
AMA484 Decision Analysis 49
Relation Between an Individual’s Utility Function and His Attitude Toward Risk
where EM V (L) is the expected monetary value of the lottery’s outcomes and CE(L)
is the cash equivalence of L.
• Let a nondegenerate lottery be any lottery in which more than one outcome can
occur.
Example
0.5
$30, 000
L
0.5
−$10, 000
Then
EM V (L) = $30, 000 × 0.5 + (−$10, 000) × 0.5 = $10, 000.
We have already see that CE(L) = −$3, 400. Thus,
Jill values L at $13, 400 less than its expected value, because she does not like the large
degree of uncertainty that is associated with the reward yielded by L.
AMA484 Decision Analysis 51
Example
Utility
RP (L)
E(U for L) Point 1
p
x1
L (Assume x1 < x2 )
1−p
x2
Suppose u(x) is strictly concave. Then, from the above figure, we see that
Since CE(L) is the value x⋆ having u(x⋆ ) = E(U for L), the figure shows that
CE(L) < EM V (L), so RP (L) > 0.
This follows because the strict concavity of u(x) implies that the line segment joining
the points (x1 , u(x1 )) and (x2 , u(x2 )) lies below the curve u(x).
AMA484 Decision Analysis 53
We can also give an algebraic proof that u(x) strictly concave implies that
RP (L) = EM V (L) − CE(L) > 0. Recall that for
p
x1
L
1−p
x2
EM V (L) = px1 + (1 − p)x2 . Now the strict concavity of u(x) implies that
Thus, the decision maker prefers px1 + (1 − p)x2 = EM V (L) with certainty to the
prospect of playing L.
The certainty equivalent of L must be less than px1 + (1 − p)x2 = EM V (L).
This implies that RP (L) = EM V (L) − CE(L), and the decision maker exhibits
risk-averse behavior.
AMA484 Decision Analysis 54
Example
1
Joan’s utility function for her asset position x is given by u(x) = x 2 .
Currently, Joan’s assets consist of $10, 000 in cash and a $90, 000 home.
During a given year, there is a 0.001 chance that Joan’s home will be destroyed by fire
or other causes.
How much would Joan be willing to pay for an insurance policy that would replace her
home if it were destroyed?
AMA484 Decision Analysis 55
Example
Let x = annual insurance premium. then Joan must choose between the following
lotteries:
Asset Position
1
Buy insurance: L1 $100, 000 − x
0.001
$100, 000 − $90, 000 = $10, 000
Do not buy insurance: L2
0.999
$100, 000
AMA484 Decision Analysis 56
Suppose both sides of the last inequality we find that L1 pL2 if and only if
Thus, Joan would pay up to $136.71 for insurance. Of course, if x = $136.71, L1 iL2 .
AMA484 Decision Analysis 57
Risk Premium
Let’s compute the risk premium for L2 :
Thus
CE(L) = 316.011542 = $99, 863.29,
and
Therefore, Joan is willing to pay for annual home insurance $46.71 more than the
expected loss of $90.
(Recall that Joan was willing to pay up to $90 + $46.71 = $136.71 to avoid the risk
involved in her home being destroyed.)
Joan exhibits risk-averse behavior (RP (L) > 0). Since
− 32
−x
u′′ (x) = < 0,
4
u(x) is strictly concave, and RP (L) > 0 would hold for any nondegenerate lottery.
AMA484 Decision Analysis 59
Exponential Utility
• One important class is called exponential utility and has been used in many
financial investment analyzes.
• An exponential utility function has only one adjustable numerical parameter, and
there are straightforward ways to discover the most appropriate value of this
parameter for a particular individual or company.
• A person with a large value of R is more willing to take risks than a person with a
small value R.
AMA484 Decision Analysis 61
It has been shown that the risk tolerance ≈ dollar amount R such that the decision
maker is (seem to be) indifferent between the following two options
• u(0) = 0
A second tip for finding R is based on empirical evidence found by Ronald Howard, a
prominent decision analyst.
He discovered tentative relationships between risk tolerance and several financial
variables – net sales, net income, and equity.
specifically, he found that R was approximately 6.4% of net sales, 124% of net
income, and 15.7% of equity for the companies he studied.
For example, according to to this prescription, a company with net sales of $30 million
should have a risk tolerance of approximately $1.92 million.
Howard admits that these percentages are only guidelines.
However, they do indicate that larger and more profitable companies tend to have
larger values of R, which means that they are more willing to take risks involving
given dollar amounts.
AMA484 Decision Analysis 63
In reality, many people exhibit both risk-seeking behavior (they purchase lottery
tickets, go to Las Vegas) and risk-averse behavior (they buy home insurance).
A person whose utility function contains both convex and concave segments may
exhibit both risk-averse and risk-seeking behavior.
AMA484 Decision Analysis 64
Example
Consider a decision maker whose utility function u(x) for change in current asset
position is given the figure below.
Utility
1 •
u(−2, 000) = 0
u(−300) = 0.18
0.8 u(−200) = 0.19
u(0) = 0.2
u(2, 500) = 0.5
0.6
•
0.4
0.2 •••
0• x
−2, 000 0 2, 000 4, 000 6, 000
AMA484 Decision Analysis 65
Now suppose the decision maker can, for $200, insure himself against a loss of
$2, 000, which occurs with probability 0.08. Then he must choose between
0.08
−$2, 000
1
L3 −$200 and L4
0.92
$0
From the figure, u(−200) = 0.19, u(0) = 0.20 and u(−2, 000) = 0. Thus,
E(U for L3 ) = 0.19 and E(U for L4 ) = 0.80 × 0 + 0.92 × (0.20) = 0.184, and
L3 pL4 .
This shows that CE(L4 ) < −$200. Since EM V (L4 ) = 0.08(−2, 000) + 0.92(0) =
−$160, RP (L4 ) = EM V (L4 ) − CE(L4 ) > 0, and the decision maker is exhibiting
risk-averse behavior, because u(x) is concave for −2, 000 < x < 0.
Thus, if his utility function has both convex and concave segments, a person can
exhibit both risk-seeking and risk-averse behavior.
AMA484 Decision Analysis 67
Prospect Theory
Here is one example of a decision that cannot be explained by EM U . Ask a person to
choose between lottery 1 and lottery 2:
Now let u(0) = 0 and u(45) = 1. A decision maker following EM U will choose
lottery 1 over lottery 2 if and only if
A decision maker following EM U will choose lottery 3 over lottery 4 if and only if
This implies that a believer in EMU cannot choose lottery 1 over lottery 2 and lottery 3
over lottery 4. Thus, for this situation, the choices of most people contradict EMU.
AMA484 Decision Analysis 70
π(p)
0.8 p π(p)
0.6
0.4
0.2
p
0 0.2 0.4 0.6 0.8 1
AMA484 Decision Analysis 71
• The shape of the π(p) function in the figure implies that individuals are more
sensitive to changes in probability when the probability of an event is small (near
0) or large (near 1).
From the value of π(p) given in the figure, we can compare the expected “prospects”
of lottery 1 versus lottery 2 and lottery 3 versus lottery 4.
Thus, lottery 1 is preferred to lottery 2 if u(30) > 0.602, while lottery 3 is preferred to
0.258
lottery 4 if 0.258 > 0.293 × u(30) or u(30) < 0.293 = 0.88.
Our paradox evaporates, because for many people, u(30) will be between 0.602 and
0.88.
AMA484 Decision Analysis 73
Framing
• Kahneman and Tversky’s idea of framing is based on the fact that people often set
their utility functions from the standpoint of a frame from which they view the
current situation.
• Most people’s utility functions treat a loss of a given value as being more serious
than a gain of an identical value.
AMA484 Decision Analysis 74
Framing
Utility
outcome
Losses Gains
Reference point
AMA484 Decision Analysis 75
To see how framing can explain the failure of EM U , consider the following problem
that Kahneman and Tversky gave to a group of students. The U S is preparing for the
outbreak of a disease that is expected to kill 600 peoples. Two alternative programs
have been proposed
Most students preferred program 1, probably because with program 2 there is a large
risk of saving nobody. Since the programs are phrased in terms of lives saved, most
people take the frame or reference point for this problem to be no lives saved or 600
people dead. Since the effect of each program is expressed in gains, and the utility
function is concave for gains, we find that
2 1 1 2 1
u(200) = u(( )0 + ( 600)) > ( )u(600) + ( )u(0) = ( )u(600).
3 3 3 3 3
This implies, of course, that the person chooses program 1 over program 2.
AMA484 Decision Analysis 76
Now most people choose program 2. Note that both program1’s are identical, as are
both program 2’s. Why do most people choose program 2 for the second phrasing of
the alternatives? The second phrasing shifts most people’s reference points from “No
lives saved” (in first phrasing) to “Nobody dies”. The outcomes are expressed as losses
(deaths), so the convexity of the utility curve for losses implies that
2 2 1 2 1
( )u(−600) = ( )u(−600) + ( u(0)) > u(( )(−600) + (0)) = u(−400)
3 3 3 3 3
This implies, of course, that the person chooses program 2 over program 1.
AMA484 Decision Analysis 77
Prospect theory