Expected Utility Theory Finance

II
Fadi Zaher
University of Skövde, 2006

1
 References
• K.J. Arrow (1963) ”Uncertainty and the
Welfare Economics of Medical Care”,
American Economic Review, Vol. 53, p.941-
73.
• Mas-Colell, Winston and Green,
”Microeconomics Theory” Chapter 6.
• Varian Hall, Intermediate Microeconomics
• Varian Hall, Advance Microeconomics

2
 First part
Expected Utility Theory

- I- Introduction

- II- Risk aversion, certainty equivalent and

concavity

- III- vNM

3
I- Introduction and General Notions
• 1- Expected Utility Theory and
Intuitive questions
• 2- Risk vs. Uncertainty
• 3- Probability Theory in Finance
• 4- EUT and vNM
• 5- Lotteries

4
1- What is Expected Utility Theory?
• Expected Utility Theory (EUT) states that
the decision maker (DM) chooses between
risky or uncertain prospects by comparing
their expected utility values.

• In other words, the weighted sums

obtained by adding the utility values of
outcomes multiplied by their respective
probabilities.

5
 1- Some intuitive questions
• First, what do the utility numbers in the
formula refer to?

• In particular, do they belong to the same

value scale as do the utility numbers that
represent the DM's choices under
certainty?

6
 1- Some intuitive questions
• Second, is the weighted sum procedure
of combining probability and utility values
the only one to be considered?

• In particular, if there are indeed alternative,

intuitively attractive modeling, how will the
theorist arbitrate this conflict?

7
 1- Some intuitive questions
• Third, should it be taken for given that the DM relies on
probability values, or are there again alternative constructions that
provides theoretical comparison?

• This question is normally addressed only in the context of

uncertainty.

• Uncertainty is not the same as risk when the probabilities are not
explicitly part of the agent's decision problem. If the probabilities
are explicitly part of the agent decision problem, then
uncertainty=risk.

• For this distinction, there are two well-received versions of the

theory:
1. Subjective Expected Utility Theory (SEUT) in the case of
uncertainty,
2. von Neumann- Morgenstern Theory (VNMT) in the case of risk.
8
 2- Risk versus uncertainty
• In his seminal work Risk, Uncertainty, and Profit, Frank Knight
(1921) established the important distinction between risk and
uncertainty:

"… Uncertainty must be taken in a sense radically distinct from the

familiar notion of Risk, from which it has never been properly
separated. … The essential fact is that "risk" means in some cases a
quantity susceptible of measurement, while at other times it is
something distinctly not of this character; and there are far-reaching
and crucial differences in the bearings of the phenomena depending
on which of the two is really present and operating. … It will appear
that a measurable uncertainty, or "risk" proper, as we shall use
the term, is so far different from an unmeasurable one that it is not in
effect an uncertainty at all."

Conclusion of the text:

Risk = Measurable uncertainty => Quantifiable
Uncertainty = Unmeasurable => Not Quantifiable
9
 There are other measures of uncertainty

1. In stochastic modeling, risk is an uncertainty for which

probability can be calculated (with past statistics for
example) or at least estimated (doing projection
scenarios) mathematically.
So, it is measurable!!!

2. In insurance, risk deals only with negative uncertainty

(those bringing loss or harm).

3. In cognitive psychology, uncertainty can be real, or just a

matter of perception, such as expectations threats, etc.

10
 3- Probability theory in Finance
• In probability theory (in finance), the expected
value (or mathematical expectation, E) of a
random variable is the sum of the probability
of each possible outcome of the experiment
multiplied by its payoff ("value").

• Thus, it represents the average amount one

"expects" to win per bet if bets with identical
odds are repeated many times.

11
 4- EUT and vNM
• In the EUT, vNM proved that any "normal" preference
relation over a finite set of states can be written as an
expected utility.

• Therefore, it is also called von-Neumann Morgenstern

utility.

• A related concept is the certainty equivalent of a

gamble.

• The more risk averse a person is, the more he will be

prepared to pay to eliminate risk.

12
 Example 1
• What is best?
1. Accept €1 today or
2. having 50% chance of getting €2 tomorrow.
Note that the expected value is the same for both
1 & 2.

• People may be risk averse or risk loving

depending on the amounts involved.

• This is why people may buy an insurance policy

and a lottery ticket on the same day.

13
 Example 2: The St Petersburg Paradox
• A hypothetical gamble

• Suppose that I offers you this gamble:

• I have a “fair” coin here.

• I will flip it, and if it is tail I pay you €1 and the

gamble is over.

• If it is head, I will flip again.

14
 Example 2: The St Petersburg Paradox

• If it is tail then, I pay you €2, if not I will flip again.

• With every round, I double the amount I will pay to you if

it is tail.

• Sounds like a good deal!

• After all, you can not loose. So the question is:

How much are you willing to pay to take this gamble?

15
 The expected value of the gamble
• The gamble is risky because the payoff is
random.

• So, according to intuition, this risk should be

taken into account, meaning, you will pay
less than the expected payoff of the gamble.

• So, if the expected payoff is X, you should be

willing to pay at most X, possibly minus some
risk premium.

• BUT, the expected payoff of this gamble is

INFINITE!
16
 Infinite expected value
• With probability 1/2 you get €1. 12 1 times 20
• With probability 1/4 you get €2. 12 2 times 21
• With probability 1/8 you get €4. 12 3 times 22
• etc.
 The expected payoff is the sum of these
payoffs, weighted with their probabilities.


t 
 so 
1
 t 2)
 t 1 
1
   2 ( n 
t 1 2
  t 1
probability payoff 17
 An infinitely valuable gamble?
• You should pay everything
probability
that you own to purchase the
0,5
right to take this gamble!
0,4

• Yet, in practice, no one is 0,3

prepared to pay such a high
0,2
price.
0,1

• Why? 0
0 20 40 60 €

• Even though the expected

• With 7% probability we
payoff is infinite, the get €8, with 1%
distribution of payoffs is not probability we get €64.
attractive…
18
 5- Lotteries
• Suppose you are driving to work at University of Skövde from
Stockholm!
– If you arrive on time you get payoff= x , (prob.=95%).
– If there is a traffic jam (prob=4.8%) you get nothing.
– If you have an accident (prob =0.2%) , you get no payoff but
also have to spend to repair your car.
• This lottery can be written as:[+x,0.95; 0,0.048; -y,0.02]
• Let us consider a finite set of outcomes: [x1,..xs]
• The xi ‘s can be consumption bundles or in our case money - the
xi‘s themselves involve no uncertainty.
• We define a lottery as:
S

x 
1 1
; xS  S  ,   0,   s  1
s 1 19
 Preferences over Lotteries
• Let us call the set of all such lotteries as . (We
now assume that agents have preferences over
this set.)

• So agents have a preference relation  on  that

satisfies the usual assumptions of ordinal utility
theory; (rationality and continuity).

20
 Preferences over Lotteries
• Assumptions imply that we can represent such
preferences by a continuous utility function  :   so
that   ’  () < (’)

• We also assume that people prefer more to less (in our

case more money to less):
 1  0, a  0  V  x1 1; x2 2   V  x1  a,  1; x2 2 
• Define the expected value of a lottery as:
S
E  L    s xs
s 1

21
II- Risk Aversion and certainty
equivalent
• 1- Risk Aversion
• 2- Certainty Equivalent
• 3- Concavity

22
 1. What is risk aversion?
• Consider the lottery [{E(L},1}.

• This lottery pays E{L} with certainty or (Outcome=E(L),

probability=1).

• We define risk attitude related to this lottery and how agents prefer
outcomes relative to this lottery.

• Risk Neutral:  (L)=  ([E{L},1]). =>

• Thus, the variation in payoff between states is irrelevant to the

agent. =>

• => The agent cares only about the expectation of the prize.

23
 What is risk aversion?
• Risk Averse:  (L)<  ([E{L},1]).

• The agent would rather have the average

prize E{L} for sure than bear the risk in the
lottery L. =>

• => A risk averse agent is willing to give up

some wealth on average in order to avoid the
randomness of the prize of L.
24
 Risk - Averse/Neutral/Lover
u
i
Lottery 1: $0.5M w.p. 1
1 Risk averse
Lottery 2: $1M w.p. 0.5
$0 w.p. 0.5 Risk neutral
0.5
Agent’s strategy is the Risk seeking
choice of lottery
0
M$
0 0.5 1

25
 2. Certainty equivalent
• Let  be some utility function on  (set of all lotteries) and let L
be some lottery with expected prize E{L}.

• The certainty equivalent of L under  is defined as:

 ([CE[L],1) =  (L).

• CE(L) is the level of non-random wealth that yields the same

utility as the lottery L.

26
 Certainty equivalent
• The risk premium is the difference between the expected
prize of the lottery and its certainty equivalent: RP(L)
=E{L}-CE(L).

• All of this is the same as ordinal utility theory.

• We have not used the additional structure in the

probabilities. We now do this very soon.

27
 3. Concavity
• The certainty
equivalent is the level v(x)
of wealth that gives
the same utility as the
lottery on average.
v(x1)
Formally:
v  CE( x)   E v( x)
E{v(x)}

• We can explicitly
solve for the CE as: v(x0)
CE ( x)  v 1
 E v  x  x0 E{x} x1 x
28
v-1(E{v(x)})
 3. Concavity
• We now see that the agent is
risk averse iff  is a concave
function. v(x)

• Jensen’s inequality: strict

convex combination of two
values of a function is strictly v(x1)
below the graph of the function
then the function is concave.

E{v(x)}
• The risk premium is therefore
positive and the agent is risk
averse if  is strictly concave.
v(x0)
• If ’’ =0, then CE(x)=E{x} and
the RP=0 or risk neutrality. x0 E{x} x1 x
29
v-1(E{v(x)})=CE(x)
 Absolute Risk Aversion
• We define the coefficient of Absolute Risk Aversion (ARA) as a local
measure of the degree that an agent dislikes risk.

v( w)
A( w)  
v( w)
• A has many useful properties
− Its is invariant under an affine transformation or if u and v are
two vNM utility functions then ARA of u and v are the same.
− We can use the ARA then for interpersonal comparisons.
− Suppose Mr. X and Mr. Y have the same endowments but
different preferences. X’s utility function v is more concave than
Y’s ( say u- is more concave) so X always demands a higher
risk premium for a given level of risk.
− Here then ARA for v(w) is larger than the ARA for u(w).
30
 Precautionary Saving
• Coefficients of risk aversion measure the disutility arising from
small amount of risk imposed on agents or how much an agent
dislikes risk.
• Coefficients do not tell us about how the behavior of agents
changes when we vary the amount of risk the agent is forced to
bear.
• Example: It may be reasonable for agents to accumulate some
“precautionary” saving when facing more uncertainty.
• More risk induces a more prudent agent to accumulate
precautionary savings.
• Kimball’s coefficient of absolute prudence: v( w)
P( w)  
• An agent is prudent iff this coefficient is positive. v( w)
• The precautionary motive is important because it means that
agents save more when faced with more uncertainty.
31
 III- vNM Utility Theory
• 1- Expected Utility Representation
• 2- State Independence
• 3- Consequentialism
• 4- Irrelevance of common
alternatives
• 5- Conclusion on vNM Utility Theory
32
 1. vNM: Expected utility representation
• So far we have used ordinal utility theory as well as looking deeper
into vNM theory.

• We want to represent agent’s preferences by evaluating the

expected utility of a lottery.

• We need a function  that maps the single outcome xs to

some real number  (xs), and then we compute the expected
value of  .

• Formally, function  is the expected utility representation of  if :

S
V  x 
1 1 ; xS  S     s v ( xs )
s 1
• Von Neumann and Morgenstern first developed the use of an
expected utility under some conditions. Lets look at these briefly.
33
 2. vNM axioms: State
independence
• von Neumann and Morgenstern’s have presented a model
that allows the use of an expected utility under some
conditions.

• The first assumption is state independence.

 x 1- y

~
1- 
y x

• The most important things for the agent is the statistical

distribution of outcomes.

• A state is just a label and has no particular meaning and are

interchangeable (as in x and y in the diagram). 34
 3. vNM axioms: consequentialism
• Consider a lottery L, whose prizes are further lotteries L1 and L2: L =
[L1,p1;L2,p2] => a compound lottery.

• We assume that an agent is indifferent between L and a one-shot lottery

(reduced lottery) with four possible prizes and compounded
probabilities.

• An agent is indifferent between the two lotteries shown in the diagram below.

• Agents are only interested in the distribution of the resulting prize, but not in
the process of gambling itself.

11 x1 1 11 x1
L1
1
12 1 12 x2
x2
21 ~ 2 21
x3 x3
2
L2 2 22
22 x4 x4 35
4. vNM axioms: irrelevance of common
alternatives

• This axiom says that the ranking of two lotteries should depend
only on those outcomes where they differ.

• If L1 is better than L2, and if we compound each of these

lotteries with some third common outcome x, then it should be
true that [L1,p;x,1-p] is still better than [L2,p;x,1-p]. The common
alternative x should not matter.

36
 5. Conclusion on vNM utility theory
• State-independence, consequentialism and the irrelevance of
common alternatives + the assumptions on preferences  give rise
to the famous results of vNM- The utility function  has an expected
utility representation  such that:
S
V  x 1 1 ; xS  S    s v ( xs )
s 1
• The utility function is on the space of lotteries  which represents
the preference relation between lotteries and is an ordinal utility
function.
•  (L) is an ordinal measure of satisfaction and can be compared
only in the sense of ranking lotteries.
•  is also invariant to monotonic transformations.
37
 5. Conclusion on vNM utility theory
• The vNM utility function  has more structure.
• It represents  as a linear function of probabilities.
• As a result  is not invariant under an arbitrary monotonic transformation.
• It is invariant only under positive affine transformations.
• Hence vNM utility is cardinal.
• What does this mean?
• Cardinal numbers are measurements that are ordinal but whose difference can
also be ordered.
• (Not necessarily the case with ordinal utility) With cardinal utility we can have
the following:

v  x1   v  x2  , v  x3   v  x4   v  x1   v  x2   v  x3   v  x4 

38