Anda di halaman 1dari 29

Chapter 2 Duality and Revealed Preferences

Economics 5113 Microeconomic Theory

Kam Yu

Winter 2017
Outline
1 Duality Theory: From Expenditure to Utility
Expenditure and Utility Functions
Convexity and Monotonicity
2 Duality Between Direct and Indirect Utility
From Indirect to Direct Utility Functions
Inverse Demand Functions
3 Duality Between Demand and Preferences
Important Characterization of the Demand Function
The Integrability Problem
4 Revealed Preference
Weak Axiom
Strong Axiom

Kam Yu (Lakehead) Chapter 2 Duality and Revealed Preferences Winter 2017 2 / 29


Duality Theory: From Expenditure to Utility Expenditure and Utility Functions

Recovering the Utility Function

Suppose that we are given a function E : Rn++ R+ R+ that


satisfies the seven properties of an expenditure function.
That is, E (p, u) is continuous, strictly increasing and unbounded
above in u, increasing, linearly homogeneous, concave, and
differentiable in p.
Can we find an utility function U that gives E in an expenditure
minimization problem?
The question is analogous to the Minkowski theorem in convex
analysis (see section 6.6 in Yu): A closed convex set is the
intersection of its supporting half-spaces.

Kam Yu (Lakehead) Chapter 2 Duality and Revealed Preferences Winter 2017 3 / 29


Duality Theory: From Expenditure to Utility Expenditure and Utility Functions

Another Maximization Problem


Consider the following constrained maximization problem:

max u
u
subject to pT x E (p, u), for all p  0,
u 0.

The control variable is u. The vector x is a parameter.


The vector p is a condition imposed on the maximization problem
that the first inequality constraint holds for all p R++ .
Effectively we have infinitely many constraints, each p represents a
half space containing the upper contour set % (x).
If a solution exists, the optimal choice u is a function of x, that is,
u = U(x), which is also the value function.

Kam Yu (Lakehead) Chapter 2 Duality and Revealed Preferences Winter 2017 4 / 29


at as more and Duality
moreTheory:
prices are
From considered
Expenditure to Utility and more and
Expenditure setsUtility
areFunctions
added to the intersection,
aded area will more closely resemble a superior set for some quasiconcave real-val
Minkowski
nction. One mightTheorem at Work
suspect, therefore, that these sets can be used to construct someth

x2 x2

A(p0, u0) !A(p, u0)

{x|p0 x ! E(p0, u 0)}

x1 x1

(a) (b)

Figure 2.1. (a) The closed half-space A(p0 , u0 ). (b) The intersection of a
finite collection of the sets A(p, u0 ).

Kam Yu (Lakehead) Chapter 2 Duality and Revealed Preferences Winter 2017 5 / 29


Duality Theory: From Expenditure to Utility Expenditure and Utility Functions

Duality Theorem
We claim that if E satisfies the seven properties of an expenditure
function, then the value function U(x) is increasing, unbounded above,
and quasi-concave.

Sketch of the proof (see pages 7476 in JR for details)


Existence of u : Since E is increasing and bounded above in u, the
feasible set,
n o
G (x) = u 0 : pT x E (p, u) p  0 ,

is bounded. Since it is also closed, the solution exists by the


Weierstrass theorem.
Note: Since E is linearly homogeneous in p, G (x) depends on the
relative prices, not the absolute price of each good.

Kam Yu (Lakehead) Chapter 2 Duality and Revealed Preferences Winter 2017 6 / 29


Duality Theory: From Expenditure to Utility Expenditure and Utility Functions

Monotonicity: Suppose that x1 x2 . Then pT x1 pT x2 for all


p  0. Let u 2 = U(x2 ). That is, u 2 is the solution of the
maximization problem for x2 . Then pT x2 E (p, u 2 ) and therefore
pT x1 E (p, u 2 ) for all p  0. This means that u 2 G (x1 ) and so
U(x1 ) u 2 = U(x2 ) as required.
Quasi-concavity: Suppose that U(x2 ) > U(x1 ). We want to show
that for any convex combination x of the two bundles,

U(x ) min U(x1 ), U(x2 ) = U(x1 ).




From the definition of G (x), we know that pT x1 E (p, U(x1 )) and


pT x2 E (p, U(x2 )) for all p  0. Therefore for 0 1,

pT x1 + (1 )pT x2 E (p, U(x1 )) + (1 )E (p, U(x2 )).

Since E is increasing in u, pT x E (p, U(x1 )). We conclude that


U(x ) U(x1 ).
The proof of unboundedness is left as an exercise.
Kam Yu (Lakehead) Chapter 2 Duality and Revealed Preferences Winter 2017 7 / 29
Duality Theory: From Expenditure to Utility Expenditure and Utility Functions

Confirmation that U is a Utility Function


Now we show that the value function U(x) in the above maximization
problem is the utility function dual to expenditure function. That is,
n o
E (p, u) = min pT x : U(x) u, x 0 .
x

Suppose that p0  0 and u 0 0. For any x Rn+ such that


U(x) u 0 , we have, by the definition of U and monotonicity of E in
u,
pT x E (p, U(x)) E (p, u 0 )
for all p  0.
Let p = p0 . The above inequality can be stated as
E (p0 , u 0 ) (p0 )T x for all x Rn+ such that U(x) u 0 . This implies
that n o
E (p0 , u 0 ) min (p0 )T x : U(x) u 0 , x 0 . (1)
x

Kam Yu (Lakehead) Chapter 2 Duality and Revealed Preferences Winter 2017 8 / 29


Duality Theory: From Expenditure to Utility Expenditure and Utility Functions

Our goal is to show that (1) is an equality. To achieve this we shall


find an x0 Rn+ such that U(x0 ) u 0 and (p0 )T x0 E (p0 , u 0 ).
Since E is concave in p, by theorem 6.5 in Yu,

E (p, u 0 ) E (p0 , u 0 ) + p E (p0 , u 0 )T (p p0 )


= E (p0 , u 0 ) + p E (p0 , u 0 )T p p E (p0 , u 0 )T p0
= E (p0 , u 0 ) + p E (p0 , u 0 )T p E (p0 , u 0 ) (Eulers thm)
0 0 T
= p E (p , u ) p.

Define x0 = p E (p0 , u 0 ) so that we have E (p, u 0 ) pT x0 for all


p  0. First, x0 Rn+ because E is increasing in p. Second, by the
definition of U, U(x0 ) u 0 .
Finally, applying Eulers theorem to E at the point (p0 , u 0 ) gives

E (p0 , u 0 ) = p E (p0 , u 0 )T p0 = (p0 )T x0

as required.
Kam Yu (Lakehead) Chapter 2 Duality and Revealed Preferences Winter 2017 9 / 29
Duality Theory: From Expenditure to Utility Convexity and Monotonicity

Is U Generated by E Unique?

Recall that when we prove the existence and the seven properties of
E , we do not use the properties that U is quasi-concave and
increasing.
Now when we use E to generate an utility function by the above
process, say,
n o
W (x) = max u 0 : pT x E (p, u) p  0 ,
u

are we sure that W = U if U is not increasing and quasi-concave?


The answer is, not necessary. The duality theorem we proved above
guarantees that W is increasing and quasi-concave. So if U does not
possess these properties, then W 6= U.

Kam Yu (Lakehead) Chapter 2 Duality and Revealed Preferences Winter 2017 10 / 29


Duality Theory: From Expenditure to Utility Convexity and Monotonicity

General Relation Between W and U

In general, W (x) U(x) for all x 0.

Proof : By the definition of E , we have E (p, U(x)) pT x for all


p  0. This means that U(x) is in the feasible set of finding W .
Therefore W (x) U(x).

Let %U and %W be the preference relations induced by U and W


respectively. The above result implies that for any utility level v ,
%U (v ) %W (v ).

Kam Yu (Lakehead) Chapter 2 Duality and Revealed Preferences Winter 2017 11 / 29


Duality Theory: From Expenditure to Utility Convexity and Monotonicity

u(x) ! uCHAPTER 2
When U is Not Increasing x 1 and Not Quasi-Concave x1
0
(a) (b)

x2 x2

S(u)
u(x) ! u u(x) ! u
u(x) ! u

w(x) ! u

x1 x1 x1
0 0
(c) (b) (d)

x2
x2

Kam Yu (Lakehead) Chapter 2 Duality and Revealed Preferences Winter 2017 12 / 29


Duality Theory: From Expenditure to Utility Convexity and Monotonicity

When is U = W ?

If U is increasing and quasi-concave, then W = U.

Proof : Suppose that x0 0 and let U(x0 ) = u. By monotonicity and


quasi-concavity there exists a price vector p0 such that

E (p0 , u) = (p0 )T x0 ,

which is a supporting hyperplane of the convex set % (x0 ). By definition,


W (x0 ) satisfies the constraint pT x0 E (p, W (x0 )) for all p  0. When
p = p0 , we have (p0 )T x0 E (p0 , W (x0 )), which means

E (p0 , u) E (p0 , W (x0 )).

Since E is increasing in u, U(x0 ) = u W (x0 ).

Kam Yu (Lakehead) Chapter 2 Duality and Revealed Preferences Winter 2017 13 / 29


Duality Between Direct and Indirect Utility From Indirect to Direct Utility Functions

Just Another Minimization Problem


Recall that in the utility maximization problem, given any
consumption bundle x, and imagine that y = pT x, we have, for any
p  0,
V (p, pT x) U(x). (2)
Is there some prices p that turns the inequality into an equality?
The answer is yes. In fact if % is strictly convex, p is unique. The
budget constraint pT z = y is a supporting hyperplane for % (x) at the
bundle x.
Inequality (2) implies that
n o
U(x) = min V (p, pT x) : p  0 .
p

This provides a mean to recover the utility function from the indirect
utility function.

Kam Yu (Lakehead) Chapter 2 Duality and Revealed Preferences Winter 2017 14 / 29


Duality Between Direct and Indirect Utility From Indirect to Direct Utility Functions

From V to U
Suppose that the utility function U of a consumer is quasi-concave and
the marginal utilities of all goods are positive, that is, U(x)  0. Let
V (p, y ) be the indirect utility function generated by U. Then for all
consumption bundle x,
n o
U(x) = min V (p, pT x) : p  0 . (T.1)
p

Proof : Given any x 0, we have, for all p  0,

V (p, pT x) U(x).

Let p = U(x), = 1, and y = pT x. Then we have

U(x) = p and pT x = y ,

which are the sufficient conditions for the utility maximization problem.
Kam Yu (Lakehead) Chapter 2 Duality and Revealed Preferences Winter 2017 15 / 29
Duality Between Direct and Indirect Utility Inverse Demand Functions

Variations
Since V is homogeneous of degree zero in (p, y ), the duality relation
can be written as
n o
U(x) = min V (p, 1) : pT x = 1, p  0 . (T.1)
p

In practice, the problem in (T.1) may be analytically more convenient


then that in (T.1).
Suppose that y = 1. Then the inverse demand function is
U(x)
p = d 1 (x) = .
U(x)T x
Proof : In the UMP, the Lagrange multiplier theorem implies that
U(x) = p. Together with the budget constraint pT x = 1, we have
U(x) p p
T
= T
= T = p.
U(x) x p x p x
Kam Yu (Lakehead) Chapter 2 Duality and Revealed Preferences Winter 2017 16 / 29
Duality Between Demand and Preferences Important Characterization of the Demand Function

Can We Get U from d?

Since consumer behaviours are observable, we can in principle


estimate the ordinary demand function d.
An important question is, what properties of d must possess so that
we can derive the utility function from it?
So far we have derived the following properties of the demand
function:
1 Budget balancedness: pT d(p, y ) = y .
2 The Slutsky matrix S(p, y ) is symmetric.
3 The Slutsky matrix is negative semidefinite.
4 Homogeneity of degree zero in (p, y ).
5 Cournot and Engel aggregations.
We have shown that property 1 implies 5. Are there any more
redundancy?

Kam Yu (Lakehead) Chapter 2 Duality and Revealed Preferences Winter 2017 17 / 29


Duality Between Demand and Preferences Important Characterization of the Demand Function

Budget Balancedness and Symmetry Imply Homogeneity

Properties 1 and 2 imply property 4.

Proof : Differentiate the budget constraint with respect to pi and y , we


get, for i = 1, . . . , n,
n
X dj (p, y )
pj = di (p, y ), (3)
pi
j=1
n
X dj (p, y )
pj = 1. (4)
y
j=1

For any p and y , define a single variable function fi (t) = di (tp, ty ) for
t > 0. Our goal is to show that fi (t) = di (p, y ), which means that fi is
independent of t, or fi 0 (t) = 0 for all t > 0.

Kam Yu (Lakehead) Chapter 2 Duality and Revealed Preferences Winter 2017 18 / 29


Duality Between Demand and Preferences Important Characterization of the Demand Function

Using the product rule,


n
X di (tp, ty ) di (tp, ty )
fi 0 (t) = pj +y . (5)
pj y
j=1

Budget balancedness means that tpT d(tp, ty ) P


= ty , or simply
p d(tp, ty ) = y , which can be written as y = nj=1 pj dj (tp, ty ).
T

Substitute this y into equation (5), we get


n  
X di (tp, ty ) di (tp, ty )
fi 0 (t) = pj + dj (tp, ty ) . (6)
pj y
j=1

Notice that the expression inside the square bracket is the i-j term of the
Slutsky matrix.

Kam Yu (Lakehead) Chapter 2 Duality and Revealed Preferences Winter 2017 19 / 29


Duality Between Demand and Preferences Important Characterization of the Demand Function

Since the Slutsky matrix is symmetric, we can switch the i and the j inside
the bracket in equation (6):
n  
0
X dj (tp, ty ) dj (tp, ty )
fi (t) = pj + di (tp, ty )
pi y
j=1

n n
X dj (tp, ty ) X dj (tp, ty )
= pj + di (tp, ty ) pj
pi y
j=1 j=1

n n
1 X dj (tp, ty ) 1 X dj (tp, ty )
= tpj + di (tp, ty ) tpj
t pi t y
j=1 j=1
1 1
= di (tp, ty ) + di (tp, ty ) (by equations (3) and (4))
t t
= 0.

Kam Yu (Lakehead) Chapter 2 Duality and Revealed Preferences Winter 2017 20 / 29


Duality Between Demand and Preferences The Integrability Problem

We Only Need Three Properties for d

Now we are back to the original question. Given the observable


ordinary demand function that satisfies budget balancedness, with a
symmetric and negative semidefinite Slutsky matrix, can we recover
the utility function?
The answer is yes, and it is called the integrability problem.
We do this by showing that given the ordinary demand function, the
expenditure function exists. With the expenditure function, the utility
function can be derived.
The process is called integrability probably because the solution
involves a system of partial differential equations, hence the need for
integrations.
A sketch of the theory for the n = 2 case is discussed here.
For details, see The Integrability Problem by K.C. Border.

Kam Yu (Lakehead) Chapter 2 Duality and Revealed Preferences Winter 2017 21 / 29


Duality Between Demand and Preferences The Integrability Problem

The Integrability Problem


Suppose that a demand function d for two goods satisfies budget
balancedness, with a symmetric and negative semidefinite Slutsky
matrix.
Using the homogeneity property we can normalize the price of good 2,
that is, we can set p2 = 1.
For any price-income point (p1 , 1, y ), the demand function is
d(p1 , 1, y ), with utility u = V (p1 , 1, y ). The expenditure function
can be expressed as E (p1 , 1, u ) = e(p1 ).
The function e is the solution of the following first-order ordinary
differential equation:

de(p1 )
= d1 (p1 , 1, e(p1 )), (7)
dp1

with the initial condition e(p10 ) = y 0 .

Kam Yu (Lakehead) Chapter 2 Duality and Revealed Preferences Winter 2017 22 / 29


Duality Between Demand and Preferences The Integrability Problem

Existence of Solution

We invoke the Picard Theorem for the sufficient condition for


existence of a unique solution for e: If d1 (p1 , y ) and d1 /y are
continuous on a neighbourhood B of the initial point (p10 , y 0 ), then
there exists an interval I centred at p10 and a unique function e(p1 )
that solve the differential equation.
Given that d satisfies the three conditions, the solution e(p1 ) has all
the properties of an expenditure function:
1 Continuous as a solution to an ODE.
2 Increasing in p1 , since d1 (p1 , 1, e(p1 )) 0.
3 Concave in p1 :

d2 e(p1 ) d1 (p1 , 1, e(p1 )) d1 (p1 , 1, e(p1 ))


= + d1 (p1 , 1, e(p1 ))
dp12 p1 y
= s11 0.

Kam Yu (Lakehead) Chapter 2 Duality and Revealed Preferences Winter 2017 23 / 29


Duality Between Demand and Preferences The Integrability Problem

Example of a First-Order Ordinary Differential Equation


Suppose that y = f (t) and satisfies the equation

dy
= ky , (8)
dt
where k is a given constant. We also know that at time t = 0, y = y0 .
We know from calculus that if y = ce kt , then the derivative satisfies
equation (8), where c is an arbitrary constant.
Substitute the initial condition in the solution, we have

y0 = ce k0 ,

which gives c = y0 .
Therefore the solution for this ODE is

y = y0 e kt .

Kam Yu (Lakehead) Chapter 2 Duality and Revealed Preferences Winter 2017 24 / 29


Duality Between Demand and Preferences The Integrability Problem

General n-Commodity Case

The problem becomes a system of first-order partial differential


equations:
e(p) = d(p, e(p)),
with initial condition e(p0 ) = y 0 .
The problem is as if we try to recover the expenditure function from
the demand function using Shephards lemma.
Frobenius theorem: A necessary and sufficient condition for the
solution to exist is the symmetry of the Slutsky matrix.
Negative semi-definiteness of the Slutsky matrix implies that e has
the properties of an expenditure function.

Kam Yu (Lakehead) Chapter 2 Duality and Revealed Preferences Winter 2017 25 / 29


Revealed Preference Weak Axiom

Consumer Theory without Axioms


Paul Samuelson suggested that we could derive a consumer theory
from observing their behaviours.
Weak axiom of revealed preference (WARP): Suppose that two
distinct bundles x0 and x1 are chosen by a consumer when prices are
p0 and p1 respectively. Then

(p0 )T x1 (p0 )T x0 (9)

implies that
(p1 )T x0 > (p1 )T x1 . (10)
When prices are p0 , the consumer chooses bundle x0 . Inequality (9)
means that bundle x1 is feasible but not chosen. It is natural to
conclude that x0 % x1 .
When prices are p1 , the consumer chooses bundle x1 . Inequality (10)
means that bundle x0 is not feasible, otherwise it would be chosen by
the consumer.
Kam Yu (Lakehead) Chapter 2 Duality and Revealed Preferences Winter 2017 26 / 29
oice made by this consumer when faced with
Revealed Preference
prices p and income y. Note well that th
Weak Axiom
not a demand function because we have not mentioned utility or utility maximisation
justWARP
denotes the
andquantities
Not WARPthe consumer chooses facing p and y. To keep this point cle
our minds, we refer to x(p, y) as a choice function. In addition to WARP, we make o

x2 x2

x0

x1 x0 x1
p 0 p0
p1 p1
x1 x1

(a) (b)

Figure 2.3. The Weak Axiom of Revealed Preference (WARP).

Kam Yu (Lakehead) Chapter 2 Duality and Revealed Preferences Winter 2017 27 / 29


Revealed Preference Weak Axiom

Two Observations

1 The weak axiom can be alternatively stated as follows. If the


Laspeyres quantity index QL between period 0 and period 1 is less
than or equal to 1, then the Paasche qauntity index QP is less than 1,
where
(p0 )T x1 (p1 )T x1
QL = 0 T 0 , QP = 1 T 0 .
(p ) x (p ) x
2 A continuous, increasing, and quasi-concave utility function satisfies
the weak axiom.

Question: Is the converse of statement 2 also true? If it is, then the weak
axiom of revealed preference characterizes a well-defined utility function.

Kam Yu (Lakehead) Chapter 2 Duality and Revealed Preferences Winter 2017 28 / 29


Revealed Preference Strong Axiom

Revealed Preference and Integrability

If the revealed preferences of a consumer satisfy the weak axiom and


budget balancedness, then the resulting choice function (see pages
9395 in JR)
1 is homogeneous of degree 0 in (p, y ),
2 has a negative semidefinite Slutsky matrix,
3 (when n = 2) has a symmetric Slutsky matrix.
Therefore, in the case of n = 2, we can invoke integrability to recover
the utility function.
For the general case where n > 2, we need the strong axiom of
revealed preference: Suppose that a sequence of distinct bundles
x0 , x1 , . . . , xk , where xi1 is revealed preferred to xi (in the sense of
the weak axiom) for all i = 1, . . . , k. Then xk cannot be preferred to
x0 .

Kam Yu (Lakehead) Chapter 2 Duality and Revealed Preferences Winter 2017 29 / 29

Anda mungkin juga menyukai