BEHAVIORAL ECONOMICS

ECON F345

Dushyant Kumar
BITS Pilani, Hyderabad Campus
Decision Making Under Uncertainty

Decision Making Under Uncertainty

Risk & Uncertainty

I Risk: agreed upon probability distribution.

I Uncertainty: subjective probability distribution.
I Many a times, even in literature, these terms are used
interchangeably.
I We will be confining ourselves to the risk part only here.
I Some basics first- recap..
Basics: Recap

I Mutually exclusive events

P(A ∪ B) = P(A) + P(B)

P(A ∩ B) = 0
I Exhaustive events

P(A) + P(B) = 1

I Independent events

P(A&B) = P(A) × P(B)

Basics: Recap

I Conditional probability

P(A&B)
P(A/B) =
P(B)

P(A&B) = P(A/B) × P(B) = P(B/A) × P(A).

For independent events,

P(A/B) = P(A), P(B/A) = P(B), P(A&B) = P(A)×P(B).

Basics: Recap

I Bayes’ Rule
I Suppose two procedures of producing a particular good-
A & B.
I From procedure A, the good can be defective with probability
P(D/A), similarly P(D/B) for procedure B.

P(D) = P(D/A)P(A) + P(D/B)P(B).

I If a good is defective, what is the probability that it was

produced through A?
Basics: Recap

P(A&D)
P(A/D) =
P(D)
P(D/A)P(A)
= .
P(D/A)P(A) + P(D/B)P(B)

I Example: Upon the tests and diagnosis, suppose the doctor

infers that a patient has either type A (with prob. 13 ) or type
B cancer (with prob. 23 ). (mutually exclusive and exhaustive
events)
type A is deadly- 4 out of 5 patients die within a year,
type B is relatively less deadly- 1 out of 5 patients die within a
year.
Bayesian Updation: Example Cont...
I What is the prboability of a patient dying within a year?

4 1 1 2 2
× + × = .
5 3 5 3 5
I Prior probability and posterior probability..
I Now suppose the patient die within a year, what is the
posterior probability that the patient was suffering from type
A cancer?
P(D/A)P(A)
P(A/D) =
P(D/A)P(A) + P(D/B)P(B)
4
×1 2
= 4 15 31 2 = .
5 × 3 + 5 × 3
3

favorable event prob ↑; unfavorable event prob ↓

P(A/D) > P(A), P(B/D) < P(B)..
Bayesian Updation

I Processing new information-

Prior probability −→ posterior probability..
I Example: unfair coin toss..
prior belief- the coin is fair with probability 0.30 and unfair
with probability 0.70..
unfair in the sense that its head both sides.
consider a game where you get Rs. 100 if its H, 0 otherwise.
I What is the (max) money that you are ready to pay as entry
fee?

0.30 × (0.50 × 100 + 0.50 × 0) + 0.70 × (1 × 100) = 85.

Bayesian Updation: Example Cont..

I Now suppose you observe that in the first toss, its H.

Now what is the (max) money that you are ready to pay as
entry fee?
Notations-
U− coin being unfair.
F − coin being fair.
G − getting head in first toss.

P(G /U) = 1, P(G /F ) = 0.50

P(G /U)P(U)
P(U/G ) = = 0.82
P(G /U)P(U) + P(G /F )P(F )
Bayesian Updation: Example Cont..

I So the expected value after observing G -

0.82 × 100 + (1 − 0.82) × 50 = 92 (approx.)

I What if G was observance of T in first toss?

I Different biases....
Conjuction and Disjuction Baises

I Conjuction fallacy: Example

I Consider a Boeing 747-400 aircraft...
it has roughly around 6 million parts!
I Each part is extremely reliable, fails with probability 0.000,001
probability.. works almost always, with probability 0.999,999
I Suppose the design is such that even if one part fails, the
aircraft won’t work! (not robust)
I What is the probability of aircraft not working?
1− Prob (all parts working) = 1 − (0.999999)6,000,000 = 0.75
(approx.)
Conjuction and Disjuction Baises

I Conjuction fallacy is also referred as planning fallacy

many-a-times.
I While planning mega projects, governments typically makes
this kind of mistakes..
delay in completion of government projects..
I It is basically tendency to overestimate the probability of the
event A ∩ B.
I Disjuction fallacy: tendency to underestimate A ∪ B.
I Why is it so tough to guard against terrorist attacks?
I Cricket: a game biased against the bowlers?
Conjuction and Disjuction Baises

I Example: Birthday problem- Suppose there are 30 students in

a class. What is the probability that no two students have
same birthday?
assume births are randomly distributed over the year, non-leap
year..

365 364 336

× × ··· = 29.4 approx.
365 365 365

I Next- base rate neglect..

Base Rate Neglect

I Lets consider an example. Suppose around 1 percent of the

population have certain disease.
There is a really good (highly accurate) test available- 90
percent accuracy.
Suppose someone is dignosed positive in the test, what is the
probability that he indeed has the disease?

P(+ve/D)P(D)
P(D/ + ve) =
P(+ve/D)P(D) + P(+ve/ND)P(ND)

0.08 (approx)
I Arguements for random testing for Covid during very early
stage..
Decision Thoeries Under Uncertainty

I Till now we have covered the ways to estimate probabilities

and what are some common mistakes that we make in the
process.
I Now we move on to cover decision making using these
probabilities!
I Various criteria/approaches..
1. Maximin, Maximax, Minimax-risk, Expected Value..
2. Expected Utility Theory
3. Prospect Theory
Maximin Criterion

I First see what is the minimum payoff/outcome attached to

every action..
Choose the action which has the maximum of these min
values- maximum guranteed payoff.
I This approach captures the extreme risk averse attitude..
I Consider following example-
Maximin Criterion

I So as per Maximin principle, Carrying Umbrella is optimal..

I Notice risk aversion, extreme one infact..
I What if its a pleasant sunny morning- Feb in BITSH campus-
chances of rain very close to zero..
Analysis and prescription remain same..
I What if (Don’t Carry, No Rain) payoff is insanely high?
Again analysis and prescription remain same..
Maximin Criterion

I So, maximin principle is good and insightful to start with but

it has significant drawbacks-
1. Don’t use a significant portion of available information..
2. No diversity in terms of risk preferences- all individuals using
maximin approach are exactly same in terms of risk preferences
or attitude..
I Maximax Criterion
I Now first see what is the maximum payoff/outcome attached
to every action..
Choose the action which has the maximum of these max
values- a shot at best possible payoff.
I This approach captures the risk taking attitude.. extremme
one..
Maximax Criterion

I So as per Maximax principle, not carrying umbrella is optimal..

I Notice risk taking behaviour, extreme one infact..
I What if its extreme likely to rain, say with probability very
close to 1.
Analysis and prescription remain same..
I What if (Don’t Carry, Rain) payoff is insanely low?
Again analysis and prescription remain same..
Maximax Criterion

I So once again, maximax principle is good and insightful to

start with but it has significant drawbacks-
1. Don’t use a significant portion of available information..
2. No diversity in terms of risk preferences- all individuals using
maximax approach are exactly same in terms of risk
preferences or attitude..
I Minimax-risk Criterion
I Here we choose the one with lowest (potential) ‘regret’.
Minimax-risk Criterion

I So here carrying umbrella is optimal..

I Once again we are not considering probabilities and
magnitudes of payoffs in many states..
I Do minimax-risk criterion and maximin criterion always
recommends same action?
in this example, change the numbers such that both are
different.
I Lets consider another example to check these three solution
concepts.
Minimax-risk Criterion

I Consider a bicycle shop owner- need to stock before the

season, demand uncertain..
I Cost side: bicycles can be procured in multiple of 20 units-
20, 40, 60,....
I Per unit cost decreases with an increase in the number of
units procured:
$70 for 20 units, $67 for 40 units,
$65 for 60 units, $64 for 80 units.
I Further supoose demand can be of 10 units (with prob 0.20),
30 units (with prob. 0.40), 50 units (with prob. 0.30) and, 70
units (with prob. 0.10).
Minimax-risk Criterion: Example Contd..

I Further, retail price of a bicycle is $ 100.

Leftover stock can be returned to the wholeseller at $45 each..
I In case of shortage of stock, a returned customer attracts bad
reputation and all.. value say a loss of $ 5 per returned
customer..
I How much should the shop owner stock as per maximin,
maximax, and minimax-risk criterion?
Minimax-risk Criterion: Example Contd..
Minimax-risk Criterion: Example Contd..

I Maximin- stock just 20 units.

Maximax- stock 80 units.
Minimax-risk- stock 60 units.
I Minimax-risk criterion provide a middle or balanced path..
although it also suffers from essentially similar weaknesses-
1. probabilities are not considered at all.
2. magnitudes of payoffs of many states ingnored..
I Overall in all these three approaches, a lot information in not
taken into account..
Maxmin Criterion: Harsanyi Challenge

Next stop: Expected value approach..

Expected Value Approach

I Set of potential outcomes: {x1 , x2 , · · · , xn }.

associated probability distribution: {p1 , p2 , · · · , pn }.
Lottery- L = {x1 , p1 ; x2 , p2 ; · · · ; xn , pn }.
I Expected value- X
EV (L) = xi pi
i

notice here both magnitudes and probabilities are

accommodated..
I Expected value approach is quite widely used-
law and economics- magnitude of punishment times prob of
getting punished..
expected profit
insurance- actuarially fair..
Expected Value Approach

I Biggest strength of expected value approach- simplicity!

I Extremely desirable characteristic for any decision theory..
I However notice that we still don’t have diversity of risk
attitudes here..
I The expected value is linear in probabilities and outcome..
linear in outcome- caters just monetary outcomes; even for
monetary outcomes, its possible to have non-linearity in
satisfaction/value/utility..
I Other issue- dealing with extreme probabilities..
Expected Value Approach

I Consider the following examples-

A if you hit a particular point in a circle, you get Rs. 10 lakhs;
zero otherwise..
B if you hit a particular point in a circle, you have to pay a
penalty of Rs. 10 lakhs; zero otherwise..
I Which one do you prefer?
I Expected value approach suggest indifference..
I Suggested approach: dominance..
I Tail risk: Consider an event with extreme low probability but
catastrophic effect.. difficult to deal with in expected value
approach.. such event should be properly hedged..
Expected Value Approach: St. Petersberg Paradox

I A fair coin is tossed-

if it takes n tosses to get first head (H), you get Rs. 2n .
I What is the max entry fee that you are ready to pay for this
game?

I What is the expected value of this game?

Expected Value Approach: St. Petersberg Paradox

I A fair coin is tossed-

if it takes n tosses to get first head (H), you get Rs. 2n .
I What is the max entry fee that you are ready to pay for this
game?

I What is the expected value of this game?

Expected Value Approach: St. Petersberg Paradox

I St. Petersberg Paradox thought experiment- why such a huge

divergence between your willingness to pay and the expected
value?
I The expected value approach also does not caters for diversity
in risk attitudes?
I What is risk attitudes? We have not not defined yet!
I Before moving forward, lets first discuss risk attitudes briefly.
I Different types of risk attitudes- risk averse, risk neutral, risk
taker..
I Lets define these formally..
I Consider the lottery L and expected value EV (L) as
described earlier..
Risk Attitudes

I Will you like play the lottery i.e. take a chance with the
lottery or, rather settle with EV (L)?
I Risk averse- EV (L)  L
I Risk neutral- EV (L) ∼ L
I Risk taker- EV (L) ≺ L.
I So the expected value approach just covers the risk neutrality
part. We need a theory that can explain or accommodates all
three behaviours..
I Risk premium- Consider a risk averse individual. The
individual strictly prefers EV (L) compared to playing the
lottery and taking risk.
Risk Attitudes

I What if the individual is offered a bit less than expected value,

EV (L) − δ?
I Notice that there exist an amount such that the individual
prefer to play the lottery instead of taking that amount with
certainty- worst outcome xn ..
I So if we keep on reducing from EV (L), there exist an amount
where the individual is going to be just indifferent between
playing the lottery and taking that amount with certainty-
certainty equivalent (CE)..

L ∼ CE
Risk Attitudes

I What is the amount that this risk averse individual is

sacrificing in order to avoid the risk?

EV (L) − CE

this is known as risk premium (RP)..

I For a risk averse individual, CE < EV (L) and RP > 0,
For a risk neutral individual, CE = EV (L) and RP = 0
For a risk taker individual, CE > EV (L) and RP < 0

I Next Stop: Expected Utility Theory..

Expected Utility Theory

I John von Neumann and Oskar Morgenstern- Theory of Games

and Economic Behavior (1944)
I Consider the lottery described earlier-

L = {x1 , p1 ; x2 , p2 ; · · · ; xn , pn }

I First look for a functional representation of individual’s

preference over the set of outcomes- u(xi ).
I Existence of such representation?
I Define the expected utility function (f : L −→ R) as
X
U= pi (xi )u(xi )
i
Expected Utility Theory

I First observe that expected utility theory allows for the

non-linearity of satisfaction from outcomes (u(xi )).
This ensures that now an individual can be risk averse, risk
taker or, risk neutral..
I Example: Consider the lottery- [A.] flip a (fair) coin- if it is
head, you get Rs. 1000; if it is tail, you get nothing.

L = {100, 0.50; 0, 0}
1
suppose u(xi ) = xi2 .
EV (L) = 50
Expected Utility Theory: Example

1
EU(L) = 0.50 × 100 2 + 0.50 × 0 = 5
Utility that the individual is going to get from getting the
expected value i.e. 50 with certainty-
1
50 2 = 7.07

So the individual here prefers to have 50 with certainty rather

than playing the lottery- risk averse..
What if the utility function is

u(xi ) = xi2

Verify- risk taker..

Expected Utility Theory: Example

1
I Lets continue with u(xi ) = xi2 . What is CE and RP?
1
u(CE ) = CE 2 = EU(L) = 5
or, CE = 52 = 25.
RP = EV (L) − CE = 50 − 25 = 25.
I CE and RP for u = x 2 and u = kx, k > 0..
I The expected utility framework is a significant development
over the earlier approaches- significantly richer setup..
I What are we assuming when we are working with EUT?
Axiomatic foundation..
crucial for further developments..
Expected Utility Theory: Axiomatic Foundation

I Completeness- Any two lottery can be compared.. no

indecisiveness
I Transitivity- If L1  L2 and L2  L3 then L1  L3 .
** Best and worst lottery- if x1 is the best outcome and xn is the
worst outcome then L1 is the best lottery and Ln is the worst
lottery,
where L1 = {x1 , 1; x2 , 0; · · · ; xn , 0} and
Ln = {x1 , 0; x2 , 0; · · · ; xn , 1}
I Continuity- For every lottery Li there exist a p ∈ [0, 1] such
that
Li ∼ {x1 , p; x2 , 0; · · · ; xn , 1 − p}
or,
Li ∼ {x1 , p; xn , 1 − p}
Expected Utility Theory: Axiomatic Foundation

I Reduction of Compound Lotteries- the individual is indifferent

between a compound lottery and its reduced simple lottery.
I Independence- Suppose L1  L2 , then

{L1 , p; L3 , 1 − p}  {L2 , p; L3 , 1 − p}

I Theorem- Let preferences satisfy above axioms, then there

exist a function U : L −→ R such that L1  L2 if and only if

U(L1 ) ≥ U(L2 ).
P
U= i pi u(xi ).
Expected Utility Theory: Violations

Violations of independence axioms

I Example:
b1 = (x0 , 0; x0 + 300000, 1)
b2 = (x0 , 0.2; x0 + 400000, 0.8)
b3 = (x0 , 0.75; x0 + 300000, 0.25)
b4 = (x0 , 0.8; x0 + 400000, 0.2)
for illustration, say x0 = 0.

I b1 vs. b2 ?
b1
I b3 vs. b4 ?
b4
Violations of Independence Axioms

I It seems that while comparing b3 and b4 , many of us treat a

small difference in probabilities (0.05 here) as insignificant,
hence prefer b4 ;
but while comparing b1 and b2 , often we treat a small
difference in probabilities (0.20 here) as significant, hence
prefer b1 .
I Non-linearity in probability..
? Near the certainty- why take even a slight ‘unnecessary’ risk?
? In deep uncertainty (away from shore)- a bit more risk does
not matter..
Expected Utility Theory: Violations

I Later it was found that Allais paradox is a part of larger class

of paradoxes- common ratio violation and common
consequence violation.
I Common ratio violation Consider three outcomes-
0 < x1 < x2 . Consider following lotteries-
b1 = (x1 , p1 ; x0 , 1 − p1 ) and b2 = (x2 , p2 ; x0 , 1 − p2 )
b3 = (x1 , αp1 , ; x0 , 1 − αp1 ) and b4 = (x2 , αp2 ; x0 , 1 − αp2 )
I Notice in both cases, the common ratio- pp1 .
2
I Now if someone prefers b1 over b2 , and b4 over b3 , this is
referred as ‘common ratio violation’.
Expected Utility Theory: Violations

I In the previous example,

p1 = 1, x1 = x0 + 3000, p2 = 0.8, x2 = x0 + 4000, α = 0.25.
I Proposition: The common ratio violation is a violation of the
independence axiom, or the reduction axiom, or of both
axioms.
I Common consequence violation- Let A, B, C, D be lotteries
and α ∈ [0, 1]. Suppose that we have the following four
compound lotteries-
a1 = (A, α; D, 1 − α), and a2 = (B, α; D, 1 − α)
a3 = (A, α; C , 1 − α), and a4 = (B, α; C , 1 − α)
Expected Utility Theory: Violations

I Common consequence violation occurs if someone prefers a1

over a2 and, a4 over a3 (or, a2 over a1 and, a3 over a4 ).
I Various modifications of expected utility theory have been
tried to relax the independence axiom.
I EU is still quite popular! the default concept while dealing
with risk and uncertainty..
I Other violations- we are going to briefly discuss couple of
other violations in the context of EU theory.
I Frame independence- our decision should be independent of
framing.. example-
Expected Utility Theory: Violations

I Framing effect

Most people choose A in the case of positive framing and, B

in the case of negative framing..
Expected Utility Theory: Violations

I As per the expected utility theory, the framing should have no

role to play in our decisions.
I Framing and communication- often they play a huge role..
I Reduction axiom- many a times, we don’t treat compound
and ‘equivalent’ simple lottery in exactly same way-
compound risk premium- Abdellaoui et al. (2014)
I Rabin’s paradox- Rabin (2000)- under expected utility, risk
aversion over small stakes implies unreasonable high risk
aversion over large stakes..
I Fair bet- a risk averse individual always rejects a fair bet, a
risk taker always takes it!
Expected Utility Theory: Violations- Rabin (2000)