Lecture 1: Preferences and Choice

0.1 Preferences
1. Let X denote a set of alternatives/objects/goods. A preference relation
of an economic agent is a binary relation, denoted by <, over X. The
relation describes what the agent wants. For any x; x0 2 X,

x < x0

means that the economic agent consider x to be at least as good as x0 . (I

will call < the weak preference relation.)
2. The standard approach in economics is to learn what people learn from
their choices and avoid making any direct assumptions about preferences
over any two alternatives.
3. Instead, we assume that, overall, people preferences are internal consistent
or rational. (In economics the two concepts more or less mean the same
thing.)
4. Axioms of Rational Preferences:

(a) Completeness: For any two bundles x and x0 , it is that either x < x0 ,
or x0 < x, or both.
(b) Transitivity: For any three choices x; x0 , and x00 ,

if x < x0 and x0 < x00 ; then x < x00 :

If an agent likes x better than x0 , and x0 better than x00 , then he

must like x better than x00 .

5. If both x < x0 and x0 < x, then the agent is indi¤erent between x and
x0 . We use the notation to represent the indi¤erence relation. If x < x0
and x x0 , then the agent strictly prefers x to x0 . We use to represent
the strict preference relation.
6. The transitivity of < implies the transitivity of and the transitivity of
, and vice versa. So, we can equivalently state our axioms as

(a) Completeness: For any two bundles x and x0 , it is that either x x0 ,

x0 x, or x x0 .
(b) Transitivity: For any three choices x; x0 , and x00 ,

if x x0 and x0 x00 ; then x x00 ;

and
if x x0 and x0 x00 ; then x x00 :

1
 7. It should be clear that if x x0 , then if x0 x (since neither x x0 nor
x0 x:)
8. Note that (6) implies that for any three choices x; x0 , and x00

(a) if x x0 and x0 x00 ; then x x00 ;

(b) if x x0 and x0 x00 , then x x00 .

9. To prove 7a, note that if by way of contradiction x00 x, the transitivity of

would imply x00 x0 . On the other hand, if x00 x, then the transitivity
of would imply that x x0 .

10. Showing the equivalence or 4 and 6 are not hard. First, 4(a) and 6(a) are
equivalent. To show the transitivity of < implies the transitivity of , we
suppose, by way of contradiction, that there are x; x0 , and x00 such that

x x0 and x0 x00 ; but x x00 :

(Note that x is not strictly better x0 is equivalent to x0 is as good as x.)

By the transitivity of weak preference,

x00 < x < x0 ) x00 < x0 )( :

To show the transitivity of < implies the transitivity of , note that

x x0 and x0 x00

imply
x < x0 < x00
and
x00 < x0 < x:

11. It follows from 7 that 6b implies 4b.

12. How restrictive and realistic are these axioms?

13. Examples of Non-Transitive Decision Rule (Rubinstein):

(a) The Condorcet Paradox of Majority Rule: Suppose each object x has
three characteristics: x1 ; x2 ; x3 . The rule states that x x0 i¤ there
exists i; j 2 f1; 2; 3g, i 6= j such that xi > x0i and xj > x0j . And
x x0 if x x0 and x0 x:
(b) x x0 if x x0 > 1; and x x0 if x x0 and x0 x:

2
0.2 Utility Function
1. A utility function de…ned over X assigns a real number to each member of
X. We say a utility function u : Rn ! R represents a preference relation
< if for any objects x and x0 ,

x < x0 i¤ u(x) u(x0 ):

2. Utility function is a convenient way to describe a preference relation. For

example, if I tell you that my preferences over apples and bananas is
u (a; b) = a0:5 + b, then you would know how I would choose between any
combinations of apples and bananas.
3. Theorem. A preference relation < can be represented by a utility function
only if it satis…es completeness and transitivity.

4. Proof: Suppose < can be represented by u. Since for any x and x0 , either
u (x) u (x0 ) ; or u (x0 ) u (x). Hence < must be complete. Since
the ordering on real number is transitive, the preference relation must be
transitive.
5. The converse of theorem is also true whenever X is countable. The idea
is obvious, but it does not hurt to write down a formal proof.
6. Proof: Since X is countable, we can order it in some way:

X = fx1 ; x2 ; :::g

Let u (x1 ) = 0. De…ne u (xi ), i > 1 recursively. For all i > 1, re-order
the objects such that xj denote the element with the j-th highest utility
among all xj , j < i. (By construction xj xj+1 .)
8
>
> (u (x1 ) + 1) =2 if xi x1
>
>
< u (x1 ) if xi x1
u (xi ) = u xj + u xj 1 =2 if xj 1 xi xj for some j :
>
>
>
> u (xi ) if xi xi 1
:
(u (xi ) 1) =2 if xi xi 1

The point is that there is always a rational number between any two
rational numbers. Note that by construction u 2 [ 1; 1].

7. The above argument will not work is X is uncountable. (For example

when X is a convex subset of <n .) For in that case we cannot order X
(or, mathematically speaking, construct a one-to-one mapping between X
and the set of integers).
8. To ensure that a preference relation over a convex subset of <n is repre-
sentable by a utility function, we need an extra axiom:

3
 9. Axiom 3: (Continuity) There are two (equivalent) ways to state this ax-
ioms.

(a) If y x, then there exist neighborhoods By of y and Bx of x such

that any y 0 x0 for any y 0 2 By and x0 2 Bx . (We are using the
metric of the Euclidean space.)
(b) The preference relation is continuous if it is preserved under limits.
That is, for any two sequences of consumption bundles, xn and y n ,
n = 1; 2; ::::with xn y n for all n, we have lim xn lim y n .

10. 9a says that if y is strictly better than x, then anything su¢ ciently close
to y will also be strictly better than anything su¢ ciently close to x.
11. Assume 9a. If 9b is false and that lim xn lim y n . Then there would exist
a neighborhood of lim xn and a neighborhood of lim y n such that anything
in the former would be strictly better than anything in the latter. But
this would contradict the supposition that xn y n for all n.

12. Assume 9b. If 9a is false, then there must exist a sequence xn converging
to x and a sequence y n converging to y such that y n xn for all n. By
9b, y x, contradicting the supposition that y x.
13. Not all rational preferences are continuous: Example: Lexicographic pref-
erences. Suppose there are two goods: x1 and x2 . A consumer prefers x
to x0 if either

x1 > x01
or x1 = x01 and x2 > x02

and is indi¤erent between x and x0 if x = x0 .

14. Consider the sequence xn = ((1 + 1=n) ; 1), n = 1; 2; :::: For all n, xn is
strictly preferable to (1; 2). [Try to plot the utilities on a real line.] For all
n, xn must lie between the utility xn 1 and the utility of (1; 2), and the
utility of (1; 1) is less than that of (1; 2). But you can see that as n goes
to in…nity, the utility is going to converge to that of (1; 2). Somehow the
lexicographic preference relation requires that the utility to make a jump
at (1; 1).

15. It is straightforward to see that the lexicographic preference relation is

complete and transitive.
16. For an elegant formal proof that it could not be represented by a utility
function, see Mas Colell pp. 46 and Rubinstein pp.16.

17. Theorem. A preference relation < that is complete, transitive, continuous,

and de…ned over a convex subset of <n , can be represented by a continuous
utility function.

4
18. Note that the continuity axiom is su¢ cient but not necessary for a pref-
erence relation to be representable by a utility function.
19. This theorem due to Debreu is classic result in economics. But it is mainly
a mathematical result and does not convey any economic insight.
20. We will consider two proofs. One follows Reny and Mas-Colell that makes
use of an additional axiom. The second, which follows Rubinstein, proves
existence but not continuity.

MGW
1. Axiom 4: Strict Monotonicity. For all x and x0 , if x x0 , then x x0 , and
if, in addition, xi < x0i for some i, then x x0 . The weakly-better-than-set
for x is
fx0 jx0 xg :
Axiom 4 means that the indi¤erence set will be a line which always have
a negative slope. We call these lines indi¤erence curves. Note that indif-
ference curves never crosses each other.
2. Step 1. Let e (1; 1; :::; 1) denote a unit consumption bundle (which
contains one unit of each good). For any consumption bundle x, de…ne
u (x) by the equation
u (x) e x:
(u (x) e (u (x) ; :::; u (x)), i.e., u (x) e is the consumption bundle that
consists of u (x) units of each good.)
Step 2. u (x) is uniquely de…ned for all x. Let xmax max fxi ; i = 1; :::; ng.
By strict monotonicity, it must be that

xmax e < x < 0:

By the continuity of preferences, the sets

A ft 2 <+ : te < xg

and
B ft 2 <+ : x < teg
are both closed. By monotonicity,

A = ft 2 [t; 1)g

and
B = t 2 [0; t) :
By completeness, any t must belong to either A or B; thus A+ [A = <+ .
We therefore can conclude that t t; or that the intersection of A and B
is non-empty. Uniqueness follows from the monotonicity axiom.

5
 Step 3. For any two consumption bundles x and x0 ,
x x0 , u (x) e u (x0 ) e (transitivity)
, u (x) u (x0 ) (strict monotonicity).
Hence, x < x0 if and only if u (x) u (x0 ).
Step 4. (Technical) A function u is continuous at x0 i¤ for all " > 0, there
exists some such that for all x0 2 fx : d (x; x0 ) g, ju (x) u (x0 )j ".
Intuitively, if u is continuous at x, a small movement from x will produce
only a small change in u. To prove continuity, we need to make use of the
following mathematical result: A function is continuous if for any open
set (a; b) <; the inverse mapping u 1 ((a; b)) is open in <n+ . (Recall
that a set in <n is open if and only for any point in the set, an open ball
including the point also belongs to the set. A set is closed if its complement
is open.) Explain brie‡y the “if” part: Roughly, if the inverse image of
an open set is open, then pick any x and an open ball B" (u (x)), an open
ball B (x) B" (u (x)). Now, in our case,
1
u ((a; b)) = fxja < u (x) < b)g
= fxjae u (x) e beg
= fxjae x beg :
Finally, the intersection of two open set is open.
3. This proof makes use of the natural utility function de…ned on the diagonal
and uses the continuity property to identify each x with a unique element
on the diagonal.

Rubinstein
1. The drawback of the …rst proof is that it requires a unnecessary axiom.
2. Recall we have constructed a utility function de…ned over all rational z
on ( 1; 1). Now, the set of rational numbers are dense in the set of <n .
That is, for any real number x, you can …nd a rational number that is as
close to z as you want. So we can just assign the utility of x to be equal
to the utility of a rational number very very close to it.
3. First, we show that if is continuous, then for any x y, there exists z
such that x z y.
4. Suppose not.
5. Set x1 = x and y1 = y.
6. Given (xt ; yt ). Consider the midpoint (xt + yt ) =2. By supposition, it is
either that (xt + yt ) =2 x or y (xt + yt ) =2. In the …rst case, set
xt+1 = (xt + yt ) =2;
yt+1 = yt :

6
 In the second case, set

xt+1 = xt ;
yt+1 = (xt + yt ) =2:

7. Note that by construction for all t

xt x y yt ,

and xt and yt converges to the same limit, say, q, and by continuity y q

x, a contradiction.
8. Let v be utility function de…ned over all rational x on ( 1; 1). Let Y be
the set of all rational consumption bundles. For any x, we de…ne

u (x) = sup fv (z) jz 2 Y and x zg .

Set u (x) = 1 if the set is empty.

9. We now show that u is a utility function that represents .
10. First, note that if x x0 , then u (x) = u (x0 ).
11. Suppose x x0 . Since is continuous and rational numbers are dense in
<n , there are some rational z and z 0 such that

x z z0 x0 .

12. By construction u (x) v (z) > v (z 0 ) u (x0 ). (The …rst inequality

follows from the de…nition of u and that fact that x z. The second
follows from the fact that v (z) u (z) for any z 2 Y (because z does not
belong to the set fz 0 2 Y and z z 0 g. A related point: it is possible that
v (z) > u (z) for any z 2 Y:)

0.2.1 Ordinality v.s. Cardinality

1. A monotonic transformation is a function f that is strictly increasing.
Examples: adding a constant, take square root.
2. Many utility functions represent the same underlying preferences. In fact,
if u represents a preference relation then any monotonic transformation
of u will also represents the same preference relation.
3. The only information conveyed by a utility function is about where a
bundle has a higher utility than another bundle. It doesn’t tell you by
how much you like the …rst bundle over the second.
4. We call a measure that conveys information about ranking an ordinal
measure. By contrast, a cardinal measure measures intensity as well as
ranking.