You are on page 1of 32

Utility maximization:

A Representation Theorem

ECON 404
Behavioral Economics

1 Econ 404 – Brian W. Rogers © 2015 by Brian W. Rogers


Announcements

• First lab this Friday

• Please come to the section for which you’re registered

2 Econ 404 – Brian W. Rogers


Roadmap

• Continuing: revealed preference on


menus

• Monday: survey of ways in which people


often fail to be utility maximizers
 Leading, in due course, to a range of
behavioral models
 This will relate closely to Friday’s lab

3 Econ 404 – Brian W. Rogers


Big picture

• Observable data: choices from menus


• Unobservable traits: preferences

• Question: When can we find a utility function that:


 Represents the DM’s preferences
 “Explains” choices as arising from utility-maximization?

• This is probably the most canonical setting in which


economics frames the question of rationality

6 Econ 404 – Brian W. Rogers


Property A (reminder)

• Independence of Irrelevant Alternatives (IIA):


 If x is chosen from menu A [ x ϵ c(A) ]
 And if B is a subset of A that includes x [x ϵ B and B A]
 Then x must be chosen from B [ x ϵ c(B) ]

.x
A
B

7 Econ 404 – Brian W. Rogers


IIA is a necessary condition
for utility maximization

• Means: If IIA is violated, then there does not exist any


utility function that explains the choices
 Equivalently: If a DM is a utility-maximizer, their choices
must satisfy IIA

• In the abstract, the following are equivalent:


• Q is a necessary condition for P
• If P then Q Q
P
• P implies Q
• [not Q] implies [not P]
• (Draw the Venn diagram)
10 Econ 404 – Brian W. Rogers
IIA is a necessary condition
for utility maximization

• Means: If IIA is violated, then there does not exist any


utility function that explains the choices
 Equivalently: If a DM is a utility-maximizer, their choices
must satisfy IIA

• In this application:
 Consider a utility maximizer
 If x is chosen from A, x has higher utility than other
alternatives in A
 B is a subset of A.
 So x must have higher utility than alternatives in B
 So x must indeed be chosen from B if it’s available
11 Econ 404 – Brian W. Rogers
A deductive logic puzzle
(How to think about “If P then Q”)

• Each card has a number on one side and a color on the other

• Question: Which card(s) must you turn over in order to test


that if a card shows an even number on one face, then its
opposite face is blue?

12 Econ 404 – Brian W. Rogers


Solution
• “If EVEN then BLUE” [If P then Q]

• You don’t have to turn over the 3


 [Not P], i.e., odd, needn’t imply anything

• You DO have to turn over the 8

• You don’t have to turn over the BLUE card


 If it is EVEN it satisfies the rule; if it is ODD, see above
 A sort of “confirmation bias”

• You DO have to turn over the RED card


 If it is EVEN it violates the rule
13 Econ 404 – Brian W. Rogers
Context matters

• Rule: If you are drinking alcohol, you must be over 21

• Many more people get this version right (57% vs 24%)


 Inspect 16 and BEER

14 Econ 404 – Brian W. Rogers


Take aways

• If P then Q has us inspect the 8


• If [not Q] then [not P] has us inspect the RED

• In our application, we may wish to see if a given data


set satisfies IIA

• Note that we only have to inspect cases where the


antecedent conditions of IIA are satisfied
 In particular, in a data set with no nested menus, IIA is
satisfied “trivially”

15 Econ 404 – Brian W. Rogers


IIA and transitivity

• IIA satisfied Problem Available shows Chosen show


1 GoT, WD GoT
2 WD, PLL WD
3 PLL, GoT PLL

Problem Available shows Chosen show


• IIA violated 1 GoT, WD GoT
2 WD, PLL WD
3 PLL, GoT PLL
4 GoT, WD, PLL GoT

16 Econ 404 – Brian W. Rogers


Exposition and logic

• Suppose I tell you “If P then Q”

 This is probably a silly, even misleading, thing to say if Q is


always true, even though my statement is logically correct
 “If tax rates are raised then the demand curve is downward sloping.”

 This is also confusing if P is never true (especially if Q is also


never true), although, again, the statement is logically correct
 “If 2 is odd then 1+1=3”
 “If the demand curve is upward sloping then firms prefer to make less
profit”

17 Econ 404 – Brian W. Rogers


Property B (of choices)

• Take menu A and a subset B A


• If x and y are both chosen from B [x,y ϵ c(B)]
• And if y is chosen from A [y ϵ c(A)]
• Then x must also be chosen from A [x ϵ c(A)]

.x
.y
A
B

18 Econ 404 – Brian W. Rogers


Property B (of choices)

• Take menu A and a subset B A


• If x and y are both chosen from B [x,y ϵ c(B)]
• And if y is chosen from A [y ϵ c(A)]
• Then x must also be chosen from A [x ϵ c(A)]

• Intuition:
 Since x and y are chosen together (from B) they are equally good
 Since y is chosen from A, it is at least as good as anything in A
 So x, being equally as good as y, is also at least as good as anything
from A

• Note: If the DM is never indifferent, then Property B holds


trivially (“automatically”)

19 Econ 404 – Brian W. Rogers


Main result

• When our model has latent (unobserved) variables, we seek a


representation
 A data set
 A model
 A set of conditions on the data which are necessary and sufficient for it
to be consistent with the model

• This means that testing the model is equivalent to testing the


conditions

• Goal:
 Choice data satisfies Properties A&B if, and only if, it is consistent
with utility maximization

20 Econ 404 – Brian W. Rogers


Example

• Choice data that satisfies Property A (IIA), but not


Property B, and, hence, for which no utility function
can explain the data
Available shows Chosen show(s)
GoT, WD GoT, WD
WD, PLL WD
PLL, GoT GoT
GoT, WD, PLL GoT

21 Econ 404 – Brian W. Rogers


Formalizing the data
• The data is encoded by a choice correspondence
 A correspondence is like a function (a mapping), but where
the result is allowed to be multi-valued

• X: the set of all possible choice alternatives


• 2X: the set of all subsets of X (power set of X)
• Ø: empty set
• 2X/Ø: all non-empty subsets of X (i.e., menus)

• Definition: A choice correspondence is a mapping


c: 2X/Ø  2X/Ø
such that c(A)⸦A for all non-empty A⸦X
22 Econ 404 – Brian W. Rogers
Examples and notes

• Don’t worry! This is just formalizing old ideas


 c({WD, PLL}) = {WD}
 c({WD, PLL, GoT}) = {PLL, GoT}

• The DM must choose something from any set


 Can’t give up and choose nothing
 Can’t choose something not in the set
 Can choose multiple items: this is like the DM saying “I am
equally happy to choose any of these alternatives”

23 Econ 404 – Brian W. Rogers


Formalizing the model

• There exists a utility function u: X  R such that for


all A, c(A) = arg max u(x), among all xϵA
 I.e., the chosen alternative(s) maximize utility

• Definition: When this holds, we say that


u rationalizes c
 If c can be rationalized by some u, then we say that c has a
utility representation

24 Econ 404 – Brian W. Rogers


The main representation theorem

• Theorem: A choice correspondence has a utility


representation if and only if it satisfies Properties A &
B
 If: A & B imply utility representation
 Only if: The existence of a rationalizing utility function
implies A & B are satisfied

• Recall: the model has unobservable variables

• But the properties can be directly tested on a given data


set
25 Econ 404 – Brian W. Rogers
How to proceed
• In order to prove this, we must introduce another model based
on preferences

• A binary relation R on a set X is a set of ordered pairs of


elements of X
 If (x,y) ϵ R, we write xRy
 Interpret as “x is at least as good as y”

• A binary relation is a preference relation if it is:


 Transitive: for all x,y,z in X, if [xRy and yRz] then [xRz]
 Reflexive: for all x in X, xRx

• A preference relation is complete if:


 for all x,y in X, either xRy or yRx (possibly both)

26 Econ 404 – Brian W. Rogers


An example

• Let X = {x,y,z}
• Let R = {(x,x), (x,y), (y,y), (y,z), (z,z)}

• R is reflexive: (x,x), (y,y), and (z,z) all belong to R


• R is not transitive: (x,y) and (y,z) are in R, but (x,z) is
not
• R is not complete: neither (x,z) nor (z,x) are in R

27 Econ 404 – Brian W. Rogers


Preference relations and utility

• The binary relation is our device to encode preferences


 Answers questions like “how do you feel about x vs. y?”

• Definition: A utility function u represents a binary


relation R if:
 For all x,y in X, u(x) ≥ u(y) if and only if xRy

• Theorem 1: X finite. A binary relation R on X has a


utility representation if and only if it is a complete
preference relation
 I.e., it is complete, transitive and reflexive
28 Econ 404 – Brian W. Rogers
Preference relations and choices

• A preference relation R rationalizes a choice


correspondence c if for every subset A of X, we have:
 c(A) = {xϵA | xRy for all y ϵ A}

• Theorem 2: Fix a finite X and a (complete) choice


correspondence c on X.
There exists a complete preference relation R that
rationalizes c if and only if c satisfies Properties A
and B.

29 Econ 404 – Brian W. Rogers


Putting everything together

• Theorems 1 & 2 characterize the choice data that can


be viewed as coming from a utility-maximizing
decision maker

• utility  preferences  choices


Thm 1 Thm 2

• We do not take the stance that the thought process of


the DM is that of utility maximization per se
 It is instead a tool to allow the researcher to understand the
properties of the DM’s choices
30 Econ 404 – Brian W. Rogers
Proof outline for Theorem 1

• Theorem 1: X finite. A binary relation R on X has a utility


representation if and only if R is a complete preference relation

• First, a word on “if and only if”


• “P if and only if Q” means:
 If Q then P (or “Q implies P), and
 If P then Q (i.e., P only if Q)

• So we have two things to prove:


 1) If [R has utility rep’n] then [R is complete preference relation]
 2) If [R is complete preference relation] then [R has a utility rep’n]

31 Econ 404 – Brian W. Rogers


Proving first part

• 1) If R has a utility representation then R is a


complete preference relation

• Suppose R has a utility representation:


 There exists u(.) such that u(x) ≥ u(y) if and only if xRy
 For every x, clearly u(x) ≥ u(x), so xRx (r is reflexive)
 Take any x,y. Either u(x) ≥ u(y) or u(y) ≥ u(x)
 So either xRy or yRx (R is complete)
 Take any x,y,z. If xRy and yRz then u(x) ≥ u(y) ≥ u(z)
 So u(x) ≥ u(z), implying xRz (R is transitive)

32 Econ 404 – Brian W. Rogers


Second part

• 2) If R is a complete preference relation, then R has a


utility representation

• Assume R is complete, reflexive and transitive.


• Must show that there exists a function u(.) with the
property that u(x) ≥ u(y) if and only if xRy

• This direction uses mathematical induction


 I will only outline the technique
 See HW 1

33 Econ 404 – Brian W. Rogers


Induction example

• The sum of the first n odd numbers equals n2.


 Example 1+3+5 = 9 = 32

• Proof:
• Verify a base case -- n=1: 1 = 12
• Next: Assume it’s true for n: 1+3+5+…+2n-1= n2
• WTS: It’s true for n+1:
 (1+3+…+2n-1)+2n+1 = n2 +2n+1 = (n+1)2

• Done!

34 Econ 404 – Brian W. Rogers


Proof by induction: strategy

• Induct on the size of set X, |X|.

• First, show there exists a utility representation when |X| = 1.


 Trivial, but does require reflexivity

• Next assume the result is true for |X| = n, i.e., that given the
properties, there exists a utility representation.

• Prove that the result must then hold for |X| = n+1.

• This proves the result for all finite |X|.

35 Econ 404 – Brian W. Rogers


What’s coming

• (Partial) proof of Theorem 2

• Big picture: rationality is characterized by our


Properties A&B

• (When) do people behave accordingly?


• What does this suggest about
 their rationality?
 Our model?

• How should we proceed?


36 Econ 404 – Brian W. Rogers

You might also like