Comparative Statics: 1 The Maximum Theorems

c 2001 Michael Carter
All rights reserved
Lecture notes based on

Foundations of Mathematical Economics
Comparative statics
1 The maximum theorems
max (x, )
x()
Let
() = max (x, )
() = arg max (x, )
x()
Objective
function
Constraint
correspondence
Value
function
Solution
correspondence
Monotone
maximum
theorem
Theorem 2.1
supermodular,
increasing
weakly
increasing
increasing
increasing
x()
Continuous
maximum
theorem
Theorem 2.3
continuous
Convex
maximum
theorem
Theorem 3.10
concave
Smooth
maximum
theorem
Theorem 6.1
smooth
continuous,
compact-valued
continuous
convex
compact-valued
nonempty, uhc
convex-valued
smooth
regular
locally
smooth
locally
smooth
2 The envelope theorems

2.1 Envelope theorem 1
() = max
(x, )
= (x (), )
so that
() = x
x
+
The rst-order conditions determining x are

x = x
concave
All rights reserved

Moveover, x () satises the constraint as a identity

(x ()) = 0 = x
x
=0
Substituting, we conclude that

() =
Example 1 (Chip producer) It is characteristic of microchip production technology that a proportion of output is defective. Consider a small producer for
whom the price of good chips is xed. Suppose that proportion 1 of the
rms chips are defective and cannot be sold. Let () denote the rms total
cost function where is the number of chips (including defectives) produced.
Suppose that with experience, the yield of good chips increases. How does this
aect the rms production ? Does the rm compensate for the increased yield
by reducing production, or does it celebrate by increasing production?
The rms optimization problem is
() = max ()
= ( )

( )
() = +
= + ( ( ))
But the rst-order condition dening () is

( ) = 0
so that
() = > 0
Further, we can deduce that

() =
so that
()
()
()
=
0
since the prot function is convex.
All rights reserved

2.2 Envelope theorem 2
() = max (x, )
()
= (x (), )
x
() = x
+
The rst-order conditions determining x are

x = x
Moveover, x () satises the constraint as a identity
(x (), ) = 0 = x
or
x
+ = 0
x
=
x
Substituting, we conclude that

() = =
Example 2 (Consumer problem)
(p, ) = max (x)
(1)
subject to p x =
(2)
= (x) (p x )
= =
= =
which leads immediately to Roys identity

x (p, )

All rights reserved
2.3 Smooth envelope theorem (Corollary 6.1.1)

Assume that x0 is a strict local maximum of
max (x, )
x()
where () = { x : g(x, ) 0 }. By the smooth maximum theorem,

there exists a neighbourhood around 0 and function x such that
() = (x (), ) for every
and is dierentiable. Applying the chain rule
[] = x x +
indirect direct
What do we know of the indirect eect?

First If x is optimal, it must satisfy the Kuhn-Tucker conditions
x = 0 gx and 0 g(x, ) = 0
(3)
at (x0 , 0 ) where 0 is the unique Lagrange multiplier associated with

x0 .
Second The solution x () satises the constraint g(x (), ) = 0 for all
. Another application of the chain rule gives
gx x + g = 0 = 0 gx x = g
(4)
Using (3) and (4), the indirect eect is x x = 0 x x = g and therefore

[] = 0 g =
(5)
where denotes the Lagrangean (x, , ) = (x, ) g(x, ). This is the
envelope theorem, which states that the derivative of the value function
is equal to the partial derivative of the Lagrangean evaluated at the optimal
solution (x0 , 0 ).
In the special case in which the feasible set is independent of the parameters, g = 0 and (5) becomes
[] =
The indirect eect is zero, and the only impact on of a change in is the
direct eect f .
4

All rights reserved
2.4 General envelope theorem (Theorem 6.2)

The assumptions required for Corollary 6.1.1 are stringent. Where the feasible set is independent of the parameters, a more general result can be given.
Let x be the solution correspondence of the constrained optimization problem
max (x, )
x
in which : is continuous and compact. Suppose that is

continuously dierentiable in , that is [x, ] is continuous in .
Then the value function
() = sup (x, )
is dierentiable wherever x is single-valued with [] = [x(), ].

Proof.
To simplify the proof, assume that x is single-valued for every
Then
() = (x (), ) for every
For any = 0
)
(
)
(
() ( 0 ) = x (), x ( 0 ), 0
)
(
)
(
x ( 0 ), x ( 0 ), 0
= [x ( 0 ), 0 ]( 0 ) + () 0
with () 0 as 0 . On the other hand, by the mean value theorem
(, 0 ) such that
(Theorem 4.1) there exist
)
(
)
(
() ( 0 ) = x (), x ( 0 ), 0
)
(
)
(
x (), x (), 0
0)
= [x (), ](
Letting 0
[x ( 0 ), 0 ]( 0 )
() ( 0 )
[x ( 0 ), 0 ]( 0 )
lim
lim
lim
0
0
0
0
0
0
is dierentiable (Exercise 4.3) and
[] = [x (), ]
where [x (), ] denotes the partial derivative of with respect to
holding x constant at x = x ().

5
All rights reserved

Note that there is no requirement in Theorem 6.2 that is dierentiable with respect to the decision variables x, only with respect to the
parameters. The practical importance of dispensing with dierentiability with respect to x is that Theorem 6.2 applies even when the feasible
set is discrete (See Example 6.2).
()
(1 , )
(2 , )
(3 , )
3 Comparative statics of optimization models

There are four dierent approaches to comparative statics of optimization
models
Revealed preference approach
Envelope theorem approach
Monotone maximum theorem approach
Implicit function theorem approach
3.1 Revealed preference approach
A competitive rms optimization problem is to choose a feasible production
plan y to maximize total prot
max p y
y
Consequently, if y1 maximizes prot when prices are p1 , then

p1 y1 p y for every y
Similarly, if y2 maximizes prot when prices are p2 , then
p2 y2 p y for every y
6
All rights reserved

In particular
p1 y1 p1 y2
and
p2 y2 p2 y1
Adding these inequalities

p1 y1 + p2 y2 p1 y2 + p2 y1
Rearranging
p2 (y2 y1 ) p1 (y2 y1 )
and therefore
(p2 p1 ) (y2 y1 ) 0
or
(1 2 )(2 2 ) 0
(6)
=1
If prices change from p1 to p2 , the optimal production plan must change in

such a way as to satisfy the inequality (6). For a change in the price of a
single good (2 = 1 for every = ), (6) implies that
(2 1 )(2 1) 0
or
2 > 1 = 2 1
3.2 The envelope theorem approach

Letting (y, p) = p y denote the objective function, the competitive rm
solves
max (y, p)
y
Note that is dierentiable with p [y, p] = y. Applying the envelope

theorem 6.2, the prot function
(p) = sup (y, p)
y
is dierentiable wherever the supply correspondence y is single-valued with

p [p] = p [y (p), p] = y (p)
or
(7)
y (p) = (p)
which is known as Hotellings lemma.

The practical signicance of Hotellings lemma is that, if we know the
prot function, we can calculate the supply function by straightforward
dierentiation instead of solving a constrained optimization problem.
7
All rights reserved

Its theoretical signicance is more important. Hotellings lemma enables us to deduce the properties of the supply function y from the
already established properties of the prot function. In particular, we
know that the prot function is convex (Example 3.42).
From Hotellings lemma (7), we deduce that the derivative of the supply
function is equal to the second derivative of the prot function
y [p] = 2 [p]
or equivalently that the Jacobian of the supply function is equal to the Hessian of the prot function.
y (p) = (p)
Since is smooth and convex, its Hessian (p) is symmetric (Theorem 4.2)
and nonnegative denite (Proposition 4.1) for all p. Consequently, the Jacobian of the supply function y is also symmetric and nonnegative denite.
This implies for all goods and
[p] 0
[p] = [p]
Nonnegativity
Symmetry
In a similar fashion, we can deduce

Shephards lemma (Example 6.7)
Roys identity (Example 6.8)
From the latter, we can easily derive the Slutsky equation (Example 6.9).
3.3 The implicit function theorem approach
The rst-order conditions of an equality constrained optimization problem
constitute a system of equations.
(x; ) = 0
Provided the Jacobian (x [x; ]) of this system is non-singular, we can use
the implicit function theorem to solve for x in terms of . We illustrate by
means of an example.

All rights reserved
Example Recall again the chip maker, whose optimization problem is

max ()
The rst-order and second-order conditions for prot maximization are

(, , ) = () = 0 and [, , ] = () < 0
The second-order condition requires increasing marginal cost. Assuming is
2 , the rst-order condition implicitly denes a function (). Dierentiating
the rst-order condition with respect to , we deduce that
= ()
or
()
which is positive by the second-order condition. An increase in yield is

analogous to an increase in product price , inducing an increase in output
.
Examples 6.15 and 6.16 apply the same technique to deduce the comparative statics of a competitive multi-input rm.
4 References
Milgrom, P., and I. Segal (2000), Envelope Theorems for Arbitrary
Choice Sets. Department of Economics, Stanford University: mimeo.
Silberberg, E. (1990), The Structure of Economics: A Mathematical
Analysis (2nd edition). New York, NY: McGraw-Hill.
All rights reserved

5 Homework
1. Prove Proposition 5.2, that is if and g are 2 and [x ] is of full
rank, then the value function
(c) = sup{ (x) : g(x) = c }
is dierentiable with (c) = , where = (1 , 2 , . . . , ) are the
Lagrange multipliers associated with x .
2. Suppose that the cost function of a monopolist changes from 1 () to
2 () in such a way that
0 < 1 () < 2 () for every > 0
Let 1 denote the prot maximizing price with the cost function 1 ()
and let 1 be the corresponding output. Similarly let 2 and 2 be the
prot maximizing price and output when the costs are given by 2 ().
(a) Show that
2 (1 ) 2 (2 ) 1 (1 ) 1 (2 )
(8)
(b) The Fundamental Theorem of Calculus states: If () is a

continuous function on [a,b], then

()
() () =
Apply this to inequality (8) to deduce that 1 2 and therefore

that 1 2 .
(c) State concisely the proposition you have just proved.
3. Assume that a competitive rm produces a single output from
inputs x = (1 , 2 , . . . , ) according to the production function =
(x) so as to maximize prot
(w, ) = max (x) w x
x
Assume that there is a unique optimum for every and w. Show

that the input demand (w, ) and supply (w, ) functions have the
following properties:
[w, ] 0
[w, ] 0
[w, ] = [w, ]
[w, ] = [w, ]
10
Upward sloping supply

Downward sloping demand
Symmetry
Reciprocity
All rights reserved

Solutions 7
1 The Lagrangean for this problem is
(
)
= (x) g(x) c
By Corollary 6.1.1
2
(c) = c =
(a) With cost function 1 (1 ), the rms prot is

= 1 ()
Since this is maximised at 1 and 1 (although the monopolist could
have sold 2 at price 2 )
1 1 1 (1 ) 2 2 1 (2 )
Rearranging
1 1 2 2 1 (1 ) 1 (2 )
Similarly
(1)
2 2 2 (2 ) 1 1 2 (1 )
which can be rearranged to yield

2 (1 ) 2 (2 ) 1 1 2 2
Combining the previous inequality with (1) yields
2 (1 ) 2 (2 ) 1 (1 ) 1 (2 )
(b) Applying the Fundamental Theorem of Calculus to both sides, this
implies
1
1
2 ()
1 ()
or
2 ()
2 ()
1 ()
1 ()
1
2
(2 () 1 ()) 0
0 for every (by assumption), this implies that

Since
2 1 . Assuming the demand curve is downward sloping, this implies
2 1 .

All rights reserved
(c) There is an implicit requirement to utilize the Fundamental Theorem of

Calculus, namely that () is continuous. With this proviso, we have
shown that the monopoly price is increasing in marginal cost. Specifically we have shown: Assuming that a monopolists cost function is
continously dierentiable (in output), the prot maximizing monopoly
price is an increasing (i.e. nondecreasing) function of marginal cost.
3 By Theorem 6.2
w [w, ] = x and [w, ] =
and therefore
2
(, w) =
(, w) 0
(, w) = 2 (, w) 0
(, w) = 2 (, w) = (, w)
(, w) = 2 (, w) = (, w)
since is convex and therefore (w, ) is symmetric (Theorem 4.2) and
nonnegative denite (Proposition 4.1).

Comparative Statics: 1 The Maximum Theorems

Uploaded by

Copyright:

Available Formats

Comparative Statics: 1 The Maximum Theorems

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Comparative Statics: 1 The Maximum Theorems

Uploaded by

Copyright:

Available Formats

c 2001 Michael Carter

All rights reserved

Lecture notes based on

() = arg max (x, )

2 The envelope theorems

The rst-order conditions determining x are

c 2001 Michael Carter

All rights reserved

Lecture notes based on

Moveover, x () satises the constraint as a identity

Substituting, we conclude that

But the rst-order condition dening () is

Further, we can deduce that

since the prot function is convex.

c 2001 Michael Carter

All rights reserved

Lecture notes based on

The rst-order conditions determining x are

Substituting, we conclude that

which leads immediately to Roys identity

Lecture notes based on

c 2001 Michael Carter

All rights reserved

2.3 Smooth envelope theorem (Corollary 6.1.1)

where () = { x : g(x, ) 0 }. By the smooth maximum theorem,

What do we know of the indirect eect?

at (x0 , 0 ) where 0 is the unique Lagrange multiplier associated with

Using (3) and (4), the indirect eect is x x = 0 x x = g and therefore

Lecture notes based on

c 2001 Michael Carter

All rights reserved

2.4 General envelope theorem (Theorem 6.2)

in which : is continuous and compact. Suppose that is

is dierentiable wherever x is single-valued with [] = [x(), ].

holding x constant at x = x ().

c 2001 Michael Carter

All rights reserved

Lecture notes based on

3 Comparative statics of optimization models

Consequently, if y1 maximizes prot when prices are p1 , then

c 2001 Michael Carter

All rights reserved

Lecture notes based on

Adding these inequalities

If prices change from p1 to p2 , the optimal production plan must change in

3.2 The envelope theorem approach

Note that is dierentiable with p [y, p] = y. Applying the envelope

is dierentiable wherever the supply correspondence y is single-valued with

which is known as Hotellings lemma.

c 2001 Michael Carter

All rights reserved

Lecture notes based on

In a similar fashion, we can deduce

Lecture notes based on

c 2001 Michael Carter

All rights reserved

Example Recall again the chip maker, whose optimization problem is

The rst-order and second-order conditions for prot maximization are