Prospect Theory

&
Reference-Dependent
Preferences

AE6307
Behavioral Economics for Policy Analysis

(C) Yohanes E. Riyanto 1

Introduction
 Suppose you are an event manager and are planning
an outdoor sport event.
 The probability of rain tomorrow is 30% and no rain is
70%.
 If it does not rains you will make S$5000, but if it rains
you will only make S$1000.
 Expected value (EV) = 0.70(5000)+0.3(1000)=S$3800.
 Now, you can instead choose an indoor stadium, and
not be affected by the weather (for sure not affected
by rain). Your income would be S$2000.
 What should you do?  can compare the EV and the
sure payoff of S$2000.
Introduction
 Problem: comparing EVs (e.g. the monetary value) of
prospects can be misleading.
 Consider the following option (St. Petersburg game).
 You flip a coin repeatedly until a heads first appears.
The pot starts at S$1 and doubles every time a tails
appears. You win whatever is in the pot the first time
you throw heads and the game ends.
 For example:
 Heads on the 1st toss: get S$1
 Heads on the 2nd toss: get S$2
 Heads on the 3rd toss: get S$4
 Heads on the 4th toss: get S$8

(C) Yohanes E. Riyanto 3

Introduction
 Now suppose you face the following choice, which
would you prefer?
A.S$10 for sure
B.The St. Petersburg Game.

 Expected Value of the St. Petersburg Game can be

calculated  Prob. of getting a heads on a given
round:
1st round: ½
2nd round: Pr(Tails)*Pr(Heads)= ¼
3rd round: Pr(Tails)*Pr(Tail)*Pr(Heads)=1/8
kth round: (½)k

 EV =(1/2)*1+(1/4)*2+(1/8)*4+(1/16)*8+…
(C) Yohanes E. Riyanto 4
Introduction
 St Petersburg Game …

(C) Yohanes E. Riyanto 5

Expected Utility
 EV =(1/2)*1+(1/4)*2+(1/8)*4+(1/16)*8+…
EV =(1/2)+(1/2) +(1/2) +(1/2) +…
EV =∞
 The expected value of the game is infinite, and yet few
people will be willing take it  St. Petersburg Paradox.
 Daniel Bernoulli (1738)  the value of a gamble is not its
expected monetary value, but instead some subjective
value (satisfaction/ benefit/ consumption value/
usefulness) from consuming a good or service or an
outcome), or utility, that people attach to this monetary
value.
U($) = # of utils  ∂U($)/ ∂$>0 (more money is better)
$  used to consume goods x  U($)=U(goods)
 Thus, people maximize expected utility.
(C) Yohanes E. Riyanto 6
Expected Utility
 Another example showing that people do not value an
uncertain prospect (gamble/ lottery) by its expected
value:
 Option A  80% chance to win $100 and 20% chance
to win $10.
 Option B  $80 for sure.
 Which will you choose?
 EVA=$82 & EVB=$80
 People almost always pick the sure thing (option B)
rather then the gamble (option A).
 If evaluation is based on expected value  option
A should have been chosen.

(C) Yohanes E. Riyanto 7

Expected Utility
 Thus, people maximize expected utility.
 If there are 2 possible outcomes:
Expected Utility= Pr. (outcome 1)*U(outcome 1) +
Pr. (outcome 2)*U(outcome 2)
 Suppose my current wealth (X) is S$20,000 and my
utility is represented by the function U(X)=(3/5)X.
 Someone offers me a risky prospect (gamble): with
50% chance I’ll get S$10,000 and with 50% chance I’ll
get S$30,000. Should I accept this gamble?
 EU(X0)=U(S$20,000)=(3/5)*(S$20,000)=S$12,000.
 EU(X1) = 0.50*U(10,000)+0.50*U(30,000)
=0.50*(3/5)(10,000)+0.50*(3/5)(30,000)=S$12,000.
(C) Yohanes E. Riyanto 8
Expected Utility
 The two options have exactly the same Expected Value
(Expected Utility), so I should be indifferent between
the two of them  I am RISK NEUTRAL.
E
Utility 18 I am risk
neutral and is indifferent
between certain events
and uncertain events
with the same
C expected income.
12
U(x)=3/5 X
My marginal utility of
wealth (∂U(x)/ ∂X)=3/5 
A constant  linear utility
6 function.

Income ($1,000)
0 10 20 30
(C) Yohanes E. Riyanto 9
Expected Utility

 Suppose my utility is U(X)=√X.

 My expected utilities from the two options:
EU(X0)=U(S$20,000)= √(S$20,000)=S$141.42.
EU(X1)=0.50*U(10,000)+0.50*U(30,000)
=0.50*√10,000+0.50* √30,000)
=S$136.6
 Even though both options have the same expected
value, my current wealth gives me higher expected
utility, so I should not accept the gamble.
 I am RISK AVERSE.

(C) Yohanes E. Riyanto 10

Expected Utility

Utility
E
173.21
D
141.42 I am risk averse
because I would prefer
136.6 a certain income of
F
A $20,000 to an uncertain
100 expected income =
$20,000

U(x)= √X
My marginal utility of
wealth (∂U(x)/ ∂X)=(1/2)(1/√X) > 0
and (∂U2(x)/ ∂2X)=(-1/4) X-3/2 < 0
 concave utility function
(increasing at a decreasing rate)

Income ($1,000)
0 10 20 30

(C) Yohanes E. Riyanto 11

Expected Utility

 Suppose my utility is U(X)=X2.

 My expected utilities from the two options:
EU(X0)=U(S$20,000)=(S$20,000)2=S$4*108.
EU(X1)=0.50*U(10,000)+0.50*U(30,000)
=0.50* (10,000)2+0.50* (30,000)2
=S$5*108
 Even though both options have the same expected
value, the gamble gives me higher expected utility, so
I should accept the gamble.
 I am a RISK SEEKER.

(C) Yohanes E. Riyanto 12

Expected Utility

Utility
(in 108) E
9
I am risk
seeker because I
would prefer the gamble
F to a certain income with
the same EV.
5

4 C U(x)=X2
My marginal utility of
wealth (∂U(x)/ ∂X)=2X > 0 
A (∂U2(x)/ ∂2X)=2>0
1  convex utility function.
Increasing at an increasing rate

Income ($1,000)
10 20 30

(C) Yohanes E. Riyanto 13

Expected Utility

 Often in reality we may exhibit both risk-seeking

behavior and risk-averse behavior at the same time 
e.g., we buy 3D (lottery) but at the same time buy
home insurance.
 9

 We are often risk-neutral or even risk-seeking with

regards to small gambles, and risk-averse with
regards to large gambles.

 In order to be able to be represented by a utility

function, preferences over outcomes (lotteries) must
satisfy the following axioms (Von Neumann and
Morgenstern, 1946).

(C) Yohanes E. Riyanto 14

Expected Utility

 Axioms of expected utility:

Completeness: we can decide between two
alternatives  for every lotteries A and B, we have
either A ≺ B , A ≻ B or A ∼ B (A is worse than B, or
better, or equally good).
Transitivity:
for every lotteries A, B and C in which A
≽ B and B ≽ C, we must have A ≽ C.
Continuity: Let A, B and C be lotteries with A ≻ B ≻
C; then there exists a probability p such that a
combination of A and C; (pA + (1 − p)C) ∼B 
equally preferred to B.

(C) Yohanes E. Riyanto 15

Expected Utility

• Continuity: Let A, B and C be lotteries with A ≻ B ≻ C;

then there exists a probability p such that a
combination of A and C (pA + (1 − p)C) ∼B  equally
preferred to B.
• 3 outcomes  {get $1M, get nothing, die}
Get $1M) Get Nothing Die

(C) Yohanes E. Riyanto 16

Expected Utility
• Continuity: Let A, B and C be lotteries with A ≻ B ≻ C;
then there exists a probability p such that a
combination of A and C (pA + (1 − p)C) ∼B  equally
preferred to B.
• 3 outcomes  {get $1M, get nothing, die}
• 3 lotteries  A={2/3,1/3,0}, B={0,1,0} and
C={0,0,1}.
• It is natural to think that A≻B ≻C  nobody wants
to die  completeness and transitivity are satisfied.
Lottery (Gamble) A Lottery (Gamble) B Lottery (Gamble) C
66.67% 33.33% 100% 100%
chance chance chance chance

> >

(C) Yohanes E. Riyanto 17

Expected Utility
• Continuity: Let A, B and C be lotteries with A ≻ B ≻ C;
then there exists a probability p such that a
combination of A and C (pA + (1 − p)C) ∼B  equally
preferred to B.
• Now consider; A’=pA+(1- p)C  chance of getting
$1M with a strictly positive probability of death is
equally preferred to stay alive for sure, as long as p
is sufficiently large.

Lottery (Gamble) A’
p = 98% Lottery (Gamble) B
A’ = 98%*A + (1-98%)*C
100%
65.67% 32.67% 2% chance

≥
chance chance chance
•

(C) Yohanes E. Riyanto 18

Expected Utility
 Axioms of expected utility:
Independence: if we mix two lotteries with a third
one, the preference ordering of the two mixtures will
not change (is independent of) the particular third
lottery used.
If A≻B, then pA+(1-p)C ≻ pB+(1-p)C.
p = 98% p = 98%
A’ = 98% * A + (1-98%) * C B’ = 98% * B + (1-98%) * C

65.67% 32.67% 2% 98% 2%

chance chance chance chance chance

>

 Adding some positive risk of death to the chance of

getting $1M would still be preferred to adding some
positive risk of death to the chance of (C)
getting 0$).
Yohanes E. Riyanto 19
Expected Utility
 Independence axiom is the one that is most often violated
in reality  we will look at this in detail later.
 If those 4 axioms are satisfied, then we can formulate
expected utility theory for a rational decision maker
 Those 4 axioms imply:
1. Linear expectation of a utility function:
U(x1,p1;…; xn,pn)=p1u(x1)+…+pn u(xn)

The overall utility of a prospect (U) is the expected utility

of its outcomes.

2. Asset integration: gambles are evaluated by

integrating gains or losses with current wealth (w).
U(w+x1,p1;…; w+xn,pn)=p1u(w+x1)+…+ pn u(w+xn) > U(w)
(C) Yohanes E. Riyanto 20
Violation of the EU Theory (Allais Paradox):

 Consider the following options:

A: 98% chance of $500M
2% chance of $0
B: $100M for sure
Maurice Allais
 What about the following options:
X: 0.98% chance of $500M
99.02% chance of $0
Y: 1% chance of $100M
99% chance of 0.
 This pattern of choice is ‘irrational’  it violates the
independence axiom  why? (C) Yohanes E. Riyanto 21
Violation of the EU Theory (Allais Paradox):
 The second problem: The first problem:
\

X: 0.98% chance of $500M A: 98% chance of $500M

99.02% chance of $0 2% chance of $0

Y: 1% chance of $100M B: $100M for sure

99% chance of 0.
The 2nd problem = 1% chance
 Re-write the second problem: of reaching the first problem

X: 0.98% chance of $500M X: 98% chance of $500M

0.02% chance of $0 2% chance of $0
99% chance of $0 99% chance of $0

Y: 1% chance of $100M Y: 100% chance of 100M

99% chance of 0. 99% chance of 0.
(C) Yohanes E. Riyanto 22
Violation of the EU Theory (Allais Paradox):
 Independence axiom is violated here
If A≺ B, then 𝐗𝐗 = pA + (1 − p)C ≺ 𝐘𝐘 = pB + (1 − p)C.
The first problem:
\

A: 98% chance of $500M

2% chance of $0
B: $100M for sure
p = 1%
X = 1% * A + (100%-1%) *C

1% * A: 99% * C:
1% * 98% chance of $500M + 1% 99% * chance of $ 0M
* 2% chance of $0

0.98 % chance of $ 500 M +

0.02% chance of $0 (C) Yohanes E. Riyanto 23
Violation of the EU Theory (Allais Paradox):
 Independence axiom is violated here
If A≺ B, then 𝐗𝐗 = pA + (1 − p)C ≺ 𝐘𝐘 = pB + (1 − p)C.
The first problem:
\

A: 98% chance of $500M

2% chance of $0
B: $100M for sure
p = 1%
Y = 1% * B + (100%-1%) *C

1% * B: 99% * C:
1% * 100% chance of $100M 99% * chance of $ 0M

1 % chance of $ 100 M

(C) Yohanes E. Riyanto 24

Violation of the EU Theory (Allais Paradox):

If A ≺ B, then X=pA+(1-p)C ≺ Y=pB+(1-p)C

A: 98% chance of $500M
2% chance of $0
B: $100M for sure

We know from the above choice that B ≻ A, hence

by independence axion, we should also have Y ≻
X, but we don’t; instead we have X ≻ Y

X: 0.98% chance of $500M Y: 1% chance of $100M

0.02% chance of $0 99% chance of $0
99% chance of $0

(C) Yohanes E. Riyanto 25

Violation of the EU Theory (Allais Paradox):

More Examples (Kahneman and Tversky, 1979):

 The certainty effects  Which would you prefer?
A: 80% chance of $4,000. C: 20% chance of $4,000.
N=95 [20] [35]
B: $3,000 for sure D: 25% chance of $3,000
[80] N=95 [65]

 EU Theory: If you prefer B to A (B≻A), then you

should also prefer D to C (D≻C)

(C) Yohanes E. Riyanto 26

More Examples (K&T, 1979):
 More examples  which would you prefer?
A: 50% chance to win a 3-week tour of Europe.
N=72 [22]
B: A one week tour of Europe for sure.
[78]
 What about this?
C : 5% chance to win a 3-week tour of Europe.
N=72 [67]
D: 10% chance to win one week tour of Europe.
[33]
 If you prefer B to A (B≻A), then according to EU theory
you should also prefer D to C (D≻C)

(C) Yohanes E. Riyanto 27

More Examples (K&T, 1979):
 Which would you prefer?
A: 45% chance of $6,000. C: 0.001% chance of $6,000.
N=66 [14] N=66 [73]
B: 90% chance of $3,000. D: 0.002% chance of $3,000
[86] [27]
 In options A and B, the prob. of winning are large 
most people choose the option where winning is more
probable.
 In options C and D, there is possibility of winning but
the chances are extremely small  most people would
choose the option which gives larger gains.
 Possibility and Probability effects.

(C) Yohanes E. Riyanto 28

More Examples (K&T, 1979):
 The Reflection Effects  Gains and Losses:
A: 80% chance of $4,000. A: 80% chance of -$4,000.

B: 100% chance of $3,000. B: 100% chance of -$3,000.

A: 20% chance of $4,000. A: 20% chance of -$4,000.

B: 25% chance of $3,000. B: 25% chance of -$3,000.

A: 90% chance of $3,000. A: 90% chance of -$3,000.

B: 45% chance of $6,000. B: 45% chance of -$6,000.

A: 0.002% chance of $3,000. A: 0.002% chance of -$3,000.

B: 0.001% chance of $6,000. B: 0.001% chance of -$6,000.

(C) Yohanes E. Riyanto 29
More Examples (K&T, 1979):
 The Isolation Effects:
People like to simplify choices (e.g. between 2 options) 
zoom in (isolate) only on components that distinguish them 
may lead to inconsistent preferences.

 Example: Suppose you have the following sequential

(2-stage) game:
Inthe 1st stage: 75% chance of ending the game with nothing
($0), and 25% chance of moving to the 2nd stage.

If you reach the 2nd stage, you face a choice between 2 options:

A: 80% chance of getting $4000

B: getting $3000 for sure.

You must made your choice between A and B before the game
starts.
(C) Yohanes E. Riyanto 30
More Examples (K&T, 1979):
 The Isolation Effects … You have a choice between a risky
gamble and a sure gamble.
$3000 A: 80% chance of getting $4000
100%
B: getting $3000 for sure.
25%
80%
More Risk-Averse
$4000

20%
$0 You essentially have a choice
75% between 25%*80%=20% chance
of getting $4000, and 25%*100%=
$0 25% chance of getting $3000.
$3000
You can present the game
25%
differently as shown in this tree
c  Standard form
75% $0 C: 25% chance of $3,000.
20% $4000 N=141 [22]
c D: 20% chance of $4,000.
[78]
80% $0 More Risk-Seeking  choice
between 2 gambles.
(C) Yohanes E. Riyanto 31
Framing of Outcomes (K&T, 1984):
 Framing of outcomes matters.

Imagine that the U.S. is preparing for the outbreak of

an unusual Asian disease, which is expected to kill 600
people. Two alternative programs to combat the
disease have been proposed. Assume that the exact
scientific estimates of the consequences of the
programs are as follows: framed as gain

If program A is adopted 200 people will be 72%

saved.

If program B is adopted, there is a 1/3

probability that 600 people will be saved and a 28%
2/3 probability that no people will be saved.

(C) Yohanes E. Riyanto 32

Framing of Outcomes (K&T, 1984):

Imagine that the U.S. is preparing for the outbreak of

an unusual Asian disease, which is expected to kill 600
people. Two alternative programs to combat the
disease have been proposed. Assume that the exact
scientific estimates of the consequences of the
programs are as follows:

If program A is adopted 400 people will die. 20%

If program B is adopted, there is a 1/3

probability that nobody will die and a 2/3 78%
probability that 600 people will die.

framed as loss

(C) Yohanes E. Riyanto 33

Framing of Outcomes (K&T, 1984):

 Kahneman and Tversky (1979).

 Suppose you are asked to choose between the
followings:

 In addition to whatever you own, you have been

given 1000. You are now asked to choose between:
A: (1000, prob.=0.5) and B: (500)
N=70 [16] [84]
 In addition to whatever you own, you have been
given 2000. You are now asked to choose between:
C: (-1000, prob.=0.5)and D: (-500)
N=68 [69] [31]

(C) Yohanes E. Riyanto 34

Reference Point

 People evaluate a prospect based on a reference

point  consider the following example:
 Today Jack and Jill each have a wealth of $5
million and they have the same utility function.
 Yesterday Jack had $ 1 million and Jill had $9
million.
 Do they have the same utility? (equally happy?).
 Under EU theory, they should be equally happy
today (the same wealth).
 But, it is obvious that Jack would be happier than
Jill.
(C) Yohanes E. Riyanto 35
Prospect Theory
 Different from the standard EU theory, people do not
integrate gains and losses  instead evaluate a
prospect (x1,p1;…; xn,pn) by using 2 distinct thinking
phases  editing phase & evaluation phase.
 Editing Phase  consists of 4 components.
Coding: people perceive outcomes as either gains or
losses, relative to some neutral reference position.
Reference point  e.g. current asset position, status quo,
expectation, social norms, etc).

Combination:prospects are sometimes simplified by

combining probabilities for identical outcomes.
For example  ($200, 25% & $200, 25%)  will just be
evaluated in the form of ($200, 50%).
(C) Yohanes E. Riyanto 36
Prospect Theory
 Editing phase…
Segregation: Riskless components of are segregated out
of the risky components.
For example  ($300, 80% & $200, 20%)  will be
decomposed into a sure gain of $200 and a risky
prospect of ($100, 80%).

Cancellation: components that are shared by 2 prospects

are discarded.
For example in the isolation effect discussed before the
1st stage of the game ($0, 75% & 2nd stage, 25%) was
common for both prospects (A and B) and thus ignored..

(C) Yohanes E. Riyanto 37

Prospect Theory
 Evaluation Phase:
Under EU theory, the utility obtained from a prospect is
represented in a linear expectation form:

U(x1,p1;…; xn,pn)=p1u(x1)+…+pnu(xn)
Incontrast, under prospect theory, people do not weigh
probabilities linearly, but instead attach a decision weight
π(p) to each probability.
U(x1,p1;…; xn,pn)= π(p) u(x1)+…+ π(pn) u(xn)
 Decision weight (π(p)):
Example: a cash gift worth $300  assign a value v=1.
suppose you are only given a ticket to a lottery that has a
prize of $300. How does the value of the ticket vary with
the probability of winning?
(C) Yohanes E. Riyanto 38
Prospect Theory
 Decision weight (π(p)) …
 Example: a sure cash gift worth $300  assign a value
v=1. suppose you are only given a ticket to a lottery
that has a winning prize of $300. How does the
desirability of the ticket (v) vary with the probability of
winning?
 Intuition: the value should be v ε [0,1], and is likely
not to be a linear function of the prob.
 Consider an increase in the probability:
Prob. Δprob. A change from impossibility to
possibility (0%5%) OR from
0%5% 5%
possibility to certainty (95%100%)
30%35% 5% should have a bigger impact than
a comparable change in the middle
95%100% 5%
(30%35%).

(C) Yohanes E. Riyanto 39

Prospect Theory
 Decision weight (π(p))
 Another example: you are forced to
play a Russian roulette game  the gun
is loaded with 6 bullets. You are given
the opportunity to pay for the removal
of 1 bullet from this loaded gun.
 Would you pay as much as to reduce the number of
bullets from 4 to 3 as you would reduce the number of
bullets from 1 to 0?
 Would you pay as much as to reduce the number of
bullets from 4 to 3 as you would reduce the number of
bullets from 6 to 5?
 Removing 1 bullet from 4 is not really an improvement
from removing the only bullet in the gun, or removing 1
bullet when all chambers are loaded.
(C) Yohanes E. Riyanto 40
Prospect Theory
 Russian roulette…
 In other words, changing the probability from 3/6 to 4/6
has a smaller impact on the value of prospect than
changing them from 0 to 1/6, or from 5/6 to certainty.
 It suggests that the probability weighting function π(p)
must be concave near zero and convex near one.

45o line  π(p) =p*

prob. weighting
function π(p)

(C) Yohanes E. Riyanto 41

 Prospect Theory – Decision Weight
 How much would you pay to remove 1 bullet when
there are 4 bullets in your gun chambers?

From Drazen Prelec’s Lecture Notes (MIT) (C) Yohanes E. Riyanto 42

 Prospect Theory – Decision Weight
 How much would you pay to remove 1 bullet when
there is only 1 bullet in your gun chambers?

From Drazen Prelec’s Lecture Notes (MIT) (C) Yohanes E. Riyanto 43

 Prospect Theory – Decision Weight
 How much would you pay to remove 1 bullet when
there are 6 bullets in your gun chambers?

From Drazen Prelec’s Lecture Notes (MIT) (C) Yohanes E. Riyanto 44

Prospect Theory – Decision Weight
 Summary of Properties of the probability weighting
function:
 Overweighting of small probabilities: For all p<p*,
we have π(p)>p.

Rare events are given greater psychological weight

in our minds than is normatively appropriate  we
tend to overestimate small probability event.

 Underweighting of large probabilities: For all p>p*,

we have π(p)<p.

 π(p)=p at around 20% - 40%.

(C) Yohanes E. Riyanto 45

Prospect Theory – Decision Weight
 Functional form that fits well the described probability
weighting function.
pγ
π p = 1 for 1 ≥ γ > 0
pγ + 1 − p γ γ

γ = 0.5
associated weight

γ = 0.60

γ = 0.75

γ=1

The function is more s-shape

for affect-rich items  airplane
crashes, emotional outcomes,
disease deaths.

probability (C) Yohanes E. Riyanto 46

Prospect Theory – Decision Weight
 Imagine that you hold the following lottery ticket, which would you
prefer?
A. Or B.

 If the chance of getting either A) or B) is 1%?

 If it is for certain they can get either A) or B)?

 “Money, Kisses and Electric Shocks: On the Affective Psychology of

Risk, Psychological Science, 12(3), 185-190 (Rottenstreich, Y. and C.
Hsee)
(C) Yohanes E. Riyanto 47
Prospect Theory – VALUE FUNCTION
 In addition to the probability weighting function, the
shape of value function is another important aspect
that distinguishes prospect theory from EU theory.
 People treat gains and losses separately instead of
integrating them into the current wealth levels.
 Gains and losses are defined relative to reference
point (e.g.: status quo, current assets, social norms, or
expectations).
 I am gambling and losing  getting back even may be my
reference point, or at least covering my entry fee $100
could be my reference point.
 My GPA improves by 0.1. but if others’ GPA improves by
0.3 I may view an increase by 0.1 as a loss.
(C) Yohanes E. Riyanto 48
Prospect Theory – VALUE FUNCTION
 Loss Aversion: losses loom larger than gains  risk
seeking in losses and risk averse in gains.
risk-averse

−𝑉𝑉 −𝑋𝑋 > 𝑉𝑉 𝑋𝑋

losses
reference
point

Xα for X ≥ 0
V X =� α with λ > 1
−λ −X for X < 0

λ=coefficient of loss aversion

For simplicity we assume π(p)=p.
risk-seeking
(C) Yohanes E. Riyanto 49
Prospect Theory – VALUE FUNCTION
 Putting everything together  combining the
probability weighting function and value function:

V X, p; Y, 1 − p =π p V X +π 1−p V Y
with X ≥ 0 ≥ Y
 Back to our example of Asian Disease: Saving 200 lives
for sure is better than gambling to save 600.

(C) Yohanes E. Riyanto 50

Prospect Theory – VALUE FUNCTION
 Having 400 deaths for sure looks worse than having
some chance of zero deaths.

 Example: suppose the reference point is 0, and for

simplicity assume π(p)=p.
X 0.8 for X ≥ 0
V X =� 0.8 with λ = 2
−2 −X for X < 0
(C) Yohanes E. Riyanto 51
Prospect Theory
 Suppose an individual is thinking of playing ‘flip a coin
game’  heads wins $600 and tails losses $500.
 Do nothing: U(0)=0
 Flip: V=0.5*6000.8+(-2)*0.5*5000.8=-60

 Loss aversion  makes this favorable gamble

unattractive.
 Another example: you are used to making $1000 a year
risk free (this becomes your reference point R), but can
buy bonds such that 30% of the time you make $600
and 70% of the time you make $1200.
 Do nothing: U(0)=0
 bonds: V=0.7*(1200-1000)0.8+(-2)*0.3*(600-1000)0.8=-
24  unattractive (loss aversion)
(C) Yohanes E. Riyanto 52
Ambiguity Aversion
 Ellsberg Paradox (Ellsberg, 1961)  violation of the
expected utility theory  people dislike ambiguity (the
devil you know is still better than the devil you don’t
know).
 2 Urns (I and II): Blue and Red balls.

 Subjects are asked which they would rather bet on:

a. A blue draw from urn 1 or a red draw from urn 1.
b. A blue draw from urn 2 or a red draw from urn 2.
c. A blue draw from urn 1 or a blue draw from urn 2.
d. A red draw from urn 1 or a red draw from urn 2.
(C) Yohanes E. Riyanto 53
Ambiguity Aversion
 Result:

 Since subjects prefer to choose urn 1 rather than urn 2

in choice c (betting on blue balls), it must be (correctly
or wrongly) that she thinks urn 2 has more red balls
than blue balls.
 But if this is the case, they should choose urn 2 and not
urn 1 in choice d (betting on red balls), because she
thinks (correctly or wrongly) that urn 2 has more red
balls. (C) Yohanes E. Riyanto 54
 Applications:

Prospect Theory
&
Reference-Dependent Preferences

(C) Yohanes E. Riyanto 55

Endowment Effect

 Willingness to Pay (WTP) for a good is the maximum

amount of money you would give up in order to obtain
the good.
 My WTP for Coke is $1, or $2 if I am really thirsty.
 Willingness to Accept (WTA) for a good is the minimum
amount of money you would require in order to give up
your good.
 If you give me $1 I am willing to give up my Coke,
but if I am really thirsty you have to give me $2.

WTP is how much you would buy the good for;

WTA is how much you would sell the good for

(C) Yohanes E. Riyanto 56

Endowment Effect

 Example: A big match between

Chelsea FC and Manchester
City.
 The match is sold out. I bought the ticket early at $75.
 I am a Chelsea fan, and willing to pay for the ticket up
to $150.
 Now, I learn that the ticket is sold out, and in black
market the ticket is sold at $300.
 Because I am such a big fan, I still don’t want to sell at
$300.
 My WTP (max buying price) is $150, but my WTA (max
selling price) is above $300  endowment effect.
(C) Yohanes E. Riyanto 57
Endowment Effect
 According to the standard economic theory: WTP≅WTA.
 WTP: Lose $ but gain the good  -U($)=U(good)
 WTA: Gain $ but lose the good  U($) = -U(good)

 They may differ slightly, but only because of (small)

income effects  in the WTA case: I am richer by the
amount of the good.
W
Economic Theory:
Indifference Curve is reversible

WTA ≅ WTP
No disparity
W0 B

WTP
WTA A
W1 U0
Q0 Q1 Q
(C) Yohanes E. Riyanto 58
Endowment Effect
 Another Example:

 Suppose your neighbor has a chicken coop in his

backyard. There is 0.1% chance you will get bird flu and
die because of it. How much would you be WTP for him to
eliminate this risk?

 Now suppose instead your new neighbor is moving in and

wants to install a chicken coop in his backyard. There is
0.1% chance you will get bird flu and die because of it.
How much would your neighbor have to pay you for you
to be WTA this risk?

 Most likely WTA>>WTP  disparity between WTA&WTP 

cannot be explained with the standard economic theory.

(C) Yohanes E. Riyanto 59

Endowment Effect

(C) Yohanes E. Riyanto 60

Endowment Effect
 Kahneman, Knetsch and Thaler (1990)
 Experiments with Cornell Univ. undergraduate
students.

 First step: conducted induced-value experiment

(control treatment). Half subjects were given tokens
and the other half were not. Each token is equivalent
to $X, and each individual (regardless of whether
owning the token or not) is drawn with different value
of $X (e.g. $6, $8, …)

 We can rank the valuation ($X) from low to high.

 Experimenter set a market for token and allowed a
non-owner to buy a token from an owner. Unsold
token can be exchanged for $X.

(C) Yohanes E. Riyanto 61

Endowment Effect
 Kahneman, Knetsch and Thaler (1990)
 Supply and demand schedule can be constructed  they
found that market clearing price is more or less at S=D,
for example:

(C) Yohanes E. Riyanto 62

Endowment Effect
 Kahneman, Knetsch and Thaler (1990)
 The induced-value experiment showed that subjects
understood the task.
 Second step: half subjects were given Cornell coffee
mugs (priced $6). All subjects are asked to examine their
own (if they were given) or their neighbors’ mugs.
 Subjects who did not receive a mug were asked how
much they would be WTP to get a mug. Those who were
given a mug were asked their minimum WTA to sell away
their mug.
 Prediction: since the mugs were randomly assigned, half
of the mugs lovers (those who have valuation above the
median valuation) would end up with mugs and the
others (mugs haters) would end up having no mugs 
half of the mugs should change hands.
(C) Yohanes E. Riyanto 63
Endowment Effect
 Kahneman, Knetsch and Thaler (1990)
 Fewer than expected number of mugs changing hands.
 WTA>WTP, WTA is almost double of WTP.
 Same qualitative results when the items were pens
instead of mugs.

(C) Yohanes E. Riyanto 64

Endowment Effect
 Kahneman, Knetsch and Thaler (1990)
 The coined the results as ENDOWMENT EFFECT  people
value a good or service more when they actually posses it
or have the property right over it, than when they don’t.
 Owning it is the reference point  losing (parting with) it
feels like a loss  LOSS AVERSION.
 Robustness check: conduct the same experiment, but add
1 more group called the “choosers”  did not get a mug
but asked to choose whether they would like to either get
the mug or the money at each prices that buyers and
sellers considered.
 This will allow us to check whether the low amount of
trading is due to the seller reluctance to sell or buyer
reluctance to buy.
 Result: Sellers were more reluctant to part with the good.
(C) Yohanes E. Riyanto 65
Endowment Effect
 Knetsch (1990)

 Map indifference curves. Subjects were divided into 2

groups. Subjects in one group were given pens, and
subjects in the other group were given $5.40.

 They were then presented with a series of offers which

they could accept or reject.

 If accept, those with pens (money) would receive

money (pens)  Marginal Rate of Substitution (MRS).

 Accepted and rejected offers can then be plotted  we

get indifference curves.

(C) Yohanes E. Riyanto 66

Endowment Effect
 Knetsch (1990) …

 The pens were worth more money to those subjects

who started with pens than those started with
money  because of the endowment effects.
 Indifference curves would cross  violation of the
standard economic theory. (C) Yohanes E. Riyanto 67
Endowment Effect
 Knetsch (1990) …

W
W
Economic Theory:
Indifference Curve is reversible

C
WTA ≅ WTP
No disparity
W0 B W0 B
WTA
WTP WTP
WTA A A
W1 U0 W1 U0
Q0 Q1 Q Q0 Q1 Q

(C) Yohanes E. Riyanto 68

Endowment Effect
 Prospect Theory:

 Value of getting a pen and lose $(X)

V = U pen + U −$X = A − λX

 Pay if: A-λX≥0, thus WTP=A/λ

 Value of losing a pen and getting $X

V = U −pen + U $X = −λA + X

 Accept if: -λA+X≥0, thus WTA=λA

 Thus: WTA>WTP.

(C) Yohanes E. Riyanto 69

Insurance Decision: Deductibles vs. Rebates
 Consider the following choices:
Imagine that you have just bought a new S$100,000 car and are
buying insurance for your car. The insurance package described
below includes all coverage mandated by the law including
comprehensive and collision insurance. Suppose you are offered
the policy described below:

“This policy has a deductible of $600 which will be

subtracted from the total claims against the policy.
In other words, if you make any claims against the
policy, the company will give you the total amount of
the claims minus the deductible. If your claims in
one year total less than $600, the company will pay
nothing. If your claims exceed S$600, the
company will pay all of the amount above S$600.”
Would you pay a premium of $1000 for one year of this
coverage? [44.3% YES] (C) Yohanes E. Riyanto 70
Insurance Decision: Deductibles vs. Rebates
 Consider the following choices:
Imagine that you have just bought a new S$100,000 car and are
buying insurance for your car. The insurance package described
below includes all coverage mandated by the law including
comprehensive and collision insurance. Suppose you are offered
the policy described below:
“With this policy, a rebate of S$600 minus any claims
paid will be given to you at the end of the year. In
other words, if you have no claims against the policy,
the company will give you S$600 back at the end of
the year. If you do file one or more claims, you will
get back S$600 minus the amount the company paid
out for your claims. Should your total claims exceed
S$600, the company will give you no rebate but will
pay the claims.
Would you pay a premium of $1600 for one year of this
coverage? N=187 [67.8% YES] (C) Yohanes E. Riyanto 71
Insurance Decision: Deductibles vs. Rebates

additional value of
premium rebate
deductible
rebate

value of
deductible

reference
point

 Deductible (a loss from the status quo) has bigger impact

because of loss aversion  the rebate is recorded as gains
and hence has smaller impact, even after considering the
effect of additional premium.
(C) Yohanes E. Riyanto 72
Behavioral Finance:
Myopic Loss Aversion and the Equity Premium Puzzle

 Study shows that since 1926 the average real annual returns
on stocks (equities) has been about 7%, while the real return
on treasury bills (or bonds) has been less than 1%.

 While it is normal to have the returns from stocks higher

than the returns from t-bills (bonds) because stocks are
riskier  nevertheless the substantial premium of stocks
returns to bonds is hard to justify.

 Mehra and Prescott (1985) estimate that investors would

need to have a ridiculously high coefficient of risk aversion
over 30 (they must be extremely risk averse) to justify the
substantial risk premium  equity premium puzzle.

(C) Yohanes E. Riyanto 73

Behavioral Finance:
Myopic Loss Aversion and the Equity Premium Puzzle

 This implies that a person would be indifferent

between a gamble with 50% chance of S$50,000 &
50% chance of S$100,000, and S$51,209 for sure.

Consider the following options:

∼ B: 100% chance of
A: 50% chance of $100,000
50% chance of $50,000 $51,209

< B: 100% chance of

A: 50% chance of $100,000
50% chance of $50,000 $51,500

(C) Yohanes E. Riyanto 74

Behavioral Finance:
Myopic Loss Aversion and the Equity Premium Puzzle
 It is almost impossible that somebody is this extremely risk
averse.
 Behavioral economics explanation is myopic loss aversion,
which involves loss aversion and a frequent portfolio
evaluation period (narrow framing - myopic way of
evaluating portfolio).
 Benartzi and Thaler (1995): using the prospect theory value
function (with loss aversion:

Xα for X ≥ 0
V X =� α with λ > 1
−λ −X for X < 0

 Stocks have the probability of yielding negative returns  loss

aversion makes stocks less attractive than the sure returns
from t-bills (bonds).
(C) Yohanes E. Riyanto 75
Behavioral Finance:
Myopic Loss Aversion and the Equity Premium Puzzle
 However, stocks can potentially give high returns  over long
horizon  stock returns dominate T-bills or bond returns.
 So the longer the investors intend to hold stocks, the more
attractive stocks would appear, as long the investment is not
evaluated frequently as losses will be more likely.
 Hence, two factors that contribute to an investor being unwilling to
bear the risks associated with holding equities would be loss
aversion and a short evaluation period (myopic loss aversion).
 Benartzi and Thaler (1995):

Assuming investors behave according to the

prospect theory, how often do they have to
evaluate the changes in their portfolios to make
them indifferent between the (US) historical
distributions of returns on stocks and bonds?.
(C) Yohanes E. Riyanto 76
Behavioral Finance:
Myopic Loss Aversion and the Equity Premium Puzzle
 They found out that the prospect values of stocks and
bonds investments will be about the same when;
 Stocks yield returns 8% more than bonds (just like the
observed premium from the data).
 The evaluation period is more or less 1 year.

(C) Yohanes E. Riyanto 77

Behavioral Finance: Disposition Effect
 People dislike incurring losses much more
than they like incurring gains and are willing
to gamble in the domain of losses.
 DISPOSITION EFFECT: Investors will tend to hold on losing
stocks (that have lost value relative to the purchase price)
TOO LONG, and will be TOO EAGER to sell stocks that have
risen in value.
 Odean (1998) provides convincing evidence with data obtain
from a brokerage firms  10000 randomly selected
customer accounts.
 Not enough just by simply look at the actual
number of securities sold for a gain vs. those
sold for loss  this is because in an upward
moving market, there are naturally more
winners than losers in the portfolio they sell.
(C) Yohanes E. Riyanto 78
Behavioral Finance: Disposition Effect
 So must evaluate the tendency for people to sell their
winners and losers relative to their opportunities to sell.
 For this, he divided securities into realized gains, realized
loss, paper gains and paper losses.

 Proportion of gains & loss realized (PGR & PLR):

Realized Gains
PGR = --------------------------
Realized Gains + Paper Gains

Realized Losses
PLR = --------------------------
Realized Losses + Paper Losses

 Investors held losing stocks 124 days and held winners

104 days.
Behavioral Finance: Disposition Effect
 Investors said that they held on losers because they expect
them to bounce back, BUT from Odean’s data the unsold
losers returned only 5% in the following year, whereas the
sold winners returned 11.6%.
 Prospect Theory:

 Reference point: the price of purchase.

 Loss aversion: gamble by holding on losing stocks.
 Risk averse in gains: selling winning stocks too soon.

(C) Yohanes E. Riyanto 80

Behavioral Finance: Disposition Effect

(C) Yohanes E. Riyanto 81

Loss Aversion & Disposition Effect in Housing
Market
 In housing market booms: houses (apartments) sell quickly
at prices close to (and sometimes above) sellers’ asking
prices.
 In housing busts, on the contrary, homes (apartments) stay
on the market for long periods with asking prices that are
well above expected market prices  many sellers
eventually withdraw their properties from sale, rather than
lowering their price.
 Loss aversion: prefer to have the house
unsold rather than selling at a loss.
 Genoseve and Mayer (2001) using data
from Boston area condominium markets
provide this evidence.
(C) Yohanes E. Riyanto 82
Racetrack Betting: Favorite Longshot Bias
 Perimutuel Betting: betting system where the holders of
winning tickets divide the total amount of money bet on a
race (the pool), after deductions for tax and racetrack costs.
 In a perimutuel betting the bettors itself determines the
payoff odds (e.g., if many people have bet on the actual
winner of a contest then the payoff will be low, simply
because many winners will divide the pool).
 Fixed Odds Betting: the odds for
a particular outcome is determined
and given in advance and if this
outcome occurs you are paid out
based on the odds you bet at,
regardless of what every other
individual betting on the event wagered on. You are not
betting against other bettors but instead against the odd-
maker (the house). (C) Yohanes E. Riyanto 83
Perimutuel Betting vs Fixed Odds Betting
 Longshot bias: bias toward over-betting on “longshots”,
which are horses with relatively small chance (high odd) of
winning and under-betting on “favorites”, which horses with
big chance (small odd) of winning..

 In other words: “Would you prefer to bet on a favorite but if

you win you only receive 30% profit or would you prefer to
bet on an underdog and if you win you receive 1000$
profit”.

 Many people prefer the latter

because the profit when
winning is large  but forget
that the prob. of winning is
also really small.
(C) Yohanes E. Riyanto 84
Racetrack Betting: Favorite Longshot Bias
 Snowberg and Wolfers (2010) shows that the favorite
longshot-bias is evident everywhere and remarkably
persistent.

(C) Yohanes E. Riyanto 85

Racetrack Betting: Favorite Longshot Bias
 Explanation: Prospect Theory  probability weighting
function  overweighting of small probabilities and
underweighting of large probabilities.

prob. weighting
function

 Leading to anomaly  more people are willing to bet for

large odd despite of negative rate of returns per dollar bet.
(C) Yohanes E. Riyanto 86
Loss Aversion among Golfers

 Rules for Scoring

 18 holes (full course)  each hole is categorized by
‘par’, i.e. the number of strokes required to put the ball
into the hole.
 Minimum par is 3  one long swing from the tee and
two putts in the green. Usually, a hole is either a par-4
or par-5 hole.
 The goal: get as few strokes as possible to get into the
hole. (C) Yohanes E. Riyanto 87
Loss Aversion among Golfers
 The summary of scores (source: Wikipedia):

(C) Yohanes E. Riyanto 88

Loss Aversion among Golfers
 Do golf players exhibit aversion to losses when exercising
“putt”?

 “Is Tiger Woods Loss Averse? Persistent Bias in the Face of

Experience, Competition, and High Stakes”, (Pope, D. and M.
Schweitzer, American Economic Review, (2011) 101(1), 129-
157).
 Compared putts to aim for a par or for over pars (bogey or
double bogey) and putts to aim for a birdie (one shot less
than par) or an eagle (two shots less than par) in more than
2.5 millions putts in pro golf tournaments.
(C) Yohanes E. Riyanto 89
Loss Aversion among Golfers
 Results: golfers are influenced by the reference point of par.
When golfers are having a stroke to reach either an eagle or
a birdie) they are significantly less accurate than when they
are having a stroke to reach a par, or a bogey, or a double
bogey.
 This finding is consistent with loss aversion  players invest
more effort (focus) when putting for par to avoid a loss.

 They were able to rule out other explanations such as

tournament specific differences (the terrain) and also
psychological factors.

(C) Yohanes E. Riyanto 90

Loss Aversion among Golfers

(C) Yohanes E. Riyanto 91

Reference Point: Round Numbers as Goals
 SAT test (the standardized test for college admission in the
US).
 Until 2006, SAT scores were between 400 – 1600 in
intervals of 10.
 Students are allowed to retake the test.

 If round numbers act as a

performance goal, then those who
scoring just below a round number
would be more likely retaking the
test than those scoring just above it.
 “Round numbers as Goals: Evidence
from Baseball, SAT Takers and the Lab,” Psychological
Science, 22(1), 71-79, (Pope, D. and Simonsohn, U).
(C) Yohanes E. Riyanto 92
Reference Point: Round Numbers as Goals
 Data from 4,303,906 SAT takers from 1994-2001. Because
the retake data of individual SAT takers are not observed,
the researchers focus on high school seniors and high
school juniors.
 The seniors do not have the chance to receive the scores
and then retake it before their college applications.
 In contrast, the juniors can have the opportunity to see their
scores and retake it before sending it for their college
applications.
 If indeed round numbers are performance goals, we should
expect that there is a discrete jump in the frequency of
juniors with the scores below and above the round
numbers, but not for seniors.

(C) Yohanes E. Riyanto 93

Reference Point: Round Numbers as Goals

(C) Yohanes E. Riyanto 94

Reference Point and Loss Aversion: Can
Losing Lead to Winning
 Loss aversion  position relative to a goal can motivate
individuals to work harder.
 Analyze NBA games played in 1993/1994 season and March
1 2009  18,060 games.
 How NBA teams’ likelihood of winning varies base on
halftime point differential between the two teams (winning
or losing at half time).
 Evaluation of causal relationship between the two using the
regression discontinuity design.

(C) Yohanes E. Riyanto 95

Reference Point and Loss Aversion: Can
Losing Lead to Winning
 Because winning or losing are not randomly assigned to
teams, unlike randomized experiment to evaluate causal
relationship, we need to use Regression Discontinuity
approach.
 Find teams that are on
either side of discontinuity
(barely winning or barely control group
losing). treatment group

 These teams should be

similar on all other things,
except that they are cut-off point
winning (or losing)  allow
us to examine the causal
impact of losing by a small margin. (C) Yohanes E. Riyanto 96
Reference Point and Loss Aversion: Can
Losing Lead to Winning

(C) Yohanes E. Riyanto 97