Economics 202N: Producer theory

Luke Stein

Stanford University

November 5, 2008

ABabcdfghiejkl
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Individual decision-making under certainty

Objects of inquiry

Our study of microeconomics begins with individual

decision-making under certainty

Items of interest include:

Feasible set
Objective function (Feasible set → R)
Choice correspondence (Parameters ⇒ Feasible set)
“Maximized” objective function (Parameters → R)

2 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Individual decision-making under certainty

Course outline

We will divide decision-making under certainty into three units:

1 Producer theory
Feasible set defined by technology
Objective function p · y depends on prices

2 Abstract choice theory

Feasible set totally general
Objective function may not even exist

3 Consumer theory
Feasible set defined by budget constraint and depends on prices
Objective function u(x)

3 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Producer theory: simplifying assumptions

Standard model: firms choose production plans (technologically

feasible lists of inputs and outputs) to maximize profits

Simplifying assumptions include:

1 Firms are price takers (both input and output markets)
2 Technology is exogenously given
3 Firms maximize profits; should be true as long as
The firm is competitive
There is no uncertainty about profits
Managers are perfectly controlled by owners

4 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Role of simplifying assumptions

No consensus about “correct” view

Modeling is an abstraction
Relies on simplifying but untrue assumptions
Highlight important effects by suppressing other effects
Basis for numerical calculations
Models can be useful in different ways
Relevant predictions reasonably accurate; can sometimes be
checked using data or theoretical analysis
Failure of relevant predictions can highlight which simplifying
assumptions are most relevant
“Usual” or “standard” models often fail realism checks; do
not skip validation

5 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Outline

1 Production sets

2 Profit maximization

3 Rationalizability

4 Rationalizability: the differentiable case

5 Single-output firms

6 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Outline

1 Production sets

2 Profit maximization

3 Rationalizability

4 Rationalizability: the differentiable case

5 Single-output firms

7 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Production sets

Exogenously given technology applies over n commodities (both

inputs and outputs)

Definition (production plan)

A vector y = (y1 , . . . , yn ) ∈ Rn where an output has yk > 0 and an
input has yk < 0.

Definition (production set)

Set Y ⊆ Rn of feasible production plans; generally assumed to be
non-empty and closed.

8 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Properties of production sets I

Definition (shutdown)
0 ∈ Y.

Definition (free disposal)

y ∈ Y and y 0 ≤ y imply y 0 ∈ Y .

=⇒

9 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Properties of production sets II

Definition (nonincreasing returns to scale)

y ∈ Y implies αy ∈ Y for all α ∈ [0, 1].

Implies shutdown

Definition (nondecreasing returns to scale)

y ∈ Y implies αy ∈ Y for all α ≥ 1.

Along with shutdown, implies π(p) = 0 or π(p) = +∞ for all p

Definition (constant returns to scale)

y ∈ Y implies αy ∈ Y for all α ≥ 0; i.e., nonincreasing and
nondecreasing returns to scale.

10 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Properties of production sets III

Definition (convex production set)

y , y 0 ∈ Y imply ty + (1 − t)y 0 ∈ Y for all t ∈ [0, 1].

Vaguely “nonincreasing returns to specialization”

If 0 ∈ Y , then convexity implies nonincreasing returns to scale

Strictly convex iff for t ∈ (0, 1), the convex combination is in the
interior of Y

11 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Characterizing Y : Transformation function I

Definition (transformation function)

Any function T : Rn → R with
1 T (y ) ≤ 0 ⇐⇒ y ∈ Y ; and
2 T (y ) = 0 ⇐⇒ y is a boundary point of Y .

Can be interpreted as the amount of technical progress required to

make y feasible

The set y : T (y ) = 0 is the production possibilities frontier
(a.k.a. transformation frontier)

12 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Characterizing Y : Transformation function II

When the transformation function is differentiable, we can define

the marginal rate of transformation of good l for good k:
Definition (marginal rate of transformation)
∂T (y )
∂yl
MRTl,k (y ) ≡ ∂T (y )
,
∂yk
∂T (y )
defined for points where T (y ) = 0 and ∂yk 6= 0.

Measures the extra amount of good k that can be obtained per

unit reduction of good l
Equals the slope of the PPF

13 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Outline

1 Production sets

2 Profit maximization

3 Rationalizability

4 Rationalizability: the differentiable case

5 Single-output firms

14 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

The Profit Maximization Problem

The firm’s optimal production decisions are given by

correspondence y : Rn ⇒ Rn

y (p) ≡ argmax p · y
y ∈Y

= y ∈ Y : p · y = π(p)

Resulting profits are given by profit function π : Rn → R ∪ {+∞}

π(p) ≡ sup p · y
y ∈Y

15 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

A note on maxima and suprema

We have a tendency to be fast and loose with these, but recall

that:
A maximum is the highest achieved value
A supremum is a least upper bound (which may or may not
be achieved)

Fact
We have not made sufficient assumptions to ensure that a
maximum profit is achieved (i.e., y (p) 6= ∅), and so the sup
cannot necessarily be replaced with a max.
In particular we allow for the possibility that π(p) = +∞, which
can happen if Y is unbounded.

16 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

A note on convex functions

Definition (convexity)
f : Rn → R is convex iff for all x and y ∈ Rn , and all λ ∈ [0, 1], we
have

λf (x) + (1 − λ)f (y ) ≥ f λx + (1 − λ)y .

In the differentiable case, also characterized by any of

If f : R → R, then f 00 (x) ≥ 0 for all x
Hessian ∇2 f (x) is a positive semidefinite matrix for all x
f (·) lies above its tangent hyperplanes; i.e.,

f (x) ≥ f (y ) + ∇f (y ) · (x − y ) for all x and y

17 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Convex function illustration

f (·)

λf (x) + (1 − λ)f (y )
f (λx + (1 − λ)y )

x λx + (1 − λ)y y

18 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Convexity of π(·)

Theorem
π(·) is a convex function.

Proof.
Fix any p1 , p2 and let pt ≡ tp1 + (1 − t)p2 for t ∈ [0, 1]. Then for
any y ∈ Y ,

pt · y = t p1 · y +(1 − t) p2 · y
| {z } | {z }
≤π(p1 ) ≤π(p2 )

≤ tπ(p1 ) + (1 − t)π(p2 ).

Since this is true for all pt · y , it holds for supy ∈Y pt · y = π(pt ):

π(pt ) ≤ tπ(p1 ) + (1 − t)π(p2 ).

19 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

A note on homogeneous functions

Definition (homogeneity)
f : Rn → R is homogeneous of degree k iff for all x ∈ Rn , and all
λ > 0, we have
f (λx) = λk f (x).

We will overwhelmingly rely on

Homogeneity of degree zero: f (λx) = f (x)
Homogeneity of degree one: f (λx) = λf (x)

20 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Euler’s Law I

Theorem (Euler’s Law)

Suppose f (·) is differentiable. Then it is homogeneous of degree k
iff p · ∇f (p) = kf (p).

Proof.
Homogeneous ⇒ p · ∇f (p) = kf (p) proved by differentiating
f (λp) = λk f (p) with respect to λ, and then setting λ = 1.
Homogeneous ⇐ p · ∇f (p) = kf (p) may be covered in section.

21 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Euler’s Law II

Corollary
If f (·) is homogeneous of degree one, then ∇f (·) is homogeneous
of degree zero.

Proof.
Homogeneity of degree one means

λf (p) = f (λp).

Differentiating in p,

λ∇f (p) = λ∇f (λp)

∇f (p) = ∇f (λp)

22 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Homogeneity of π(·)

Theorem
π(·) is homogeneous of degree one; i.e., π(λp) = λπ(p) for all p
and λ > 0.
That is, if you scale all (input and output) prices up or down the
same amount, you also scale profits by that amount

Proof.

π(λp) ≡ sup λp · y
y ∈Y

= λ sup p · y
y ∈Y

= λπ(p).

23 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Homogeneity of y (·)

Theorem
y (·) is homogeneous of degree zero; i.e., y (λp) = y (p) for all p
and λ > 0.
That is, a firm makes the same production choice if all (input and
output) prices are scaled up or down the same amount

Proof.


y (λp) ≡ y ∈ Y : λp · y = π(λp)

= y ∈ Y : λp · y = λπ(p)

= y ∈ Y : p · y = π(p)
= y (p).

24 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Outline

1 Production sets

2 Profit maximization

3 Rationalizability

4 Rationalizability: the differentiable case

5 Single-output firms

25 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Recovering the feasible set

Rationalizability asks for a given y (·) and/or π(·)—which we may

not observe everywhere—about properties of the underlying Y .

Suppose that we don’t know Y , but observe some supply decisions

ỹ (p) ⊆ y (p) and/or resulting profits π̃(p) = π(p) when it faces
price vectors p from a set P ⊆ Rn
1 What can we infer about the underlying production set Y ?
2 Is there any Y such that ỹ (p) and π(p) are consistent with
profit maximization?
3 Can we recover the entire production set if we have “enough
data”?

26 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Rationalizability: definitions

Definitions (rationalization)
Supply correspondence ỹ : P ⇒ Rn is rationalized by
production set Y iff ∀p ∈ P, ỹ (p) ⊆ argmaxy ∈Y p · y .
Profit function π̃ : P → R ∪ {+∞} is rationalized by
production set Y iff ∀p, π̃(p) = supy ∈Y p · y .

Definitions (rationalizability)
ỹ (·) or π̃(·) is rationalizable if it is rationalized by some
production set.
ỹ (·) and π̃(·) are jointly rationalizable if they are both
rationalized by the same production set.

27 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

When are π(·) and y (·) jointly rationalized by Y ? I

Question 1
What can we infer about the underlying production set Y ?

1 Production plans the firm actually chooses must be feasible

The set of chosen production plans gives an “inner bound”
[
I
Y ≡ ỹ (p)
p∈P

If ỹ (·) is rationalized by Y , we must have Y I ⊆ Y

28 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

When are π(·) and y (·) jointly rationalized by Y ? II

2 Production plans that yield higher profits than those chosen

must be infeasible
The set of production plans less profitable than π̃(p) at price p
gives an “outer bound”
O

Y ≡ y : p · y ≤ π̃(p) for all p ∈ P

≡ y : p · y ≤ p · ỹ (p) for all p ∈ P

If π̃(·) is rationalized by Y , we must have Y ⊆ Y O

29 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

When are π(·) and y (·) jointly rationalized by Y ? III

Theorem
A nonempty-valued supply correspondence ỹ (·) and profit function
π̃(·) on a price set are jointly rationalized by production set Y iff
1 p · y = π̃(p) for all y ∈ ỹ (p) (adding-up), and
2 Y I ⊆ Y ⊆ Y O.

Proof.
Rationalized by Y ⇒ conditions by construction of Y I and Y O as
argued above.
Rationalized by Y ⇐ conditions since for any price vector p, the
firm can achieve profit π̃(p) by choosing any y ∈ ỹ (p) ⊆ Y I ⊆ Y ,
but cannot achieve any higher profit since Y ⊆ Y O .

30 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

When are π(·) and y (·) jointly rationalizable?

Question 2
Which observations are rationalizable, i.e., consistent with profit
maximization for some production set?

Corollary
A nonempty-valued supply correspondence ỹ (·) and profit function
π̃(·) on a price set are jointly rationalizable iff
1 p · y = π̃(p) for all y ∈ ỹ (p) (adding-up), and
2 Y I ⊆ Y O ; i.e., p · y 0 ≤ π̃(p) for all p, p 0 , and all y 0 ∈ ỹ (p 0 )
(Weak Axiom of Profit Maximization).

31 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Fully recovering Y from π(·) and y (·) I

Question 3
Can we recover the entire production set if we have enough data?

Theorem
Suppose we observe profits π(·) for all nonnegative prices
(P = Rn+ \ {0}), and further assume
1 Y satisfies free disposal, and
2 Y is convex and closed.
Then Y = Y O .

32 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Fully recovering Y from π(·) and y (·) II

Why do we need convexity and closure of Y ?

Closure makes it “more likely” that π(p) is actually achieved
(i.e., the supremum is also the maximum)
Convexity is a bit trickier. . .
The outer bound is defined as the intersection of linear
half-spaces
O

Y ≡ y : p · y ≤ π(p) for all p ∈ P
\
= y : p · y ≤ π(p)
p∈P

Thus Y O is convex (since it is the intersection of convex sets)

33 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

A note on the Separating Hyperplane Theorem I

Theorem (Separating Hyperplane Theorem)

Suppose that S and T are two convex, closed, and disjoint
(S ∩ T = ∅) subsets of Rn . Then there exists θ ∈ Rn and c ∈ R
with θ 6= 0 such that

θ · s ≥ c for all s ∈ S and θ · t < c for all t ∈ T .

Means that a convex, closed set can be separated from any point
outside the set

SHT is one of a few key tools for proving many of our results

34 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

A note on the Separating Hyperplane Theorem II

θ·x <c
θ·x >c
T

θ·x =c

35 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

A note on the Separating Hyperplane Theorem III

We can’t necessarily separate nonconvex sets:

36 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Fully recovering Y from π(·) and y (·) reprise I

Question 3
Can we recover the entire production set if we have enough data?

Theorem
Suppose we observe profits π(·) for all nonnegative prices
(P = Rn+ \ {0}), and further assume
1 Y satisfies free disposal, and
2 Y is convex and closed.
Then Y = Y O .

37 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Fully recovering Y from π(·) and y (·) reprise II

Proof.
We know Y ⊆ Y O ; thus we only need to show that Y O ⊆ Y .
Take any x 6∈ Y . Y and {x} are closed, convex, and disjoint, so
we can apply the Separating Hyperplane Theorem: there exists
p 6= 0 such that p · x > supy ∈Y p · y = π(p).
By free disposal, if any component of p were negative, then
supy ∈Y p · y = +∞. So p > 0; i.e., p ∈ Rn+ \ {0} = P. But since
p · x > π(p), it must be that x 6∈ Y O .
We have showed that x 6∈ Y ⇒ x 6∈ Y O , or equivalently
YO ⊆ Y.

38 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Outline

1 Production sets

2 Profit maximization

3 Rationalizability

4 Rationalizability: the differentiable case

5 Single-output firms

39 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Loss function

We can also describe the feasible set using a “loss function”

Definition (loss function)
L(p, y ) ≡ π(p) − p · y . This is the loss from choosing y rather
than the profit-maximizing feasible production plan.

If L(p, y ) < 0, then p · y > π(p), and hence y must be infeasible

The outer bound can therefore be written

O

Y ≡ y : p · y ≤ π(p) for all p ∈ P

= y : inf L(p, y ) ≥ 0 ,
p∈P

i.e., the set of points at which losses are nonnegative at any price

40 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Hotelling’s Lemma I

Assume rationalizability
Consider any p 0 ∈ P, and any y 0 ∈ y (p 0 ):
y 0 ∈ Y I (by definition)
0 O

Thus y ∈ Y = y : inf p∈P L(p, y ) ≥ 0 (by WAPM)
That is, inf p∈P L(p, y 0 ) ≥ 0
But by adding-up, p 0 · y 0 = π(p 0 ), so L(p 0 , y 0 ) = 0
Thus the infimum is achieved, and equals the minimum:

min L(p, y 0 ) = L(p 0 , y 0 ) = 0 for all p 0 ∈ P and y 0 ∈ y (p 0 )

p∈P

Losses from making production choice y 0 at price p when the actual

price is p 0 must be nonnegative, and are exactly zero when p = p 0

41 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Hotelling’s Lemma II

Dual problem: loss minimization

The loss minimization problem minp∈P L(p, y 0 ) for
L(p, y ) ≡ π(p) − p · y is solved at p = p 0 whenever y 0 ∈ y (p 0 ):

min L(p, y 0 ) = L(p 0 , y 0 ) = 0.

p∈P

We can apply a first-order condition since

The set P is open, so all its points are interior
At any point at which π(·) is differentiable, so is L(·, y 0 )

42 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Hotelling’s Lemma III

This FOC is
Theorem (Hotelling’s Lemma)

∇p L(p, y ) p=p0 = 0 for all y 0 ∈ y (p 0 ).

0 


Dispensing with the loss function gives ∇π(p 0 ) = y 0 .

This can also be viewed as an application of the Envelope Theorem

to the Profit Maximization Problem: π(p) = supy ∈Y p · y
ETs relate the derivatives of the objective and value functions

43 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Implications of Hotelling’s Lemma

Recall
Theorem (Hotelling’s Lemma)
∇π(p) = y (p) wherever π(·) is differentiable.

Thus if π(·) is differentiable at p, y (p) is a singleton

We restrict ourselves to this case; we can call y (·) a supply
function rather than the more general supply correspondence
The notes include a section on the nondifferentiable case,
which we are going to skip

44 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Rationalization: y (·) and differentiable π(·) I

Theorem
y : P → Rn (the correspondence ensured to be a function by
Hotelling’s lemma, given differentiable π(·)) and differentiable
π : P → R on an open convex set P ⊆ Rn are jointly rationalizable
iff
1 p · y (p) = π(p) (adding-up),
2 ∇π(p) = y (p) (Hotelling’s Lemma), and
3 π(·) is convex.

Note that
Condition 2 describes the first-order condition and
Condition 3 describes the second-order condition
of the dual (loss minimization) problem
45 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Rationalization: y (·) and differentiable π(·) II

Proof.
We showed earlier that 2 and 3 follow from rationalizability.

It remains to be shown that 1–3 imply WAPM (i.e.,

π(p) ≥ p · y (p 0 )).
Noting that a convex function lies above its tangent hyperplanes,
and applying Hotelling’s Lemma and adding-up gives

π(p) ≥ π(p 0 ) + (p − p 0 ) · ∇π(p 0 )

= π(p 0 ) + (p − p 0 ) · y (p 0 )
= p 0 · y (p 0 ) + (p − p 0 ) · y (p 0 )
= p · y (p 0 ).

46 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Rationalization: differentiable y (·) I

Theorem
Differentiable y : P → Rn on an open convex set P ⊆ Rn is
rationalizable iff
1 y (·) is homogeneous of degree zero, and
2 The Jacobian Dy (p) is symmetric and positive semidefinite.

47 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Rationalization: differentiable y (·) II

Proof.
We showed earlier that if y (·) is rationalizable, it
1 Is homogeneous of degree zero; and
2 Satisfies Hotelling’s Lemma, thus Dy (p) = D 2 π(p) is
symmetric PSD (it is the Hessian of a convex function).
Now suppose conditions of the theorem hold. Take
π(p) = p · y (p). For each i = 1, . . . , n,

∂π(p) X ∂yj (p) X ∂yi (p)

= yi (p) + pj = yi (p) + pj = yi (p)
∂pi ∂pi ∂pj
j j
| {z }
=p·∇yi (p)=0

Thus D 2 π(p) = Dy (p) is PSD, hence π(·) is convex. Thus y (·)

and π(·) are jointly rationalizable.
48 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Rationalization: differentiable π(·)

Theorem
Differentiable π : P → R on a convex set P ⊆ Rn is rationalizable
iff
1 π(·) is homogeneous of degree one, and
2 π(·) is convex.

Proof.
We showed earlier that if π(·) is rationalizable, it is homogeneous
of degree one and convex.
Now suppose conditions of the theorem hold. Take y (p) = ∇π(p).
By Euler’s Law, π(p) = p · ∇π(p) = p · y (p). Thus y (·) and π(·)
are jointly rationalizable.

49 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Substitution matrix

Definition (substitution matrix)

The Jacobian of the optimal supply function,
 
∂y1 (p) ∂y1 (p)
  ∂p1 ... ∂pn
∂yi (p)  . .. .. 
Dy (p) ≡ ≡ .. . . .
∂pj i,j  ∂y (p) ∂yn (p)

n
∂p1 ... ∂pn

By Hotelling’s Lemma, Dy (p) = D 2 π(p), hence the

substitution matrix is symmetric
A “subtle conclusion of mathematical economics”

Convexity of π(·) implies positive semidefiniteness

Supply curves must be upward sloping (the “Law of Supply”)

50 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Outline

1 Production sets

2 Profit maximization

3 Rationalizability

4 Rationalizability: the differentiable case

5 Single-output firms

51 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

The single-output firm: notation

Notation will be a bit different for single-output firms:

p ∈ R+ : Price of output
w ∈ Rn−1
+ : Prices of inputs
q ∈ R+ : Output produced
z ∈ Rn−1
+ : Inputs used

Thus pold = (p, w ) and yold = (q, −z)

We will often label m ≡ n − 1

52 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Characterizing Y : Production function I

Definition (production function)

For a firm with only a single output q (and inputs −z), defined as
f (z) ≡ max q such that (q, −z) ∈ Y .


Y = (q, −z) : q ≤ f (z) , assuming free disposal

53 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Characterizing Y : Production function II

When the production function is differentiable, we can define the

marginal rate of technological substitution of good l for good k:
Definition (marginal rate of technological substitution)
∂f (z)
∂zl
MRTSl,k (z) ≡ ∂f (z)
,
∂zk
∂f (z)
defined for points where ∂zk 6= 0.

Measures how much of input k must be used in place of one unit

of input l to maintain the same level of output

54 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Dividing up the problem I

With one output, free disposal, and production function f (·),

m

Y = (q, −z) : z ∈ R+ and f (z) ≥ q

Given a positive output price p > 0, profit maximization requires

q = f (z), so firms solve

π(p, w ) = sup pf (z) − w · z

z∈Rm
+
| {z } | {z }
revenue cost
z(p, w ) = argmax pf (z) − w · z
z∈Rm
+

55 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Dividing up the problem II

We separate the profit maximization problem into two parts:

1 Find a cost-minimizing way to produce a given output level q

Cost function
c(q, w ) ≡ inf w ·z
z : f (z)≥q

Conditional factor demand correspondence

Z ∗ (q, w ) ≡ argmin w · z
z : f (z)≥q

= z : f (z) ≥ q and w · z = c(q, w )

2 Find an output level that maximizes difference between

revenue and cost
max pq − c(q, w )
q≥0

56 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

The Cost Minimization Problem I

Consider a restriction of Y that only includes output above some

fixed level q:

Rm

Yq ≡ (q, −z) : z ∈ + and f (z) ≥ q

The cost minimization problem is like a PMP over Yq with

πq (p, w ) ≡ qp − c(q, w )
∗
 
yq (w ) ≡ q − Z (q, w )

57 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

The Cost Minimization Problem II

Our results from the profit maximization section go through here

with appropriate sign changes; e.g.,
c(q, ·) is homogeneous of degree one in w
Z ∗ (q, ·) is homogeneous of degree zero in w
If Z ∗ (q, ·) is differentiable in w , then the matrix
Dw Z ∗ (q, w ) = Dw2 c(q, w ) is symmetric and negative
semidefinite
Rationalizability condition. . .

58 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

The Cost Minimization Problem III

Theorem
Conditional factor demand function z : R × W ⇒ Rn and
differentiable cost function c : R × W → R for a fixed output q on
an open convex set W ⊆ Rm of input prices are jointly
rationalizable iff
1 c(q, w ) = w · z(q, w ) (adding-up);
2 ∇w c(q, w ) = z(q, w ) (Shephard’s Lemma);
3 c(q, ·) is concave in w (for a fixed q).

59 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

First-order conditions: PMP I

Single-output profit maximization problem

maxm pf (z) − w · z
z∈R+ | {z } | {z }
revenue cost

where p > 0 is the output price and w ∈ Rm

+ are input prices.

Set up the Lagrangian and find Kuhn-Tucker conditions (assume

differentiability):

L(z, p, w , µ) ≡ pf (z) − w · z + µ · z

We get three (new) kinds of conditions. . .

60 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

First-order conditions: PMP II

1 FONCs: p∇f (z ∗ ) − w + µ = 0
2 Complementary slackness: µi zi∗ = 0 for all i
3 Non-negativity: µi ≥ 0 for all i
4 Original constraints: zi∗ ≥ 0 for all i

First three can be summarized as: for all i,

∂f (z ∗ )
p ≤ wi with equality if zi∗ > 0
∂zi

61 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

First-order conditions: CMP

Single-output cost minimization problem

minm w · z such that f (z) ≥ q.

z∈R+

L(z, q, w , λ, µ) ≡ − w · z + λ f (z) − q + µ · z

Applying Kuhn-Tucker here gives

∂f (z ∗ )
λ ≤ wi with equality if zi∗ > 0
∂zi

62 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

First-order conditions: Optimal Output Problem

Optimal output problem

max pq − c(q, w ).
q≥0

L(q, p, w , µ) ≡ pq − c(q, w ) + µq

Applying Kuhn-Tucker here gives

∂c(q ∗ , w )
p≤ with equality if q ∗ > 0
∂q

63 / 86
Introduction Production sets Profit maximization Rationalizability Differentiable case Single-output firms

Comparing the problems’ Kuhn-Tucker conditions

Profit Maximization Problem:

∂f (z ∗ )
p ≤ wi with equality if zi∗ > 0
∂zi
Cost Minimization Problem:
∂f (z ∗ )
λ ≤ wi with equality if zi∗ > 0
∂zi
Optimal Output Problem:

∂c(q ∗ , w )
p≤ with equality if q ∗ > 0
∂q

∂c(q ∗ ,w )
If (q ∗ , z ∗ ) > 0, then p, λ, and ∂q are all “the same”
64 / 86
Appendix

Multivariate inequalities and orthants of Euclidean space

This notation can be tricky, but is often used carefully:

1 Rn+ ≡ {x : x ≥ 0} ≡ {x : xi ≥ 0 for all i}
Includes the axes and 0

2 {x : x > 0} ≡ {x : xi ≥ 0 for all i} \ 0

Includes the axes, but not 0

3 Rn++ ≡ {x : x  0} ≡ {x : xi > 0 for all i}

Includes neither the axes nor 0

65 / 86
Appendix

Separating Hyperplane Theorem I

Theorem (Separating Hyperplane Theorem)

Suppose that S and T are two convex, closed, and disjoint
(S ∩ T = ∅) subsets of Rn . Then there exists θ ∈ Rn and c ∈ R
with θ 6= 0 such that

θ · s ≥ c for all s ∈ S and θ · t < c for all t ∈ T .

Means that a convex, closed set can be separated from any point
outside the set

66 / 86
Appendix

Separating Hyperplane Theorem II

θ·x <c
θ·x >c
T

θ·x =c

67 / 86
Appendix

Separating Hyperplane Theorem III

We can’t necessarily separate nonconvex sets:

68 / 86
Appendix

Convex functions

Definition (convexity)
f : Rn → R is convex iff for all x and y ∈ Rn , and all λ ∈ [0, 1], we
have
λf (x) + (1 − λ)f (y ) ≥ f (λx + (1 − λ)y ).

   
Also characterized by EG f (x) ≥ f EG (x) for all distributions G
In the differentiable case, also characterized by any of
If f : R → R, then f 00 (x) ≥ 0 for all x
Hessian ∇2 f (x) is a positive semidefinite matrix for all x
f (·) lies above its tangent hyperplanes:

f (x) ≥ f (y ) + ∇f (y ) · (x − y ) for all x and y

69 / 86
Appendix

Homogeneity and Euler’s Law I

Definition (homogeneity)
f : Rn → R is homogeneous of degree k iff for all x ∈ Rn , and all
λ > 0, we have
f (λx) = λk f (x).

Theorem (Euler’s Law)

Suppose f (·) is differentiable. Then it is homogeneous of degree k
iff p · ∇f (p) = kf (p).

Proof.
Homogeneous ⇒ p · ∇f (p) = kf (p) proved by differentiating
f (λp) = λk f (p) with respect to λ, and then setting λ = 1.
Homogeneous ⇐ p · ∇f (p) = kf (p) may be covered in section.
70 / 86
Appendix

Homogeneity and Euler’s Law II

Corollary
If f (·) is homogeneous of degree one, then ∇f (·) is homogeneous
of degree zero.

Proof.
Homogeneity of degree one means

λf (p) = f (λp).

Differentiating in p,

λ∇f (p) = λ∇f (λp)

∇f (p) = ∇f (λp)

71 / 86
Appendix

The Kuhn-Tucker algorithm I

The Kuhn-Tucker Theorem provide the key generalization of the
“Lagrangian” method for constrained optimization

Consider the problem

v (θ) = maxn f (x, θ)

x∈R

subject to constraints

g1 (x, θ) ≥ 0, . . . , gK (x, θ) ≥ 0.

Set up a Lagrangian
K
X
L(x, θ, λ) ≡ f (x, θ) + λk gk (x, θ)
k=1

72 / 86
Appendix

The Kuhn-Tucker algorithm II

Theorem (Kuhn-Tucker)
Suppose x ∗ solves the optimization problem at parameter θ, and
f (·, θ) and g1 (·, θ), . . . , gK (·, θ) are all differentiable in x;
the constraint set is non-empty; and
constraint qualification holds.
Then there exist nonnegative λ1 , . . . , λK such that
1 first-order conditions hold:
K
X
Dx f (x ∗ , θ) + λk Dx gk (x ∗ , θ) = Dx L(x ∗ , λ, θ) = 0;
k=1

2 λk gk (x ∗ , θ) = 0 (complementary slackness); and

3 g1 (x, θ) ≥ 0, . . . , gK (x, θ) ≥ 0 (original constraints).

73 / 86
Appendix

The Kuhn-Tucker algorithm III

Kuhn-Tucker conditions are necessary and sufficient for a solution

(assuming differentiability) as long as we have a “convex problem”:

1 The constraint set is convex

If each constraint gives a convex set, the intersection is a
convex set

The set x : gk (x, θ) ≥ 0 is convex as long as gk (·, θ) is a
quasiconcave function of x

2 The objective function is concave

If we only know the objective is quasiconcave, there are other
conditions that ensure Kuhn-Tucker is sufficient

74 / 86
Appendix

Envelope Theorem I
ETs relate objective and value functions; this one relates the
derivatives of objective and value functions for smooth PCOP:

Theorem (Envelope Theorem)

Consider a constrained optimization problem v (θ) = maxx f (x, θ)
such that g1 (x, θ) ≥ 0, . . . , gK (x, θ) ≥ 0.
Comparative statics on the value function are given by:
 K  
∂v ∂f 
 X ∂gk 
 ∂L 
= + λk =
∂θi ∂θi x ∗
 ∂θi x ∗
 ∂θi x ∗
k=1
P
(for Lagrangian L(x, θ, λ) ≡ f (x, θ) + k λk gk (x, θ)) for all θ
such that the set of binding constraints does not change in an
open neighborhood.

75 / 86
Appendix

Envelope Theorem II
The proof is given for a single constraint (but is similar for K
constraints): v (x, θ) = maxx f (x, θ) such that g (x, θ) ≥ 0
Proof.
Lagrangian L(x, θ) ≡ f (x, θ) + λg (x, θ) gives FOC
   
∂f  ∂g  ∂f  ∂g 
+λ = 0 ⇐⇒ = −λ (1)
∂x ∗ ∂x ∗ ∂x ∗ ∂x ∗

∗


where the notation ·|∗ means “evaluated at x (θ), θ for some θ.”
If g x ∗ (θ), θ = 0, take the derivative in θ of this equality

condition to get
∗ ∗
     
∂g  ∂x 
  ∂g 
 ∂g 
 ∂g  ∂x 

+ = 0 ⇐⇒ =− . (2)
∂x ∗ ∂θ θ
  ∂θ ∗
 ∂θ ∗
 ∂x ∗ ∂θ θ
 

76 / 86
Appendix

Envelope Theorem III

Proof (continued).
Note that, ∂L
∂θ = ∂f
∂θ+ λ ∂g
∂θ . Evaluating at (x ∗ (θ), θ) gives

  
∂L  ∂f  ∂g 
= +λ .
∂θ ∗ ∂θ ∗ ∂θ θ

∂L  ∂f 
 
If λ = 0, this gives that =
∂θ
∗ ∂θ ∗ ;
if λ > 0, complementary
slackness ensures g x ∗ (θ), θ = 0 so we can apply equation 2. In
either case, we get that
∗
 
∂f ∂g  ∂x 
= −λ . (3)
∂θ ∂x ∗ ∂θ θ
 

77 / 86
Appendix

Envelope Theorem IV

Proof (continued).
x ∗ (θ), θ


Applying
the chain rule to v (x, θ) = f and evaluating at
x ∗ (θ), θ gives
∗
   
∂v  ∂f  ∂x  ∂f 
= +
∂θ ∗ ∂x ∗ ∂θ θ
  ∂θ ∗
∗
   
∂g  ∂x 
  ∂f  ∂L 
= −λ + = ,
∂x ∗ ∂θ θ
  ∂θ ∗ ∂θ ∗

where the last two equalities obtain by equations 1 and 3,

respectively.

78 / 86
Appendix

The Implicit Function Theorem I

A simple, general maximization problem

X ∗ (t) = argmax F (x, t)

x∈X

where F : X × T → R and X × T ⊆ R2 .

Suppose:
1 Smoothness: F is twice continuously differentiable
2 Convex choice set: X is convex
3 00 < 0
Strictly concave objective (in choice variable): Fxx
(together with convexity of X , this ensures a unique
maximizer)
4 Interiority: x(t) is in the interior of X for all t (which means
the standard FOC must hold)

79 / 86
Appendix

The Implicit Function Theorem II

The first-order condition says the unique maximizer satisfies

Fx x(t), t = 0

Taking the derivative in t:

0 Fxt x(t), t
x (t) = −

Fxx x(t), t

Note by strict concavity, the
denominator is negative, so x 0 (t) and

00 x(t), t have the same sign
the cross-partial Fxt

80 / 86
Appendix

The Implicit Function Theorem: Higher dimensions

A more general general maximization problem

X ∗ (t) = argmaxx∈X F (x, t) where F : X × T → R and
X × T ⊆ Rn .

Under certain assumptions, we can apply a FOC:

∇x F x(t), t = 0

Taking a derivative in t we get

∂2F ∂2F



x(t), t ∂x(t) x(t), t
0= · +
∂x ∂x ∂t ∂x ∂t
"
2

−1 2
#

∂x(t) ∂ F x(t), t ∂ F x(t), t
=− ·
∂t ∂x ∂x ∂x ∂t

81 / 86
Appendix

Multivariate Topkis’ Theorem

Theorem (Topkis’ Theorem)

Suppose
1 F : X × T → R (for X a lattice, T partially ordered)

is supermodular in x (i.e., ID in all (xi , xj ))

has ID in (x, t) (i.e., ID in all (xi , tj ))
2 t 0 > t,
3 x ∈ X ∗ (t) ≡ argmaxξ∈X F (ξ, t), and x 0 ∈ X ∗ (t 0 ).

Then
(x ∧ x 0 ) ∈ X ∗ (t) and (x ∨ x 0 ) ∈ X ∗ (t 0 ).
That is, X ∗ (·) is nondecreasing in t in the stronger set order.

82 / 86
Appendix

Quasiconcavity and quasiconvexity

Definition (quasiconcavity)
f : X → R is quasiconcave iff for all x ∈ X , the upper contour set
of x 
UCS(x) ≡ ξ ∈ X : f (ξ) ≥ f (x)

if f (ξ1 ) ≥ f (x) and f (ξ2 ) ≥ f (x), then

is a convex sets; i.e.,
f λξ1 + (1 − λ)ξ2 ≥ f (x) for all λ ∈ [0, 1].
f (·) is strictly quasiconcave iff for all x ∈ X , UCS(x) is a strictly
convex set; i.e., if f (ξ1 ) ≥ f (x) and f (ξ2 ) ≥ f (x), then
λξ1 + (1 − λ)ξ2 is an interior point of UCS(x) for all λ ∈ (0, 1).

Quasiconvexity and strict quasiconvexity replace “upper contour

sets” with “lower contour sets” in the above definitions, where

LCS(x) ≡ ξ ∈ X : f (ξ) ≤ f (x)
83 / 86
Appendix

Why concavity implies quasiconcavity I

Theorem
A concave function is quasiconcave. A convex function is
quasiconvex.

Note that showing a function is quasiconcave/quasiconvex is often

harder than showing it is concave/convex

84 / 86
Appendix

Why concavity implies quasiconcavity II

Proof.
Showing that concavity implies quasiconcavity is equivalent to
showing that non-quasiconcavity implies non-concavity.
Suppose f : X → R is not quasiconcave; i.e., there exists some x
such that the upper contour set of x

UCS(x) ≡ ξ ∈ X : f (ξ) ≥ f (x)

is not a convex set.

85 / 86
Appendix

Why concavity implies quasiconcavity III

Proof (continued).
For UCS(x) to be nonconvex, there must exist some x1 ,
x2 ∈ UCS(x) and λ ∈ [0, 1] such that λx1 + (1 − λ)x2 6∈ UCS(x);
that is

f (x1 ) ≥ f (x),
f (x2 ) ≥ f (x),


f λx1 + (1 − λ)x2 < f (x).

By the above inequalities,

λf (x1 ) + (1 − λ)f (x2 ) > f λx1 + (1 − λ)x2 ,

and f (·) is therefore not concave.

86 / 86