Chapter 5 Choices Under Uncertainty

Advanced
Microeconomic
Theory
Chapter 5: Choices under Uncertainty
Outline
• Simple, Compound, and Reduced Lotteries
• Independence Axiom
• Expected Utility Theory
• Money Lotteries
• Risk Aversion
• Prospect Theory and Reference-Dependent
Utility
• Comparison of Payoff Distributions
Advanced Microeconomic Theory 2

Simple, Compound, and
Reduced Lotteries

Simple Lotteries
• Consider a set of possible outcomes (or
consequences) 𝐶.
• The set 𝐶 can include
– simple payoffs 𝐶 ∈ ℝ (positive or negative)
– consumption bundles 𝐶 ∈ ℝ$
• Outcomes are finite (𝑁 elements in 𝐶, 𝑛 =
1,2, … , 𝑁)
• Probabilities of every outcome are objectively
known
– 𝑝- for outcome 1, 𝑝. for outcome 2, etc.
Simple Lotteries
• Simple lottery is a list
𝐿 = 𝑝- , 𝑝. , … , 𝑝0
with 𝑝1 ≥ 0 for all 𝑛 and ∑0 15- 𝑝1 = 1, where
𝑝1 is interpreted as the probability of
outcome 𝑛 occuring.
• In some books, lotteries are described
including the outcomes too.

Simple Lotteries
• A simple lottery with 2
possible outcomes
• “Degenerated” p2 (0,1)
probability pairs 1
– at (0,1), outcome 2 = { p R 2 : p1 + p2 = 1}
happens with certainty. p2
– at (1,0), outcome 1
happens with certainty.
(1,0)
• Strictly positive
probability pairs p1
p1 1
– Individual faces some
uncertainty, i.e., 𝑝- +
𝑝. = 1
Simple Lotteries
• A simple lottery with 3
possible outcomes (i.e., p3
3-dim. simplex). 1
(0,0,1)
• Intercepts represent
degenerated probabilities
where one outcome is
certain. p3
(1,0,0)
• Points strictly inside the (0,1,0)
hyperplane connecting 1
p2 p1
1
the three intercepts p1 p2

= {p 0 : p1 + p2 + p3 = 1}
denote a lottery where
the individual faces
uncertainty. Advanced Microeconomic Theory 7
Simple Lotteries
• 2-dim. projection of
the 3-dim. simplex
• Vertices represent the 3
intercepts where x1 + x2 + x3 = 1
• The distance from a
given point to the side x2 x1
of the triangle
measures the x3
probability that the 1 2
outcome represented L = ( p1 , p2 , p3 )
at the opposite vertex

occurs.
Simple Lotteries
• A lottery lies on one
of the boundaries of
3
the triangle:
– We can only
construct segments
connecting the
lottery to two of the x1
x2
outcomes.
– The probability 1 x3 = 0 2
associated with the x1 + x2 = 1

third outcome is zero.
Compound Lotteries
• Given simple lotteries
7
𝐿7 = 𝑝-7 , 𝑝.7 , … , 𝑝0 for 𝑘 = 1,2, … , 𝐾
and probabilities 𝛼7 ≥ 0 with ∑; 15- 𝛼7 = 1, then the
compound lottery 𝐿- , 𝐿. , … , 𝐿; ; 𝛼- , 𝛼. , … , 𝛼; is the
risky alternative that yields the simple lottery 𝐿7 with
probability 𝛼7 for 𝑘 = 1,2, … , 𝐾.
– Think about a compound lottery as a “lottery of lotteries”:
first, I have probability 𝛼7 of playing lottery 1, and if that
happens, I have probability 𝑝-7 of outcome 1 occurring.
– Then, the joint probability of outcome 1 is
𝑝- = 𝛼- = 𝑝-- + 𝛼. = 𝑝-. + ⋯ + 𝛼; = 𝑝-;

Compound and Reduced Lotteries
• Given that interpretation, the following result
should come at no surprise:
– For any compound lottery
𝐿- , 𝐿. , … , 𝐿; ; 𝛼- , 𝛼. , … , 𝛼; , we can calculate its
corresponding reduced lottery as the simple lottery
𝐿 = 𝑝- , 𝑝. , … , 𝑝0 that generates the same ultimate
probability distribution of outcomes.
• The reduced lottery 𝐿 of any compound lottery
can be obtained by
𝐿 = 𝛼- 𝐿- + 𝛼. 𝐿. + ⋯ + 𝛼; 𝐿; ∈ ∆
• Example 1:
– All three lotteries are equally likely
- - - - - -
– P outcome 1 = = 1 + = + = =
H H I H I .
- - H - H -
– P outcome 2 = = 0 + = + = =
H H J H J I
- - H - H -
– P outcome 3 = = 0 + = + = =
H H J H J I
/ 3 L1 = (1, 0, 0)
=1
1 Reduced Lottery
= 1/ 3 1 3 3 1 1 1
2
L2 = , , , ,
3 = 1/ 3 4 8 8 2 4 4
1 3 3
L3 = , ,
4 8 8
• Example 1 (continued):
– Probability simplex of 3
the reduced lottery of

a compound lottery
– Reduced lottery 𝐿
assigns the same L2 = L3
probability weight to x2
each simple lottery. 1

1 1
2
1 1 1 1
L= L1 + L2 + L3 = , ,
L1 = (1, 0, 0) 3 3 3 2 4 4

• Example 2:
– Both lotteries are equally likely
1 1
/2 L4 = , ,0
=1 2 2
1 Reduced Lottery
1 1 1
, ,
2 = 1/ 2 2 4 4
1 1
L5 = , 0,
2 2
1 1 1 1 1
Outcome 1⟶ + =
x1
2 2 2 2 2 Outcome 3⟶ 1 0 + 1 1 = 1
1 1 1 1 2 2 2 4
Outcome 2⟶ + 0=
2 2 2 4
• Example 2 (continued):
– Probability simplex of 3
(0, 0,1)
the reduced lottery of L5 =

1 1
, 0,
a compound lottery 2 2
1 1 1 1 1
L4 + L5 = L = , ,
2 2 2 2 4
1 2
1 1 (0,1, 0)
(1, 0, 0) L4 = , ,0
2 2

• Consumer is indifferent between the two compound
lotteries which induce the same reduced lottery
– This was illustrated in the previous Examples 1 and 2
where, despite facing different compound lotteries,
the consumer obtained the same reduced lottery.
• We refer to this assumption as the Consequentialist
hypothesis:
– Only consequences, and the probability associated to
every consequence (outcome) matters, but not the
route that we follow in order to obtain a given
consequence.
Preferences over Lotteries
• For a given set of outcomes 𝐶, consider the
set of all simple lotteries over 𝐶, ℒ.
• We assume that the decision maker has a
complete and transitive preference relation ≿
over lotteries in ℒ, allowing him to compare
any pair of simple lotteries 𝐿 and 𝐿′.
– Completeness: For any two lotteries 𝐿 and 𝐿′,
either 𝐿 ≿ 𝐿′ or 𝐿O ≿ 𝐿, or both.
– Transitivity: For any three lotteries 𝐿, 𝐿′ and 𝐿′′, if
𝐿 ≿ 𝐿′ and 𝐿O ≿ 𝐿′′, then 𝐿 ≿ 𝐿′′.
• Extreme preference for certainty:
– 𝐿 ≿ 𝐿′ if and only if
max 𝑝1 ≥ max 𝑝1O
1∈0 1∈0
– The decision maker is only concerned about the

probability associated with the most likely
outcome.

• Smallest size of the support:
– 𝐿 ≿ 𝐿′ if and only if
supp(𝐿) ≤ supp(𝐿′)
where supp 𝐿 = 𝑛 ∈ 𝑁: 𝑝1 > 0 .
– The decision maker prefers the lottery whose
probability distribution is concentrated over the
smallest set of possible outcomes.

• Lexicographic preferences:
– First, order outcomes from most preferred (outcome
1) to least preferred (outcome 𝑛).
– Then 𝐿 ≿ 𝐿′, if and only if
𝑝- > 𝑝-O , or
If 𝑝- = 𝑝-O and 𝑝. > 𝑝.O , or
If 𝑝- = 𝑝-O and 𝑝. = 𝑝.O and 𝑝H > 𝑝HO , or
…
– The decision maker weakly prefers lottery 𝐿 to 𝐿′ if
outcome 1 is more likely to occur in lottery 𝐿 than in
lottery 𝐿′.
– If outcome 1 is as likely to occur in both lotteries, he
moves to outcome 2; and so on.
• The worst case scenario:
– First, attach a number 𝑣(𝑧) to every outcome 𝑧 ∈
𝐶, that is, 𝑣 𝑧 ∈ ℝ.
– Then 𝐿 ≿ 𝐿′ if and only if
min 𝑣 𝑧 : 𝑝 𝑧 > 0 > min 𝑣 𝑧 : 𝑝′ 𝑧 > 0
– The decision maker prefers lottery 𝐿 if the lowest
utility he can get from playing lottery 𝐿 is higher
than the lowest utility he can obtain from playing
lottery 𝐿′.
• Continuity of preferences over lotteries:
– Continuity 1: For any three lotteries 𝐿, 𝐿′, and 𝐿′′,
the sets
𝛼 ∈ 0,1 : 𝛼𝐿 + 1 − 𝛼 𝐿′ ≿ 𝐿′′ ⊂ [0,1] and
𝛼 ∈ 0,1 : 𝐿′′ ≿ 𝛼𝐿 + 1 − 𝛼 𝐿′ ⊂ [0,1]
are closed.
– Continuity 2: if 𝐿 ≻ 𝐿′, then there is a
neighborhoods of 𝐿 and 𝐿′, 𝐵(𝐿) and 𝐵(𝐿′), such
that for all 𝐿c ∈ 𝐵(𝐿) and 𝐿d ∈ 𝐵(𝐿′), we have
𝐿c ≻ 𝐿d .
• Small changes in the
probability 3
distribution of
lotteries 𝐿 and 𝐿′ do B ( L)
not change the L
preference over the B( L ') La
two lotteries. L' Lb
1 2

• Example:
3
B(L’)
1 2
If 𝐿 ≻ 𝐿′, then 𝐿c ≻ 𝐿d .
• The continuity assumption, as in consumer
theory, implies the existence of a utility
function 𝑈: ℒ → ℝ such that
𝐿 ≿ 𝐿′ if and only if 𝑈(𝐿) ≥ 𝑈(𝐿′)
• However, we first impose an additional
assumption in order to have a more
structured utility function.
– The following assumption is related with
consequentialism: the Independence axiom.
• Independence Axiom (IA): a preference
relation satisfies IA if, for any three lotteries 𝐿,
𝐿′, and 𝐿′′, and 𝛼 ∈ (0,1) we have
𝐿 ≿ 𝐿′ if and only if
𝛼𝐿 + 1 − 𝛼 𝐿′′ ≿ 𝛼𝐿′ + 1 − 𝛼 𝐿′′
• Intuition: If we mix each of two lotteries, 𝐿
and 𝐿′, with a third one (𝐿′′), then the
preference ordering of the two resulting
compound lotteries is independent of the
particular third lottery .
• 𝐿 ≿ 𝐿′ if and only if
𝛼𝐿 + 1 − 𝛼 𝐿O ≿ 𝛼𝐿′ + (1 − 𝛼)𝐿′′
3 3 3 3
1 2 2 1 1 2

• Example 1 (intuition):
– The decision maker prefers lottery 𝐿 to 𝐿O , 𝐿 ≿ 𝐿O
– Construct a compound lottery by a coin toss:
§ play lottery 𝐿 if heads comes up
§ play lottery 𝐿′′ if tails comes up
– By IA, if 𝐿 ≿ 𝐿O , then
- - OO - -
𝐿+ 𝐿 ≿ 𝐿′ + 𝐿′′
. . . .

• Example 2 (violations of IA):
– Extreme preference for certainty
– Consider two simple lotteries 𝐿 and 𝐿′ for which
𝐿 ∼ 𝐿O .
– Construct two compound lotteries for which
1 1 1 1
𝐿 + 𝐿 ≁ 𝐿′ + 𝐿
2 2 2 2
– If 𝐿 ∼ 𝐿O , then it must be that
max 𝑝- , 𝑝. , … , 𝑝1 = max 𝑝-O , 𝑝.O , … , 𝑝1O
• Example 2 (violations of IA):
- -
– Compound lottery 𝐿 + 𝐿 coincides with simple
. .
lottery 𝐿.
– Hence, max 𝑝- , 𝑝. , … , 𝑝1 is used to evaluate
lottery 𝐿.
- -
– But compound lottery 𝐿′ + 𝐿 is a reduced
. .
lottery with associated probabilities
1 O 1 1 O 1
max 𝑝- + 𝑝- , … , 𝑝1 + 𝑝1
2 2 2 2
which might differ from max 𝑝-O , 𝑝.O , … , 𝑝1O .
• Example 2 (violations of IA, a numerical example):
– Consider two simple lotteries
𝐿 = (0.4, 0.5, 0.1), 𝐿 = (0.5, 0, 0.5)
– Hence,
max 0.4, 0.5, 0.1 = 0.5 = max 0.5, 0, 0.5
implying that 𝐿 ∼ 𝐿O .
- -
– However, the compound lottery 𝐿′ + 𝐿 entails
. .
probabilities
0.4 + 0.5 0.5 + 0 0.1 + 0.5
, , = 0.45, 0.25, 0.3
2 2 2
implying that max 0.45, 0.25, 0.3 = 0.45.
• Example 2 (violations of IA, a numerical example):
– Therefore,
max 0.4, 0.5, 0.1 = 0.5 > 0.45 = max 0.45, 0.25, 0.3
- - - O -
and thus 𝐿 = 𝐿 + 𝐿 ≻ 𝐿 + 𝐿.
. . . .
– This violates the IA, which requires

- - - -
𝐿 + 𝐿 ∼ 𝐿′ + 𝐿
. . . .
• Example 3 (violations of IA, “worst case scenario”):
– Consider 𝐿 ≻ 𝐿O .
- -
– Then, the compound lottery 𝐿 + 𝐿 does not need
. .
- -
to be preferred to 𝐿′ + 𝐿.
. .
– Example:
§ Consider the simple lotteries 𝐿 = (1,3) and 𝐿′ = (10,0),
with probabilities (𝑝-, 𝑝.) and (𝑝-O, 𝑝.O), respectively.
§ This implies
min 𝑣 𝑧 : 𝑝 𝑧 > 0 = 1 for lottery 𝐿
min 𝑣 𝑧 : 𝑝′ 𝑧 > 0 = 0 for lottery 𝐿′
§ Hence, 𝐿 ≻ 𝐿O .
• Example 3 (violations of IA, “worst case scenario”):
– Example (continued):
- - -- H
§ However, the compound lottery 𝐿 + 𝐿′ is , ,
. . . .
H
whose worst possible outcome is , which is preferred
.
- -
to that of 𝐿 + 𝐿, which is 1.
. .
§ Hence, despite 𝐿 ≻ 𝐿O over simple lotteries,
- - - -
𝐿= 𝐿 + 𝐿 ≺ 𝐿 + 𝐿′,
. . . .
which violates the IA.

Expected Utility Theory

• The utility function 𝑈: ℒ → ℝ has the expected
utility (EU) form if there is an assignment of
numbers 𝑢- , 𝑢. , … , 𝑢0 to the 𝑁 possible
outcomes such that, for every simple lottery 𝐿 =
𝑝- , 𝑝. , … , 𝑝0 ∈ ℒ we have
𝑈 𝐿 = 𝑝- 𝑢- + ⋯ + 𝑝0 𝑢0
– A utility function with the EU form is also referred to
as a von-Neumann-Morgenstern (vNM) expected
utility function.
– Note that this function is linear in the probabilities.
• Hence, a utility function 𝑈: ℒ → ℝ has the expected
utility form if and only if it is linear in the probabilities,
i.e.,
; ;
𝑈 n 𝛼 7 𝐿7 = n 𝛼7 = 𝑈(𝐿7 )
75- 75-
for any 𝐾 lotteries 𝐿7 ∈ ℒ, 𝑘 = 1,2, … , 𝐾 and

probabilities 𝛼- , 𝛼. , … , 𝛼; ≥ 0 and ∑;
75- 𝛼7 = 1.
• Intuition: the utility of the expected value of the 𝐾

lotteries, 𝑈 ∑; 75- 𝛼7 𝐿7 , coincides with the expected
utility of the 𝐾 lotteries, ∑;75- 𝛼7 𝑈(𝐿7 ).

• Note that the utility of the expected value of playing
the 𝐾 lotteries is
;
𝑈 n 𝛼7 𝐿7 = n 𝑢1 = n 𝛼7 𝑝17
75- 1
7
where ∑7 𝛼7 𝑝17 is the total joint probability of

outcome 𝑛 occurring.

• Note that the expected utility from playing the 𝐾
lotteries is
;
n 𝛼7 = 𝑈(𝐿7 ) = n 𝛼7 = n 𝑢1 𝑝17
75- 7
1
where ∑1 𝑢1 𝑝17 is the expected utility from playing a

given lottery 𝑘.

• The EU property is a cardinal property:
– Not only rank matters, the particular number
resulting form 𝑈: ℒ → ℝ also matters.
• Hence, the EU form is preserved only under
increasing linear transformations (a.k.a. affine
transformations).
p: ℒ → ℝ is
– Hence, the expected utility function 𝑈
another vNM utility function if and only if
p 𝐿 = 𝛽𝑈 𝐿 + 𝛾
𝑈
for every 𝐿 ∈ ℒ, where 𝛽 > 0.
Expected Utility Theory:
Representability
• Suppose that the preference relation ≿ satisfies
rationality, continuity and independence. Then, ≿
admits a utility representation of the EU form.
• That is, we can assign a number 𝑢1 to every outcome
𝑛 = 1,2, … , 𝑁 in such a manner that for any two
lotteries
O
𝐿 = 𝑝- , 𝑝. , … , 𝑝0 and 𝐿′ = 𝑝-O , 𝑝.O … , 𝑝0
we have 𝐿 ≿ 𝐿O if and only if 𝑈 𝐿 ≥ 𝑈 𝐿′ , or
0 0
n 𝑝1 𝑢1 ≥ n 𝑝1O 𝑢1
15- 15-
• Notation: 𝑢1 is the utility that the decision maker
assigns to outcome 𝑛. It is usually referred as the
Bernoulli utility function.
Indifference Curves
• Let us next analyze the effect of the IA on
indifference curves over lotteries.
1) Indifference curves must be straight lines:
Recall that from the IA, 𝐿 ~ 𝐿O implies that
𝛼𝐿 + 1 − 𝛼 𝐿 ~ 𝛼𝐿 + (1 − 𝛼)𝐿O
$
for all 𝛼 ∈ 0,1 .

Indifference Curves
3
If L ~ L ', then L ~ L + (1 )L '
L'
1 2
Straight indifference curves

Indifference Curves
• Why indifference curves must be straight?
O - - O
– We have that 𝐿 ~ 𝐿 , but 𝐿 ≺ 𝐿 + 𝐿 . This is
. .
equivalent to
1 1 1 1 O
𝐿 + 𝐿 ≺ 𝐿 + 𝐿
2 2 2 2
– But from the IA we must have
1 1 1 1 O
𝐿 + 𝐿 ~ 𝐿 + 𝐿
2 2 2 2
– Hence, indifference curves must be straight lines
in order to satisfy the IA.
Indifference Curves
• Curvy indifference curves over lotteries are
incompatible with the IA
- -
– The compound lottery 𝐿 + 𝐿′ would not lie on
. .
the same indifference curve as lottery 𝐿 and 𝐿′.
– Hence, the decision maker is not indifferent
- -
between the compound lotteries 𝐿 + 𝐿′ and
. .
- -
𝐿 + 𝐿′.
. .

Indifference Curves
3
1 1
L '+ L
2 2
L L'
1 2
Curvy indifference curve

Indifference Curves
2) Indifference curves must be parallel lines:
If we have that 𝐿 ~ 𝐿O , then by the IA
1 2 1 2 OO
𝐿 + 𝐿′′ ~ 𝐿′ + 𝐿
3 3 3 3
§ That is, the convex combination of 𝐿 and 𝐿O with
a third lottery 𝐿OO should also lie on the same
indifference curve.
§ This implies that the indifference curves must be
parallel lines in order to satisfy the IA.
Indifference Curves
§ Nonparallel indifference curves are incompatible
with the IA.
– If compound lotteries 3
- . -
𝐿 + 𝐿′′ and 𝐿′ +
H H H
.
𝐿′′ lie on different
H
(nonparallel)
indifference curves,
then
1 2 1 2 OO
𝐿 + 𝐿′′ ≺ 𝐿′ + 𝐿
3 3 3 3 1 2
which violates the IA.

Violations of the IA:

– Despite the intuitive appeal of the IA, we
encounter several settings in which decision
makers violate it.
– We next elaborate on these violations.

Violations of the IA
• Allais’ paradox:
– Consider a lottery over three possible monetary
outcomes:
1st prize 2nd prize 3rd prize
$2.5mln $500,000 $0
– First choice set:

-t Ju -
𝐿- = (0,1,0) and 𝐿O- =( , , )
-tt -tt -tt
– Second choice set:
-- Ju -t ut
𝐿. = (0, , ) and 𝐿O. =( , 0, )
-tt -tt -tt -tt
– About 50% students surveyed expressed 𝐿- ≻ 𝐿O- and
𝐿O. ≻ 𝐿. .
– These choices violate the IA.
– To see this, consider that the decision maker’s
preferences over lotteries have a EU form. Hence, 𝐿- ≻
𝐿O- implies
10 89 1
𝑢v > 𝑢.v + 𝑢v + 𝑢t
100 100 100
Ju Ju
– By the IA, we can add 𝑢t − 𝑢 on both sides
-tt -tt v
89 89
𝑢v + 𝑢t − 𝑢v >
100 100
10 89 1 89 89
𝑢.v + 𝑢v + 𝑢t + 𝑢t − 𝑢v
100 100 100 100
Advanced Microeconomic Theory 100 51
– Simplifying
11 89 10 90
𝑢v + 𝑢t > 𝑢.v + 𝑢t
100 100 100 100
yz {| $} yz {| $~}
which implies 𝐿. ≻ 𝐿O. .

– Did your own choices violate the IA?

• Reactions to the Allais’ Paradox:
– Approximation to rationality: people adapt their
choices as they go.
– Little economic significance: the lotteries involve
probabilities that are close to zero and one.
– Regret theory: the reason why 𝐿- ≻ 𝐿O- is because
I didn’t want to regret a sure win of $500,000.
– Give up the IA in favor of a weaker assumption:
the betweenness axiom.

• Machina’s paradox:
– Consider that
Trip to Barcelona ≻ Movie about Barcelona ≻ Home
– Now, consider the following two lotteries
uu - uu -
𝐿- = ( , , 0) and 𝐿. = ( , 0, )
-tt -tt -tt -tt
– From the previous preferences over certain

outcomes, how can we know this individual’s
preferences over lotteries?
§ Using the IA.
– From 𝑇 ≻ 𝑀 and the IA, we can construct the
compound lotteries
99 1 99 1
𝑇+ 𝑀≻ 𝑀+ 𝑀
100 100 100 100
– From 𝑀 ≻ 𝐻 and the IA, we have
99 1 99 1
𝑀+ 𝑀≻ 𝑀+ 𝐻
100 100 100 100
– By transitivity,
99 1 99 1
𝑇+ 𝑀≻ 𝑇+ 𝐻
100 100 100 100
$‚ $}
– Hence, 𝐿- ≻ 𝐿. . Advanced Microeconomic Theory 55
– Therefore, for preferences over lotteries to be
consistent with the IA, we need 𝐿- ≻ 𝐿. .
– Many subjects in experimental settings would rather
prefer 𝐿. , thus violating the IA.
– Many people explain choosing 𝐿. over 𝐿- on
grounds of the disappointment they would
experience in the case of losing the trip to
Barcelona, and having to watch a movie instead.
§ Similar to regret theory.

• Dutch books:
– In the above two anomalies, actual behavior is
inconsistent with the IA.
– Can we then rely on the IA?
– What would happen to individuals whose
behavior violates the IA?
– They would be weeded out of the market because
they would be open to the acceptance of so-called
Dutch books, leading them to a sure loss of
money.
– Consider that 𝐿 ≻ 𝐿′. By the IA, we should have
𝛼𝐿 + 1 − 𝛼 𝐿 ≻ 𝛼𝐿 + (1 − 𝛼)𝐿O
$
– If, instead, the IA is violated, then
𝐿 ≺ 𝛼𝐿 + (1 − 𝛼)𝐿O
– Consider an individual with these preferences, who
initially owns lottery 𝐿.
– If we offer him the compound lottery 𝛼𝐿 + (1 −
𝛼)𝐿O , for a small fee $𝑥, he would accept such a
trade.
– After the realization stage, he owns either 𝐿 or 𝐿′
§ If 𝐿′, then we offer 𝐿 again for $𝑦.
§ If 𝐿, then we offer 𝛼𝐿 + (1 − 𝛼)𝐿O for $𝑦.
– Either way, he is at the same position as he started
(owning 𝐿 or 𝛼𝐿 + (1 − 𝛼)𝐿O ), but having lost
$𝑥 + $𝑦 in the process.
– We can repeat this process ad infinitum.
– Hence, individuals with preferences that violate
the IA would be exploited by microeconomists
(they would be a “money pump”).
• Further reading:
– “Developments in non-expected utility theory:
The hunt for a descriptive theory of choice under
risk” (2000) by Chris Starmer, Journal of Economic
Literature, vol. 38(2)
– Choices, Values and Frames (2000) by Nobel prize
winners Daniel Kahneman and Amos Tversky,
Cambridge University Press.
– Theory of Decision under Uncertainty (2009) by
Itzhak Gilboa, Cambridge University Press.
Theories Modifying
1) Weighted utility theory:
– The payoff function from playing lottery 𝐿 is
𝑉 𝐿 = n 𝑤‡ = 𝑢(𝑥‡ )
ˆ∈‰
where
Š ˆ‹ Œ(ˆ‹ )
𝑤‡ = ∑•∈Ž Š(ˆ‹ )Œ(ˆ‹ )
and 𝑔: 𝐶 → ℝ
– The utility of outcome 𝑥‡ ∈ 𝐶 is weighted
according to:
a) its probability 𝑝 𝑥‡
b) outcome 𝑥‡ itself through function 𝑔: 𝐶 → ℝ
Theories Modifying
– Example: Consider a lottery with two payoffs 𝑥- and 𝑥.
with probabilities 𝑝 and 1 − 𝑝. Then, the weighted utility
is
𝑉 𝐿 = 𝑤- 𝑢 𝑥- + 𝑤. 𝑢 𝑥.
𝑔 𝑥- 𝑝
= 𝑢 𝑥-
𝑔 𝑥- 𝑝 + 𝑔 𝑥. 1 − 𝑝
𝑔 𝑥. 1 − 𝑝
+ 𝑢(𝑥. )
𝑔 𝑥- 𝑝 + 𝑔 𝑥. 1 − 𝑝
If 𝑔(𝑥‡ ) = 𝑔(𝑥• ) for any 𝑥‡ ≠ 𝑥• , then
𝑉 𝐿 = 𝑝𝑢 𝑥- + 1 − 𝑝 𝑢 𝑥.
which is a standard expected utility function.
Theories Modifying
• The weighted utility theory relies on the same
axioms as expected utility theory, except for
the IA, which is relaxed to the “weak
independence axiom.”
– Weak independence axiom: if we have that
𝐿- ~𝐿. , we can find a pair of probabilities 𝛼 and 𝛼′
such that
𝛼𝐿- + 1 − 𝛼 𝐿H ~ 𝛼 O 𝐿. + 1 − 𝛼 O 𝐿H
– The IA becomes a special case if 𝛼 = 𝛼′.
Theories Modifying
2) Rank dependent utility theory:
– First, rank the outcomes 𝑥- , 𝑥. , … , 𝑥1 from worst (𝑥- ) to
best (𝑥1 )
– Second, apply a probability weighting function
𝑤‡ = 𝜋 𝑝‡ + ⋯ + 𝑝1 − 𝜋 𝑝‡“- + ⋯ + 𝑝1
𝑤1 = 𝜋 𝑝1
where 𝜋(⋅) is a non-decreasing transformation function,
with 𝜋(0) = 0 and 𝜋(1) = 1.
– Finally, a rank-dependent utility is
𝑉 𝐿 = n 𝑤‡ ⋅ 𝑢(𝑥‡ )
ˆ∈‰
Theories Modifying
– For a lottery with two outcomes, 𝑥- and 𝑥. where
𝑥. > 𝑥- , the rank-dependent utility is
𝑉(𝐿) = 𝑤(𝑝)𝑢(𝑥- ) + (1 − 𝑤(𝑝))𝑢(𝑥. )
where 𝑝 is the probability of outcome 𝑥- .
– This model allows for different weight to be attached
to each outcome, as opposed to expected utility
theory models in which the same utility weight is
attached to all outcomes.

Theories Modifying
– Transformation function 𝜋(⋅)
π(p) π(p)
1 1
π(p) > p π(p) = p
π(p) = p
π(p) < p
p p
0 1 0 1
Pessimistic π(p) Optimistic π(p)

Theories Modifying
• Empirical evidence
suggests an S-shaped π(p)
transformation 1
function.
• Intuition: individuals
are pessimistic in rare
outcomes (i.e., 𝑝 < 𝑝),
but become optimistic
for outcomes they
p
have frequently 0 π(p) = p 1
encountered.
Theories Modifying
• The rank-dependent utility theory relies on the
same axioms as expected utility theory, except
for the IA, which is replaced by co-monotonic
independence.

Money Lotteries

Money Lotteries
• We now restrict our attention to lotteries over
monetary amounts, i.e., 𝐶 = ℝ.
• Money is continuous variable, 𝑥 ∈ ℝ, with
cumulative distribution function (CDF)
𝐹 𝑥 = 𝑃𝑟𝑜𝑏 𝑦 ≤ 𝑥 for all 𝑦 ∈ ℝ

Money Lotteries
• A uniform, continuous CDF, 𝐹 𝑥 = 𝑥
– Same probability weight to every possible payoff
F(.)
1
F(x)=x
Uniform
Distribution
1/2
45o
x
1/2 1

Money Lotteries
• A non-uniform, continuous CDF, 𝐹 𝑥
F(.)
1/2
x
1/2 1

Money Lotteries
• A non-uniform,
discrete CDF
F(.)
0 if 𝑥 < 1 1
1 3/4
if 𝑥 ∈ [1, 4)
𝐹 𝑥 = 4 1/2
3
if 𝑥 ∈ [4, 6) 1/4
4
1 if 𝑥 ≥ 6 1 2 3 4 5 6 7... x

Money Lotteries
• If 𝑓 𝑥 is a density function associated with
the continuous CDF 𝐹 𝑥 , then
ˆ
𝐹 𝑥 = ž 𝑓 𝑡 𝑑𝑡
¡¢
f(.)
x
Money Lotteries
• If 𝑓 𝑥 is a density function associated with
the discrete CDF 𝐹 𝑥 , then
𝐹 𝑥 = n𝑓 𝑡
£¤ˆ
f(.)
1/2
1/4
1 2 3 4 5 6 7 x
Money Lotteries
• We can represent simple lotteries by 𝐹 𝑥 .
• For compound lotteries:
– If the list of CDF’s 𝐹- 𝑥 , 𝐹. 𝑥 , ..., 𝐹; 𝑥
represent 𝐾 simple lotteries, each occurring with
probability 𝛼- , 𝛼. , … , 𝛼; , then the compound
lottery can be represented as
;
𝐹 𝑥 =n 𝛼7 𝐹7 𝑥
75-
– For simplicity, assume that CDF’s are distributed
over non-negative amounts of money.
Money Lotteries
– We can express EU as
𝐸𝑈 𝐹 = ∫ 𝑢 𝑥 𝑓 𝑥 𝑑𝑥 or ∫ 𝑢 𝑥 𝑑𝐹(𝑥)
where 𝑢 𝑥 is an assignment of utility value to every
non-negative amount of money.
– If there is a density function 𝑓 𝑥 associated with
the CDF 𝐹(𝑥), then we can use either of the
expressions. If there is no, we can only use the latter.
– Note: we do not need to write down the limits of
integration, since the integral is over the full range of
possible realizations of 𝑥.
Money Lotteries
– 𝐸𝑈 𝐹 is the mathematical expectation of the
values of 𝑢 𝑥 , over all possible values of 𝑥.
– 𝐸𝑈 𝐹 is linear in the probabilities
§ In the discrete probability distribution,
𝐸𝑈 𝐹 = 𝑝- 𝑢- + 𝑝. 𝑢. + ⋯
– The EU representation is sensitive not only to the
mean of the distribution, but also to the variance,
and higher order moments of the distribution of
monetary payoffs.
§ Let us next analyze this property.
Money Lotteries
• Example: Let us show that if 𝑢 𝑥 = 𝛽𝑥 . + 𝛾𝑥, then
EU is determined by the mean and the variance alone.
– Indeed,
𝐸𝑈 𝑥 = ž 𝑢 𝑥 𝑑𝐹 𝑥 = ž 𝛽𝑥 . + 𝛾𝑥 𝑑𝐹 𝑥
= 𝛽 ž 𝑥 .𝑑𝐹 𝑥 + 𝛾 ž 𝑥 𝑑𝐹 𝑥
§ ˆ} § ˆ
– On the other hand, we know that
.
𝑉𝑎𝑟 𝑥 = 𝐸 𝑥. − 𝐸 𝑥 ⟹
.
𝐸 𝑥. = 𝑉𝑎𝑟 𝑥 + 𝐸 𝑥
Money Lotteries
• Example (continued):
– Substituting 𝐸 𝑥 . in 𝐸𝑈 𝑥 ,
.
𝐸𝑈 𝑥 = 𝛽𝑉𝑎𝑟 𝑥 + 𝛽 𝐸 𝑥 + 𝛾𝐸 𝑥
ª§ ˆ }
– Hence, the EU is determined by the mean and the
variance alone.

Money Lotteries
• Recall that we refer to 𝑢 𝑥 as the Bernoulli
utility function, while 𝐸𝑈 𝑥 is the vNM
function.
• We imposed few assumptions on 𝑢 𝑥 :
– Increasing in money and continuous
• We must impose an additional assumption:
– 𝑢 𝑥 is bounded
– Otherwise, we can end up in relatively absurd
situations (St. Petersburg-Menger paradox).
Money Lotteries
• St. Petersburg-Menger paradox:
– Consider an unbounded Bernoulli utility function,
𝑢 𝑥 . Then, we can always find an amount of
money 𝑥« such that 𝑢 𝑥« > 2« , for every
integer 𝑚.
– Now consider a lottery in which we toss a coin
repeatedly until tails come up. We give a
monetary payoff of 𝑥« if tails is obtained at the
𝑚th toss.
– The probability that tails comes up in the m-th
- - - -
toss is = = =. . 𝑚 times = -.
. . . .
Money Lotteries
– Then, the EU of this lottery is
¢
1
𝐸𝑈(𝑥) = n «
𝑢(𝑥« )
«5- 2
– But, because of 𝑢 𝑥« > 2« , we have that
¢ 1 ¢ 1 «
𝐸𝑈 𝑥 = n «
𝑢 𝑥« ≥ n «
2
«5- 2 «5- 2
¢
=n 1=+∞
«5-
which implies that this individual would be willing to pay
infinite amounts of money to be able to pay this lottery.
– Hence, we assume that the Bernouilli utility function is
bounded.
Measuring Risk Preferences

• An individual exhibits risk aversion if
ž 𝑢 𝑥 𝑑𝐹 𝑥 ≤ 𝑢 ž 𝑥𝑑𝐹 𝑥
for any lottery 𝐹(=)

• Intuition:
– The utility of receiving the expected monetary value of playing
the lottery (left-hand side) is higher than…
– The expected utility from playing the lottery (right-hand side).
• If this relationship happens with
a) =, we denote this individual as risk neutral
b) <, we denote him as risk averter
c) ≥, we denote him as risk lover.

• Graphical illustration:
– Consider a lottery with two equally likely outcomes, $1
and $3, with associated utilities of 𝑢(1) and 𝑢(3),
respectively.
- -
– Expected value of the lottery is 𝐸𝑉 = = 1 + = 3 = 2, with
. .
associated utility of 𝑢(2).
- -
– Expected utility of the lottery is 𝑢 1 + 𝑢(3).
. .

• Risk averse individual
– Utility from the expected value of the lottery, 𝑢(2),
is higher than the EU from playing the lottery,
- -
𝑢 1 + 𝑢(3).
. . u(x)
u(3) u(x)
u(2)
1 1
u (1) + u (3)
2 2
u(1)
1 2 3 x
• Risk neutral individual
coincides with the EU of playing the lottery,
- -
𝑢 1 + 𝑢(3).
. .
u(x) u(x)
u(3)
1 1
u (1) + u (3) = u (2)
2 2
u(1)
1 2 3 x
• Risk loving individual
is lower than the EU from playing the lottery,
- -
𝑢 1 + 𝑢(3).
. .
u(x)
u(x)
u (3)
1 1
u (1) + u (3)
2 2
u (2)
u (1)
1 2 3 x
• Certainty equivalent, 𝑐(𝐹, 𝑢):
– An alternative measure of risk aversion
– It is the amount of money that makes the
individual indifferent between playing the lottery
𝐹(=), and accepting a certain amount 𝑐(𝐹, 𝑢).
That is,
𝑢 𝑐(𝐹, 𝑢) = ∫ 𝑢 𝑥 𝑑𝐹 𝑥 or ∑ 𝑢 𝑥 𝑓 𝑥
– 𝑐(𝐹, 𝑢) is below (above) the expected value of the
lottery for risk averse (lover) individuals, and
exactly coincides for risk neutral individuals.
• Certainty equivalent for a risk-averse individual
– 𝑐(𝐹, 𝑢) is the amount of u(x)
money (𝑥) for which utility u(3) u(x)
is equal to the EU of the u(2)

1 1
lottery 2
u (1) + u (3)
2
1 1
𝑢 𝑐(𝐹, 𝑢) = 𝑢 1 + 𝑢(3) u(1)
2 2
– Risk premium (RP): the
amount that a risk-averse
1 1.87 2 3 x
person would pay to avoid
Risk premium
taking a risk:
c(F,u), Certainty Equivalent
𝑅𝑃 = 𝐸𝑉 − 𝑐 𝐹, 𝑢 > 0
• Certainty equivalent for a risk lover
– Individual would have u(x) u(x)
to be given an amount u(3)
of money above the
expected value of the 12 u(1) + 12 u(3)
lottery in order to u(2)
convince him to “stop u(1)
playing” the lottery:

1 2 3 x
𝑅𝑃 = 𝐸𝑉 − 𝑐 𝐹, 𝑢 < 0
Risk c(F,u)
Premium

• Certainty equivalent for a risk neutral individual
– The certainty equivalent
𝑐 𝐹, 𝑢 coincides with u(x)
u(x)
the expected value of u (3)
the lottery. 1 1
u (1) + u (3) = u (2)
2 2
– Hence,
u (1)
𝑅𝑃 = 𝐸𝑉 − 𝑐 𝐹, 𝑢 = 0
1 2 3 x
c(F,u)

• Probability premium, 𝜋(𝑥, 𝜀, 𝑢):
– An alternative measure of risk aversion
– It is the excess in winning probability over fair
odds that makes the individual indifferent
between the certainty outcome 𝑥 and a gamble
between the two outcomes 𝑥 + 𝜀 and 𝑥 − 𝜀:
𝑢 𝑥
1 1
= + 𝜋 𝑥, 𝜀, 𝑢 𝑢 𝑥 + 𝜀 + − 𝜋 𝑥, 𝜀, 𝑢 𝑢 𝑥 − 𝜀
2 2
– Intuition: Better than fair odds must be given for
the individual to accept the risk.
• The “extra probability” 𝜋 that is needed to make the
EU of the lottery coincides with the utility of the
expected lottery:
1 1
𝑢 2 = +𝜋 𝑢 3 + −𝜋 𝑢 1
2 2

• The following properties are equivalent:
1) The decision maker is risk averse.
2) The Bernoulli utility function 𝑢 𝑥 is concave,
𝑢OO (𝑥) ≤ 0.
3) The certainty equivalent is lower than the
expected value of the lottery, i.e., 𝑐(𝐹, 𝑢) ≤
∫ 𝑢 𝑥 𝑑𝐹 𝑥 .
4) The risk premium is positive, 𝑅𝑃 = 𝐸𝑉 − 𝑐 𝐹, 𝑢 .
5) The probability premium is positive for all 𝑥 and 𝜀,
i.e., 𝜋(𝑥, 𝜀, 𝑢) ≥ 0.
• Arrow-Pratt coefficient of absolute risk aversion:
𝑢OO (𝑥)
𝑟³ 𝑥 = − O
𝑢 (𝑥)
– Clearly, the greater the curvature of the utility function,
𝑢OO (𝑥), the larger the coefficient 𝑟³ 𝑥 .
– But, why do not we simply have 𝑟³ 𝑥 = 𝑢OO (𝑥)?
• Because it will not be invariant to positive linear
transformations of the utility function, such as 𝑣 𝑥 = 𝛽𝑢 𝑥 .
That is, 𝑣′′ 𝑥 = 𝛽𝑢′′ 𝑥 is affected by the transformation, but
the above coefficient of risk aversion is unaffected.
𝛽𝑢OO (𝑥) 𝑢OO (𝑥)
𝑟³ 𝑥 = − O
=− O
𝛽𝑢 (𝑥) 𝑢 (𝑥)
• Example (CARA utility function).
– Take 𝑢 𝑥 = −𝑒 ¡cˆ where 𝑎 > 0. Then
𝑢OO (𝑥) −𝑎. 𝑒 ¡cˆ
𝑟³ 𝑥 = − O =− ¡cˆ
=𝑎
𝑢 (𝑥) 𝑎𝑒
which is constant in wealth 𝑥.
– The literature refers to this Bernoulli utility
function as the Constant Absolute Risk Aversion
(CARA).

• If 𝑟³ 𝑥 decreases as we increase wealth 𝑥, then
we say that such Bernoulli utility function
satisfies decreasing absolute risk aversion (DARA)
𝜕𝑟³ 𝑥
<0
𝜕𝑥
• Intuition: wealthier people are willing to bear
more risk than poorer people. Note, however,
that this is NOT due to different utility functions,
but because the same utility function is evaluated
at higher/lower wealth levels.
• A sufficient condition for DARA is 𝑢OOO 𝑥 > 0.
• Arrow-Pratt coefficient of relative risk aversion:
·~~(ˆ)
𝑟¶ 𝑥 = −𝑥 = or 𝑟¶ 𝑥 = 𝑥 = 𝑟³ 𝑥
·~ (ˆ)
– 𝑟¶ 𝑥 does not vary with the wealth level at which it is
evaluated.
– We can show that
𝜕𝑟¶ 𝑥 𝜕𝑟³ 𝑥
= 𝑟³ 𝑥 + 𝑥 =
𝜕𝑥 𝜕𝑥
“
– Therefore,
𝜕𝑟¶ 𝑥 ⇒ 𝜕𝑟³ 𝑥
< 0 <0
⇍ 𝜕𝑥
𝜕𝑥 Advanced Microeconomic Theory 100
• Example:
– Take 𝑢 𝑥 = 𝑥 d . Then
𝑏 𝑏 − 1 𝑥 d¡.
𝑟¶ 𝑥 = −𝑥 = d¡-
=1−𝑏
𝑏𝑥
for all 𝑥.
– The literature refers to this Bernoulli utility
function as the Constant Relative Risk Aversion
(CRRA).

– Consider a CRRA Utility
utility function 1.0

Increasing degree
𝑢 𝑥 = 𝑥 d for 𝑏 = 0.8
of risk aversion x1/4
- - - x1/3
1, , , . 0.6 x1/2
. H I x
– 𝑟¶ 𝑥 increases, 0.4
- . H
respectively, to , , , 0.2
. H I
making utility 0.2 0.4 0.6 0.8 1.0 Money, x
function more
concave.
• A utility function 𝑢³ (=) exhibits more strong risk aversion
than another utility function 𝑢º (=) if, there is a constant
𝜆 > 0,
𝑢³OO (𝑥- ) 𝑢³O (𝑥. )
OO ≥𝜆≥ O
𝑢º (𝑥- ) 𝑢º (𝑥. )
• In addition, if 𝑥- = 𝑥. , the above condition can be re-
written as
𝑢³OO (𝑥- ) 𝑢ºOO (𝑥- )
O ≥ O
𝑢³ (𝑥- ) 𝑢º (𝑥- )
• Then, 𝑢³ (=) also exhibits more risk aversion than 𝑢º (=).

• For two utility functions 𝑢- and 𝑢. , where 𝑢.
is a concave transformation of 𝑢- , the
following properties are equivalent:
1) There exists an increasing concave function 𝜑(=)
such that 𝑢. 𝑥 = 𝜑(𝑢- (𝑥)) for any 𝑥. That is,
𝑢. = is more concave than 𝑢- = .
2) 𝑟³ 𝑥, 𝑢. ≥ 𝑟³ 𝑥, 𝑢- for any 𝑥.
3) 𝑐(𝐹, 𝑢. ) ≤ 𝑐(𝐹, 𝑢- ) for any lottery 𝐹(=).
4) 𝜋(𝑥, 𝜀, 𝑢. ) ≥ 𝜋(𝑥, 𝜀, 𝑢- ) for any 𝑥 and 𝜀.
5) Whenever 𝑢. = finds a lottery 𝐹(=) at least as
good as a riskless outcome 𝑥̅ , then 𝑢- = also
finds such a lottery 𝐹(=) at least as good as 𝑥̅ .
That is
𝐸𝑈. = ž 𝑢. 𝑥 𝑑𝐹 𝑥 ≥ 𝑢. 𝑥̅ ⟹
𝐸𝑈- = ž 𝑢- 𝑥 𝑑𝐹 𝑥 ≥ 𝑢- 𝑥̅

• Different degrees of risk u(x)
aversion u1(x)
• 𝑢- = and 𝑢. = are u2(x)
evaluated at the same u (x)=u (x)
1 2
wealth level 𝑥. EU 1
• The same lottery yields a EU 2
larger expected utility for

the individual with less
risk averse preferences,
𝐸𝑈- > 𝐸𝑈. . 1 x 3 x
• 𝑐 𝐹, 𝑢. < 𝑐(𝐹, 𝑢- ), c ( F , u2 ) c ( F , u )
1
reflecting that individual
2 is more risk averse.
Prospect Theory and Reference-
Dependent Utility

Prospect Theory
• Prospect theory: a decision maker’s total value from
a list of possible outcomes 𝑥 = (𝑥- , 𝑥. , … , 𝑥1 ) with
associated probabilities 𝑝 = (𝑝- , 𝑝. , … , 𝑝1 ) is
1
𝑣 𝑥, 𝑝 = n 𝑤(𝑝‡ ) = 𝑣(𝑥‡ )
‡5-
where
– 𝑤(𝑝‡ ) is a “probability weighting function”
– 𝑣 𝑥‡ is the “value function” the individual
obtains from outcome 𝑥‡
Prospect Theory
• Three main differences relative to standard
expected utility theory:
• First, 𝑤 𝑝‡ ≠ 𝑝‡ :
– if 𝑤 𝑝‡ > 𝑝‡ , individuals overestimate the
likelihood of outcome 𝑥‡
– if 𝑤 𝑝‡ < 𝑝‡ , individuals underestimate the
likelihood of outcome 𝑥‡
– if 𝑤 𝑝‡ = 𝑝‡ , the model coincides with standard
expected utility theory.
Prospect Theory
• Second, every payoff 𝑥‡ is evaluated relative to a
“reference point” 𝑥t , with the value function
𝑣 𝑥‡ , which is
– Increasing and concave, 𝑣 OO 𝑥‡ < 0, for all 𝑥‡ > 𝑥t ,
• That is, the individual is risk averse for gains.
– Decreasing and convex, 𝑣 OO 𝑥‡ > 0, for all 𝑥‡ < 𝑥t
• That is, the individual is risk lover for losses.
– Extremes:
• if 𝑥t = 0, the individual is risk averse for all payoffs;
• if 𝑥t = +∞, he is risk lover for all payoffs.
Prospect Theory
• Third, value function 𝑣 𝑥‡ has a kink at the
reference point 𝑥t .
– The curve becomes steeper for losses (to the left of
𝑥t ) than for gains (to the right of 𝑥t ).
• Loss aversion:
• A given loss of $a produces a larger disutility than a gain
of the same amount.

Prospect Theory
• Value function in prospect theory

Prospect Theory
• Example:
– Consider as in Tversky and Kahneman (1992)
Œ¾
𝑤 𝑝 = ‚ and 𝑣 𝑥 = 𝑥 ¿
Œ¾ “(-¡Œ)¾ ¾
where 0 < 𝛽 < 1, and 0 < 𝛼 < 1.
• Note that this implies probability weighting,
but does not consider a value function with
loss aversion relative to a reference point.

Prospect Theory
– Depicting the probability weighting function

Prospect Theory
• Example:
– A common value function is
𝑣 𝑥‡ = 𝑥‡ ¿ if 𝑥‡ ≥ 𝑥t , and
= −𝜆(−𝑥‡ )¿ if 𝑥‡ < 𝑥t
where 0 < 𝛼 ≤ 1, and 𝜆 ≥ 1 represents loss
aversion.
• If 𝜆 = 1 the individual does not exhibit loss
aversion.

Prospect Theory
• Example:
• Common simplifications, assume 𝛼 = 𝛽 = 1
(which implies no probability weighting, and
linear value functions), to estimate 𝜆.
• Average estimates 𝜆 = 2.25 and 𝛽 = 0.88

Prospect Theory
– Nicholas Barberis (2013) “Thirty Years of Prospect
Theory in Economics: A Review and Assessment,”
Journal of Economic Perspectives, 27(1), pp. 173-96.
– R. Duncan Luce and Peter C. Fishburn (1991) “Rank
and sign-dependent linear utility models for binary
gambles.” Journal of Economic Theory, 53, pp. 75–100.
– Daniel Kahneman and Amos Tversky (1992) “Advances
in prospect theory: Cumulative representation of
uncertainty” Journal of Risk and Uncertainty, 5(4), pp.
297–323.
– Peter Wakker and Amos Tversky (1993) “An
axiomatization of cumulative prospect theory.”
Journal of Risk and Uncertainty, 7, pp. 147–176.
Reference-Dependent Utility
• Individual preferences are affected by reference
points. Thus, gains and loses can be evaluated
differently.
• Consider a consumption vector 𝑥 ∈ ℝ1 which is
evaluated against a 𝑛-dimensional reference
vector 𝑟 ∈ ℝ1 . Utility function is
𝑢 𝑥 𝑟 = 𝑚 𝑥 + 𝑛(𝑥|𝑟)
where 𝑛 𝑥7 𝑟7 = 𝜇 𝑚7 𝑥7 − 𝑚7 (𝑟7 )
measures the gain/loss of consuming 𝑥7 units of
good 𝑘 relative to its reference amount 𝑟7 .
• For lotteries with cumulative distribution
function 𝐹(𝑥),
𝑈 𝐹 𝑟 = ∫ 𝑢 𝑥 𝑟 𝑑𝐹(𝑥)
• For lotteries over the set of reference points
𝑢 𝐹 𝐺 = ∫ ∫ 𝑢 𝑥 𝑟 𝑑𝐺(𝑟)𝑑𝐹(𝑥)

– “Reference-Dependent Consumption Plans”
(2009) by Koszegi and Rabin, American Economic
Review, vol. 99(3).
– “Rational Choice with Status Quo Bias” (2005) by
Masatlioglu and Ok, Journal of Economic Theory,
vol. 121(1).
– “On the complexity of rationalizing behavior”
(2007) Apesteguia and Ballester, Economics
Working Papers 1048.

Comparison of Payoff
Distributions

Comparison of Payoff Distributions
• So far we compared utility functions, but not
the distribution of payoffs.
• Two main ideas:
1) 𝐹(=) yields unambiguously higher returns than
𝐺(=). We will explore this idea in the definition
of first order stochastic dominance (FOSD);
2) 𝐹(=) is unambiguously less risky than 𝐺(=). We
will explore this idea in the definition of second
order stochastic dominance (SOSD).

• FOSD: 𝐹(=) FOSD 𝐺(=) if, for every non-decreasing
function 𝑢: ℝ → ℝ, we have
ž 𝑢 𝑥 𝑑𝐹 𝑥 ≥ ž 𝑢 𝑥 𝑑𝐺 𝑥
• The distribution of monetary payoffs 𝐹(=) FOSD the

distribution of monetary payoffs 𝐺(=) if and only if
𝐹(𝑥) ≤ 𝐺 𝑥 or 1 − 𝐹 𝑥 ≥ 1 − 𝐺 𝑥
for every 𝑥.
• Intuition: For every amount of money 𝑥, the probability
of getting at least 𝑥 is higher under 𝐹(=) than under
𝐺(=).
• At any given outcome 𝑥, the probability of
obtaining prizes above 𝑥 is higher with lottery 𝐹(=)
than with lottery 𝐺(=), i.e., 1 − 𝐹 𝑥 ≥ 1 − 𝐺 𝑥 .
1
1 G( x )
G( x ) G(x)
1 F (x )
F(x)
F (x )
x x
• Example:
– Let us take lotteries 𝐹(=) and 𝐺(=) over discrete
outcomes.
$1 $2 $3 $4 $5 Dollars
1 1
G(∙) 0 0 0
2 2
0 1 1 0 1
F(∙)
4 4 2
How can we know if 𝐹(=) FOSD 𝐺(=)?

– 𝐹(=) lies below lottery 𝐺 = . Hence, 𝐹(=) concentrates
more probability weight on higher monetary outcomes.
– Thus, 𝐹(=) FOSD 𝐺(=).
F(x)
1
G(.)
3/4 1/2 F(.)
1/2
1/4
1/4 1/2
1/4
$1 $2 $3 $4 $5 x
• Example (Binomial distribution):
– Consider the binomial distribution
𝑁 ˆ
𝐹 𝑥; 𝑁, 𝑝 = 𝑝 (1 − 𝑝)0¡ˆ
𝑝
- -
– where 𝑥 ∈ 0, 𝑁 . Assuming 𝑁 = 100 and parameter 𝑝 increasing from 𝑝 = to 𝑝 = .
I .
Then, 𝐹 𝑥; 100,1/2 FOSD 𝐹 𝑥; 100,1/4 .
1.0
0.8
0.6

0.4
1 1
p= p=
0.2 4 2
20 40 60 80 100
• We now focus on the riskiness or dispersion of a
lottery, as opposed to higher/lower returns of
lottery (FOSD).
• To focus on riskiness, we assume that the CDFs
we compare have the same mean (i.e., same
expected return).
• SOSD: 𝐹(=) SOSD 𝐺(=) if, for every non-decreasing
function 𝑢: ℝ → ℝ, we have
ž 𝑢 𝑥 𝑑𝐹 𝑥 ≥ ž 𝑢 𝑥 𝑑𝐺 𝑥

• Example (Mean-Preserving Spread):
– Let us take lotteries 𝐹(=) and 𝐺(=) over discrete outcomes.
– Lottery 𝐺(=) spreads the probability weight of lottery 𝐹(=)
over a larger set of monetary outcomes.
– The mean is nonetheless unaltered (2.5).
– For these two reasons, we say that a CDF is a mean
preserving spread of the other.
$1 $2 $3 $4 $5 Dollars
𝐹(=)
G(∙) 0 1 1 0 0
2 2
𝐺(=) 1 1 1 1
F(∙) 0
4 4 4 4
• 𝐺(=) is a mean-preserving spread of 𝐹(=), but
it is riskier than 𝐹(=) in the SOSD sense.
• Note that neither FOSD the other
– 𝐹(=) is not above/below 𝐺(=) for all 𝑥
F(.)
F(.)
1
3/4 1
G(.)
2
1/2
1
1/4
2
$1 $2 $3 $4 $5 Dollars
• Example (Elementary increase in risk):
– 𝐺(=) is an Elementary Increase in Risk (EIR) of
another CDF 𝐹(=) if 𝐺(=) takes all the probability
weight of an interval 𝑥 O , 𝑥 OO and transfers it to
the end points of this interval, 𝑥 O and 𝑥 OO , such
that the mean of the original lottery is preserved.
– EIR is a mean-preserving spread (MPS), but the
converse is not necessarily true:
⇒
𝐸𝐼𝑅 𝑀𝑃𝑆
⇍
– Hence, if 𝐺(=) is an EIR of 𝐹(=), then 𝐹(=) SOSD
𝐺(=).
– both CDFs 𝐹(=) and F(x), G(x)
𝐺(=) maintain the
same mean.
1
G(x)
– 𝐺(=) concentrates F(x)
more probability at
the end points of the
interval 𝑥 O , 𝑥 OO than Areas of same
size
𝐹(=). x' x '' x

• Hazard rate dominance: The hazard rate of
lottery 𝐹(𝑥) is
𝑓(𝑥)
𝐻𝑅Æ (𝑥) =
1 − 𝐹(𝑥)
– Intuition: It measures the instantaneous
probability of an event happening at time 𝑥 given
that it did not happen before 𝑥.
– Example: a computer stops working at exactly 𝑥
– If 𝐻𝑅Æ (𝑥) ≤ 𝐻𝑅Ç (𝑥), lottery 𝐹 𝑥 dominates
𝐺 𝑥 in terms of the hazard rate.
– Since −𝐻𝑅Æ (𝑥) can be expressed as
𝑑
−𝐻𝑅Æ 𝑥 = ln 1 − 𝐹(𝑥)
𝑑𝑥
– Solving for 𝐹(𝑥),
ˆ
𝐹 𝑥 = 1 − exp − ž 𝐻𝑅Æ 𝑡 𝑑𝑡
t
– Then,
ˆ
𝐹 𝑥 = 1 − exp − ž 𝐻𝑅Æ 𝑡 𝑑𝑡
t
ˆ
≤ 1 − exp − ž 𝐻𝑅Æ 𝑡 𝑑𝑡 = 𝐺 𝑥
t
– Thus, 𝐻𝑅Æ (𝑥) ≤ 𝐻𝑅Ç (𝑥) implies that 𝐹 𝑥 FOSD 𝐺 𝑥 .
• Reverse hazard rate: The reverse hazard rate of
lottery 𝐹(𝑥) is
𝑓(𝑥)
𝑅𝐻𝑅Æ (𝑥) =
𝐹(𝑥)
– Intuition: It measures the probability that,
conditional on the realized payoff in the lottery
being equal or lower than 𝑥, the payoff you receive
is exactly 𝑥.
– If 𝑅𝐻𝑅Æ (𝑥) ≥ 𝑅𝐻𝑅Ç (𝑥), lottery 𝐹 𝑥 dominates
𝐺 𝑥 in terms of the reverse hazard sense.
– Let us express 𝑅𝐻𝑅Æ (𝑥) as
𝑑
𝑅𝐻𝑅Æ 𝑥 = ln 𝐹(𝑥)
𝑑𝑥
– Solving for 𝐹(𝑥),
ˆ
𝐹 𝑥 = exp − ž 𝑅𝐻𝑅Æ 𝑡 𝑑𝑡
t
– Then,
ˆ ˆ
𝐹 𝑥 = exp − ž 𝑅𝐻𝑅Æ 𝑡 𝑑𝑡 ≤ exp − ž 𝑅𝐻𝑅Æ 𝑡 𝑑𝑡 = 𝐺 𝑥
t t
– Thus, 𝑅𝐻𝑅Æ (𝑥) ≥ 𝑅𝐻𝑅Ç (𝑥) implies that 𝐹 𝑥 FOSD 𝐺 𝑥 .

• Likelihood ratio: The likelihood ratio of a
lottery 𝐹 𝑥 is
𝑓(𝑦)
𝐿𝑅Æ =
𝑓(𝑥)
for any two payoffs 𝑥 and 𝑦, where 𝑦 > 𝑥.
– 𝐹 𝑥 dominates 𝐺 𝑥 in terms of likelihood ratio if
𝑓(𝑥) 𝑓(𝑦)
≤
𝑔(𝑥) 𝑔(𝑦)

• 𝐿𝑅 dominance implies 𝐻𝑅 dominance:
– Let us rewrite 𝐿𝑅 dominance as
𝑔(𝑦) 𝑓(𝑦)
≤
𝑔(𝑥) 𝑓(𝑥)
– Then, for all 𝑥,
¢ ¢
𝑔(𝑦) 𝑓(𝑦)
ž 𝑑𝑦 ≤ ž 𝑑𝑦
t 𝑔(𝑥) t 𝑓(𝑥)
– Simplifying
-¡Ç(ˆ) -¡Æ(ˆ) É(ˆ) Š(ˆ)
≤ or ≤
Š(ˆ) É(ˆ) -¡Æ(ˆ) -¡Ç(ˆ)
which implies 𝐻𝑅 Æ 𝑥 ≤ 𝐻𝑅Ç 𝑥 .
• Summary:
– 𝐿𝑅 dominance implies 𝐻𝑅 dominance
– 𝐻𝑅 and 𝑅𝐻𝑅 dominance imply FOSD.

Appendix 5.1:
State-Dependent Utility

• So far the decision maker only cared about the
payoff arising from every outcome of the
lottery.
• Now we assume that the decision maker cares
not only about his monetary outcomes, but
also about the state of nature that causes
every outcome.
– That is, 𝑢ÊËÌËÍ - (𝑥) ≠ 𝑢ÊËÌËÍ . (𝑥) for given 𝑥.

• Let us assume that each of the possible monetary
payoffs in a lottery is generated by an underlying
cause (i.e., an underlying state of nature).
• Examples:
– The monetary payoff of an insurance policy is
generated by a car accident
§ State of nature = {car accident, no car accident}
– The monetary payoff of a corporate stock is
generated by the state of the economy
§ State of nature = {economic growth, economic
depression} Advanced Microeconomic Theory 142
• Generally, let 𝑠 ∈ 𝑆 denote a state of nature,
where 𝑆 is a finite set.
• Every state 𝑠 has a well-defined, objective
probability 𝜋Ï ≥ 0.
• A random variable is function 𝑔: 𝑆 → ℝ, that
maps states into monetary payoffs.

• Examples (revisited):
– Car accident: the random variable assigns a
monetary value to the state of nature care
accident, and to the state of nature no accident.
State of nature Probability Monetary payoff

Car accident 𝜋ÌÐÐÑÒÍÓË Damage + Deductible – Premium = $1,000
No car accident 𝜋Ó{ ÌÐÐÑÒÍÓË Premium = -$50

• Examples (revisited):
– Corporate stock: the random variable assigns a
monetary value to the state of nature econ.
growth, and to the state of nature eco.
depression.
State of nature Probability Monetary payoff

Economic growth 𝜋ÔÕ{ÖË× Dividends, higher price of shares = $250
Economic depression 𝜋ÒÍØÕÍÊÊÑ{Ó No dividends, loss if we sell shares = -$125

• Every random variable 𝑔(=) can be used to represent
lottery 𝐹(=) over monetary payoffs as
𝐹 𝑥 = n 𝜋Ï
Ï: Š(Ï)Ùˆ
where {𝑠: 𝑔(𝑠) ≤ 𝑥} represents all those states of
nature 𝑠 that generate a monetary payoff 𝑔(𝑠) ∈ ℝ
below a cutoff payoff 𝑥.
• The random variable 𝑔(=) generates a monetary payoff
for every state of nature 𝑠 ∈ 𝑆, and since 𝑆 is finite, we
can represent this list of monetary payoffs as
Ü
(𝑥- , 𝑥. , … , 𝑥Ü ) ∈ ℝ“
where 𝑥Ï is the monetary payoff corresponding to state
of nature 𝑠. Advanced Microeconomic Theory 146
• Example:
– A random variable 𝑔(= Prob. =0
) describes the
3
1
monetary outcome 1
1 =
1
2
associated to the four 3/4 2 =
4 1
=
states of nature 𝑆 = 1/2 4
4
{1,2,3,4}. 1/4
– Outcomes are ordered x1 x2 x3 x4 g(s)

from lower to higher
𝑥- ≤ 𝑥. ≤ 𝑥H ≤ 𝑥I .
– Hence,
-
𝐹(𝑥- ) = 𝜋- =
.
- - H
𝐹(𝑥. ) = 𝜋- + 𝜋. = + =
. I I
- - H
𝐹(𝑥H ) = 𝜋- + 𝜋. + 𝜋H = + + 0 =
. I I
𝐹(𝑥I ) = 𝜋- + 𝜋. + 𝜋H + 𝜋I = 1
• Disadvantage of 𝐹(𝑥):
– For a given 𝑥, we cannot keep track of which
state(s) of nature generated 𝑥.
State-Dependent Utility:
Extended EU representation
• We now have a preference relation ≿ ranks
Ü
lists of monetary payoffs (𝑥- , 𝑥. , … , 𝑥Ü ) ∈ ℝ“ .
• Note the similarity of this setting with that in
consumer theory:
– Preferences over bundles then, preferences over
lists of monetary payoffs here.
– Since (𝑥- , 𝑥. , … , 𝑥Ü ) ∈ ℝ“Ü specifies one payoff for
each state of nature, this list is referred to as
contingent commodities.
• Preference relation ≿ has an Extended EU
representation if for every 𝑠 ∈ 𝑆, there is a function
Ü
𝑢Ü : ℝ“ → ℝ“ (mapping the monetary outcome of state
𝑠, 𝑥Ï , into a utility value in ℝ), such that for any two
Ü
lists of monetary outcomes (𝑥- , 𝑥. , … , 𝑥Ü ) ∈ ℝ“ and
(𝑥-O , 𝑥.O , … , 𝑥ÜO ) ∈ ℝ“
Ü
,
(𝑥- , 𝑥. , … , 𝑥Ü ) ≿ (𝑥-O , 𝑥.O , … , 𝑥ÜO ) iff
n 𝜋Ï 𝑢Ï (𝑥Ï ) ≥ n 𝜋Ï 𝑢Ï (𝑥ÏO )
Ï Ï
• The main difference with the previous sections is that
now the Bernoulli utility function is state-dependent,
𝑢Ï (=), whereas in the previous sections it was state-
independent, 𝑢(=).
• Graphical representation:
– First, at the “certainty line” the decision maker
receives the same monetary amount, regardless the
state of nature, 𝑥- = 𝑥. .
– Second, all the (𝑥- , 𝑥. ) pairs on a given ind. curve
satisfy 𝜋- = 𝑢- 𝑥- + 𝜋. = 𝑢. 𝑥. = 𝑈 Ý
– Third, the upper contour set of an ind. curve that
passes through point (𝑥̅- , 𝑥̅. ) satisfy
𝜋- = 𝑢- 𝑥- + 𝜋. = 𝑢. 𝑥.
≥ 𝜋- = 𝑢- 𝑥̅- + 𝜋. = 𝑢. 𝑥̅.
or, more generally, ∑Ï 𝜋Ï 𝑢Ï (𝑥Ï ) ≥ ∑Ï 𝜋Ï 𝑢Ï (𝑥̅Ï ).
– Fourth, movement along a given ind. curve does
not change the decision maker’s utility level. Hence,
totally differentiating
𝜕𝑢- 𝑥̅- 𝜕𝑢. 𝑥̅.
𝜋- = 𝑑𝑥- + 𝜋. = 𝑑𝑥. = 0
𝜕𝑥- 𝜕𝑥.
and re-arranging,
𝜕𝑢- 𝑥̅-
𝑑𝑥. 𝜋- = 𝜋- = 𝑢-O (𝑥̅- )
𝜕𝑥-
=− =−
𝑑𝑥- 𝜕𝑢. 𝑥̅. 𝜋. = 𝑢.O (𝑥̅. )
𝜋. =
𝜕𝑥.
which represents the slope of the ind. curve,
evaluated at point (𝑥̅- , 𝑥̅. ). This is really similar to
MRS. Advanced Microeconomic Theory 152
– The slope of the ind.
curve at (𝑥̅- , 𝑥̅. ) is x2 45o line(certainty line)
𝑑𝑥. 𝜋- = 𝑢-O(𝑥̅-) x1 = x 2
=−
𝜋. = 𝑢.O(𝑥̅.)
( x1 , x2 ) such that
𝑑𝑥- u ( x1 ) + u ( x2 )
1 1 2 2
u ( x1 ) + u ( x2 )
– If the Bernoulli utility is
1 1 2 2
x2
state-independent, i.e., u ( x1 ) +
1 1 u ( x2 ) = u
2 2
𝑢- = = 𝑢. = = ⋯ = x1 x1
𝑢Ü = , then the slope is
𝑑𝑥. 𝜋-
=−
𝑑𝑥- 𝜋. Advanced Microeconomic Theory 153
• Example (Insurance with state-dependent
utility):
– Start from an initial situation of 𝑤, 𝑤 − 𝐷
without insurance, where 𝐷 is loss from accident.
– After insurance is purchased, the decision maker
gets a payment of 𝑧- in state 1, and 𝑧. in state 2,
where 𝑧- ≶ 0 and 𝑧. ≶ 0,
𝑤 + 𝑧- , 𝑤 − 𝐷 + 𝑧.
– Moreover, if the policy is actuarially fair, then its
expected payoff is zero,
𝜋- 𝑧- + 𝜋. 𝑧. = 0
à‚
– The budget line is 𝑧. = − 𝑧
à} -
(Accident)
x2 x1 = x 2
slope = 1
w D
{( w + z1 , w D + z2 ) : 1 z1 + z = 0}
2 2
w x1
(No Accident)
• Without state dependency:
– Indifference curves are tangent to the budget line
at the certainty line, since the slope of the
à‚
indifference curve is − .
à}
– Hence, the decision maker would insure
completely since his consumption level is
unaffected by the possibility of suffering an
accident.

• With state dependency:
– Indifference curves are NOT tangent to the budget
line at the certainty line.
– The decision-maker prefers a point such as
(𝑥-O , 𝑥.O ) to the certain outcome (𝑥̅ , 𝑥̅ ).
– That is, at (𝑥̅ , 𝑥̅ ) he prefers higher payoffs in state
1 than in state 2 if 𝑢-O 𝑥̅ > 𝑢.O (𝑥̅ ). Otherwise, he
would prefer higher payoffs in state 2 than in state
1.
·‚~ ˆ̅
– Note that 𝑢-O 𝑥̅ > 𝑢.O (𝑥̅ ) implies that >1
·}~ ˆ̅
à‚ ⋅·‚~ ˆ̅ à‚
and − <− .
à} ⋅·}~ ˆ̅ à}
45o line(certainty line),
x2 x1 = x 2
Slope of Ind. Curve at
( x1 , x2 )
slope = 1
( x1' , x2' )
u2
u1
x1
( x1 , x2 ) (No Accident)
• Let us now allow for the possibility that the
monetary payoff under state 𝑠, 𝑥Ï , is not a certain
amount of money, but a random amount with
distribution function 𝐹Ï (⋅).
• Hence, all monetary outcomes arising from the 𝑆
states of world can be described as a lottery 𝐿 =
(𝐹- , 𝐹. , … , 𝐹Ü ).
• Given this “extended” definition of lotteries, we
can then re-write the IA, as the “extended” IA.
• Extended IA: The preference relation satisfies
the extended IA if, for any three lotteries 𝐿, 𝐿O ,
and 𝐿OO and 𝛼 ∈ (0,1), we have that
𝐿 ≿ 𝐿O iff
𝛼𝐿 + (1 − 𝛼)𝐿OO ≿ 𝛼𝐿O + (1 − 𝛼)𝐿OO
• Hence, “extended” IA is a mere extension of the
standard IA to the case of “extended” lotteries
𝐿 = (𝐹- , 𝐹. , … , 𝐹Ü ).
• Extended EU theorem: Suppose preferences
relation satisfies continuity and the extended
IA. Then we can assign a utility function 𝑢Ï (=)
for money in every state 𝑠 such that for any
two lotteries 𝐿 = (𝐹- , 𝐹. , … , 𝐹Ü ) and 𝐿O =
(𝐹-O , 𝐹.O , … , 𝐹ÜO ) we have
𝐿 ≿ 𝐿O iff
n ž 𝑢Ï 𝑥Ï 𝑑𝐹Ï (𝑥Ï ) ≥ n ž 𝑢Ï 𝑥Ï 𝑑𝐹ÏO (𝑥Ï )
Ï Ï
Appendix 5.2:
Subjective Probability Theory

• So far we were assuming that probabilities
were objective and observable.
• This is not the case in certain cases. Instead
people might hold probabilistic beliefs about
the likelihood of a certain event: subjective
probability.

• Can we deduce subjective probability from
actual behavior? Yes!
• Imagine a decision maker who prefers a
gamble
($1 in state 1, $0 in state 2) ≿
($0 in state 1, $1 in state 2)
• If the value of money is the same across
states, then he must be assigning a higher
subjective probability to state 1 than to state
2.
• Let us start with some definitions.
• First, we define state 𝑠 preferences, ≿Ï , on
state 𝑠 lotteries 𝐹Ï (=) by 𝐹Ï (=) ≿ 𝐹ÏO (=) if
ž 𝑢Ï 𝑥Ï 𝑑𝐹Ï (𝑥Ï ) ≥ ž 𝑢Ï 𝑥Ï 𝑑𝐹ÏO (𝑥Ï )
• Hence, the state preferences (≿- , ≿. , … , ≿Ü )

on state lotteries (𝐹- , 𝐹. , … , 𝐹Ü ) are state
uniform if
≿Ï =≿Ï~ for any two states 𝑠 and 𝑠 ′
• That is, preferences over lotteries are state
uniform if for any two states 𝑠 and 𝑠 ′, the
ranking of any two lotteries 𝐹Ï (=) and 𝐹ÏO (=)
coincides in both states, i.e.,
𝐹Ï (=) ≿ 𝐹ÏO (=) or
𝐹ÏO (=) ≿ 𝐹Ï (=) or
𝐹Ï = ~𝐹ÏO (=)

• With state uniformity, 𝑢Ï = and 𝑢Ï~ = can differ
only up to an increasing linear transformation.
• That is, there is a utility function 𝑢 = such that
𝑢Ï = = 𝜋Ï 𝑢 = + 𝛽Ï
𝑢Ï~ = = 𝜋Ï~ 𝑢 = + 𝛽Ï~
for every state 𝑠 and 𝑠 O , and for every 𝜋Ï , 𝜋Ï~ > 0
and 𝛽Ï , 𝛽Ï~ > 0.
• In words, the ranking between the expected
utility of state 𝑠 and 𝑠 O remains unaffected.
• Subjective probabilities EU theorem:
– Suppose that a preference relation satisfies continuity
and the extended IA, and that preferences over
lotteries are state uniform.
– Then, there are subjective probabilities
(𝜋- , 𝜋. , … , 𝜋Ü ) ≫ 0 and a utility function 𝑢 = on
certain amounts of money, such that for any two lists
of monetary amounts (𝑥- , 𝑥. , … , 𝑥Ü ) and
(𝑥-O , 𝑥.O , … , 𝑥ÜO ),
(𝑥- , 𝑥. , … , 𝑥Ü ) ≿ (𝑥-O , 𝑥.O , … , 𝑥ÜO ) iff
n 𝜋Ï 𝑢Ï (𝑥Ï ) ≥ n 𝜋Ï 𝑢Ï (𝑥ÏO )
Ï Ï
• Intuition: a decision maker prefers the first list
of monetary outcomes to the second if the
“subjective” expected utility from the first list
is larger than that from the second.
• The predictions of the subjective EU theorem
are not necessarily satisfied in all
experimental settings.
– Example: Ellsberg paradox

• Ellsberg paradox:
– An urn contains 300 balls: 100 are red and the
remaining 200 are either blue or green.
– We first present the following two gambles to a
group of students, asking each of them to choose
either gamble A or B.
§ Gamble A: $1000 if the ball is red
§ Gamble B: $1000 if the ball is blue
– We next present the following two gambles to the
same group of students, asking each of them to
choose either gamble C or D.
§ Gamble C: $1000 if the ball is not red
§ Gamble D: $1000 if the ball is not blue
• Ellsberg paradox (continued):
– Common choices: people choose A to B, and C to D.
– But these choices violate subjective EU theory!
– We know that
𝑝 Red = 1 − 𝑝(not Red)
𝑝 Blue = 1 − 𝑝(not Blue)
– If gamble A is preferred to B, then we must have
𝑝 Red 𝑢 $1000 > 𝑝 Blue 𝑢 $1000 ⟹
𝑝 Red > 𝑝 Blue
– And if gamble C is preferred to D, then we must have
𝑝 not Red 𝑢 $1000 > 𝑝 not Blue 𝑢 $1000 ⟹
𝑝 not Red > 𝑝 not Blue
– But the above two expressions are incompatible.
Appendix 5.3:
Ambiguity and Ambiguity
Aversion

Ambiguity and Ambiguity Aversion
• Alternative theories that account for the anomaly
in the Ellsberg paradox:
1) expected utility theory with multiple priors (also
referred to as maxmin expected utility)
2) rank-dependent expected utility (or Choquet
expected utility)
• Individuals have ambiguous (unclear) beliefs,
rather than objective or subjective beliefs.
• Let 𝑓 denote an act 𝑓: 𝑠 → 𝑥 from the set of
states to the set of outcomes.

• Maxmin expected utility (MEU):
– If subjects have too little information to form their
priors, one could alternatively allow them to
consider a set of priors.
– If an individual is uncertainty averse, he will
choose lottery 𝑓 over another lottery 𝑔 if the
former provides a higher expected utility than the
latter according to his worst possible prior.

• Uncertainty aversion: Consider an individual
who is indifferent between two lotteries 𝑓 and
𝑔. Then, he is uncertainty averse if he weakly
prefers the compound lottery 𝛼𝑓 + 1 − 𝛼 𝑔
to lottery 𝑓, where 𝛼 ∈ (0,1).
– Intuition: a decision maker who is uncertainty
averse has a preference for mixing (or hedging),
since the compound lottery becomes at least as
valuable as either of the two lotteries alone.

• Certainty-independence: For any two lotteries
𝑓 and 𝑔 and a constant act 𝑘 (i.e., a certain
outcome or a lottery that remains constant
across all states), the decision maker weakly
prefers lottery 𝑓 to 𝑔 if and only if he prefers
𝛼𝑓 + 1 − 𝛼 𝑘 to 𝛼𝑔 + 1 − 𝛼 𝑘 , where 𝛼 ∈
(0,1).
– Certainty-independence axiom relaxes the IA as it
only requires that preferences over two lotteries
to be unaffected when each lottery is mixed with a
certain outcome 𝑘.
– A decision maker weakly prefers lottery 𝑓 to 𝑔 if
and only if
min ž 𝑢 𝑓 𝑠 𝑑𝑝 𝑠 ≥ min ž𝑢 𝑔 𝑠 𝑑𝑝(𝑠)

Œ∈‰ Ü Œ∈‰ Ü
– That is, the individual evaluates the expected utility

of lotteries 𝑓 and 𝑔 according to each of his
multiple priors 𝑝 ∈ 𝐶, and then selects the lottery
that yields the highest of the worst possible
expected utilities.

• Example:
– Consider a decision maker with Bernoulli utility
function 𝑢 𝑥 = 𝑥 , where 𝑥 ≥ 0 denotes monetary
amounts.
– Assume that the decision maker faces two lotteries
𝐿³ = $1, $100
𝐿º = ($3, $5)
– Also, assume that the decision maker’s priors are
(𝑝³ , 1 − 𝑝³ ) for 𝐿³
(𝑝º , 1 − 𝑝º ) for 𝐿º
– According to MEU, the decision maker chooses lottery
𝐿º if
min[𝑝º 3 + (1 − 𝑝º ) 5]
Œæ
≥ min[𝑝³ 1 + (1 − 𝑝³ ) 100]
Œç
– If the decision maker does not have any available
information with which to update his priors, priors can
take values (𝑝³ , 𝑝º ) ∈ (0,1).
– It is possible that in his most pessimistic belief, he
receives the lowest monetary amount with probability
one.
– Then, with argmin 𝑝º = 1,
min[𝑝º 3 + (1 − 𝑝º ) 5] = 3
Œæ
– Similarly, with argmin 𝑝³ = 1,
min[𝑝³ 1 + (1 − 𝑝³ ) 100] = 1
Œç
– Hence a decision maker with MEU preferences
selects lotter 𝐿º because 3 ≥ 1.

• Choquet expected utility (CEU):
– Define beliefs with the use of capacities.
– A capacity is defined as a real-valued function 𝑣(=)
from a subset of the state space 𝑆 to [0,1], with
the normalization 𝑣 ∅ = 0 and 𝑣 𝑆 = 1.
– If the capacity 𝑣(=) satisfies monotonicity, 𝑣(𝐴) ≥
𝑣(𝐵), where 𝐴 is a superset of 𝐵.
– We cannot use a standard integral over states
since the capacity 𝑣(=) does not correspond to our
notion of beliefs.
– A decision maker weakly prefers 𝑓 to 𝑔 if the
Choquet integrals satisfy
ž 𝑢 𝑓 𝑆 𝑑𝑣 𝑆 ≥ ž𝑢(𝑔 𝑆 )𝑑𝑣(𝑆)
Ü Ü
– The CEU and MEU models are connected if we

impose the uncertainty aversion axiom in CEU
context. For that we need that capacity 𝑣(=)
satisfies supermodularity, i.e.,
𝑣 𝐴 ∪ 𝐵 − 𝑣(𝐵) ≥ 𝑣 𝐴 ∪ 𝐶 − 𝑣(𝐶)
where 𝐶 is a subset of 𝐵, i.e., 𝐶 ⊂ 𝐵.
• Example:
– While the use of Choquet integrals is involved, the
literature often uses “simple” capacities.
– A simple capacity on state space 𝑆 can be understood as a
convex combination between two extreme capacities:
1. a standard probability weight on 𝐴, 𝑝 𝐴 ∈ 0,1 .
2. the “complete ignorance” capacity 𝑤, where 𝑤 𝑆 = 1 and
𝑤 𝐴 = 0 for every 𝐴 ⊆ 𝑆.

– Formally, simple capacities are defined as
𝑣 𝐴 = 𝜆𝑝 𝐴 + 1 − 𝜆 𝑤(𝐴)
for every 𝐴 ⊆ 𝑆 and where 𝜆 ∈ 0,1 .
– Parameter 𝜆 denotes the individual’s degree of confidence
on 𝑝 𝐴 , while (1 − 𝜆) captures his degree of ambiguity
about 𝑝 𝐴 .
– For further reading, see Haller (2000) and Aflaki (2013).

– Choquet, G. (1953). Theory of capacities. Ann. Inst. Fourier (Grenoble) 5 131-
295.
– Dow, J. and S. Werlang. (1992). Uncertainty aversion, risk aversion, and the
optimal choice of portfolio. Econometrica, (1), 197.
– Epstein, L. and T. Wang. (1994). Intertemporal Asset Pricing under Knightian
Uncertainty. Econometrica, (2), 283-322.
– Hansen, L., Sargent, T. (2001). “Robust Control and Model Uncertainty”.
American Economic Review 91, 60-66.
– Machina, M. (2014). Handbook of the economics of risk and uncertainty (First
edition). Elsevier.
– Mukerji, S. and J. Tallon (2004). Ambiguity aversion and the absence of wage
indexation. Journal of Monetary Economics, (3), 653-670.
– Nishimura, K. and H. Ozaki. (2004). Search and knightian uncertainty. Journal
of Economic Theory, (2), 299-333.
– Schmeidler, D. (1989). Subjective Probability and Expected Utility Without
Additivity. Econometrica, 57, 571-587.
– Uppal, R. and T. Wang. (2003). Model Misspecification and
Underdiversification. Journal of Finance, (6), 2465-2486.

Chapter 5 Choices Under Uncertainty

Uploaded by

Copyright:

Available Formats

Chapter 5 Choices Under Uncertainty

Uploaded by

Document Information

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

Chapter 5 Choices Under Uncertainty

Uploaded by

Copyright:

Available Formats

Advanced

Advanced Microeconomic Theory 2

Advanced Microeconomic Theory 3

Advanced Microeconomic Theory 5

• Points strictly inside the (0,1,0)

the three intercepts p1 p2

probability that the 1 2

at the opposite vertex

associated with the x1 + x2 = 1

Advanced Microeconomic Theory 10

the reduced lottery of

each simple lottery. 1

Advanced Microeconomic Theory 13

the reduced lottery of L5 =

Advanced Microeconomic Theory 15

– The decision maker is only concerned about the

Advanced Microeconomic Theory 18

Advanced Microeconomic Theory 19

not change the L

preference over the B( L ') La

two lotteries. L' Lb

Advanced Microeconomic Theory 23

Advanced Microeconomic Theory 27

Advanced Microeconomic Theory 28

– This violates the IA, which requires

Advanced Microeconomic Theory 34

Advanced Microeconomic Theory 35

for any 𝐾 lotteries 𝐿7 ∈ ℒ, 𝑘 = 1,2, … , 𝐾 and

• Intuition: the utility of the expected value of the 𝐾

Advanced Microeconomic Theory 37

where ∑7 𝛼7 𝑝17 is the total joint probability of

Advanced Microeconomic Theory 38

where ∑1 𝑢1 𝑝17 is the expected utility from playing a

Advanced Microeconomic Theory 39

for all 𝛼 ∈ 0,1 .

Advanced Microeconomic Theory 42

If L ~ L ', then L ~ L + (1 )L '

Straight indifference curves

Advanced Microeconomic Theory 45

Curvy indifference curve

which violates the IA.

Violations of the IA:

Advanced Microeconomic Theory 49

– First choice set:

which implies 𝐿. ≻ 𝐿O. .

Advanced Microeconomic Theory 52

Advanced Microeconomic Theory 53

– From the previous preferences over certain

Advanced Microeconomic Theory 56

Advanced Microeconomic Theory 65

Advanced Microeconomic Theory 66

Advanced Microeconomic Theory 68

Advanced Microeconomic Theory 69

Advanced Microeconomic Theory 70

Advanced Microeconomic Theory 71

Advanced Microeconomic Theory 72

Advanced Microeconomic Theory 73

Advanced Microeconomic Theory 80

Advanced Microeconomic Theory 84

for any lottery 𝐹(=)