Uncertainty

Dushyant Kumar
BITS Pilani, Hyderabad Campus
Introduction

I Till now, we have covered the cases where consumer makes a

decision when faced with ‘certain’ prospects- you buy and
consume some quantity of x which gives you some particular
level of utility..
I But many a times we deal with uncertainties..
I Insurance and financial sectors are major examples, but we
face uncertainties in all walks of life-
which course to register for,
whether pursue higher degree or join a job, etc..
Introduction

I Whether to carry an umbrella or not!

I When we are deciding to buy any durable good, utility that
we are going to get from it depends on many other event
which are uncertain.
I How should we deal with such cases..
I Can economic theory do any value addition to the analysis for
such cases..
Lottery

I Earlier we had preferences defined over set of definite

alternatives.
I Now we need to deal with uncertain prospects..
I We will be using the term ‘lottery’ for such uncertain
prospects.. gambles..
I Let A be the set of outcomes- A = {a1 , a2 , · · · , an }.
I These outcomes can be consumption bundles or, simply
money.
I These outcomes are fixed, no uncertainties here..
Lottery

I A lottery is a probability distribution over this set A- you get

P
ai with probability pi , and i pi = 1.
I So, a certain prospect ai is a lottery where one gets ai with
probability 1.
I We can have lottery over lotteries!
I Consider two lotteries- p and p 0 .
I We can construct a compound or upper level lottery- a lottery
which gives p with probability α and p 0 with probability 1 − α.
Lottery

I We can have multi-layered compound lotteries... infinitely

layered lotteries?
I We assume that the consumer treats simple and compound
lotteries in same way- only net or effective probabilities
matter..
I Lets denote sets of all lotteries by L.
I L is convex.. Also closed and bounded..
I Also assume that order of lottery does not matter-
p ◦ x, (1 − p) ◦ y ∼ (1 − p) ◦ y , p ◦ x.
Expected Utility Theory

I Now, we want an utility representation U : L −→ R such that

p  q ⇐⇒ U(p) ≥ U(q).

I Also, we want the utility representation to be related to the

‘utility’ from outcomes (u(ai ))..
I As earlier, we assume the preferences to be complete and
transitive.
I We can also define continuity as earlier-
Definition (continuity)
{p ∈ [0, 1] : p ◦ x, (1 − p) ◦ y  z} and
{p ∈ [0, 1] : p ◦ x, (1 − p) ◦ y  z} are closed sets for all x, y and z
in L.
Expected Utility Theory

I From completeness and transitivity, we can arrange the

elements of A − a1  a2  · · ·  an (wlog)
I Also we restrict ourselves to cases where a1  an .
I Then an alternative definition of continuity is-
Definition (continuity)
For any lottery L ∈ L, there is some probability α ∈ [0, 1], such
that L ∼ {α ◦ a1 , (1 − α) ◦ an }.
I Is it ‘obvious’ ?
Expected Utility Theory

I A = {Rs.1000, Rs.10, death}, 1000  10  death.

I Continuity implies- there exists α such that
10 ∼ (α ◦ 1000, (1 − α)death)!
I It seems outrageous..
I But think gain!
Expected Utility Theory

Definition
A utility function U : L −→ R has the expected utility property
if there exists u : A −→ R such that
X
U(L) = pi u(ai ).
i

I As per expected utility property, we should be thinking in

terms of ‘utilities of outcomes’.
I What restriction do we need to put on the preferences?
I Preferences must be complete and transitive (rational) to
have any kind of utility representation.
I Preferences on a compact and convex set must be continuous
to have a continuous utility representation.
Independence Axiom

I Additionally we need independence axiom-

Definition
A preference relation  satisfies independence if, for every
p, q, r ∈ L and α ∈ (0, 1),

p  q ⇐⇒ αp + (1 − α)r  αq + (1 − α)r

I Also referred as independence of irrelevant alternatives (IIA)..

I This seems reasonable?
I Consider follwoing Preference: A = {x, y }, you prefer a
lottery that gives you (closest to) x with probability 0.5.
I This preference is rational- complete and transitive..
Independence Axiom

I What about independence axiom- consider three lotteries-

p = (0.5, 0.5), q = (0, 1) and, r = (1, 0).
I Clearly, p  q. So, we should have

αp + (1 − α)r  αq + (1 − α)r .

Lets check the same for α = 0.5,

0.5p + 0.5r = 34 , 14

0.5q + 0.5r = 21 , 12 .

I Here we have, 0.5p + 0.5r ≺ 0.5q + 0.5r .

I This preference does not satisfy independence axiom, it does
not have a expected utility representation..
Independence Axiom: Allais Paradox

I Cosider two lotteries l1 and l2 .

I l1 gives 1 million with probability 1!
I l2 gives 5 million with probability 0.10,
1 million with probability 0.89, and
nothing with probability 0.01.
I Which one do you prefer?
I In many experiments, people mostly prefer l1 to l2 .
Independence Axiom: Allais Paradox

I Now, consider two lotteries l3 and l4 -

I l3 gives 1 million with probability 0.11, and
nothing with probability .89.
I l4 gives 5 million with probability 0.10, and
nothing with probability 0.90.
I Now which one do prefer between l3 and l4 ?
I In many experiments, here people seems to prefer l4 to l3 ..
Independence Axiom: Allais Paradox

I l1 is preferred to l2 implies-

u(1) > 0.1u(5) + 0.89u(1) + 0.01u(0),

or,
0.11u(1) > 0.1u(5) + 0.01u(0)
adding 0.89u(0) on both sides,

0.11u(1) + 0.89u(0) > 0.1u(5) + 0.90u(0)!

I So, contexts matters..

I Framing effects- Amos Tversky and Daniel Kahneman..
Independence Axiom: Allais Paradox

Framing Treatment A Treatment B

A one-third chance of sav-
Positive Saves 200 lives ing all 600 people, two-third
possibility of saving no one.
Independence Axiom: Allais Paradox

Framing Treatment A Treatment B

A one-third chance that no
people will die, two-third
Negative 400 people will die
probability that all 600 will
die.
Independence Axiom: Allais Paradox

Framing Treatment A Treatment B

A one-third chance of sav-
Positive Saves 200 lives ing all 600 people, two-third
possibility of saving no one.
A one-third chance that no
people will die, two-third
Negative 400 people will die
probability that all 600 will
die.

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

case of negative framing..
Ellsberg Paradox: Ambiguity Aversion

I Consider an urn that contains 99 balls, colored Red, Black

and Green. We know that there are exactly 33 Red balls, but
the exact number of the other colors is not known. A ball is
randomly drawn from this urn. You choose a color.
I If the ball is of the color you choose, you win $1. What color
would you choose?
I If the ball is not of the color you choose, you win $1. What
color would you choose?
I In various experiments, people typically chooses red in both
cases.. Its inconsistent..
I Implication: Individual typically can’t assign probabilities
consistently, that limits the application of any theories based
on such probabilities!
Expected Utility Theory

Theorem (Expected Utility Theorem)

A preference relation  has an expected utility representation iff it
satisfies rationality, continuity, and independence.
Properties of Expected Utility Function

I Linear in probabilities..
I Uniqueness-
Theorem
Suppose that the vNM utility function u() represents  . Then the
vNM utility function, v(), represents those same preferences if and
only if for some scalar α and some scalar β > 0,

v () = α + βu(),

for all lotteries l.

Proof.
Expected Utility Theory: The St. Petersburg Paradox

I Introduced by Nicolaus Bernoulli in 1713!

I A fair coin is flipped until it comes up heads the first time. At
that point the player wins Rs. 2n , where n is the number of
times the coin was flipped. How much should one be willing
to pay for playing this game?
I ∞?
I Maybe diminishing marginal utility of money can explain this..
Cramer (1728) and Daniel Bernoulli (1738)..
I But notice we can adjust the reward to the monetary
equivalent of 2n units of utility... and the puzzle is back!
Risk Attitudes and Measurements
I Consider follwoing example due to Daniel Bernoulli (1738) (in
Latin, translated later on in English in 1954)!
I “Sempronius owns goods at home worth a total of 4000
ducats and in addition possesses 8000 ducats worth of
commodities in foreign countries from where they can only be
transported by sea. However, our daily experience teaches us
that of [two] ships one perishes.”

1
E (x) = 4000 + × 8000 = 8000.
2
I Suppose Sempronius ensure thats his 2 foreign consignments
follow two different independent but equally dangerous routes,
1 1 1
E (x) = × 4000 + × 8000 + × 12000 = 8000.
4 2 4
I Should Sempronius be indifferent between these two as the
expected values are equal?
Risk Aversion

I Essentially we are shifting some probability (0.25) from 4000

to 8000 and same probability (0.25) from 12000 to 8000.
I So, now it all depends on comparison between
u(8000) − u(4000) and u(12000) − u(8000).
I Faced with same lottery, different individuals can potentially
have different responses, depending on their risk attitudes.
I Do you prefer 10k with certainty or 5k with probability half
and 15k with probability half?
I This risk attitude is represented through the vNM utility
function.
I Consider a simple lottery which gives wealth level w1 with
probability p1 and w2 with probability p2 .
Risk Attitudes

I Expected Payoff- E (L) = p1 w1 + p2 w2 .

I Expected Utility- U(L) = p1 u(w1 ) + p2 u(w2 ).
I Utility from Expected Payoff- U(E (L)) = u(p1 w1 + p2 w2 ).

I An individual is risk averse at lottery L if U(E (L)) > u(L).

I An individual is risk neutral at lottery L if U(E (L)) = u(L).

I An individual is risk loving at lottery L if U(E (L)) < u(L).

Risk Attitudes

I If an individual is risk averse for every L ∈ L, the individual is

simply said to be risk averse.
I Risk averse if and only if vNM utility function is strictly
concave.
Also, a risk averse person has decreasing marginal utility of
money.
I Risk neutral if and only if vNM utility function is linear.
A risk neutral person has constant marginal utility of money.
I Risk lover if and only if vNM utility function is strictly convex.
A risk loving person has increasing marginal utility of money.
Risk Attitudes
Certainty Equivalent and Risk Premium

I The Certainty Equivalent (CE) of a lottery L is the

minimum amount that a person is ready to trade the lottery L
with-
u(L) = u(CE ).
I The Risk Premium (RP) of a lottery is the amount such that

u(L) = u(E (L) − RP)

or,
RP = E (L) − CE .
Example

I Consider a vNM utility function: u(w ) = ln w .

I Consider a lottery which gives amount h with probability half
and takes amount h with probability half.
I Expected payoff of the lottery:- E (L) = w .
I Expected utility from this lottery-
u(L) = 0.5ln(w + h) + 0.5ln(w − h) = 0.5 ln(w 2 − h2 ).
I Certainty Equivalent (CE):- u(L) = u(CE ), so

CE = (w 2 − h2 )0.5 < w = E (L).

I Risk Premium (RP):-

RP = E (L) − CE = w − (w 2 − h2 )0.5 > 0.
Measure of Risk Aversion

I Given two vNM utility function u() and v (), which one
represent a relatively more risk averse preference? What can
be a good measure of risk aversion?
I Clearly, concavity of a function capture this idea- if u() is
more concave than v (), u() represents a more risk averse
preference..
I Second derivative of a function can acts as a measure for risk
aversion.
I But as a absolute index, this has problems- if I just multiple
the utility function u() by some positive constant k, my
preference remains same whereas risk aversion index changes!
Arrow-Pratt Measure

I Arrow-Pratt measure of absolute risk aversion- Arrow (1965)

and Pratt (1964)-

u 00 (w )
r (w ) = − .
u 0 (w )
I Now this measure is invariant with any affine transformation.
I r (w ) > 0 - risk averse,
r (w ) = 0 - risk neutral, and
r (w ) < 0 - risk loving.
Arrow-Pratt Measure

I Arrow-Pratt measure is kind of normalized second derivative-

so it captures the risk attitudes..
I Lets verify that this measure indeed can be used for
comparison of risk attitudes.. (to be covered in the tutorial)
I Consider an individual with wealth level w .
I Suppose he is offered a lottery which pays x1 with probability
p and x2 with probability 1 − p, will he accept the bet?
I He will accept if

pu(w + x1 ) + (1 − p)u(w + x2 (x1 )) ≥ u(w ).

I Indifference set will be given by equality sign-

pu(w + x1 ) + (1 − p)u(w + x2 (x1 )) = u(w ) (1)

Arrow-Pratt Measure

I We want to characterize the Acceptance Set- set of all

lotteries that this individual is going to accept at the wealth
level w .
I Whats the slope of equation 1, at (0,0)? Differentiate it with
respect to x1 , and evaluate at x1 = 0..

pu 0 (w + x1 ) + (1 − p)u 0 (w + x2 (x1 ))x20 (x1 ) = 0. (2)

I Evaluating this at x1 = 0, we get-

p
x20 (0) = − .
1−p
Arrow-Pratt Measure
Arrow-Pratt Measure

I Suppose we are comparing individual i and j and, i’s

acceptance set is contained in j’s acceptance set, what does
this say about their risk attitudes?
I Individual i is more risk averse compared to j.
I Locally, i is more risk averse than j if in the neighborhood of
(0, 0), Ai (w ) is more ‘curved’.
I To check for the same, we differentiate equation 2 with
p
respect to x1 - (use x20 (0) = − 1−p )

u 00 (w )
 
p p
x200 (0) = − 0 = r (w ).
(1 − p)2 u (w ) (1 − p)2
Arrow-Pratt Measure

I Hence r (w ) is an obvious measure for risk aversion.

Tutorial Section Ends.......
I r (w ) is a local measure for the risk aversion, it changes with
wealth level.
I Do you expect r (w ) to increase, remain constant or decrease
with increasing wealth levels?
I Decreasing Absolute Risk Aversion (DARA) seems more
reasonable.. not ruling out others however..
Arrow-Pratt Measure for Relative Risk Aversion

I Arrow-Pratt Measure for Relative Risk Aversion-

u 00 (w )
− w.
u 0 (w )
I In case of relative risk aversion, its no longer seems reasonable
to assume that its decreasing in wealth..
I Often its assumed to be constant in applied and empirical
research..
I State dependent utility function-

pup () + quq ()
Standard Utility Functions

I Mean-variance utility function-

u(w ) = w − bw 2

E (u(w )) = w̄ − b w̄ 2 − bσw2 .
I This utility function has nice form but it has two
shortcomings-
∂u(w )
1 ∂w < 0 for some w .
2 r (w ) is increasing in w .
I Another popular form- u(w ) = e −rw .
I Constant absolute risk aversion..
Global Risk Aversion

I Up till now, we have worked with local risk aversion- at one

particular wealth level..
I It can be easily extended to global risk aversion..
I Consider 2 individuals with utility functions A(w ) and B(w ).
When can we say that individual A is more risk averse than
B?...
I Three ways to conceptualize:
00 00
1. − AA0 (w
(w ) B (w )
) > − B 0 (w ) for all levels of w .
2. A(w ) is ‘more concave’ than B(w ) or, A(w ) is a ‘concave
transformation’ of B(w ).. Formally,

A(w ) = G (B(w ))

for some increasing strictly concave function G .

Global Risk Aversion

3. Let  be a random variable with E () = 0. Lets denote the

maximum amount that A is willing to give up in order to
avoid the risk  by πA ()-

A(w − πA ()) = EA(w + )

Now we need,
πA () > πB ()
for all  and w .
I What can be shown is, all these three properties are
equivalent..
Examples: Insurance

I Consider an risk averse individual with initial wealth level w ..

I Suppose he faces a risk- with probability p he is going to
suffer a loss of L.
I There is an insurance company offering insurance against this
risk at the premium of π per unit.
I Will this customer buy the insurance? If yes, of what amount?
I Suppose he purchases insurance of q unit.. Lets choose q
optimally..
Examples: Insurance

I max pu(w − L − πq + q) + (1 − p)u(w − πq).

q>0
I Lets assume actuarially fair premium- π = p..
I This maximization exercise gives q ∗ = L. Fully insured..
I What role does risk-aversion play here- s.o.c for maximization..
I So with insurance, the individual enjoys the wealth level of
w − pL irrespective of the states of the nature.
I Check that the expected utility in the case optimal insurance
here is higher than that in the case of no insurance, for risk
averse individual..
I Optimal insurance q ∗ when π > p..