This action might not be possible to undo. Are you sure you want to continue?
Mathematics for Economists
ChienFu CHOU
September 2006
Contents
Lecture 1 Static Economic Models and The Concept of Equilibrium 1
Lecture 2 Matrix Algebra 5
Lecture 3 Vector Space and Linear Transformation 10
Lecture 4 Determinant, Inverse Matrix, and Cramer’s rule 16
Lecture 5 Diﬀerential Calculus and Comparative Statics 25
Lecture 6 Comparative Statics – Economic applications 36
Lecture 7 Optimization 44
Lecture 8 Optimization–multivariate case 61
Lecture 9 Optimization with equality constraints and Nonlinear Programming 74
Lecture 10 General Equilibrium and Game Theory 89
1
1 Static Economic Models and The Concept of Equilibrium
Here we use three elementary examples to illustrate the general structure of an eco
nomic model.
1.1 Partial market equilibrium model
A partial market equilibrium model is constructed to explain the determination of
the price of a certain commodity. The abstract form of the model is as follows.
Q
d
= D(P; a) Q
s
= S(P; a) Q
d
= Q
s
,
Q
d
: quantity demanded of the commodity D(P; a): demand function
Q
s
: quantity supplied to the market S(P; a): supply function
P: market price of the commodity
a: a factor that aﬀects demand and supply
Equilibrium: A particular state that can be maintained.
Equilibrium conditions: Balance of forces prevailing in the model.
Substituting the demand and supply functions, we have D(P; a) = S(P; a).
For a given a, we can solve this last equation to obtain the equilibrium price P
∗
as
a function of a. Then we can study how a aﬀects the market equilibrium price by
inspecting the function.
Example: D(P; a) = a
2
/P, S(P) = 0.25P. a
2
/P
∗
= 0.25P
∗
⇒ P
∗
= 2a, Q
∗
d
= Q
∗
s
=
0.5a.
1.2 General equilibrium model
Usually, markets for diﬀerent commodities are interrelated. For example, the price
of personal computers is strongly inﬂuenced by the situation in the market of micro
processors, the price of chicken meat is related to the supply of pork, etc. Therefore,
we have to analyze interrelated markets within the same model to be able to capture
such interrelationship and the partial equilibrium model is extended to the general
equilibrium model. In microeconomics, we even attempt to include every commodity
(including money) in a general equilibrium model.
Q
d1
= D
1
(P
1
, . . . , P
n
; a)
Q
s1
= S
1
(P
1
, . . . , P
n
; a)
Q
d1
= Q
s1
Q
d2
= D
2
(P
1
, . . . , P
n
; a)
Q
s2
= S
2
(P
1
, . . . , P
n
; a)
Q
d2
= Q
s2
. . . Q
dn
= D
n
(P
1
, . . . , P
n
; a)
Q
sn
= S
n
(P
1
, . . . , P
n
; a)
Q
dn
= Q
sn
Q
di
: quantity demanded of commodity i
Q
si
: quantity supplied of commodity i
P
i
: market price of commodity i
a: a factor that aﬀects the economy
D
i
(P
1
, . . . , P
n
; a): demand function of commodity i
S
i
(P
1
, . . . , P
n
; a): supply function of commodity i
We have three variables and three equations for each commodity/market.
2
Substituting the demand and supply functions, we have
D
1
(P
1
, . . . , P
n
; a) −S
1
(P
1
, . . . , P
n
; a) ≡ E
1
(P
1
, . . . , P
n
; a) = 0
D
2
(P
1
, . . . , P
n
; a) −S
2
(P
1
, . . . , P
n
; a) ≡ E
2
(P
1
, . . . , P
n
; a) = 0
.
.
.
.
.
.
D
n
(P
1
, . . . , P
n
; a) −S
n
(P
1
, . . . , P
n
; a) ≡ E
n
(P
1
, . . . , P
n
; a) = 0.
For a given a, it is a simultaneous equation in (P
1
, . . . , P
n
). There are n equations
and n unknown. In principle, we can solve the simultaneous equation to ﬁnd the
equilibrium prices (P
∗
1
, . . . , P
∗
n
).
A 2market linear model:
D
1
= a
0
+ a
1
P
1
+ a
2
P
2
, S
1
= b
0
+ b
1
P
1
+ b
2
P
2
, D
2
= α
0
+ α
1
P
1
+ α
2
P
2
, S
2
=
β
0
+β
1
P
1
+β
2
P
2
.
(a
0
−b
0
) + (a
1
−b
1
)P
1
+ (a
2
−b
2
)P
2
= 0
(α
0
−β
0
) + (α
1
−β
1
)P
1
+ (α
2
−β
2
)P
2
= 0.
1.3 National income model
The most fundamental issue in macroeconomics is the determination of the national
income of a country.
C = a +bY (a > 0, 0 < b < 1)
I = I(r)
Y = C +I +
¯
G
S = Y −C.
C: Consumption Y : National income
I: Investment S: Savings
¯
G: government expenditure r: interest rate
a, b: coeﬃcients of the consumption function.
To solve the model, we substitute the ﬁrst two equations into the third to obtain
Y = a +bY +I
0
+
¯
G ⇒ Y
∗
= (a +I(r) +
¯
G)/(1 −b).
1.4 The ingredients of a model
We set up economic models to study economic phenomena (causeeﬀect relation
ships), or how certain economic variables aﬀect other variables. A model consists of
equations, which are relationships among variables.
Variables can be divided into three categories:
Endogenous variables: variables we choose to represent diﬀerent states of a model.
Exogenous variables: variables assumed to aﬀect the endogenous variables but are
not aﬀected by them.
Causes (Changes in exogenous var.) ⇒ Eﬀects (Changes in endogenous var.)
Parameters: Coeﬃcients of the equations.
3
End. Var. Ex. Var. Parameters
Partial equilibrium model: P, Q
d
, Q
s
a Coeﬃcients of D(P; a), S(P; a)
General equilibrium model: P
i
, Q
di
, Q
si
a
Income model: C, Y , I, S r,
¯
G a, b
Equations can be divided into three types:
Behavioral equations: representing the decisions of economic agents in the model.
Equilibrium conditions: the condition such that the state can be maintained (when
diﬀerent forces/motivations are in balance).
Deﬁnitions: to introduce new variables into the model.
Behavioral equations Equilibrium cond. Deﬁnitions
Partial equilibrium model: Q
d
= D(P; a), Q
s
= S(P; a) Q
d
= Q
s
General equilibrium model: Q
di
= D
i
(P
1
, . . . , P
n
), Q
di
= Q
si
Q
si
= S
i
(P
1
, . . . , P
n
)
Income model: C = a +bY , I = I(r) Y = C +I +
¯
G S = Y −C
1.5 The general economic model
Assume that there are n endogenous variables and m exogenous variables.
Endogenous variables: x
1
, x
2
, . . . , x
n
Exogenous variables: y
1
, y
2
, . . . , y
m
.
There should be n equations so that the model can be solved.
F
1
(x
1
, x
2
, . . . , x
n
; y
1
, y
2
, . . . , y
m
) = 0
F
2
(x
1
, x
2
, . . . , x
n
; y
1
, y
2
, . . . , y
m
) = 0
.
.
.
F
n
(x
1
, x
2
, . . . , x
n
; y
1
, y
2
, . . . , y
m
) = 0.
Some of the equations are behavioral, some are equilibrium conditions, and some are
deﬁnitions.
In principle, given the values of the exogenous variables, we solve to ﬁnd the
endogenous variables as functions of the exogenous variables:
x
1
= x
1
(y
1
, y
2
, . . . , y
m
)
x
2
= x
1
(y
1
, y
2
, . . . , y
m
)
.
.
.
x
n
= x
1
(y
1
, y
2
, . . . , y
m
).
If the equations are all linear in (x
1
, x
2
, . . . , x
n
), then we can use Cramer’s rule (to
be discussed in the next part) to solve the equations. However, if some equations are
nonlinear, it is usually very diﬃcult to solve the model. In general, we use comparative
statics method (to be discussed in part 3) to ﬁnd the diﬀerential relationships between
x
i
and y
j
:
∂x
i
∂y
j
.
4
1.6 Problems
1. Find the equilibrium solution of the following model:
Q
d
= 3 −P
2
, Q
s
= 6P −4, Q
s
= Q
d
.
2. The demand and supply functions of a twocommodity model are as follows:
Q
d1
= 18 −3P
1
+P
2
Q
d2
= 12 +P
1
−2P
2
Q
s1
= −2 + 4P
1
Q
s2
= −2 + 3P
2
Find the equilibrium of the model.
3. (The eﬀect of a sales tax) Suppose that the government imposes a sales tax of
t dollars per unit on product 1. The partial market model becomes
Q
d
1
= D(P
1
+t), Q
s
1
= S(P
1
), Q
d
1
= Q
s
1
.
Eliminating Q
d
1
and Q
s
1
, the equilibrium price is determined by D(P
1
+ t) =
S(P
1
).
(a) Identify the endogenous variables and exogenous variable(s).
(b) Let D(p) = 120−P and S(p) = 2P. Calculate P
1
and Q
1
both as function
of t.
(c) If t increases, will P
1
and Q
1
increase or decrease?
4. Let the nationalincome model be:
Y = C +I
0
+G
C = a +b(Y −T
0
) (a > 0, 0 < b < 1)
G = gY (0 < g < 1)
(a) Identify the endogenous variables.
(b) Give the economic meaning of the parameter g.
(c) Find the equilibrium national income.
(d) What restriction(s) on the parameters is needed for an economically rea
sonable solution to exist?
5. Find the equilibrium Y and C from the following:
Y = C +I
0
+G
0
, C = 25 + 6Y
1/2
, I
0
= 16, G
0
= 14.
6. In a 2good market equilibrium model, the inverse demand functions are given
by
P
1
= Q
−2
3
1
Q
1
3
2
, P
2
= Q
1
3
1
Q
−2
3
2
.
(a) Find the demand functions Q
1
= D
1
(P
1
, P
2
) and Q
2
= D
2
(P
1
, P
2
).
(b) Suppose that the supply functions are
Q
1
= a
−1
P
1
, Q
2
= P
2
.
Find the equilibrium prices (P
∗
1
, P
∗
2
) and quantities (Q
∗
1
, Q
∗
2
) as functions
of a.
5
2 Matrix Algebra
A matrix is a two dimensional rectangular array of numbers:
A ≡
¸
¸
¸
¸
a
11
a
12
. . . a
1n
a
21
a
22
. . . a
2n
.
.
.
.
.
.
.
.
.
.
.
.
a
m1
a
m2
. . . a
mn
There are n columns each with m elements or m rows each with n elements. We say
that the size of A is m×n.
If m = n, then A is a square matrix.
A m×1 matrix is called a column vector and a 1 ×n matrix is called a row vector.
A 1 ×1 matrix is just a number, called a scalar number.
2.1 Matrix operations
Equality: A = B ⇒ (1) size(A) = size(B), (2) a
ij
= b
ij
for all ij.
Addition/subtraction: A+B and A−B can be deﬁned only when size(A) = size(B),
in that case, size(A+B) = size(A−B) = size(A) = size(B) and (A+B)
ij
= a
ij
+b
ij
,
(A−B)
ij
= a
ij
−b
ij
. For example,
A =
1 2
3 4
, B =
1 0
0 1
⇒A+B =
2 2
3 5
, A−B =
0 2
3 3
.
Scalar multiplication: The multiplication of a scalar number α and a matrix A,
denoted by αA, is always deﬁned with size (αA) = size(A) and (αA)
ij
= αa
ij
. For
example, A =
1 2
3 4
, ⇒4A =
4 8
12 16
.
Multiplication of two matrices: Let size(A) = m × n and size(B) = o × p, the
multiplication of A and B, C = AB, is more complicated. (1) it is not always deﬁned.
(2) AB = BA even when both are deﬁned. The condition for AB to be meaningful
is that the number of columns of A should be equal to the number of rows of B, i.e.,
n = o. In that case, size (AB) = size (C) = m×p.
A =
¸
¸
¸
¸
a
11
a
12
. . . a
1n
a
21
a
22
. . . a
2n
.
.
.
.
.
.
.
.
.
.
.
.
a
m1
a
m2
. . . a
mn
, B =
¸
¸
¸
¸
b
11
b
12
. . . b
1p
b
21
b
22
. . . a
2p
.
.
.
.
.
.
.
.
.
.
.
.
b
n1
b
n2
. . . a
np
⇒C =
¸
¸
¸
¸
c
11
c
12
. . . c
1p
c
21
c
22
. . . c
2p
.
.
.
.
.
.
.
.
.
.
.
.
c
m1
c
m2
. . . c
mp
,
where c
ij
=
¸
n
k=1
a
ik
b
kj
.
Examples:
1 2
0 5
3
4
=
3 + 8
0 + 20
=
11
20
,
1 2
3 4
5 6
7 8
=
5 + 14 6 + 16
15 + 28 18 + 32
=
19 22
43 50
,
6
5 6
7 8
1 2
3 4
=
5 + 18 10 + 24
7 + 24 14 + 32
=
23 34
31 46
.
Notice that
1 2
3 4
5 6
7 8
=
5 6
7 8
1 2
3 4
.
2.2 Matrix representation of a linear simultaneous equation system
A linear simultaneous equation system:
a
11
x
1
+ . . . + a
1n
x
n
= b
1
.
.
.
.
.
.
a
n1
x
1
+ . . . + a
nn
x
n
= b
n
Deﬁne A ≡
¸
¸
a
11
. . . a
1n
.
.
.
.
.
.
.
.
.
a
n1
. . . a
nn
, x ≡
¸
¸
x
1
.
.
.
x
n
and b ≡
¸
¸
b
1
.
.
.
b
n
. Then the equation
Ax = b is equivalent to the simultaneous equation system.
Linear 2market model:
E
1
= (a
1
−b
1
)p
1
+ (a
2
−b
2
)p
2
+ (a
0
−b
0
) = 0
E
2
= (α
1
−β
1
)p
1
+ (α
2
−β
2
)p
2
+ (α
0
−β
0
) = 0
⇒
a
1
−b
1
a
2
−b
2
α
1
−β
1
α
2
−β
2
p
1
p
2
+
a
0
−b
0
α
0
−β
0
=
0
0
.
Income determination model:
C = a +bY
I = I(r)
Y = C +I
⇒
¸
1 0 −b
0 1 0
1 1 −1
¸
C
I
Y
=
¸
a
I(r)
0
.
In the algebra of real numbers, the solution to the equation ax = b is x = a
−1
b.
In matrix algebra, we wish to deﬁne a concept of A
−1
for a n × n matrix A so that
x = A
−1
b is the solution to the equation Ax = b.
2.3 Commutative, association, and distributive laws
The notations for some important sets are given by the following table.
N = nature numbers 1, 2, 3, . . . I = integers . . . , −2, −1, 0, 1, 2, . . .
Q = rational numbers
m
n
R = real numbers
R
n
= ndimensional column vectors M(m, n) = m×n matrices
M(n) = n ×n matrices
A binary operation is a law of composition of two elements from a set to form a
third element of the same set. For example, + and × are binary operations of real
numbers R.
Another important example: addition and multiplication are binary operations of ma
trices.
7
Commutative law of + and × in R: a +b = b +a and a ×b = b ×a for all a, b ∈ R.
Association law of + and × in R: (a+b) +c = a+(b +c) and (a×b) ×c = a×(b ×c)
for all a, b, c ∈ R.
Distributive law of + and × in R: (a+b)×c = a×c+b×c and c×(a+b) = c×a+c×b
for all a, b, c ∈ R.
The addition of matrices satisﬁes both commutative and associative laws: A + B =
B + A and (A + B) + C = A + (B + C) for all A, B, C ∈ M(m, n). The proof is
trivial.
In an example, we already showed that the matrix multiplication does not satisfy the
commutative law AB = BA even when both are meaningful.
Nevertheless the matrix multiplication satisﬁes the associative law (AB)C = A(BC)
when the sizes are such that the multiplications are meaningful. However, this de
serves a proof!
It is also true that matrix addition and multiplication satisfy the distributive law:
(A + B)C = AC + BC and C(A + B) = CA + CB. You should try to prove these
statements as exercises.
2.4 Special matrices
In the space of real numbers, 0 and 1 are very special. 0 is the unit element of + and
1 is the unit element of ×: 0+a = a+0 = a, 0×a = a×0 = 0, and 1×a = a×1 = a.
In matrix algebra, we deﬁne zero matrices and identity matrices as
O
m,n
≡
¸
¸
0 . . . 0
.
.
.
.
.
.
.
.
.
0 . . . 0
I
n
≡
¸
¸
¸
¸
1 0 . . . 0
0 1 . . . 0
.
.
.
.
.
.
.
.
.
.
.
.
0 0 . . . 1
.
Clearly, O+A = A+O = A, OA = AO = O, and IA = AI = A. In the multiplication
of real numbers if a, b = 0 then a×b = 0. However,
1 0
0 0
0 0
0 1
=
0 0
0 0
=
O
2,2
.
Idempotent matrix: If AA = A (A must be square), then A is an idempotent matrix.
Both O
n,n
and I
n
are idempotent. Another example is A =
0.5 0.5
0.5 0.5
.
Transpose of a matrix: For a matrix A with size m × n, we deﬁne its transpose A
′
as a matrix with size n × m such that the ijth element of A
′
is equal to the jith
element of A, a
′
ij
= a
ji
.
A =
1 2 3
4 5 6
then A
′
=
¸
1 4
2 5
3 6
.
Properties of matrix transposition:
(1) (A
′
)
′
= A, (2) (A+B)
′
= A
′
+B
′
, (3) (AB)
′
= B
′
A
′
.
Symmetrical matrix: If A = A
′
(A must be square), then A is symmetrical. The
8
condition for A to be symmetrical is that a
ij
= a
ji
. Both O
n,n
and I
n
are symmetrical.
Another example is A =
1 2
2 3
.
Projection matrix: A symmetrical idempotent matrix is a projection matrix.
Diagonal matrix: A symmetrical matrix A is diagonal if a
ij
= 0 for all i = j. Both
I
n
and O
n,n
are diagonal. Another example is A =
¸
λ
1
0 0
0 λ
2
0
0 0 λ
3
2.5 Inverse of a square matrix
We are going to deﬁne the inverse of a square matrix A ∈ M(n).
Scalar: aa
−1
= a
−1
a = 1 ⇒ if b satisﬁes ab = ba = 1 then b = a
−1
.
Deﬁnition of A
−1
: If there exists a B ∈ M(n) such that AB = BA = I
n
, then we
deﬁne A
−1
= B.
Examples: (1) Since II = I, I
−1
= I. (2) O
n,n
B = O
n,n
⇒ O
−1
n,n
does not exist. (3)
If A =
a
1
0
0 a
2
, a
1
, a
2
= 0, then A
−1
=
a
−1
1
0
0 a
−1
2
. (4) If a
1
= 0 or a
2
= 0,
then A
−1
does not exist.
Singular matrix: A square matrix whose inverse matrix does not exist.
Nonsingular matrix: A is nonsingular if A
−1
exists.
Properties of matrix inversion:
Let A, B ∈ M(n), (1) (A
−1
)
−1
= A, (2) (AB)
−1
= B
−1
A
−1
, (3) (A
′
)
−1
= (A
−1
)
′
.
2.6 Problems
1. Let A = I −X(X
′
X)
−1
X
′
.
(a) If the dimension of X is m×n, what must be the dimension of I and A.
(b) Show that matrix A is idempotent.
2. Let A and B be n ×n matrices and I be the identity matrix.
(a) (A+B)
3
= ?
(b) (A +I)
3
= ?
3. Let B =
0.5 0.5
0.5 0.5
, U = (1, 1)
′
, V = (1, −1)
′
, and W = aU +bV , where
a and b are real numbers. Find BU, BV , and BW. Is B idempotent?
4. Suppose A is a n ×n nonsingular matrix and P is a n ×n idempotent matrix.
Show that APA
−1
is idempotent.
5. Suppose that A and B are n ×n symmetric idempotent matrices and AB = B.
Show that A−B is idempotent.
6. Calculate (x
1
, x
2
)
3 2
2 5
x
1
x
2
.
9
7. Let I =
1 0
0 1
and J =
0 1
−1 0
.
(a) Show that J
2
= −I.
(b) Make use of the above result to calculate J
3
, J
4
, and J
−1
.
(c) Show that (aI +bJ)(cI +dJ) = (ac −bd)I + (ad +bc)J.
(d) Show that (aI + bJ)
−1
=
1
a
2
+b
2
(aI − bJ) and [(cos θ)I + (sin θ)J]
−1
=
(cos θ)I −(sin θ)J.
10
3 Vector Space and Linear Transformation
In the last section, we regard a matrix simply as an array of numbers. Now we are
going to provide some geometrical meanings to a matrix.
(1) A matrix as a collection of column (row) vectors
(2) A matrix as a linear transformation from a vector space to another vector space
3.1 Vector space, linear combination, and linear independence
Each point in the mdimensional Euclidean space can be represented as a mdimensional
column vector v =
¸
¸
v
1
.
.
.
v
m
, where each v
i
represents the ith coordinate. Two points
in the mdimensional Euclidean space can be added according to the rule of matrix
addition. A point can be multiplied by a scalar according to the rule of scalar multi
plication.
Vector addition:
¸
¸
v
1
.
.
.
v
m
+
¸
¸
w
1
.
.
.
w
m
=
¸
¸
v
1
+w
1
.
.
.
v
m
+w
m
.
Scalar multiplication: α
¸
¸
v
1
.
.
.
v
m
=
¸
¸
αv
1
.
.
.
αv
m
.
x
1
`
x
2
v
1
´
´
´
´
´
´´´
´
´
´
´
´
´´´
v
2
v
1
+v
2
x
1
`
x
2
v
2v
With such a structure, we say that the mdimensional Euclidean space is a vec
tor space.
11
mdimensional column vector space: R
m
=
¸
¸
v
1
.
.
.
v
m
, v
i
∈ R
.
We use superscripts to represent individual vectors.
A m×n matrix: a collection of n mdimensional column vectors:
¸
¸
¸
¸
a
11
a
12
. . . a
1n
a
21
a
22
. . . a
2n
.
.
.
.
.
.
.
.
.
.
.
.
a
m1
a
m2
. . . a
mn
=
¸
¸
¸
¸
a
11
a
21
.
.
.
a
m1
,
¸
¸
¸
¸
a
12
a
22
.
.
.
a
m2
, . . . ,
¸
¸
¸
¸
a
1n
a
2n
.
.
.
a
mn
Linear combination of a collection of vectors {v
1
, . . . , v
n
}: w =
n
¸
i=1
α
i
v
i
, where
(α
1
, . . . , α
n
) = (0, . . . , 0).
Linear dependence of {v
1
, . . . , v
n
}: If one of the vectors is a linear combination of
others, then the collection is said to be linear dependent. Alternatively, the collection
is linearly dependent if (0, . . . , 0) is a linear combination of it.
Linear independence of {v
1
, . . . , v
n
}: If the collection is not linear dependent, then
it is linear independent.
Example 1: v
1
=
¸
a
1
0
0
, v
2
=
¸
0
a
2
0
, v
3
=
¸
0
0
a
3
, a
1
a
2
a
3
= 0.
If α
1
v
1
+α
2
v
2
+α
3
v
3
= 0 then (α
1
, α
2
, α
3
) = (0, 0, 0). Therefore, {v
1
, v
2
, v
3
} must be
linear independent.
Example 2: v
1
=
¸
1
2
3
, v
2
=
¸
4
5
6
, v
3
=
¸
7
8
9
.
2v
2
= v
1
+v
3
. Therefore, {v
1
, v
2
, v
3
} is linear dependent.
Example 3: v
1
=
¸
1
2
3
, v
2
=
¸
4
5
6
.
α
1
v
1
+ α
2
v
2
=
¸
α
1
+ 4α
2
2α
1
+ 5α
2
3α
1
+ 6α
2
=
¸
0
0
0
⇒ α
1
= α
2
= 0. Therefore, {v
1
, v
2
} is
linear independent.
Span of {v
1
, . . . , v
n
}: The space of linear combinations.
If a vector is a linear combination of other vectors, then it can be removed without
changing the span.
12
Rank
¸
¸
v
11
. . . v
1n
.
.
.
.
.
.
.
.
.
v
m1
. . . v
mn
≡ Dimension(Span{v
1
, . . . , v
n
}) = Maximum # of indepen
dent vectors.
3.2 Linear transformation
Consider a m × n matrix A. Given x ∈ R
n
, Ax ∈ R
m
. Therefore, we can deﬁne a
linear transformation from R
n
to R
m
as f(x) = Ax or
f : R
n
→R
m
, f(x) =
¸
¸
y
1
.
.
.
y
m
=
¸
¸
a
11
. . . a
1n
.
.
.
.
.
.
.
.
.
a
m1
. . . a
mn
¸
¸
x
1
.
.
.
x
n
.
It is linear because f(αx +βw) = A(αx +βw) = αAx +βAw = αf(x) +βf(w).
Standard basis vectors of R
n
: e
1
≡
¸
¸
¸
¸
1
0
.
.
.
0
, e
2
≡
¸
¸
¸
¸
0
1
.
.
.
0
, . . . , e
n
≡
¸
¸
¸
¸
0
0
.
.
.
1
.
Let v
i
be the ith column of A, v
i
=
¸
¸
a
1i
.
.
.
a
mi
.
v
i
= f(e
i
):
¸
a
11
. . .
a
m1
=
¸
¸
a
11
. . . a
1n
.
.
.
.
.
.
.
.
.
a
m1
. . . a
mn
¸
¸
1
.
.
.
0
⇒v
1
= f(e
1
) = Ae
1
, etc.
Therefore, v
i
is the image of the ith standard basis vector e
i
under f.
Span{v
1
, . . . , v
n
} = Range space of f(x) = Ax ≡ R(A).
Rank(A) ≡ dim(R(A)).
Null space of f(x) = Ax: N(A) ≡ {x ∈ R
n
, f(x) = Ax = 0}.
dim(R(A)) + dim(N(A)) = n.
Example 1: A =
1 0
0 2
. N(A) =
0
0
, R(A) = R
2
, Rank(A) = 2.
Example 2: B =
1 1
1 1
. N(B) =
k
−k
, k ∈ R
, R(B) =
k
k
, k ∈ R
,
Rank(B) = 1.
The multiplication of two matrices can be interpreted as the composition of two
linear transformations.
f : R
n
→R
m
, f(x) = Ax, g : R
p
→R
n
, g(y) = By, ⇒f(g(x)) = A(By), f◦g : R
p
→R
m
.
13
The composition is meaningful only when the dimension of the range space of g(y)
is equal to the dimension of the domain of f(x), which is the same condition for the
validity of the matrix multiplication.
Every linear transformation f : R
n
→R
m
can be represented by f(x) = Ax for
some m×n matrix.
3.3 Inverse transformation and inverse of a square matrix
Consider now the special case of square matrices. Each A ∈ M(n) represents a linear
transformation f : R
n
→R
n
.
The deﬁnition of the inverse matrix A
−1
is such that AA
−1
= A
−1
A = I. If we re
gard A as a linear transformation from R
n
→R
n
and I as the identity transformation
that maps every vector (point) into itself, then A
−1
is the inverse mapping of A. If
dim(N(A)) = 0, then f(x) is one to one.
If dim(R(A)) = n, then R(A) = R
n
and f(x) is onto.
⇒if Rank(A) =n, then f(x) is one to one and onto and there exits an inverse mapping
f
−1
: R
n
→R
n
represented by a n×n square matrix A
−1
. f
−1
f(x) = x ⇒A
−1
Ax = x.
⇒ if Rank(A) = n, then A is nonsingular.
if Rank(A) < n, then f(x) is not onto, no inverse mapping exists, and A is singular.
Examples: Rank
¸
a
1
0 0
0 a
2
0
0 0 a
3
= 3 and Rank
¸
1 4 7
2 5 8
3 6 9
= 2.
Remark: O
n,n
represents the mapping that maps every point to the origin. I
n
rep
resents the identity mapping that maps a point to itself. A projection matrix repre
sents a mapping that projects points onto a linear subspace of R
n
, eg.,
0.5 0.5
0.5 0.5
projects points onto the 45 degree line.
x
1
`
x
2
´
x
Ax
x =
1
2
Ax =
.5 .5
.5 .5
1
2
=
1.5
1.5
x
′
=
k
−k
, Ax
′
=
0
0
14
3.4 Problems
1. Let B =
¸
0 1 0
0 0 1
0 0 0
and T
B
the corresponding linear transformation T
B
: R
3
→ R
3
, T
B
(x) = Bx,
where x =
¸
x
1
x
2
x
3
∈ R
3
.
(a) Is v
1
=
¸
a
0
0
, a = 0, in the null space of T
B
? Why or why not?
(b) Is v
2
=
¸
0
0
b
, b = 0, in the range space of T
B
? Why or why not? How
about v
3
=
¸
c
d
0
?
(c) Find Rank(B).
2. Let A be an idempotent matrix.
(a) Show that I −A is also idempotent.
(b) Suppose that x = 0 is in the null space of A, i.e., Ax = 0. Show that x
must be in the range space of I −A, i.e., show that there exists a vector y
such that (I −A)y = x. (Hint: Try y = x.)
(c) Suppose that y is in the range space of A. Show that y must be in the null
space of I −A.
(d) Suppose that A is n × n and Rank[A] = n − k, n > k > 0. What is the
rank of I −A?
3. Let I =
¸
1 0 0
0 1 0
0 0 1
, A =
¸
1
3
1
3
1
3
1
3
1
3
1
3
1
3
1
3
1
3
, x =
¸
1
a
b
, y =
¸
1
α
β
, and
B = I −A.
(a) Calculate AA and BB.
(b) If y is in the range space of A, what are the values of α and β?
(c) What is the dimension of the range space of A?
(d) Determine the rank of A.
(e) Suppose now that x is in the null space of B. What should be the values
of a and b?
(f) What is the dimension of the null space of B?
15
(g) Determine the rank of B?
4. Let A =
1/5 2/5
2/5 4/5
and B =
¸
1 1 1
0 1 1
0 0 1
.
(a) Determine the ranks of A and B.
(b) Determine the null space and range space of each of A and B and explain
why.
(c) Determine whether they are idempotent.
16
4 Determinant, Inverse Matrix, and Cramer’s rule
In this section we are going to derive a general method to calculate the inverse of
a square matrix. First, we deﬁne the determinant of a square matrix. Using the
properties of determinants, we ﬁnd a procedure to compute the inverse matrix. Then
we derive a general procedure to solve a simultaneous equation.
4.1 Permutation group
A permutation of {1, 2, . . . , n} is a 11 mapping of {1, 2, . . . , n} onto itself, written as
π =
1 2 . . . n
i
1
i
2
. . . i
n
meaning that 1 is mapped to i
1
, 2 is mapped to i
2
, . . ., and
n is mapped to i
n
. We also write π = (i
1
, i
2
, . . . , ı
n
) when no confusing.
Permutation set of {1, 2, . . . , n}: P
n
≡ {π = (i
1
, i
2
, . . . , i
n
) : π is a permutation}.
P
2
= {(1, 2), (2, 1)}.
P
3
= {(1, 2, 3), (1, 3, 2), (2, 1, 3), (2, 3, 1), (3, 1, 2), (3, 2, 1)}.
P
4
: 4! = 24 permutations.
Inversions in a permutation π = (i
1
, i
2
, . . . , i
n
): If there exist k and l such that
k < l and i
k
> i
l
, then we say that an inversion occurs.
N(i
1
, i
2
, . . . , ı
n
): Total number of inversions in (i
1
, i
2
, . . . , i
n
).
Examples: 1. N(1, 2) = 0, N(2, 1) = 1.
2. N(1, 2, 3) = 0, N(1, 3, 2) = 1, N(2, 1, 3) = 1,
N(2, 3, 1) = 2, N(3, 1, 2) = 2, N(3, 2, 1) = 3.
4.2 Determinant
Determinant of A =
¸
¸
¸
¸
a
11
a
12
. . . a
1n
a
21
a
22
. . . a
2n
.
.
.
.
.
.
.
.
.
.
.
.
a
n1
a
n2
. . . a
nn
:
A ≡
¸
(i
1
,i
2
,...,in)∈Pn
(−1)
N(i
1
,i
2
,...,in)
a
1i
1
a
2i
2
. . . a
nin
.
n = 2:
a
11
a
12
a
21
a
22
= (−1)
N(1,2)
a
11
a
22
+ (−1)
N(2,1)
a
12
a
21
= a
11
a
22
−a
12
a
21
.
n = 3:
a
11
a
12
a
13
a
21
a
22
a
23
a
31
a
32
a
33
=
(−1)
N(1,2,3)
a
11
a
22
a
33
+ (−1)
N(1,3,2)
a
11
a
23
a
32
+ (−1)
N(2,1,3)
a
12
a
21
a
33
+
(−1)
N(2,3,1)
a
12
a
23
a
31
+ (−1)
N(3,1,2)
a
13
a
21
a
32
+ (−1)
N(3,2,1)
a
13
a
22
a
31
= a
11
a
22
a
33
−a
11
a
23
a
32
−a
12
a
21
a
33
+a
12
a
23
a
31
+a
13
a
21
a
32
−a
13
a
22
a
31
.
17
a
11
a
12
a
21
a
22
n = 2:
a
11
a
12
a
13
a
11
a
12
a
21
a
22
a
23
a
21
a
22
a
31
a
32
a
33
a
31
a
32
n = 3:
4.3 Properties of determinant
Property 1: A
′
 = A.
Proof: Each term of A
′
 corresponds to a term of A of the same sign.
By property 1, we can replace “column vectors” in the properties below by “row vec
tors”.
Since a n × n matrix can be regarded as n column vectors A = {v
1
, v
2
, . . . , v
n
}, we
can regard determinants as a function of n column vectors A = D(v
1
, v
2
, . . . , v
n
),
D : R
n×n
→R.
By property 1, we can replace “column vectors” in the properties below by “row vec
tors”.
Property 2: If two column vectors are interchanged, the determinant changes sign.
Proof: Each term of the new determinant corresponds to a term of A of opposite
sign because the number of inversion increases or decreases by 1.
Example:
1 2
3 4
= 1 ×4 −2 ×3 = −2,
2 1
4 3
= 2 ×3 −1 ×4 = 2,
Property 3: If two column vectors are identical, then the determinant is 0.
Proof: By property 2, the determinant is equal to the negative of itself, which is
possible only when the determinant is 0.
Property 4: If you add a linear combination of other column vectors to a column
vector, the determinant does not change.
Proof: Given other column vectors, the determinant function is a linear function of
v
i
: D(αv
i
+βw
i
; other vectors ) = αD(v
i
; other vectors ) +βD(w
i
; other vectors ).
Example:
1 + 5 ×2 2
3 + 5 ×4 4
=
1 2
3 4
+
5 ×2 2
5 ×4 4
=
1 2
3 4
+ 5 ×
2 2
4 4
=
1 2
3 4
+ 5 ×0.
Submatrix: We denote by A
ij
the submatrix of A obtained by deleting the ith
row and jth column from A.
Minors: The determinant A
ij
 is called the minor of the element a
ij
.
Cofactors: C
ij
≡ (−1)
i+j
A
ij
 is called the cofactor of a
ij
.
18
Property 5 (Laplace theorem): Given i =
¯
i, A =
n
¸
j=1
a¯
ij
C¯
ij
.
Given j =
¯
j, A =
¸
n
i=1
a
i
¯
j
C
i
¯
j
.
Proof: In the deﬁnition of the determinant of A, all terms with a
ij
can be put to
gethere to become a
ij
C
ij
.
Example:
1 2 3
4 5 6
7 8 0
= 1 ×
5 6
8 0
−2 ×
4 6
7 0
+ 3 ×
4 5
7 8
.
Property 6: Given i
′
=
¯
i,
n
¸
j=1
a
i
′
j
C¯
ij
= 0.
Given j
′
=
¯
j, =
¸
n
i=1
a
ij
′ C
i
¯
j
= 0.
Therefore, if you multiply cofactors by the elements from a diﬀerent row or column,
you get 0 instead of the determinant.
Proof: The sum becomes the determinant of a matrix with two identical rows (columns).
Example: 0 = 4 ×
5 6
8 0
−5 ×
4 6
7 0
+ 6 ×
4 5
7 8
.
4.4 Computation of the inverse matrix
Using properties 5 and 6, we can calculate the inverse of A as follows.
1. Cofactor matrix: C ≡
¸
¸
¸
¸
C
11
C
12
. . . C
1n
C
21
C
22
. . . C
2n
.
.
.
.
.
.
.
.
.
.
.
.
C
n1
C
n2
. . . C
nn
.
2. Adjoint of A: Adj A ≡ C
′
=
¸
¸
¸
¸
C
11
C
21
. . . C
n1
C
12
C
22
. . . C
n2
.
.
.
.
.
.
.
.
.
.
.
.
C
1n
C
2n
. . . C
nn
.
3. ⇒AC
′
= C
′
A =
¸
¸
¸
¸
A 0 . . . 0
0 A . . . 0
.
.
.
.
.
.
.
.
.
.
.
.
0 0 . . . A
⇒ if A = 0 then
1
A
C
′
= A
−1
.
Example 1: A =
a
11
a
12
a
21
a
22
then C =
a
22
−a
21
−a
12
a
11
.
A
−1
=
1
A
C
′
=
1
a
11
a
22
−a
12
a
21
a
22
−a
12
−a
21
a
11
; if A = a
11
a
22
−a
12
a
21
= 0.
Example 2: A =
¸
1 2 3
4 5 6
7 8 0
⇒A = 27 = 0 and
C
11
=
5 6
8 0
= −48, C
12
= −
4 6
7 0
= 42, C
13
=
4 5
7 8
= −3,
19
C
21
= −
2 3
8 0
= 24, C
22
=
1 3
7 0
= −21, C
23
= −
1 2
7 8
= 6,
C
31
=
2 3
5 6
= −3, C
32
= −
1 3
4 6
= 6, C
33
=
1 2
4 5
= −3,
C =
¸
−48 42 −3
24 −21 6
−3 6 −3
, C
′
=
¸
−48 24 −3
42 −21 6
−3 6 −3
, A
−1
=
1
27
¸
−48 24 −3
42 −21 6
−3 6 −3
.
If A = 0, then A is singular and A
−1
does not exist. The reason is that A =
0 ⇒AC
′
= 0
n×n
⇒C
11
¸
¸
a
11
.
.
.
a
n1
+· · · +C
n1
¸
¸
a
1n
.
.
.
a
nn
=
¸
¸
0
.
.
.
0
. The column vectors
of A are linear dependent, the linear transformation T
A
is not onto and therefore an
inverse transformation does not exist.
4.5 Cramer’s rule
If A = 0 then A is nonsingular and A
−1
=
C
′
A
. The solution to the simultaneous
equation Ax = b is x = A
−1
b =
C
′
b
A
.
Cramer’s rule: x
i
=
¸
n
j=1
C
j
b
j
A
=
A
i

A
, where A
i
is a matrix obtained by replacing
the ith column of A by b, A
i
= {v
1
, . . . , v
i−1
, b, v
i+1
, . . . , v
n
}.
4.6 Economic applications
Linear 2market model:
a
1
−b
1
a
2
−b
2
α
1
−β
1
α
2
−β
2
p
1
p
2
=
b
0
−a
0
β
0
−α
0
.
p
1
=
b
0
−a
0
a
2
−b
2
β
0
−α
0
α
2
−β
2
a
1
−b
1
a
2
−b
2
α
1
−β
1
α
2
−β
2
, p
2
=
a
1
−b
1
b
0
−a
0
α
1
−β
1
β
0
−α
0
a
1
−b
1
a
2
−b
2
α
1
−β
1
α
2
−β
2
.
Income determination model:
¸
1 0 −b
0 1 0
1 1 −1
¸
C
I
Y
=
¸
a
I(r)
0
.
C =
a 0 −b
I(r) 1 0
0 1 −1
1 0 −b
0 1 0
1 1 −1
, I =
1 a −b
0 I(r) 0
1 0 −1
1 0 −b
0 1 0
1 1 −1
, Y =
1 0 a
0 1 I(r)
1 1 0
1 0 −b
0 1 0
1 1 −1
.
20
ISLM model: In the income determination model, we regard interest rate as given
and consider only the product market. Now we enlarge the model to include the
money market and regard interest rate as the price (an endogenous variable) deter
mined in the money market.
good market: money market:
C = a +bY L = kY −lR
I = I
0
−iR M =
¯
M
C +I +
¯
G = Y M = L
end. var: C, I, Y , R (interest rate), L(demand for money), M (money supply)
ex. var:
¯
G,
¯
M (quantity of money). parameters: a, b, i, k, l.
Substitute into equilibrium conditions:
good market: money market endogenous variables:
a +bY +I
0
−iR +
¯
G = Y , kY −lR =
¯
M, Y , R
1 −b i
k −l
Y
R
=
a +I
0
+
¯
G
¯
M
Y =
a +I
0
+
¯
G i
¯
M −l
1 −b i
k −l
, R =
1 −b a +I
0
+
¯
G
k
¯
M
1 −b i
k −l
.
Twocountry income determination model: Another extension of the income
determination model is to consider the interaction between domestic country and the
rest of the world (foreign country).
domestic good market: foreign good market: endogenous variables:
C = a +bY C
′
= a
′
+b
′
Y
′
C, I, Y ,
I = I
0
I
′
= I
′
0
M (import),
M = M
0
+mY M
′
= M
′
0
+m
′
Y
′
X (export),
C +I +X −M = Y C
′
+I
′
+X
′
−M
′
= Y
′
C
′
, I
′
, Y
′
, M
′
, X
′
.
By deﬁnition, X = M
′
and X
′
= M. Substituting into the equilibrium conditions,
(1 −b +m)Y −m
′
Y
′
= a +I
0
+M
′
0
−M
0
(1 −b
′
+m
′
)Y
′
−mY = a
′
+I
′
0
+M
0
−M
′
0
.
1 −b +m −m
′
−m 1 −b
′
+m
′
Y
Y
′
=
a +I
0
+M
′
0
−M
0
a
′
+I
′
0
+M
0
−M
′
0
.
Y =
a +I
0
+M
′
0
−M
0
−m
′
a
′
+I
′
0
+M
0
−M
′
0
1 −b
′
+m
′
1 −b +m −m
′
−m 1 −b
′
+m
′
Y
′
=
1 −b +m a +I
0
+M
′
0
−M
0
−m a
′
+I
′
0
+M
0
−M
′
0
1 −b +m −m
′
−m 1 −b
′
+m
′
.
4.7 Inputoutput table
Assumption: Technologies are all ﬁxed proportional, that is, to produce one unit of
product X
i
, you need a
ji
units of X
j
.
21
IO table: A =
¸
¸
¸
¸
a
11
a
12
. . . a
1n
a
21
a
22
. . . a
2n
.
.
.
.
.
.
.
.
.
.
.
.
a
n1
a
n2
. . . a
nn
.
Column i represents the coeﬃcients of inputs needed to produce one unit of X
i
.
Suppose we want to produce a list of outputs x =
¸
¸
¸
¸
x
1
x
2
.
.
.
x
n
, we will need a list of inputs
Ax =
¸
¸
¸
¸
a
11
x
1
+a
12
x
2
+. . . +a
1n
x
n
a
21
x
2
+a
22
x
2
+. . . +a
2n
x
n
.
.
.
a
n1
x
1
+a
n2
x
2
+. . . +a
nn
x
n
. The net output is x −Ax = (I −A)x.
If we want to produce a net amount of d =
¸
¸
¸
¸
d
1
d
2
.
.
.
d
n
, then since d = (I − A)x,
x = (I −A)
−1
d.
4.8 A geometric interpretation of determinants
Because of properties 2 and 4, the determinant function D(v
1
, . . . , v
n
) is called an
alternative linear nform of R
n
. It is equal to the volume of the parallelepiped formed
by the vectors {v
1
, . . . , v
n
}. For n = 2, A is the area of the parallelogram formed by
a
11
a
12
,
a
21
a
22
. See the diagram:
x
1
`
x
2
/
/
/
/
/
/
/
/
/ / `
v
2
v
1
A = Area of D
D
/
/
/
/
/
/
/
/
/ /
If the determinant is 0, then the volume is 0 and the vectors are linearly dependent,
one of them must be a linear combination of others. Therefore, an inverse mapping
does not exist, A
−1
does not exist, and A is singular.
22
4.9 Rank of a matrix and solutions of Ax = d when A = 0
Rank(A) = the maximum # of independent vectors in A = {v
1
, . . . , v
n
} = dim(Range
Space of T
A
).
Rank(A) = the size of the largest nonsingular square submatrices of A.
Examples: Rank
1 2
3 4
= 2. Rank
¸
1 2 3
4 5 6
7 8 9
= 2 because
1 2
4 5
is non
singular.
Property 1: Rank(AB) ≤ min{Rank(A), Rank(B)}.
Property 2: dim(Null Space of T
A
) + dim(Range Space of T
A
) = n.
Consider the simultaneous equation Ax = d. When A = 0, there exists a row of
A that is a linear combination of other rows
and Rank(A) < n. First, form the augmented matrix M ≡ [A
.
.
.d] and calculate
the rank of M. There are two cases.
Case 1: Rank(M) = Rank (A).
In this case, some equations are linear combinations of others (the equations are de
pendent) and can be removed without changing the solution space. There will be
more variables than equations after removing these equations. Hence, there will be
inﬁnite number of solutions.
Example:
1 2
2 4
x
1
x
2
=
3
6
. Rank(A) = Rank
1 2
2 4
= 1 = Rank(M) =
Rank
1 2 3
2 4 6
.
The second equation is just twice the ﬁrst equation and can be discarded. The so
lutions are
x
1
x
2
=
3 −2k
k
for any k. On x
1
x
2
space, the two equations are
represented by the same line and every point on the line is a solution.
Case 2: Rank(M) = Rank(A) + 1.
In this case, there exists an equation whose LHS is a linear combination of the LHS
of other equations but whose RHS is diﬀerent from the same linear combination of
the RHS of other equations. Therefore, the equation system is contraditory and there
will be no solutions.
Example:
1 2
2 4
x
1
x
2
=
3
7
. Rank(A) = Rank
1 2
2 4
= 1 < Rank(M) =
Rank
1 2 3
2 4 7
= 2.
Multiplying the ﬁrst equation by 2, 2x
1
+ 4x
2
= 6, whereas the second equation
says 2x
1
+ 4x
2
= 7. Therefore, it is impossible to have any
x
1
x
2
satisfying both
equations simultaneously. On x
1
x
2
space, the two equations are represented by two
parallel lines and cannot have any intersection points.
23
4.10 Problems
1. Suppose v
1
= (1, 2, 3)
′
, v
2
= (2, 3, 4)
′
, and v
3
= (3, 4, 5)
′
. Is {v
1
, v
2
, v
3
} linearly
independent? Why or why not?
2.. Find the inverse of A =
¸
6 5
8 7
.
3. Given the 3 × 3 matrix A =
2 1 6
5 3 4
8 9 7
¸
¸
,
(a) calculate the cofactors C
11
, C
21
, C
31
,
(b) use Laplace expansion theorem to ﬁnd A,
(c) and use Cramer’s rule to ﬁnd X
1
of the following equation system:
2 1 6
5 3 4
8 9 7
¸
¸
X
1
X
2
X
3
¸
¸
=
1
2
3
¸
¸
.
(Hint: Make use of the results of (a).)
4. Use Cramer’s rule to solve the nationalincome model
C = a +b(Y −T) (1)
T = −t
0
+t
1
Y (2)
Y = C +I
0
+G (3)
5. Let A =
¸
0 1 0
0 0 1
0 0 0
.
(a) Find AA and AAA.
(b) Let x = (1, 2, 3)
′
, compute Ax, AAx, and AAAx.
(c) Find Rank[A], Rank[AA], and Rank[AAA].
6. Let X =
¸
1 −1
1 0
1 1
.
(a) Find X
′
X and (X
′
X)
−1
.
(b) Compute X(X
′
X)
−1
X
′
and I −X(X
′
X)
−1
X
′
.
(c) Find Rank[X(X
′
X)
−1
X
′
] and Rank[I −X(X
′
X)
−1
X
′
].
7. A =
¸
1 2
3 6
, B =
¸
1 2 1
3 6 1
, and C =
¸
1 2 4
3 6 12
.
(a) Find the ranks of A, B, and C.
(b) Use the results of (a) to determine whether the following system has any solution:
¸
1 2
3 6
¸
X
1
X
2
=
¸
1
1
.
(c) Do the same for the following system:
¸
1 2
3 6
¸
X
1
X
2
=
¸
4
12
.
24
8. Let A =
3 2
1 2
, I the 2 ×2 identity matrix, and λ a scalar number.
(a) Find A−λI. (Hint: It is a quadratic function of λ.)
(b) Determine Rank(A − I) and Rank(A − 4I). (Remark: λ = 1 and λ = 4 are the
eigenvalues of A, that is, they are the roots of the equation A − λI = 0, called the
characteristic equation of A.)
(c) Solve the simultaneous equation system (A − I)x = 0 assuming that x
1
= 1.
(Remark: The solution is called an eigenvector of A associated with the eigenvalue
λ = 1.)
(d) Solve the simultaneous equation system (A−4I)y = 0 assuming that y
1
= 1.
(e) Determine whether the solutions x and y are linearly independent.
25
5 Diﬀerential Calculus and Comparative Statics
As seen in the last chapter, a linear economic model can be represented by a matrix
equation Ax = d(y) and solved using Cramer’s rule, x = A
−1
d(y). On the other hand,
a closed form solution x = x(y) for a nonlinear economic model is, in most applica
tions, impossible to obtain. For general nonlinear economic models, we use diﬀerential
calculus (implicit function theorem) to obtain the derivatives of endogenous variables
with respect to exogenous variables
∂x
i
∂y
j
:
f
1
(x
1
, . . . , x
n
; y
1
, . . . , y
m
) = 0
.
.
.
f
n
(x
1
, . . . , x
n
; y
1
, . . . , y
m
) = 0
⇒
¸
¸
¸
¸
¸
∂f
1
∂x
1
. . .
∂f
1
∂x
n
.
.
.
.
.
.
.
.
.
∂f
n
∂x
1
. . .
∂f
n
∂x
n
¸
¸
¸
¸
¸
∂x
1
∂y
1
. . .
∂x
1
∂y
m
.
.
.
.
.
.
.
.
.
∂x
n
∂y
1
. . .
∂x
n
∂y
m
= −
¸
¸
¸
¸
¸
∂f
1
∂y
1
. . .
∂f
1
∂y
m
.
.
.
.
.
.
.
.
.
∂f
n
∂y
1
. . .
∂f
n
∂y
m
.
⇒
¸
¸
¸
¸
¸
∂x
1
∂y
1
. . .
∂x
1
∂y
m
.
.
.
.
.
.
.
.
.
∂x
n
∂y
1
. . .
∂x
n
∂y
m
= −
¸
¸
¸
¸
¸
∂f
1
∂x
1
. . .
∂f
1
∂x
n
.
.
.
.
.
.
.
.
.
∂f
n
∂x
1
. . .
∂f
n
∂x
n
−1
¸
¸
¸
¸
¸
∂f
1
∂y
1
. . .
∂f
1
∂y
m
.
.
.
.
.
.
.
.
.
∂f
n
∂y
1
. . .
∂f
n
∂y
m
.
Each
∂x
i
∂y
j
represents a causeeﬀect relationship. If
∂x
i
∂y
j
> 0 (< 0), then x
i
will increase
(decrease) when y
j
increases. Therefore, instead of computing x
i
= x
i
(y), we want to
determine the sign of
∂x
i
∂y
j
for each ij pair. In the following, we will explain how it
works.
5.1 Diﬀerential Calculus
x = f(y) ⇒f
′
(y
∗
) =
dx
dy
y=y
∗
≡ lim
∆y→0
f(y
∗
+ ∆y) −f(y
∗
)
∆y
.
On yx space, x = f(y) is represented by a curve and f
′
(y
∗
) represents the slope of
the tangent line of the curve at the point (y, x) = (y
∗
, f(y
∗
)).
Basic rules:
1. x = f(y) = k,
dx
dy
= f
′
(y) = 0.
2. x = f(y) = y
n
,
dx
dy
= f
′
(y) = ny
n−1
.
3. x = cf(y),
dx
dy
= cf
′
(y).
26
4. x = f(y) +g(y),
dx
dy
= f
′
(y) +g
′
(y).
5. x = f(y)g(y),
dx
dy
= f
′
(y)g(y) +f(y)g
′
(y).
6. x = f(y)/g(y),
dx
dy
=
f
′
(y)g(y) −f(y)g
′
(y)
(g(y))
2
.
7. x = e
ay
,
dx
dy
= ae
ay
. x = ln y,
dx
dy
=
1
y
.
8. x = sin y,
dx
dy
= cos y. x = cos y,
dx
dy
= −sin y.
Higher order derivatives:
f
′′
(y) ≡
d
dy
d
dy
f(y)
=
d
2
dy
2
f(y), f
′′′
(y) ≡
d
dy
d
2
dy
2
f(y)
=
d
3
dy
3
f(y).
5.2 Partial derivatives
In many cases, x is a function of several y’s: x = f(y
1
, y
2
, . . . , y
n
). The partial
derivative of x with respect to y
i
evaluated at (y
1
, y
2
, . . . , y
n
) = (y
∗
1
, y
∗
2
, . . . , y
∗
n
) is
∂x
∂y
i
(y
∗
1
,y
∗
2
,...,y
∗
n
)
≡ lim
∆y
i
→0
f(y
∗
1
, . . . , y
∗
i
+ ∆y
i
, . . . , y
∗
n
) −f(y
∗
1
, . . . , y
∗
i
, . . . , y
∗
n
)
∆y
i
,
that is, we regard all other independent variables as constant (f as a function of y
i
only) and take derivative.
9.
∂x
n
1
x
m
2
∂x
1
= nx
n−1
1
x
m
2
.
Higher order derivatives: We can deﬁne higher order derivatives as before. For the
case with two independent variables, there are 4 second order derivatives:
∂
∂y
1
∂x
∂y
1
=
∂
2
x
∂y
2
1
,
∂
∂y
2
∂x
∂y
1
=
∂
2
x
∂y
2
∂y
1
,
∂
∂y
1
∂x
∂y
2
=
∂
2
x
∂y
1
∂y
2
,
∂
∂y
2
∂x
∂y
2
=
∂
2
x
∂y
2
2
.
Notations: f
1
, f
2
, f
11
, f
12
, f
21
, f
22
.
∇f ≡
¸
¸
f
1
.
.
.
f
n
: Gradient vector of f.
H(f) ≡
¸
¸
f
11
. . . f
1n
.
.
.
.
.
.
.
.
.
f
n1
. . . f
nn
: second order derivative matrix, called Hessian of f.
Equality of crossderivatives: If f is twice continously diﬀerentiable, then f
ij
= f
ji
and H(f) is symmetric.
5.3 Economic concepts similar to derivatives
Elasticity of X
i
w.r.t. Y
j
: E
X
i
,Y
j
≡
Y
j
X
i
∂X
i
∂Y
j
, the percentage change of X
i
when Y
j
increases by 1 %. Example: Q
d
= D(P), E
Q
d
,P
=
P
Q
d
dQ
d
dP
27
Basic rules: 1. E
X
1
X
2
,Y
= E
X
1
,Y
+E
X
2
,Y
, 2. E
X
1
/X
2
,Y
= E
X
1
,Y
−E
X
2
,Y
,
3. E
Y,X
= 1/E
X,Y
.
Growth rate of X = X(t): G
X
≡
1
X
dX
dt
, the percentage change of X per unit of
time.
5.4 Mean value and Taylor’s Theorems
Continuity theorem: If f(y) is continuous on the interval [a, b] and f(a) ≤ 0, f(b) ≥ 0,
then there exists a c ∈ [a, b] such that f(c) = 0.
Rolle’s theorem: If f(y) is continuous on the interval [a, b] and f(a) = f(b) = 0, then
there exists a c ∈ (a, b) such that f
′
(c) = 0.
Mean value theorem: If f(y) is continously diﬀerentiable on [a, b], then there exists a
c ∈ (a, b) such that
f(b) −f(a) = f
′
(c)(b −a) or
f(b) −f(a)
b −a
= f
′
(c).
Taylor’s Theorem: If f(y) is k + 1 times continously diﬀerentiable on [a, b], then for
each y ∈ [a, b], there exists a c ∈ (a, y) such that
f(y) = f(a)+f
′
(a)(y −a)+
f
′′
(a)
2!
(y −a)
2
+. . . +
f
(k)
(a)
k!
(y −a)
k
+
f
(k+1)
(c)
(k + 1)!
(y −a)
k+1
.
5.5 Concepts of diﬀerentials and applications
Let x = f(y). Deﬁne ∆x ≡ f(y + ∆y) −f(y), called the ﬁnite diﬀerence of x.
Finite quotient:
∆x
∆y
=
f(y + ∆y) −f(y)
∆y
⇒∆x =
∆x
∆y
∆y.
dx, dy: Inﬁnitesimal changes of x and y, dx, dy > 0 (so that we can divid something
by dx or by dy) but dx, dy < a for any positive real number a (so that ∆y→dy).
Diﬀerential of x = f(y): dx = df = f
′
(y)dy.
Chain rule: x = f(y), y = g(z) ⇒x = f(g(z)),
dx = f
′
(y)dy, dy = g
′
(z)dz ⇒dx = f
′
(y)g
′
(z)dz.
dx
dz
= f
′
(y)g
′
(z) = f
′
(g(z))g
′
(z).
Example: x = (z
2
+ 1)
3
⇒x = y
3
, y = z
2
+ 1 ⇒
dx
dz
= 3y
2
2z = 6z(z
2
+ 1)
2
.
Inverse function rule: x = f(y), ⇒y = f
−1
(x) ≡ g(x),
dx = f
′
(y)dy, dy = g
′
(x)dx ⇒dx = f
′
(y)g
′
(x)dx.
dy
dx
= g
′
(x) =
1
f
′
(y)
.
Example: x = ln y ⇒y = e
x
⇒
dx
dy
=
1
e
x
=
1
y
.
28
5.6 Concepts of total diﬀerentials and applications
Let x = f(y
1
, y
2
). Deﬁne ∆x ≡ f(y
1
+ ∆y
1
, y
2
+ ∆y
2
) − f(y
1
, y
2
), called the ﬁnite
diﬀerence of x.
∆x = f(y
1
+ ∆y
1
, y
2
+ ∆y
2
) −f(y
1
, y
2
)
= f(y
1
+ ∆y
1
, y
2
+ ∆y
2
) −f(y
1
, y
2
+ ∆y
2
) +f(y
1
, y
2
+ ∆y
2
) −f(y
1
, y
2
)
=
f(y
1
+ ∆y
1
, y
2
+ ∆y
2
) −f(y
1
, y
2
+ ∆y
2
)
∆y
1
∆y
1
+
f(y
1
, y
2
+ ∆y
2
) −f(y
1
, y
2
)
∆y
2
∆y
2
dx = f
1
(y
1
, y
2
)dy
1
+f
2
(y
1
, y
2
)dy
2
.
dx, dy =
¸
¸
dy
1
.
.
.
dy
n
: Inﬁnitesimal changes of x (endogenous), y
1
, . . . , y
n
(exogenous).
Total diﬀerential of x = f(y
1
, . . . , y
n
):
dx = df = f
1
(y
1
, . . . , y
n
)dy
1
+. . .+f
n
(y
1
, . . . , y
n
)dy
n
= (f
1
, . . . , f
n
)
¸
¸
dy
1
.
.
.
dy
n
= (∇f)
′
dy.
Implicit function rule:
In many cases, the relationship between two variables are deﬁned implicitly. For
example, the indiﬀerence curve U(x
1
, x
2
) =
¯
U deﬁnes a relationship between x
1
and
x
2
. To ﬁnd the slope of the curve
dx
2
dx
1
, we use implicit function rule.
dU = U
1
(x
1
, x
2
)dx
1
+U
2
(x
1
, x
2
)dx
2
= d
¯
U = 0 ⇒;
dx
2
dx
1
= −
U
1
(x
1
, x
2
)
U
2
(x
1
, x
2
)
.
Example: U(x
1
, x
2
) = 3x
1
3
1
+ 3x
1
3
2
= 6 deﬁnes an indiﬀerence curve passing through
the point (x
1
, x
2
) = (1, 1). The slope (Marginal Rate of Substitution) at (1, 1) can be
calculated using implicit function rule.
dx
2
dx
1
= −
U
1
U
2
= −
x
−
2
3
1
x
−
2
3
2
= −
1
1
= −1.
Multivariate chain rule:
x = f(y
1
, y
2
), y
1
= g
1
(z
1
, z
2
), y
2
= g
2
(z
1
, z
2
), ⇒x = f(g
1
(z
1
, z
2
), g
2
(z
1
, z
2
)) ≡ H(z
1
, z
2
).
We can use the total diﬀerentials dx, dy
1
, dy
2
to ﬁnd the derivative
∂x
∂z
1
.
dx = (f
1
, f
2
)
dy
1
dy
2
= (f
1
, f
2
)
g
1
1
g
1
2
g
2
1
g
2
2
dz
1
dz
2
= (f
1
g
1
1
+f
2
g
2
1
, f
1
g
1
2
+f
2
g
2
2
)
dz
1
dz
2
.
29
⇒
∂x
∂z
1
=
∂H
∂z
1
= f
1
g
1
1
+f
2
g
2
1
,
∂x
∂z
2
=
∂H
∂z
2
= f
1
g
1
2
+f
2
g
2
2
.
Example: x = y
6
1
y
7
2
, y
1
= 2z
1
+ 3z
2
, y
2
= 4z
1
+ 5z
2
,
∂x
∂z
1
= 6y
5
1
y
7
2
(2) + 7y
6
1
y
6
2
(4).
Total derivative:
x = f(y
1
, y
2
), y
2
= h(y
1
), x = f(y
1
, h(y
1
)) ≡ g(y
1
),
⇒dx = f
1
dy
1
+f
2
dy
2
= f
1
dy
1
+f
2
h
′
dy
1
= (f
1
+f
2
h
′
)dy
1
.
Total derivative:
dx
dy
1
y
2
=h(y
1
)
= f
1
+f
2
h
′
.
Partial derivative (direct eﬀect of y
1
on x):
∂x
∂y
1
=
∂f
∂y
1
= f
1
(y
1
, y
2
).
Indirect eﬀect through y
2
:
∂x
∂y
2
dy
2
dy
1
= f
2
h
′
.
Example: Given the utility function U(x
1
, x
2
) = 3x
1
3
1
+ 3x
1
3
2
, the MRS at a point
(x
1
, x
2
) is m(x
1
, x
2
) =
dx
2
dx
1
= −
U
1
(x
1
, x
2
)
U
2
(x
1
, x
2
)
= −
x
−
2
3
1
x
−
2
3
2
. The rate of change of MRS
w.r.t. x
1
along the indiﬀerence curve passing through (1, 1) is a total derivative
dm
dx
1
3x
1/3
1
+3x
1/3
2
=6
=
d
2
x
2
dx
2
1
3x
1/3
1
+3x
1/3
2
=6
=
∂m
∂x
1
+
∂m
∂x
2
dx
2
dx
1
=
∂m
∂x
1
+
∂m
∂x
2
−
x
−
2
3
1
x
−
2
3
2
.
5.7 Inverse function theorem
In Lecture 3, we discussed a linear mapping x = Ay and its inverse mapping y = A
−1
x
when A = 0.
x
1
x
2
=
a
11
a
12
a
21
a
22
y
1
y
2
y
1
y
2
=
a
22
A
−a
12
A
−a
21
A
a
11
A
x
1
x
2
.
Therefore, for a linear mapping with A = 0, an 11 inverse mapping exists and the
partial derivatives are given by the inverse matrix of A. For example, ∂x
1
/∂y
1
= a
11
where∂y
1
/∂x
1
=
a
22
A
etc. The idea can be generalized to nonlinear mappings.
A general nonlinear mapping from R
n
to R
n
, y =
¸
¸
y
1
.
.
.
y
n
→ x =
¸
¸
x
1
.
.
.
x
n
, is
represented by a vector function
x =
¸
¸
x
1
.
.
.
x
n
=
¸
¸
f
1
(y
1
, . . . , y
n
)
.
.
.
f
n
(y
1
, . . . , y
n
)
≡ F(y).
30
Jacobian matrix: J
F
(y) ≡
¸
¸
¸
¸
¸
∂x
1
∂y
1
. . .
∂x
1
∂y
n
.
.
.
.
.
.
.
.
.
∂x
n
∂y
1
. . .
∂x
n
∂y
n
=
¸
¸
f
1
1
. . . f
1
n
.
.
.
.
.
.
.
.
.
f
n
1
. . . f
n
n
.
Jacobian:
∂(x
1
, . . . , x
n
)
∂(y
1
, . . . , y
n
)
≡ J
F
(y).
Inverse function theorem: If x
∗
= F(y
∗
) and J
F
(y
∗
) = 0 (J
F
(y
∗
) is nonsingular),
then F(y) is invertible nearby x
∗
,
that is, there exists a function G(x) ≡
¸
¸
g
1
(x
1
, . . . , x
n
)
.
.
.
g
n
(x
1
, . . . , x
n
)
such that y = G(x) if
x = F(y). In that case, J
G
(x
∗
) = (J
F
(y
∗
))
−1
.
Reasoning:
¸
¸
dx
1
.
.
.
dx
n
=
¸
¸
f
1
1
. . . f
1
n
.
.
.
.
.
.
.
.
.
f
n
1
. . . f
n
n
¸
¸
dy
1
.
.
.
dy
n
⇒
¸
¸
dy
1
.
.
.
dy
n
=
¸
¸
g
1
1
. . . g
1
n
.
.
.
.
.
.
.
.
.
g
n
1
. . . g
n
n
¸
¸
dx
1
.
.
.
dx
n
=
¸
¸
f
1
1
. . . f
1
n
.
.
.
.
.
.
.
.
.
f
n
1
. . . f
n
n
−1
¸
¸
dx
1
.
.
.
dx
n
Example:
x
1
x
2
= F(r, θ) =
r cos θ
r sin θ
. J
F
(r, θ) =
cos θ −r sin θ
sin θ r cos θ
.
J = J
F
 = r(cos
2
θ + sin
2
θ) = r > 0, ⇒ r =
x
2
1
+x
2
2
, θ = tan
−1
x
2
x
1
and
J
G
= (J
F
)
−1
. When r = 0, J = 0 and the mapping is degenerate, i.e., the whole set
{r = 0, −π ≤ θ < π} is mapped to the origin (0, 0), just like the case in Lecture 3
when the Null space is a line.
Notice that g
1
1
= 1/(f
1
1
) in general.
5.8 Implicit function theorem and comparative statics
Linear model: If all the equations are linear, the model can be represtned in matrix
form as
Ax+By = c ⇔
¸
¸
a
11
· · · a
1n
.
.
.
.
.
.
.
.
.
a
n1
· · · a
nn
¸
¸
x
1
.
.
.
x
n
+
¸
¸
b
11
· · · b
1m
.
.
.
.
.
.
.
.
.
b
n1
· · · a
nm
¸
¸
y
1
.
.
.
y
m
−
¸
¸
c
1
.
.
.
c
n
=
¸
¸
0
.
.
.
0
.
If A = 0, then the solution is given by x = −A
−1
(By + c). The derivative matrix
[∂x
i
/∂y
j
]
ij
= A
−1
B. Using total diﬀerentials of the equations, we can derive a similar
derivative matrix for general nonlinear cases.
31
We can regard the LHS of a nonlinear economic model as a mapping from R
n+m
to R
n
:
¸
¸
f
1
(x
1
, . . . , x
n
; y
1
, . . . , y
m
) = 0
.
.
.
f
n
(x
1
, . . . , x
n
; y
1
, . . . , y
m
) = 0
⇔ F(x; y) = 0.
Jacobian matrix: J
x
≡
¸
¸
f
1
1
. . . f
1
n
.
.
.
.
.
.
.
.
.
f
n
1
. . . f
n
n
.
Implicit function theorem: If F(x
∗
; y
∗
) = 0 and J
x
(x
∗
; y
∗
) = 0 (J
x
(x
∗
; y
∗
) is
nonsingular), then F(x; y) = 0 is solvable nearby (x
∗
; y
∗
), that is, there exists a
function x =
¸
¸
x
1
.
.
.
x
n
= x(y) =
¸
¸
x
1
(y
1
, . . . , y
m
)
.
.
.
x
n
(y
1
, . . . , y
m
)
such that x
∗
= x(y
∗
) and
F(x(y); y) = 0. In that case,
¸
¸
¸
¸
¸
∂x
1
∂y
1
. . .
∂x
1
∂y
m
.
.
.
.
.
.
.
.
.
∂x
n
∂y
1
. . .
∂x
n
∂y
m
= −
¸
¸
f
1
1
. . . f
1
n
.
.
.
.
.
.
.
.
.
f
n
1
. . . f
n
n
−1
¸
¸
¸
¸
¸
∂f
1
∂y
1
. . .
∂f
1
∂y
m
.
.
.
.
.
.
.
.
.
∂f
n
∂y
1
. . .
∂f
n
∂y
m
.
Reasoning:
¸
¸
df
1
.
.
.
df
n
=
¸
¸
0
.
.
.
0
⇒
¸
¸
f
1
1
. . . f
1
n
.
.
.
.
.
.
.
.
.
f
n
1
. . . f
n
n
¸
¸
dx
1
.
.
.
dx
n
+
¸
¸
¸
¸
¸
∂f
1
∂y
1
. . .
∂f
1
∂y
m
.
.
.
.
.
.
.
.
.
∂f
n
∂y
1
. . .
∂f
n
∂y
m
¸
¸
dy
1
.
.
.
dy
m
= 0
⇒
¸
¸
dx
1
.
.
.
dx
n
= −
¸
¸
f
1
1
. . . f
1
n
.
.
.
.
.
.
.
.
.
f
n
1
. . . f
n
n
−1
¸
¸
¸
¸
¸
∂f
1
∂y
1
. . .
∂f
1
∂y
m
.
.
.
.
.
.
.
.
.
∂f
n
∂y
1
. . .
∂f
n
∂y
m
¸
¸
dy
1
.
.
.
dy
m
.
Example: f
1
= x
2
1
x
2
− y = 0, f
2
= 2x
1
− x
2
− 1 = 0, When y = 1, (x
1
, x
2
) = (1, 1)
is an equilibrium. To calculate
dx
1
dy
and
dx
2
dy
at the equilibrium we use the implicit
function theorem:
¸
¸
dx
1
dy
dx
2
dy
= −
f
1
1
f
1
2
f
2
1
f
2
2
−1
¸
¸
¸
∂f
1
∂y
∂f
2
∂y
= −
2x
1
x
2
x
2
1
2 −1
−1
−1
0
= −
2 1
2 −1
−1
−1
0
=
1/4
1/2
.
32
5.9 Problems
1. Given the demand function Q
d
= (100/P) −10, ﬁnd the demand elasticity η.
2. Given Y = X
2
1
X
2
+ 2X
1
X
2
2
, ﬁnd ∂Y/∂X
1
, ∂
2
Y/∂X
2
1
, and ∂
2
Y/∂X
1
∂X
2
, and
the total diﬀerential DY .
3. Given Y = F(X
1
, X
2
)+f(X
1
)+g(X
2
), ﬁnd ∂Y/∂X
1
, ∂
2
Y/∂X
2
1
, and ∂
2
Y/∂X
1
∂X
2
.
4. Given the consumption function C = C(Y −T(Y )), ﬁnd dC/dY .
5. Given that Q = D(q ∗ e/P), ﬁnd dQ/dP.
6. Y = X
2
1
X
2
, Z = Y
2
+ 2Y −2, use chain rule to derive ∂Z/∂X
1
and ∂Z/∂X
2
.
7. Y
1
= X
1
+2X
2
, Y
2
= 2X
1
+X
2
, and Z = Y
1
Y
2
, use chain rule to derive ∂Z/∂X
1
and ∂Z/∂X
2
.
8. Let U(X
1
, X
2
) = X
1
X
2
2
+ X
2
1
X
2
and X
2
= 2X
1
+ 1, ﬁnd the partial derivative
∂U/∂X
1
and the total derivative dU/dX
1
.
9. X
2
+Y
3
= 1, use implicit function rule to ﬁnd dY/dX.
10. X
2
1
+ 2X
2
2
+Y
2
= 1, use implicit function rule to derive ∂Y/∂X
1
and ∂Y/∂X
2
.
11. F(Y
1
, Y
2
, X) = Y
1
−Y
2
+X −1 = 0 and G(Y
1
, Y
2
, X) = Y
2
1
+Y
2
2
+X
2
−1 = 0.
use implicit function theorem to derive dY
1
/dX and dY
2
/dX.
12. In a Cournot quantity competition duopoly model with heterogeneous products,
the demand functions are given by
Q
1
= a −P
1
−cP
2
, Q
2
= a −cP
1
−P
2
; 1 ≥ c > 0.
(a) For what value of c can we invert the demand functions to obtain P
1
and
P
2
as functions of Q
1
and Q
2
?
(b) Calculate the inverse demand functions P
1
= P
1
(Q
1
, Q
2
) and P
2
= P
2
(Q
1
, Q
2
).
(c) Derive the total revenue functions TR
1
(Q
1
, Q
2
) = P
1
(Q
1
, Q
2
)Q
1
and TR
2
(Q
1
, Q
2
) =
P
2
(Q
1
, Q
2
)Q
2
.
13. In a 2good market equilibrium model, the inverse demand functions are given
by
P
1
= A
1
Q
α−1
1
Q
β
2
, P
2
= A
2
Q
α
1
Q
β−1
2
; α, β > 0.
(a) Calculate the Jacobian matrix
∂P
1
/∂Q
1
∂P
1
/∂Q
2
∂P
2
/∂Q
1
∂P
2
/∂Q
2
and Jacobian
∂(P
1
, P
2
)
∂(Q
1
, Q
2
)
.
What condition(s) should the parameters satisfy so that we can invert the
functions to obtain the demand functions?
(b) Derive the Jacobian matrix of the derivatives of (Q
1
, Q
2
) with respect to
(P
1
, P
2
),
∂Q
1
/∂P
1
∂Q
1
/∂P
2
∂Q
2
/∂P
1
∂Q
2
/∂P
2
.
33
5.10 Proofs of important theorems of diﬀerentiation
Rolle’s theorem: If f(x) ∈ C[a, b], f
′
(x) exists for all x ∈ (a, b), and f(a) = f(b) =
0, then there exists a c ∈ (a, b) such that f
′
(c) = 0.
Proof:
Case 1: f(x) ≡ 0 ∀x ∈ [a, b] ⇒f
′
(x) = 0 ∀x ∈ (a, b)./
Case 2: f(x) ≡ 0 ∈ [a, b] ⇒ ∃e, c such that f(e) = m ≤ f(x) ≤ M = f(c) and
M > m. Assume that M = 0 (otherwise m = 0 and the proof is similar). It is easy
to see that f
′
−
(c) ≥ 0 and f
′
+
(c) ≤ 0. Therefore, f
′
(c) = 0. Q.E.D.
Mean Value theorem: If f(x) ∈ C[a, b] and f
′
(x) exists for all x ∈ (a, b). Then
there exists a c ∈ (a, b) such that
f(b) −f(a) = f
′
(c)(b −a).
Proof:
Consider the function
φ(x) ≡ f(x) −
¸
f(a) +
f(b) −f(a)
b −a
(x −a)
.
It is clear that φ(x) ∈ C[a, b] and φ
′
(x) exists for all x ∈ (a, b). Also, φ(a) = φ(b) = 0
so that the conditions of Rolle’s Theorem are satisﬁed for φ(x). Hence, there exists
a c ∈ (a, b) such that φ
′
(c) = 0, or
φ
′
(c) = f
′
(c) −
f(b) −f(a)
b −a
= 0 ⇒f
′
(c) =
f(b) −f(a)
b −a
= 0.
Q.E.D.
Taylor’s Theorem: If f(x) ∈ C
r
[a, b] and f
(r+1)
(x) exists for all x ∈ (a, b). Then
there exists a c ∈ (a, b) such that
f(b) = f(a)+f
′
(a)(b−a)+
1
2
f
′′
(a)(b−a)
2
+. . .+
1
r!
f
(r)
(a)(b−a)
r
+
1
(r + 1)!
f
(r+1)
(c)(b−a)
r+1
.
Proof:
Deﬁne ξ ∈ R
(b −a)
r+1
(r + 1)!
ξ ≡ f(b)−
¸
f(a) +f
′
(a)(b −a) +
1
2
f
′′
(a)(b −a)
2
+. . . +
1
r!
f
(r)
(a)(b −a)
r
.
Consider the function
φ(x) ≡ f(b) −
¸
f(x) +f
′
(x)(b −x) +
1
2
f
′′
(x)(b −x)
2
+. . . +
1
r!
f
(r)
(x)(b −x)
r
+
ξ
(r + 1)!
(b −x)
r+1
.
34
It is clear that φ(x) ∈ C[a, b] and φ
′
(x) exists for all x ∈ (a, b). Also, φ(a) = φ(b) = 0
so that the conditions of Rolle’s Theorem are satisﬁed for φ(x). Hence, there exists
a c ∈ (a, b) such that φ
′
(c) = 0, or
φ
′
(c) =
ξ −f
(r+1)
(c)
r!
= 0 ⇒f
(r+1)
(c) = ξ.
Q.E.D.
Inverse Function Theorem: Let E ⊆ R
n
be an open set. Suppose f : E → R
n
is C
1
(E), a ∈ E, f(a) = b, and A = J(f(a)), A = 0. Then there exist open sets
U, V ⊂ R
n
such that a ∈ U, b ∈ V , f is one to one on U, f(U) = V , and f
−1
: V →U
is C
1
(U).
Proof:
(1. Find U.) Choose λ ≡ A/2. Since f ∈ C
1
(E), there exists a neighborhood U ⊆ E
with a ∈ U such that J(f(x)) −A < λ.
(2. Show that f(x) is one to one in U.) For each y ∈ R
n
deﬁne φ
y
on E by
φ
y
(x) ≡ x + A
−1
(y − f(x)). Notice that f(x) = y if and only if x is a ﬁxed point of
φ
y
. Since J(φ
y
(x)) = I − A
−1
J(f(x)) = A
−1
[A − J(f(x))] ⇒ J(φ
y
(x)) <
1
2
on U.
Therefore φ
y
(x) is a contraction mapping and there exists at most one ﬁxed point in
U. Therefore, f is one to one in U.
(3. V = f(U) is open so that f
−1
is continuous.) Let V = f(U) and y
0
= f(x
0
) ∈ V
for x
0
∈ U. Choose an open ball B about x
0
with radius ρ such that the closure
[B] ⊆ U. To prove that V is open, it is enough to show that y ∈ V whenever
y −y
0
< λρ. So ﬁx y such that y −y
0
< λρ. With φ
y
deﬁned above,
φ
y
(x
0
) −x
0
= A
−1
(y −y
0
) < A
−1
λρ =
ρ
2
.
If x ∈ [B] ⊆ U, then
φ
y
(x) −x
0
≤ φ
y
(x) −φ
y
(x
0
) +φ
y
(x
0
) −x
0
<
1
2
x −x
0
+
ρ
2
≤ ρ.
That is, φ
y
(x) ∈ [B]. Thus, φ
y
(x) is a contraction of the complete space [B] into itself.
Hence, φ
y
(x) has a unique ﬁxed point x ∈ [B] and y = f(x) ∈ f([B]) ⊂ f(U) = V .
(4. f
−1
∈ C
−1
.) Choose y
1
, y
2
∈ V , there exist x
1
, x
2
∈ U such that f(x
1
) = y
1
,
f(x
2
) = y
2
.
φ
y
(x
2
) −φ
y
(x
1
) = x
2
−x
1
+A
−1
(f(x
1
) −f(x
2
)) = (x
2
−x
1
) −A
−1
(y
2
−y
1
).
⇒(x
2
−x
1
)−A
−1
(y
2
−y
1
) ≤
1
2
x
2
−x
1
⇒
1
2
x
2
−x
1
≤ A
−1
(y
2
−y
1
) ≤
1
2λ
y
2
−y
1
or x
2
−x
1
≤
1
λ
y
2
−y
1
. It follows that (f
′
)
−1
exists locally about a. Since
f
−1
(y
2
) −f
−1
(y
1
) −(f
′
)
−1
(y
1
)(y
2
−y
1
) = (x
2
−x
1
) −(f
′
)
−1
(y
1
)(y
2
−y
1
)
= −(f
′
)
−1
(y
1
)[−f
′
(x
1
)(x
2
−x
1
) +f(x
2
) −f(x
1
)],
35
We have
f
−1
(y
2
) −f
−1
(y
1
) −(f
′
)
−1
(y
1
)(y
2
−y
1
)
y
2
−y
1
≤
(f
′
)
−1
λ
f(x
2
) −f(x
1
) −f
′
(x
1
)(x
2
−x
1
)
x
2
−x
1
.
As y
2
→y
1
, x
2
→x
1
. Hence (f
1−
)
′
(y) = {f
′
[f
−1
(y)]} for y ∈ V . Since f
−1
is diﬀer
entiable, it is continuous. Also, f
′
is continuous and its inversion, where it exists, is
continuous. Therefore (f
−1
)
′
is continuous or f
−1
∈ C
1
(V ). Q.E.D.
Implicit Function Theorem: Let E ⊆ R
(n+m)
be an open set and a ∈ R
n
, b ∈ R
m
,
(a, b) ∈ E. Suppose f : E →R
n
is C
1
(E) and f(a, b) = 0, and J(f(a, b)) = 0. Then
there exist open sets A ⊂ R
n
and B ⊂ R
m
a ∈ A and b ∈ B, such that for each
x ∈ B, there exists a unique g(x) ∈ A such that f(g(x), x) = 0 and g : B → A is
C
1
(B).
Proof:
Deﬁning F : R
n+m
→R
n+m
by F(x, y) ≡ (x, f(x, y)). Note that since
J(F(a, b)) =
¸
¸
¸
∂x
i
∂x
j
1≤i,j≤n
∂x
i
∂x
n+j
1≤i≤n,1≤j≤m
∂f
i
∂x
j
1≤i≤m,1≤j≤n
∂f
i
∂x
n+j
1≤i,j≤m
=
I O
N M
,
J(F(a, b)) = M = 0. By the Inverse Function Theorem there exists an open set
V ⊆ R
n+m
containing F(a, b) = (a, 0) and an open set of the form A × B ⊆ E
containing (a, b), such that F : A × B→V has a C
1
inverse F
−1
: V →A × B. F
−1
is of the form F
−1
(x, y) = (x, φ(x, y)) for some C
1
function φ. Deﬁne the projection
π : R
n+m
→R
m
by π(x, y) = y. Then π ◦ F(x, y) = f(x, y). Therefore
f(x, φ(x, y)) = f ◦ F
−1
(x, y) = (π ◦ F) ◦ F
−1
(x, y) = π ◦ (F ◦ F
−1
)(x, y) = π(x, y) = y
and f(x, φ(x, 0)) = 0. So, deﬁne g : A→B by g(x) = φ(x, 0). Q.E.D.
36
6 Comparative Statics – Economic applications
6.1 Partial equilibrium model
Q = D(P, Y )
∂D
∂P
< 0,
∂D
∂Y
> 0 end. var: Q, P.
Q = S(P) S
′
(P) > 0 ex. var: Y..
f
1
(P, Q; Y ) = Q−D(P, Y ) = 0
df
1
dY
=
dQ
dY
−
∂D
∂P
dP
dY
−
∂D
∂Y
= 0
f
2
(P, Q; Y ) = Q−S(P) = 0
df
2
dY
=
dQ
dY
−S
′
(P)
dP
dY
= 0
.
1 −
∂D
∂P
1 −S
′
(P)
¸
¸
dQ
dY
dP
dY
=
∂D
∂Y
0
, J =
1 −
∂D
∂P
1 −S
′
(P)
= −S
′
(P) +
∂D
∂P
< 0.
dQ
dY
=
∂D
∂Y
−
∂D
∂P
0 −S
′
(P)
J
=
−
∂D
∂Y
S
′
(P)
J
> 0,
dP
dY
=
1
∂D
∂Y
1 0
J
=
−
∂D
∂Y
J
> 0.
6.2 Income determination model
C = C(Y ) 0 < C
′
(Y ) < 1.
I = I(r) I
′
(r) < 0 end. var. C, Y, I
Y = C +I +
¯
G ex. var.
¯
G, r.
Y = C(Y ) +I(r) +
¯
G ⇒dY = C
′
(Y )dY +I
′
(r)dr +d
¯
G ⇒dY =
I
′
(r)dr +d
¯
G
1 −C
′
(Y )
.
∂Y
∂r
=
I
′
(r)
1 −C
′
(Y )
< 0,
∂Y
∂
¯
G
=
1
1 −C
′
(Y )
> 0.
6.2.1 Income determination and trade
Consider an income determination model with import and export:
C = C(Y ) 1 > C
y
> 0, I =
¯
I,
M = M(Y, e) M
y
> 0, M
e
< 0 X = X(Y
∗
, e), X
y∗
> 0, X
e
> 0
C +I +X −M = Y,
where import M is a function of domestic income and exchange rate e and export
X is a function of exchange rate and foreign income Y
∗
, both are assumed here as
exogenous variables. Substituting consumption, import, and export functions into
the equilibrium condition, we have
C(Y )+
¯
I+X(Y
∗
, e)−M(Y, e) = Y, ⇒F(Y, e, Y
∗
) ≡ C(Y )+
¯
I+X(Y
∗
, e)−M(Y, e)−Y = 0..
37
Use implicit function rule to derive
∂Y
∂
¯
I
and determine its sign:
∂Y
∂
¯
I
= −
F
I
F
y
=
1
1 −C
′
(Y ) +M
y
.
Use implicit function rule to derive
∂Y
∂e
and determine its sign:
∂Y
∂e
= −
F
e
F
y
=
X
e
−M
e
1 −C
′
(Y ) +M
y
.
Use implicit function rule to derive
∂Y
∂Y
∗
and determine its sign:
∂Y
∂Y
∗
= −
F
y
∗
F
y
=
X
y
∗
1 −C
′
(Y ) +M
y
.
6.2.2 Interdependence of domestic and foreign income
Now extend the above income determination model to analyze the joint dependence
of domestic income and foreign income:
C(Y ) +
¯
I +X(Y
∗
, e) −M(Y, e) = Y C
∗
(Y
∗
) +
¯
I
∗
+X
∗
(Y, e) −M
∗
(Y
∗
, e) = Y
∗
,
with a similar assumption on the foreigner’s consumption function: 1 > C
∗
y∗
> 0.
Since domestic import is the same as foreigner’s export and domestic export is for
eigner’s import, X
∗
(Y, e) = M(Y, e) and M
∗
(Y
∗
, e) = X(Y
∗
, e) and the system be
comes:
C(Y ) +
¯
I +X(Y
∗
, e) −M(Y, e) = Y C
∗
(Y
∗
) +
¯
I
∗
+M(Y, e) −X(Y
∗
, e) = Y
∗
,
Calculate the total diﬀerential of the system (Now Y
∗
becomes endogenous):
1 −C
′
+M
y
−X
y∗
−M
y
1 −C
∗′
+X
y∗
dY
dY
∗
=
(X
e
−M
e
)de +d
¯
I
(M
e
−X
e
)de +d
¯
I
∗
,
J =
1 −C
′
+M
y
−X
y∗
−M
y
1 −C
∗′
+X
y∗
= (1 −C
′
+M
y
)(1 −C
∗′
+M
y∗
) −M
y
X
y
∗ > 0.
Use Cramer’s rule to derive
∂Y
∂e
and
∂Y
∗
∂e
and determine their signs:
dY =
(X
e
−M
e
)de +d
¯
I −X
y∗
(M
e
−X
e
)de +d
¯
I
∗
1 −C
∗′
+X
y∗
J
=
(X
e
−M
e
)(1 −C
∗′
+X
y∗
−X
y∗
)de + (1 −C
∗′
+X
y∗
)d
¯
I +X
y∗
d
¯
I
∗
J
,
38
dY
∗
=
1 −C
′
+M
y
(X
e
−M
e
)de +d
¯
I
−M
y
(M
e
−X
e
)de +d
¯
I
∗
J
=
−(X
e
−M
e
)(1 −C
′
+M
y
−M
y
)de + (1 −C
′
+M
y
)d
¯
I
∗
+X
y
d
¯
I
J
.
∂Y
∂e
=
(X
e
−M
e
)(1 −C
∗′
)
J
> 0,
∂Y
∗
∂e
=
−(X
e
−M
e
)(1 −C
′
)
J
< 0.
Derive
∂Y
∂
¯
I
and
∂Y
∗
∂
¯
I
and determine their signs:
∂Y
∂
¯
I
= −
1 −C
∗′
+M
y∗
J
> 0,
∂Y
∗
∂
¯
I
=
M
y
J
< 0.
6.3 ISLM model
C = C(Y ) 0 < C
′
(Y ) < 1 M
d
= L(Y, r)
∂L
∂Y
> 0,
∂L
∂r
< 0
I = I(r) I
′
(r) < 0 M
s
=
¯
M
Y = C +I +
¯
G M
d
= M
s
.
end. var: Y , C, I, r, M
d
, M
s
. ex. var:
¯
G,
¯
M.
Y −C(Y ) −I(r) =
¯
G
L(Y, r) =
¯
M
⇒
(1 −C
′
(Y ))dY −I
′
(r)dr = d
¯
G
∂L
∂Y
dY +
∂L
∂r
dr = d
¯
M
1 −C
′
−I
′
L
Y
L
r
dY
dr
=
d
¯
G
d
¯
M
, J =
1 −C
′
−I
′
L
Y
L
r
= (1−C
′
)L
r
+I
′
L
Y
< 0.
dY =
d
¯
G −I
′
d
¯
M L
r
J
=
L
r
d
¯
G+I
′
d
¯
M
J
, dr =
1 −C
′
d
¯
G
L
Y
d
¯
M
J
=
−L
Y
d
¯
G+ (1 −C
′
)d
¯
M
J
∂Y
∂
¯
G
=
L
r
J
> 0,
∂Y
∂
¯
M
=
I
′
J
> 0,
∂r
∂
¯
G
= −
L
Y
J
> 0,
∂r
∂
¯
M
=
(1 −C
′
)
J
< 0.
6.4 Twomarket general equilibrium model
Q
1d
= D
1
(P
1
, P
2
) D
1
1
< 0, Q
2d
= D
2
(P
1
, P
2
) D
2
2
< 0, D
2
1
> 0.
Q
1s
=
¯
S
1
Q
2s
= S
2
(P
2
) S
′
2
(P
2
) > 0
Q
1d
= Q
1s
Q
2d
= Q
2s
end. var: Q
1
, Q
2
, P
1
, P
2
. ex. var:
¯
S
1
.
D
1
(P
1
, P
2
) =
¯
S
1
⇒ D
1
1
dP
1
+D
1
2
dP
2
= d
¯
S
1
D
2
(P
1
, P
2
) −S
2
(P
2
) = 0 D
2
1
dP
1
+D
2
2
dP
2
−S
′
2
dP
2
= 0.
39
D
1
1
D
1
2
D
2
1
D
2
2
−S
′
2
dP
1
dP
2
=
d
¯
S
1
0
, J =
D
1
1
D
1
2
D
2
1
D
2
2
−S
′
2
= D
1
1
(D
2
2
−S
′
2
)−D
1
2
D
2
1
.
Assumption: D
1
1
 > D
1
2
 and D
2
2
 > D
2
1
 (owneﬀects dominate), ⇒J > 0.
dP
1
d
¯
S
1
=
D
2
2
−S
′
2
J
< 0,
dP
2
d
¯
S
1
=
−D
2
1
J
< 0
From Q
1s
=
¯
S
1
,
∂Q
1
∂
¯
S
1
= 1. To calculate
∂Q
2
∂
¯
S
1
, we have to use chain rule:
∂Q
2
∂
¯
S
1
= S
′
2
dP
2
d
¯
S
1
< 0.
6.4.1 Car market
Suppose we want to analyze the eﬀect of the price of used cars on the market of
new cars. The demand for new cars is given by Q
n
= D
n
(P
n
; P
u
), ∂D
n
/∂P
n
< 0
∂D
n
/∂P
u
> 0, where Q
n
is the quantity of new cars and P
n
(P
u
) the price of a new
(used) car. The supply function of new cars is Q
n
= S(P
n
), S
′
(P
n
) > 0.
end. var: P
n
, Q
n
. ex. var: P
u
.
D
n
(P
n
; P
u
) = S(P
n
); ⇒
∂D
n
∂P
n
dP
n
+
∂D
n
∂P
u
dP
u
= S
′
(P
n
)dP
n
.
dP
n
dP
u
=
∂D
n
/∂P
u
S
′
(P
n
) −∂D
n
/∂P
n
> 0,
dQ
n
dP
u
= S
′
(P
n
)
dP
n
dP
u
=
S
′
(P
n
)∂D
n
/∂P
u
S
′
(P
n
) −∂D
n
/∂P
n
> 0.
The markets for used cars and for new cars are actually interrelated. The demand
for used cars is Q
u
= D
u
(P
u
, P
n
), ∂D
n
/∂P
u
< 0, ∂D
n
/∂P
n
> 0. In each period,
the quantity of used cars supplied is ﬁxed, denoted by
¯
Q
u
. Instead of analyzing the
eﬀects of a change in P
u
on the new car market, we want to know how a change in
¯
Q
u
aﬀects both markets.
end. var: P
n
, Q
n
, P
u
, Q
u
. ex. var:
¯
Q
u
.
Q
n
= D
n
(P
n
, P
u
), Q
n
= S(P
n
); Q
u
= D
u
(P
u
, P
n
), Q
u
=
¯
Q
u
⇒D
n
(P
n
, P
u
) = S(P
n
), D
u
(P
u
, P
n
) =
¯
Q
u
; D
n
n
dP
n
+D
n
u
dP
u
= S
′
dP
n
, D
u
n
dP
n
+D
u
u
dP
u
=
¯
Q
D
n
n
−S
′
D
n
u
D
u
n
D
u
u
dP
n
dP
u
=
0
d
¯
Q
u
, J =
D
n
n
−S
′
D
n
u
D
u
n
D
u
u
= (D
n
n
−S
′
)D
u
u
−D
n
u
D
u
n
.
Assumption: D
n
n
 > D
n
u
 and D
u
u
 > D
u
n
 (owneﬀects dominate), ⇒J > 0.
dP
n
d
¯
Q
u
=
−D
n
u
J
< 0,
dP
u
d
¯
Q
u
=
D
n
n
−S
′
J
< 0.
From Q
u
=
¯
Q
u
,
∂Q
u
∂
¯
Q
u
= 1. To calculate
∂Q
n
∂
¯
Q
u
, we have to use chain rule:
∂Q
n
∂
¯
Q
u
=
S
′
dP
n
d
¯
Q
n
< 0.
40
6.5 Classic labor market model
L = h(w) h
′
> 0 labor supply function
w = MPP
L
=
∂Q
∂L
= F
L
(K, L) F
LK
> 0, F
LL
< 0 labor demand function.
endogenous variables: L, w. exogenous variable: K.
L −h(w) = 0 ⇒ dL −h
′
(w)dw = 0
w −F
L
= 0 dw −F
LL
dL −F
LK
dK = 0
.
1 −h
′
(w)
−F
LL
1
dL
dw
=
0
F
LK
dK
, J =
1 −h
′
(w)
−F
LL
1
= 1−h
′
(w)F
LL
> 0.
dL
dK
=
0 −h
′
(w)
F
LK
1
J
=
h
′
F
LK
J
> 0,
dw
dK
=
1 0
−F
LL
F
LK
J
=
F
LK
J
> 0.
6.6 Problem
1. Let the demand and supply functions for a commodity be
Q = D(P) D
′
(P) < 0
Q = S(P, t) ∂S/∂P > 0, ∂S/∂t < 0,
where t is the tax rate on the commodity.
(a) Derive the total diﬀerentail of each equation.
(b) Use Cramer’s rule to compute dQ/dt and dP/dt.
(c) Determine the sign of dQ/dt and dP/dt.
(d) Use the Q−P diagram to explain your results.
2. Suppose consumption C depends on total wealth W, which is predetermind, as
well as on income Y . The ISLM model becomes
C = C(Y, W) 0 < C
Y
< 1 C
W
> 0 M
S
= M
I = I(r) I
′
(r) < 0 Y = C +I
M
D
= L(Y, r) L
Y
> 0 L
r
< 0 M
S
= M
D
(a) Which variables are endogenous? Which are exogenous?
(b) Which equations are behavioral/institutional, which are equilibrium condi
tions?
The model can be reduced to
Y −C(Y, W) −I(r) = 0, L(Y, r) = M.
(c) Derive the total diﬀerential for each of the two equations.
(d) Use Cramer’s rule to derive the eﬀects of an increase in W on Y and r, ie.,
derive ∂Y/∂W and ∂r/∂W.
(e) Determine the signs of ∂Y/∂W and ∂r/∂W.
41
3. Consider a 2industry (e.g. manufacturing and agriculture) general equilibrium
model. The demand for manufacturing product consists of two components:
private demand D
1
and goverment demand G. The agricultural products have
only private demand D
2
. Both D
1
and D
2
depend only on their own prices.
Because each industry requires in its production process outputs of the other,
the supply of each commodity depends on the price of the other commodity as
well as on its own price. Therefore, the model may be written as follows:
Q
d
1
= D
1
(P
1
) +G D
′
1
(P
1
) < 0,
Q
d
2
= D
2
(P
2
) D
′
2
(P
2
) < 0,
Q
s
1
= S
1
(P
1
, P
2
) S
1
1
> 0, S
1
2
< 0, S
1
1
> S
1
2
,
Q
s
2
= S
2
(P
1
, P
2
) S
2
1
< 0, S
2
2
> 0, S
2
2
> S
2
1
,
Q
d
1
= Q
s
1
,
Q
d
2
= Q
s
2
.
(a) Which variables are endogenous? Which are exogenous?
(b) Which equations are behavioral? Which are deﬁnitional? Which are equi
librium conditions?
4. The model can be reduced to
S
1
(P
1
, P
2
) = D
1
(P
1
) +G S
2
(P
1
, P
2
) = D
2
(P
2
)
(c) Compute the total diﬀerential of each equation.
(d) Use Cramer’s rule to derive ∂P
1
/∂G and ∂P
2
/∂G.
(e) Determine the signs of ∂P
1
/∂G and ∂P
2
/∂G.
(f) Compute ∂Q
1
/∂G and ∂Q
2
/∂G. (Hint: Use chain rule.)
(g) Give an economic interpretation of the results.
5. The demand functions of a 2commodity market model are:
Q
d
1
= D
1
(P
1
, P
2
) Q
d
2
= D
2
(P
1
, P
2
).
The supply of the ﬁrst commodity is given exogenously, ie., Q
s
1
= S
1
. The
supply of the second commodity depends on its own price, Q
s
2
= S
2
(P
2
). The
equilibrium conditions are:
Q
d
1
= Q
s
1
, Q
d
2
= Q
s
2
.
(a) Which variables are endogenous? Which are exogenous?
(b) Which equations are behavioral? Which are deﬁnitional? Which are equi
librium conditions?
The model above can be reduced to :
D
1
(P
1
, P
2
) −S
1
= 0 D
2
(P
1
, P
2
) −S
2
(P
2
) = 0
42
Suppose that both commodities are not Giﬀen good (hence, D
i
i
< 0, i = 1, 2),
that each one is a gross substitute for the other (ie., D
i
j
> 0, i = j), that
D
i
i
 > D
i
j
, and that S
′
2
(P
2
) > 0.
(c) Calculate the total diﬀerential of each equation of the reduced model.
(d) Use Cramer’s rule to derive ∂P
1
/∂S
1
and ∂P
2
/∂S
1
.
(e) Determine the signs of ∂P
1
/∂S
1
and ∂P
2
/∂S
1
.
(f) Compute ∂Q
1
/∂S
1
and ∂Q
2
/∂S
1
and determine their signs.
6. The demand functions for ﬁsh and chicken are as follows:
Q
d
F
= D
F
(P
F
−P
C
), D
′
F
< 0
Q
d
C
= D
C
(P
C
−P
F
), D
′
C
< 0
where P
F
, P
C
are price of ﬁsh and price of chicken respectively. The supply
of ﬁsh depends on the number of ﬁshermen (N) as well as its price P
F
: Q
s
F
=
F(P
F
, N), F
P
F
> 0, F
N
> 0. The supply of chicken depends only on its price
P
C
: Q
s
C
= C(P
C
), C
′
> 0. The model can be reduced to
D
F
(P
F
−P
C
) = F(P
F
, N)
D
C
(P
C
−P
F
) = C(P
C
)
(a) Find the total diﬀerential of the reduced system.
(b) Use Cramer’s rule to ﬁnd dP
F
/dN and dP
C
/dN.
(c) Determine the signs of dP
F
/dN and dP
C
/dN. What is the economic mean
ing of your results?
(d) Find dQ
C
/dN.
7. In a 2good market equilibrium model, the inverse demand functions are given
by
P
1
= U
1
(Q
1
, Q
2
), P
2
= U
2
(Q
1
, Q
2
);
where U
1
(Q
1
, Q
2
) and U
2
(Q
1
, Q
2
) are the partial derivatives of a utility function
U(Q
1
, Q
2
) with respect to Q
1
and Q
2
, respectively.
(a) Calculate the Jacobian matrix
∂P
1
/∂Q
1
∂P
1
/∂Q
2
∂P
2
/∂Q
1
∂P
2
/∂Q
2
and Jacobian
∂(P
1
, P
2
)
∂(Q
1
, Q
2
)
.
What condition(s) should the parameters satisfy so that we can invert the
functions to obtain the demand functions?
(b) Derive the Jacobian matrix of the derivatives of (Q
1
, Q
2
) with respect to
(P
1
, P
2
),
∂Q
1
/∂P
1
∂Q
1
/∂P
2
∂Q
2
/∂P
1
∂Q
2
/∂P
2
.
(c) Suppose that the supply functions are
Q
1
= a
−1
P
1
, Q
2
= P
2
,
and Q
∗
1
and Q
∗
2
are market equilibrium quantities. Find the comparative
statics
dQ
∗
1
da
and
dQ
∗
2
da
. (Hint: Eliminate P
1
and P
2
.)
43
8. In a 2good market equilibrium model with a sales tax of t dollars per unit on
product 1, the model becomes
D
1
(P
1
+t, P
2
) = Q
1
= S
1
(P
1
), D
2
(P
1
+t, P
2
) = Q
2
= S
2
(P
2
).
Suppose that D
i
i
< 0 and S
′
i
> 0, D
i
i
 > D
i
j
, i = j, i, j = 1, 2.
(a) Calculate dP
2
/dt.
(b) Calculate dQ
2
/dt.
(c) Suppose that D
i
j
> 0. Determine the signs of dP
2
/dt and dQ
2
/dt.
(d) Suppose that D
i
j
< 0. Determine the signs of dP
2
/dt and dQ
2
/dt.
(e) Explain your results in economics.
44
7 Optimization
A behavioral equation is a summary of the decisions of a group of economic agents
in a model. A demand (supply) function summarizes the consumption (production)
decisions of consumers (producers) under diﬀerent market prices, etc. The derivative
of a behavioral function represents how agents react when an independent variable
changes. In the last chapter, when doing comparative static analysis, we always
assumed that the signs of derivatives of a behavioral equation in a model are known.
For example, D
′
(P) < 0 and S
′
(P) > 0 in the partial market equilibrium model,
C
′
(Y ) > 0, I
′
(r) < 0, L
y
> 0, and L
r
< 0 in the ISLM model. In this chapter, we
are going to provide a theoretical foundation for the determination of these signs.
7.1 Neoclassic methodology
Neoclassic assumption: An agent, when making decisions, has an objective function
in mind (or has well deﬁned preferences). The agent will choose a feasible decision
such that the objective function is maximized.
A consumer will choose the quantity of each commodity within his/her budget con
straints such that his/her utility function is maximized. A producer will choose to
supply the quantity such that his proﬁt is maximizaed.
Remarks: (1) Biological behavior is an alternative assumption, sometimes more ap
propriate, (2) Sometimes an agent is actually a group of people with diﬀerent per
sonalities like a company and we have to use game theoretic equilibrium concepts to
characterize the collective behavior.
Maximization ⇒ Behavioral equations
Game equilibrium ⇒ Equilibrium conditions
x
1
, . . . , x
n
: variables determined by the agent (endogenous variables).
y
1
, . . . , y
m
: variables given to the agent (exogenous variables).
Objective function: f(x
1
, . . . , x
n
; y
1
, . . . , y
m
).
Opportunity set: the agent can choose only (x
1
, . . . , x
n
) such that (x
1
, . . . , x
n
; y
1
, . . . , y
m
) ∈
A ⊂ R
n+m
. A is usually deﬁned by inequality constriants.
max
x
1
,...,xn
f(x
1
, . . . , x
n
; y
1
, . . . , y
m
) subject to
g
1
(x
1
, . . . , x
n
; y
1
, . . . , y
m
) ≥ 0
.
.
.
g
k
(x
1
, . . . , x
n
; y
1
, . . . , y
m
) ≥ 0.
Solution (behavioral equations): x
i
= x
i
(y
1
, . . . , y
m
), i = 1, . . . , n (derived from FOC).
∂x
i
/∂y
j
: derived by the comparative static method (sometimes its sign can be deter
mined from SOC).
Example 1: A consumer maximizes his utility function U(q
1
, q
2
) subject to the
budget constraint p
1
q
1
+ p
2
q
2
= m ⇒ demand functions q
1
= D
1
(p
1
, p
2
, m) and
q
2
= D
2
(p
1
, p
2
, m).
45
Example 2: A producer maximizes its proﬁt Π(Q; P) = PQ − C(Q) where C(Q) is
the cost of producing Q units of output ⇒ the supply function Q = S(P).
one endogenous variable: this chapter.
n endogenous variables without constraints: next
n endogenous variables with equality constraints:
Nonlinear programming: n endogenous variables with inequality constraints
Linear programming: Linear objective function with linear inequality constraints
Game theory: more than one agents with diﬀerent objective functions
7.2 Diﬀerent concepts of maximization
Suppose that a producer has to choose a Q to maximize its proﬁt π = F(Q):
max
Q
F(Q). Assume that F
′
(Q) and F
′′
(Q) exist.
A local maximum Q
l
: there exists ǫ > 0 such that F(Q
l
) ≥ F(Q) for all Q ∈
(Q
l
−ǫ, Q
l
+ǫ).
A global maximum Q
g
: F(Q
g
) ≥ F(Q) for all Q.
A unique global maximum Q
u
: F(Q
u
) > F(Q) for all Q = Q
u
.
Q
`
F
Q
l
Q
l
: local max.
not global max
Q
`
F
Q
g
: a global max.
but not unique
Q
g
Q
`
F
Q
u
: unique global max
Q
u
The agent will choose only a global maximum as the quantity supplied to the mar
ket. However, it is possible that there are more than one global maximum. In that
case, the supply quantity is not unique. Therefore, we prefer that the maximization
problem has a unique global maximum.
A unique global maximum must be a global maximum and a global maximum must
be a local maximum. ⇒to ﬁnd a global maximum, we ﬁrst ﬁnd all the local maxima.
One of them must be a global maximum, otherwise the problem does not have a
solution (the maximum occurs at ∞.) We will ﬁnd conditions (eg., increasing MC or
decreasing MRS) so that there is only one local maximum which is also the unique
global maximum.
7.3 FOC and SOC for a local maximum
At a local maximum Q
l
, the slope of the graph of F(Q) must be horizontal F
′
(Q
l
) = 0.
This is called the ﬁrst order condition (FOC) for a local maximum.
A critical point Q
c
: F
′
(Q
c
) = 0.
A local maximum must be a critical point but a critical point does not have to be a
local maximum.
46
Q
`
F
F
′′
(Q
c
) < 0
local max.
Q
c
Q
`
F
F
′′
(Q
c
) > 0
local min.
Q
c
Q
`
F
F
′′
(Q
c
) = 0
degenerate
Q
c
A degenerate critical point: F
′
(Q) = F
′′
(Q) = 0.
A nondegenerate critical point: F
′
(Q) = 0, F
′′
(Q) = 0.
A nondegenerate critical point is a local maximum (minimum) if F
′′
(Q) < 0 (F
′′
(Q) >
0).
FOC: F
′
(Q
l
) = 0 SOC: F
′′
(Q
l
) < 0
Example: F(Q) = −15Q+9Q
2
−Q
3
, F
′
(Q) = −15+18Q−3Q
2
= −3(Q−1)(Q−5).
There are two critical points: Q = 1, 5. F
′′
(Q) = 18 − 6Q, F
′′
(1) = 12 > 0, and
F
′′
(5) = −12 < 0. Therefore, Q = 5 is a local maximum. It is a global maximum for
0 ≤ Q < ∞.
Remark 1 (Degeneracy): For a degenerate critical point, we have to check higher
order derivatives. If the lowest order nonzero derivative is of odd order, then it is a
reﬂect point; eg., F(Q) = (Q−5)
3
, F
′
(5) = F
′′
(5) = 0 and F
′′′
(5) = 6 = 0 and Q = 5
is not a local maximum. If the lowest order nonzero derivative is of even order and
negative (positive), then it is a local maximum (minimum); eg., F(Q) = −(Q− 5)
4
,
F
′
(5) = F
′′
(5) = F
′′′
(5) = 0, F
(4)
(5) = −24 < 0 and Q = 5 is a local maximum.
Remark 2 (Unboundedness): If lim
Q→∞
F(Q) = ∞, then a global maximum does not
exist.
Remark 3 (Nondiﬀerentiability): If F(Q) is not diﬀerentiable, then we have to use
other methods to ﬁnd a global maximum.
Remark 4 (Boundary or corner solution): When there is nonnegative restriction
Q ≥ 0 (or an upper limit Q ≤ a), it is possible that the solution occurs at Q = 0 (or
at Q = a). To take care of such possibilities, FOC is modiﬁed to become F
′
(Q) ≤ 0,
QF
′
(Q) = 0 (or F
′
(Q) ≥ 0, (a −Q)F
′
(Q) = 0).
7.4 Supply function of a competitive producer
Consider ﬁrst the proﬁt maximization problem of a competitive producer:
max
Q
Π = PQ−C(Q), FOC ⇒
∂Π
∂Q
= P −C
′
(Q) = 0.
The FOC is the inverse supply function (a behavioral equation) of the producer: P
= C
′
(Q) = MC. Remember that Q is endogenous and P is exogenous here. To ﬁnd
47
the comparative statics
dQ
dP
, we use the total diﬀerential method discussed in the last
chapter:
dP = C
′′
(Q)dQ, ⇒
dQ
dP
=
1
C
′′
(Q)
.
To determine the sign of
dQ
dP
, we need the SOC, which is
∂
2
Π
∂Q
2
= −C
′′
(Q) < 0. There
fore,
dQ
s
dP
> 0.
Remark: The result is true no matter what the cost function C(Q) is. MC = C
′
(Q)
can be nonmonotonic, but the supply curve is only part of the increasing sections of
the MC curve and can be discontinuous.
Q
`
P
MC
MC is the
supply curve.
Q
`
P
MC
Q
1
Q
2
P
c
S(P
c
) = {Q
1
, Q
2
},
supply curve has
two component.
7.5 Maximization and comparative statics: general procedure
Maximization problem of an agent: max
X
F(X; Y ).
FOC: F
X
(X
∗
; Y ) = 0, ⇒X
∗
= X(Y ) · · · · · · Behavioral Equation
Comparative statics: F
XX
dX +F
XY
dY = 0 ⇒
dX
dY
= −
F
XY
F
XX
. SOC: F
XX
< 0
Case 1: F
XY
> 0 ⇒
dX
dY
= −
F
XY
F
XX
> 0.
Case 2: F
XY
< 0 ⇒
dX
dY
= −
F
XY
F
XX
< 0.
Therefore, the sign of
dX
dY
depends only on the sign of F
XY
.
7.6 Utility Function
A consumer wants to maximize his/her utility function U = u(Q) + M = u(Q) +
(Y −PQ).
FOC:
∂U
∂Q
= u
′
(Q) −P = 0,
⇒u
′
(Q
d
) = P (inverse demand function)
⇒Q
d
= D(P) (demand function, a behavioral equation)
48
∂
2
U
∂Q∂P
= U
PQ
= −1 ⇒
dQ
d
dP
= D
′
(P) < 0, the demand function is a decreasing
function of price.
7.7 Input Demand Function
The production function of a producer is given by Q = f(x), where x is the quantity
of an input employed. Its proﬁt is Π = pf(x) − wx, where p (w) is the price of the
output (input).
The FOC of proﬁt maximization problem is pf
′
(x) −w = 0
⇒f
′
(x) = w/p (inverse input demand function)
⇒x = h(w/p) (input demand function, a behavioral equation)
∂
2
Π
∂x∂(w/p)
= −1 ⇒
dx
d(w/p)
= h
′
(w/p) < 0, the input demand is a decreasing func
tion of the real input price
w
p
.
7.8 Envelope theorem
Deﬁne the maximum function M(Y ) ≡ max
X
F(X, Y ) = F(X(Y ), Y ) then the total
derivative
dM
dY
= M
′
(Y ) =
∂F(X, Y )
∂Y
X=X(Y )
.
Proof: M
′
(Y ) = F
X
dX
dY
+ F
Y
. At the maximization point X = X(Y ), FOC implies
that the indirect eﬀect of Y on M is zero.
In the consumer utility maximization problem, V (P) ≡ U(D(P)) + Y − PD(P)
is called the indirect utility function. The envelope theorem implies that V
′
(P) =
dU
dP
Q
d
=D(P)
≡ −D(P), this is a simpliﬁed version of Roy’s identity.
In the input demand function problem, π(w, p) ≡ pf(h(w/p)) −wh(w/p) is the proﬁt
function. Let p = 1 and still write π(w) ≡ f(h(w)) −wh(w). The envelope theorem
implies that π
′
(w) =
dΠ
dw
x=h(w)
= −h(w), a simpliﬁed version of Hotelling’s lemma.
Example: The relationships between LR and SR cost curves
STC(Q; K) = C(Q, K), K: ﬁrm size. Each K corresponds to a STC.
LTC(Q) = min
K
C(Q, K), ⇒ K = K(Q) is the optimal ﬁrm size.
LTC is the envelope of STC’s. Each STC tangents to the LTC (STC = LTC) at the
quantity Q such that K = K(Q). Notice that the endogenous variable is K and the
exogenous is Q here.
By envelope theorem, LMC(Q) = dLTC(Q)/dQ = ∂C(Q, K(Q))/∂Q =SMC(Q; K(Q)).
That is, when K = K(Q) is optimal for producing Q, SMC=LMC.
49
Since LAC(Q) =LTC(Q)/Q and SAC(Q) =STC(Q)/Q, LAC is the envelope of SAC’s
and each SAC tangents to the LAC (SAC = LAC) at the quantity Q such that
K = K(Q).
7.9 Eﬀect of a Unit Tax on Monopoly Output (Samuelson)
Assumptions: a monopoly ﬁrm, q = D(P) ⇐⇒ P = f(q), (inverse functions),
C = C(q), t = unit tax
max π(q) = Pq −C(q) −tq = qf(q) −C(q) −tq
q: endogenous; t: exogenous
FOC: ∂π/∂q = f(q) +qf
′
(q) −C
′
(q) −t = 0.
The FOC deﬁnes a relationship between the monopoly output q
∗
and the tax rate t
as q
∗
= q(t) (a behavioral equation). The derivative dq
∗
/dt can be determined by the
sign of the cross derivative:
∂
2
π
∂q∂t
= −1 < 0
Therefore, we have dq
∗
/dt < 0.
The result can be obtained using the q–pdiagram. FOC ⇐⇒ MR = MC + t.
Therefore, on q–p space, an equilibrium is determined by the intersection point of
MR and MC +t curves.
Case 1: MR is downward sloping and MC is upward sloping:
When t increases, q
∗
decreases as seen from the left diagram below.
q
`
P
MC
MC+t
MR
<
q
`
P
MC
MC+t
MR
<
Case 2: Both MR and MC are downward sloping and MR is steeper. MC decreasing;
MR decreasing more ⇒t ↑ q ↓.
Case 3: Both MR and MC are downward sloping, but MC is steeper. The diagram
shows that dq
∗
/dt > 0. It is opposite to our comparative statics result. Why?
MR = MC + t violates SOC < 0, therefore, the intersection of MR and MC + t
is not a proﬁt maximizing point.
50
q
`
P
MC
MC+t
MR
>
7.10 A Price taker vs a price setter
A producer employs an input X to produce an output Q. The production function
is Q = rX. The inverse demand function for Q is P = a − Q. The inverse supply
function of X is W = b + X, where W is the price of X. The producer is the only
seller of Q and only buyer of X.
There are two markets, Q (output) and X (input). We want to ﬁnd the 2market
equilibrium, i.e., the equilibrium values of W, X, Q, and P. It depends on the pro
ducer’s power in each market.
Case 1: The ﬁrm is a monopolist in Q and a price taker in X. To the producer,
P is endogenous and W is exogenous. Given W, its object is
max
x
π = PQ−WX = (a−Q)Q−WX = (a−rX)(rX)−WX, ⇒FOC ar−2r
2
X−W = 0.
The input demand function is X = X(W) =
ar−W
2r
2
.
Equating the demand and supply of X, the input market equilibrium is X =
ar−b
2r
2
+1
and W = b +X =
ar+2r
2
b
2r
2
+1
.
Substituting back into the production and output demand functions, the output mar
ket equilibrium is Q = rX = far
2
−br2r
2
+ 1 and P = a−Q = fa +ar
2
+b2r
2
+ 1.
Case 2: The ﬁrm is a price taker in Q and a monopsony in X. To the producer,
P is exogenous and W is endogenous. Given P, its object is
max
Q
π = PQ−(b +X)X = PQ−(b + (Q/r))(Q/r), ⇒FOC P −
b
r
−
2Q
r
2
= 0.
The output supply function is Q = Q(P) =
r
2
P−br
2
.
Equating the demand and supply of Q, the output market equilibrium is Q =
ar
2
−br
r
2
−2
and P = a −Q =
2a+br
2r
2
+1
.
Substituting back into the production and output demand functions, the output mar
ket equilibrium is X = Q/r = far −br
2
−2 and W = b +X = far +br
2
−3br
2
−2.
Case 3: The ﬁrm is a monopolist in Q and a monopsony in X. To the producer,
both P and W are endogenous, its object is
max
x
π = (a −Q)Q−(b +X)X = (a −(rX))(rX) −(b +X)X.
51
(We can also eliminate X instead of Q. The two procedures are the same.) (Show
that π is strictly concave in X.) Find the proﬁt maximizing X as a function of a and
b, X = X(a, b).
Determine the sign of the comparative statics
∂X
∂a
and
∂X
∂b
and explain your results
in economics.
Derive the price and the wage rate set by the ﬁrm P and W and compare the results
with that of cases 1 and 2.
7.11 Labor Supply Function
Consider a consumer/worker trying to maximize his utility function subject to the
time constraint that he has only 168 hours a week to spend between work (N) and
leisure (L), N +L = 168, and the budget constraint which equates his consumption
(C) to his wage income (wN), C = wN, as follows:
max U = U(C, L) = U(wN, 168 −N) ≡ f(N, w)
Here N is endogenous and w is exogenous. The FOC requires that the total derivative
of U w.r.t. N be equal to 0.
FOC:
dU
dN
= f
N
= U
C
w +U
L
(−1) = 0.
FOC deﬁnes a relationship between the labor supply of the consumer/worker, N,
and the wage rate w, which is exactly the labor supply function of the individual
N
∗
= N(w). The slope the supply function N
′
(w) is determined by the sign of the
crossderivative f
Nw
U, U
c
, U
L
C ← N
ւ
ւտ
տ
L w
f
Nw
= U
C
+wNU
CC
−NU
LC
The sign of f
Nw
is indeterminate, therefore, the slope of N
∗
= N(w) is also indeter
minate.
Numerical Examples:
Example 1: U = 2
√
C + 2
√
L elasticity of substitution σ > 1
U = 2
√
C+2
√
L = 2
√
wN+2
√
168 −N,
dU
dN
=
w
√
C
−
1
√
L
=
w
√
wN
−
1
√
168 −N
= 0.
52
Therefore, the inverse labor supply function is w = N/(168 − N), which is
positively sloped.
N
`
w
σ > 1
N
`
w
σ < 1
Example 2: U =
CL
C+L
elasticity of substitution σ < 1
U =
CL
C +L
=
wN(168 −N)
wN + 168 −N
,
dU
dN
=
wL
2
−C
2
(C +L)
2
= 0.
Therefore, the inverse labor supply function is w = [(168 − N)/N]
2
, which is
negatively sloped.
Example 3: U = CL CobbDouglas( σ = 1)
U = CL = wN(168 −N)
dU
dN
= w(168 −2N) = 0
The labor supply function is a constant N = 84 and the curve is vertical.
N
`
w
σ = 1
N
`
w
Backwardbending Labor Supply Curve: It is believed that the labor supply
curve can be backwardbending.
7.12 Exponential and logarithmic functions and interest compounding
Exponential function f(x) = e
x
is characterized by f(0) = 1 and f(x) = f
′
(x) =
f
′′
(x) = . . . = f
(n)
(x). The Taylor expansion of the exponential functionat x = 0
becomes
e
x
= 1 +x +x
2
/2! +· · · +x
n
/n! + . . .
Some Rules:
d
dx
(e
x
) = e
x
,
d
dx
e
ax
= ae
ax
,
d
dx
e
f(x)
= f
′
(x)e
f(x)
.
53
The inverse of e
x
is the logarithmic function: ln x. If x = e
y
, then we deﬁne y ≡ lnx.
More generally, if x = a
y
, a > 0, then we deﬁne y ≡ log
a
x =
ln x
ln a
.
Using inverse function rule: dx = e
y
dy,
d
dx
ln x =
1
e
ln x
= 1/x.
x
`
y
y = e
x
y = ln x
a
x
= e
ln a
x
= e
(ln a)x
⇒
d
dx
a
x
= (ln a)e
(ln a)x
= (ln a)a
x
.
d
dx
ln
a
x =
1
ln a
1
x
growth rate of a function of time y = f(t):
growth rate ≡
1
y
dy
dt
=
f
′
f
=
d
dt
[ln f(t)].
Example: f(t) = g(t)h(t). ln f(t) = ln g(t) + ln h(t), therefore, growth rate of f(t) is
equal to the sum of the growth rates of g(t) and h(t).
Interest compounding
A: principal (PV), V = Future Value, r = interest rate, n = number of periods
V = A(1 +r)
n
If we compound interest m times per period, then the interest rate each time becomes
r/m, the number of times of compounding becomes mn, and
V = A[(1 +r/m)
m
]
n
lim
m→∞
(1 +r/m)
m
= 1 +r +r
2
/2! +· · · +r
n
/n! +. . . = e
r
Therefore, V →Ae
rn
, this is the formula for instantaneous compounding.
54
7.13 Timing: (when to cut a tree)
t: number of years to wait, A(t) present value after t years
V (t) = Ke
√
t
: the market value of the tree after t years
A(t) = Ke
√
t
e
−rt
is the present value
We want to ﬁnd the optimal t such that the present value is maximized.
max
t
A(t) = Ke
√
t
e
−rt
.
The FOC is
A
′
=
1
2
√
t
−r
A(t) = 0
Because A(t) = 0, FOC implies:
1
2
√
t
−r = 0, t
∗
= 1/(4r
2
)
For example, if r = 10%, t = 25, then to wait 25 years before cutting the tree is the
optimum.
Suppose that A(t) = e
f(t)
FOC becomes: A
′
(t)/A(t) = f
′
(t) = r, ⇒ f
′
(t) is the instantaneous growth rate (or
the marginal growth rate) at t, ⇒ at t = 25, growth rate = 10% = r, at t = 26,
growth rate < 10%. Therefore, it is better to cut and sell the tree at t = 25 and put
the proceed in the bank than waiting longer.
SOC: It can be shown that A
′′
< 0.
7.14 Problems
1. Suppose the total cost function of a competitive ﬁrm is C(Q) = e
aQ+b
. Derive
the supply function.
2. The input demand function of a competitive producer, X = X(W), can be
derived by maximizing the proﬁt function Π = F(X) − WX with respect to
X, where X is the quantity of input X and W is the price of X. Derive the
comparative statics dX/dW and determine its sign.
3. The utility function of a consumer is given by U = U(X) +M, where X is the
quantity of commodity X consumed and M is money. Suppose that the total
income of the consumer is $ 100 and that the price of X is P. Then the utility
function become U(X) + (100 −XP).
(a) Find the ﬁrst order condition of the utility maximization problem.
(b) What is the behavior equation implied by the ﬁrst order condition?
(c) Derive dX/dP and determine its sign. What is the economic meaning of
your result?
4. The consumption function of a consumer, C = C(Y ), can be derived by max
imizing the utility function U(C, Y ) = u
1
(C) + u
2
(Y − C), where u
′
1
(C) > 0,
u
′
2
(Y −C) > 0 and u
1
”(C) < 0, u
2
”(Y −C) < 0. Derive the comparative statics
dC/dY and determine its sign.
55
5. Consider a duopoly market with two ﬁrms, A and B. The inverse demand func
tion is P = f(Q
A
+ Q
B
), f
′
< 0, f” < 0. The cost function of ﬁrm A is TC
A
= C(Q
A
), C
′
> 0, C” > 0. The proﬁt of ﬁrm A is Π
A
= PQ
A
− TC
A
=
Q
A
f(Q
A
+Q
B
) −C(Q
A
). For a given output of ﬁrm B, Q
B
, there is a Q
A
which
maximizes ﬁrm A’s proﬁt. This relationship between Q
B
and Q
A
is called the
reaction function of ﬁrm A, Q
A
= R
A
(Q
B
).
(a) Find the slope of the reaction function R
′
A
=
dQ
A
dQ
B
.
(b) When Q
B
increases, will ﬁrm A’s output Q
A
increase or decrease?
6. The proﬁt of a monopolistic ﬁrm is given by Π = R(x) − C(x, b), where x is
output, b is the price of oil, R(x) is total revenue, and C(x, b) is the total cost
function. For any given oil price b, there is an optimal
output which maximizes proﬁt, that is, the optimal output is a function of oil
price, x = x(b). Assume that C
bx
= ∂
2
C/∂b∂x > 0, that is, an increase in oil
price will increase marginal cost. Will an increase in oil price increase output,
that is, is dx/db > 0?
7. Consider a monopsony who uses a single input, labor (L), for the production of a
commodity (Q), which he sells in a perfect competition market. His production
function is Q = F(L), (f
′
(L) > 0). The labor supply function is L = L(w), or
more convenient for this problem, w = L
−1
(L) = W(L). Given the commodity
price p, there is an optimal labor input which maximizes the monopsonist’s
total proﬁt Π = pf(L) −W(L)L. In this problem, you are asked to derive the
relation between L and p.
(a) State the FOC and the SOC of the proﬁt maximization problem.
(b) Derive the comparative statics dL/dp and determine its sign.
8. Suppose that a union has a ﬁxed supply of labor (L) to sell, that unemployed
workers are paid unemployment insurance at a rate of $u per worker, and that
the union wishes to maximize the sum of the wage bill plus the unemployment
compensation S = wD(w) +u(L −D(w)), where w is wage per
worker, D(w) is labor demand function, and D
′
(w) < 0. Show that if u in
creases, then the union should set a higher w. (Hint: w is endogenous and u is
exogenous.)
9. The production function of a competitive ﬁrm is given by Q = F(L, K). where
L is variable input and K is ﬁxed input. The short run proﬁt function is given
by Π = pQ−wL−rK, where p is output price, w is wage rate, and r is the rental
rate on capital. In the short run, given the quantity of ﬁxed input K,there is a
L which maximizes Π. Hence, the short run demand for L can be regarded as
a function of K. Assume that F
LK
> 0.
(a) State the FOC and SOC of the proﬁt maximization problem.
(b) Derive the comparative statics dL/dK and determine its sign.
56
7.15 Concavity and Convexity
The derivation of a behavioral equation X = X(Y ) from the maximization problem
max
x
F(X; Y ) is valid only if there exists a unique global maximum for every Y . If
there are multiple global maximum, then X = X(Y ) has multiple values and the
comparative static analysis is not valid. Here we are going to discuss a condition on
F(X; Y ) so that a critical point is always a unique global maximum and the compar
ative static analysis is always valid.
Convex sets
A is a convex set if ∀X
0
, X
1
∈ A and 0 ≤ θ ≤ 1, X
θ
≡ (1 − θ)X
0
+ θX
1
∈ A. (If
X
0
, X
1
∈ A then the whole line connecting X
0
and X
1
is in A.)
x
1
`
x
2
some convex sets
x
1
`
x
2
`
some nonconvex sets
(1) If A
1
and A
2
are convex, then A
1
∩ A
2
is convex but A
1
∪ A
2
is not necessar
ily convex. Also, the empty set itself is a convex set.
(2) The convex hull of A is the smallest convex set that contains A. For example, the
convex hull of {X
0
} ∪ {X
1
} is the straight line connecting X
0
and X
1
.
Convex and concave functions
Given a function F(X), we deﬁne the sets
G
+
F
≡ {(x, y) y ≥ F(x), x ∈ R}, G
−
F
≡ {(x, y) y ≤ F(x), x ∈ R}, G
+
F
, G
−
F
⊂ R
2
.
If G
+
F
(G
−
F
) is a convex set, then we say F(X) is a convex function (a concave func
tion). If F(X) is deﬁned only for nonnegative values X ≥ 0, the deﬁnition is similar.
X
`
F(X)
G
−
F
G
+
F
G
−
F
is a convex set ⇒ F(X) is concave
X
`
F(X)
G
−
F
G
+
F
G
+
F
is a convex set ⇒ F(X) is convex
Equivalent Deﬁnition: Given X
0
< X
1
, 0 ≤ θ ≤ 1, denote F
0
= F(X
0
),
F
1
= F(X
1
). Deﬁne X
θ
≡ (1 − θ)X
0
+ θX
1
, F(X
θ
) = F((1 − θ)X
0
+ θX
1
). Also
57
deﬁne F
θ
≡ (1 −θ)F(X
0
) +θF(X
1
) = (1 −θ)F
0
+θF
1
.
X
`
F(X)
X
0
X
θ
X
1
F
0
= F(X
0
)
F(X
θ
)
F
1
= F(X
1
)
X
`
X
0
X
θ
X
1
F
0
F
θ
F
1
X
θ
−X
0
X
1
−X
θ
=
θ(X
1
−X
0
)
(1 −θ)(X
1
−X
0
)
=
θ
1 −θ
,
F
θ
−F
0
F
1
−F
θ
=
θ(F
1
−F
0
)
(1 −θ)(F
1
−F
0
)
=
θ
1 −θ
.
Therefore, (X
θ
, F
θ
) is located on the straight line connecting (X
0
, F
0
) and (X
1
, F
1
)
and when θ shifts from 0 to 1, (X
θ
, F
θ
) shifts from (X
0
, F
0
) to (X
1
, F
1
) (the right
ﬁgure). On the other hand, (X
θ
, F(X
θ
)) shifts along the curve representing the graph
of F(X) (the left ﬁgure). Put the two ﬁgures together:
X
`
F(X)
X
0
X
θ
X
1
F
θ
F(X
θ
)
F(X) is strictly concave ⇒ if for all X
0
, X
1
and θ ∈ (0, 1), F(X
θ
) > F
θ
.
F(X) is concave ⇒ if for all X
0
, X
1
and θ ∈ [0, 1], F(X
θ
) ≥ F
θ
.
X
`
F(X)
F(X) is strictly
concave
X
`
F(X)
F(X) is concave
(not strictly)
Notice that these concepts are global concepts. (They have something to do with the
whole graph of F, not just the behavior of F nearby a point.) The graph of a concave
function can have a ﬂat part. For a strictly concave function, the graph should be
curved everywhere except at kink points.
F(X) is strictly convex ⇒ if for all X
0
, X
1
and θ ∈ (0, 1), F(X
θ
) < F
θ
.
F(X) is convex ⇒ if for all X
0
, X
1
and θ ∈ [0, 1], F(X
θ
) ≤ F
θ
.
58
X
`
F
F(X) is strictly
convex
X
`
F
H
F(X) is convex
(not strictly)
Remark 1: A linear function is both concave and convex since F
θ
≡ F(X
θ
).
X
`
F
>
>
>
>
>
>
>
>
>
X
`
F
/
/
/
/
/
/ /
.
.
.
.
a concave
piecewiselinear
function
X
`
F
/
/
/
/
/
/ /
a convex
piecewiselinear
function
Remark 2: A piecewiselinear function consists of linear components; for example,
the income tax schedule T = f(Y ) is a piecewiselinear function. Other examples are
concave F(X) =
2X X ≤ 1
1 +X X > 1
convex F(X) =
X X ≤ 1
2X −1 X > 1
In the following theorems, we assume that F
′′
(X) exists for all X.
Theorem 1: F(X) is concave, ⇔ F
′′
(X) ≤ 0 for all X.
F
′′
(X) < 0 for all X ⇒ F(X) is strictly concave.
Proof: By Taylor’s theorem, there exist
¯
X
0
∈ [X
0
, X
θ
] and
¯
X
1
∈ [X
θ
, X
1
] such that
F(X
1
) = F(X
θ
) +F
′
(X
θ
)(X
1
−X
θ
) +
1
2
F
′′
(
¯
X
1
)(X
1
−X
θ
)
2
F(X
0
) = F(X
θ
) +F
′
(X
θ
)(X
0
−X
θ
) +
1
2
F
′′
(
¯
X
0
)(X
0
−X
θ
)
2
⇒ F
θ
= F(X
θ
) +
1
2
θ(1 −θ)(X
1
−X
0
)
2
[F
′′
(
¯
X
0
) +F
′′
(
¯
X
1
)].
Theorem 2: If F(X) is concave and F
′
(X
0
) = 0, then X
0
is a global maximum.
If F(X) is strictly concave and F
′
(X
0
) = 0, then X
0
is a unique global maximum.
Proof: By theorem 1, X
0
must be a local maximum. If it is not a global maximum,
then there exists X
1
such that F(X
1
) > F(X
0
), which implies that F(X
θ
) > F(X
0
)
for θ close to 0. Therefore, X
0
is not a local maximum, a contradiction.
Remark 1 (boundary/corner solution): The boundary or corner condition F
′
(X) ≤ 0,
59
XF
′
(X) = 0 (or F
′
(X) ≥ 0, (X − a)F
′
(X) = 0) becomes suﬃcient for global maxi
mum.
Remark 2 (minimization problem): For the minimization problem, we replace con
cavity with convexity and F
′′
(X) < 0 with F
′′
(X) > 0. If F(X) is convex and
F
′
(X
∗
) = 0, then X
∗
is a global minimum.
If F(X) is strictly convex and F
′
(X
∗
) = 0, then X
∗
is a unique global minimum.
Remark 3: The sum of two concave functions is concave. The product of two concave
function is not necesarily concave. X
a
is strictly concave if a < 1, strictly convex if
a > 1. e
X
is strictly convex and ln X is strctly concave with X > 0.
X
`
F
X
a
, a > 1
X
`
F
X
a
, 0 < a < 1
X
`
F
X
a
, a < 0
X
`
F
e
aX
X
`
F
ln X
X
`
F
−aX
2
+bX +c
Remark 4: A concave function does not have to be diﬀerentiable, but it must be
continuous on the interior points.
7.16 Indeterminate forms and L’Hˆopital’s rule
Let f(x) =
g(x)
h(x)
, g(a) = h(a) = 0 and g(x) and h(x) be continuous at x = a. f(a) is
not deﬁned because it is
0
0
. However, lim
x→a
f(x) can be calculated.
lim
x→a
f(x) = lim
x→a
g(x)
h(x)
= lim
x→a
g(x) −g(a)
h(x) −h(a)
= lim
∆x→0
g(a + ∆x) −g(a)
h(a + ∆x) −h(a)
=
g
′
(a)
h
′
(a)
.
The same procedure also works for the case with g(a) = h(a) = ∞.
Example 1: f(x) =
g(x)
h(x)
=
ln[(a
x
+ 1)/2]
x
,
60
g(0) = h(0) = 0, h
′
(x) = 1, g
′
(x) =
(ln a)a
x
a
x
+ 1
.
⇒ h
′
(0) = 1 and g
′
(0) =
ln a
2
, ⇒ lim
x→0
f(x) =
ln a
2
.
Example 2: f(x) =
g(x)
h(x)
=
x
e
x
, h
′
(x) = e
x
, g
′
(x) = 1.
⇒ h
′
(∞) = ∞ and g
′
(∞) = 1, ⇒ lim
x→∞
f(x) =
1
∞
= 0.
Example 3: f(x) =
g(x)
h(x)
=
ln x
x
, h
′
(x) = 1, g
′
(x) =
1
x
.
⇒ lim
x→0
+ h
′
(x) = 1 and lim
x→0
+ g
′
(x) = ∞, ⇒ lim
x→0
+ f(x) =
1
∞
= 0.
7.17 Newton’s method
We can approximate a root x
∗
of a nonlinear equation f(x) = 0 using an algorithm
called Newton’s method.
x
`
x
n
x
n+1
x
∗
f(x
n
)
`
`
`
`
`
f
′
(x
n
) =
f(x
n
)
x
n
−x
n+1
Recursive formula: x
n+1
= x
n
−
f(x
n
)
f
′
(x
n
)
.
If f(x) is not too strange, we will have lim
n→∞
x
n
= x
∗
. Usually, two or three
steps would be good enough.
Example: f(x) = x
3
−3, f
′
(x) = 3x
2
, x
n+1
= x
n
−
x
3
−3
3x
2
= x
n
−
x
3
+
1
x
2
.
Starting x
0
= 1, x
1
= 1 −
1
3
+ 1 =
5
3
≈ 1.666. x
2
=
5
3
−
5
9
+
9
25
=
331
225
≈ 1.47. The
true value is x
∗
=
3
√
3 ≈ 1.44225.
61
8 Optimization–Multivariate Case
Suppose that the objective function has n variables: max
x
1
,...,xn
F(x
1
, . . . , x
n
) = F(X).
A local maximum X
∗
= (x
∗
1
, . . . , x
∗
n
): ∃ǫ > 0 such that F(X
∗
) ≥ F(X) for all X
satisfying x
i
∈ (x
∗
i
−ǫ, x
∗
i
+ǫ) ∀i.
A critical point X
c
= (x
c
1
, . . . , x
c
n
):
∂F(X
c
)
∂x
i
= 0 ∀i.
A global maximum X
∗
: F(X
∗
) ≥ F(X) ∀X.
A unique global maximum X
∗
: F(X
∗
) > F(X) ∀X = X
∗
.
The procedure is the same as that of the single variable case. (1) Use FOC to ﬁnd
critical points; (2) use SOC to check whether a critical point is a local maximum, or
show that F(X) is concave so that a critical point must be a global maximum.
If we regard variables other than x
i
as ﬁxed, then it is a single variable maximization
problem and, therefore, we have the necessary conditions
FOC:
∂F
∂x
i
= 0 and SOC:
∂
2
F
∂x
2
i
< 0, i = 1, . . . , n.
However, since there are n×n second order derivatives (the Hessian matrix H(F)) and
the SOC above does not consider the crossderivatives, we have a reason to suspect
that the SOC is not suﬃcient. In the next section we will provide a counterexample
and give a true SOC.
8.1 SOC
SOC of variablewise maximization is wrong
Example: max
x
1
,x
2
F(x
1
, x
2
) = −x
2
1
+ 4x
1
x
2
−x
2
2
.
FOC: F
1
= −2x
1
+ 4x
2
= 0, F
2
= 4x
1
−2x
2
= 0, ⇒x
1
= x
2
= 0.
SOC? F
11
= F
22
= −2 < 0.
x
1
`
x
2
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
4
1
4
1
4
1
4
1
(0,0) is
a saddle point.
x
i
`
F
When x
j
= 0,
x
i
= 0 is
a maximum.
62
F(0, 0) = 0 < F(k, k) = 2k
2
for all k = 0. ⇒(0, 0) is not a local maximum!
The true SOC should take into consideration the possibility that when x
1
and x
2
increase simultaneously F may increase even if individual changes cannot increase F.
SOC of sequential maximization:
We can solve the maximization problem sequentially. That is, for a given x
2
, we
can ﬁnd a x
1
such that F(x
1
, x
2
) is maximized. The FOC and SOC are F
1
= 0 and
F
11
< 0. The solution depends on x
2
, x
1
= h(x
2
), i.e., we regard x
1
as endogenous
variable and x
2
as an exogenous variable in the ﬁrst stage. Using implicit function
rule, dx
1
/dx
2
= h
′
(x
2
) = −F
12
/F
11
. In the second stage we maximize M(x
2
) ≡
F(h(x
2
), x
2
). The FOC is M
′
(x
2
) = F
1
h
′
+F
2
= F
2
= 0 (since F
1
= 0). The SOC is
M
′′
(x
2
) = F
1
h
′′
+F
11
(h
′
)
2
+ 2F
12
h
′
+F
22
= (−F
2
12
+F
11
F
22
)/F
11
< 0.
Therefore, the true SOC is: F
11
< 0 and F
11
F
22
−F
2
12
> 0.
For the nvariable case, the sequential argument is more complicated and we use Tay
lor’s expansion.
SOC using Taylor’s expansion:
To ﬁnd the true SOC, we can also use Taylor’s theorem to expand F(X) around a
critical point X
∗
:
F(X) = F(X
∗
) +F
1
(X
∗
)(x
1
−x
∗
1
) +F
2
(X
∗
)(x
2
−x
∗
2
)
+
1
2
(x
1
−x
∗
1
, x
2
−x
∗
2
)
F
11
(X
∗
) F
12
(X
∗
)
F
21
(X
∗
) F
22
(X
∗
)
x
1
−x
∗
1
x
2
−x
∗
2
+ higher order terms.
Since X
∗
is a critical point, F
1
(X
∗
) = F
2
(X
∗
) = 0 and we have the approximation
for X close to X
∗
:
F(X) −F(X
∗
) ≈
1
2
(x
1
−x
∗
1
, x
2
−x
∗
2
)
F
11
(X
∗
) F
12
(X
∗
)
F
21
(X
∗
) F
22
(X
∗
)
x
1
−x
∗
1
x
2
−x
∗
2
≡
1
2
v
′
H
∗
v.
True SOC: v
′
H
∗
v < 0 for all v = 0.
In the next section, we will derive a systematic method to test the true SOC.
8.2 Quadratic forms and their signs
A quadratic form in n variables is a second degree homogenous function.
f(v
1
, . . . , v
n
) =
n
¸
i=1
n
¸
j=1
a
ij
v
i
v
j
= (v
1
, . . . , v
n
)
¸
¸
a
11
· · · a
1n
.
.
.
.
.
.
.
.
.
a
n1
· · · a
nn
¸
¸
v
1
.
.
.
v
n
≡ v
′
Av.
Since v
i
v
j
= v
j
v
i
, we can assume that a
ij
= a
ji
so that A is symmetrical.
Example 1: (v
1
−v
2
)
2
= v
2
1
−2v
1
v
2
+v
2
2
= (v
1
, v
2
)
1 −1
−1 1
v
1
v
2
.
63
Example 2: 2v
1
v
2
= (v
1
, v
2
)
0 1
1 0
v
1
v
2
.
Example 3: x
2
1
+ 2x
2
2
+ 6x
1
x
3
= (x
1
, x
2
, x
3
)
¸
1 0 3
0 2 0
3 0 0
¸
x
1
x
2
x
3
.
v
′
Av is called negative semideﬁnite if v
′
Av ≤ 0 for all v ∈ R
n
.
v
′
Av is called negative deﬁnite if v
′
Av < 0 for all v = 0.
v
′
Av is called positive semideﬁnite if v
′
Av ≥ 0 for all v ∈ R
n
.
v
′
Av is called positive deﬁnite if v
′
Av > 0 for all v = 0.
(1) −v
2
1
−2v
2
2
< 0, if v
1
= 0 or v
2
= 0 ⇒ negative.
(2) −(v
1
−v
2
)
2
≤ 0, = 0 if v
1
= v
2
⇒ negative semideﬁnite.
(3) (v
1
+v
2
)
2
≥ 0, = 0 if v
1
= −v
2
⇒ positive semideﬁnite.
(4) v
2
1
+v
2
2
> 0, if v
1
= 0 or v
2
= 0 ⇒ positive.
(5) v
2
1
−v
2
2
>
<
0 if v
1

>
<
v
2
 ⇒ neither positive nor negative deﬁnite.
(6) v
1
v
2
>
<
0 if sign(v
1
)
=
=
sign(v
2
) ⇒ neither positive nor negative deﬁnite.
Notice that if v
′
Av is negative (positive) deﬁnite then it must be negative (positive)
semideﬁnite.
Testing the sign of a quadratic form:
n = 2: v
′
Av = (v
1
, v
2
)
a
11
a
12
a
21
a
22
v
1
v
2
= a
11
v
2
1
+ 2a
12
v
1
v
2
+a
22
v
2
2
=
a
11
v
2
1
+ 2
a
12
a
11
v
1
v
2
+
a
2
12
a
2
11
v
2
2
+
−
a
2
12
a
11
+a
22
v
2
2
= a
11
v
1
+
a
12
a
11
v
2
2
+
a
11
a
12
a
21
a
22
a
11
v
2
2
.
Negative deﬁnite ⇔ a
11
< 0 and
a
11
a
12
a
21
a
22
= a
11
a
22
−a
2
12
> 0.
Positive deﬁnite ⇔ a
11
> 0 and
a
11
a
12
a
21
a
22
= a
11
a
22
−a
2
12
> 0.
For negative or positive semideﬁnite, replace strict inequalities with semiinequalities.
The proofs for them are more diﬃcult and discussed in Lecture 13.
Example 1: F(v
1
, v
2
) = −2v
2
1
+ 8v
1
v
2
− 2v
2
2
= (v
1
, v
2
)
−2 4
4 −2
v
1
v
2
, a
11
=
−2 < 0 but
−2 4
4 −2
= −12 < 0 ⇒ neither positive nor negative. The matrix is
the Hessian of F of the counterexample in section 1. Therefore, the counterexample
violates the SOC.
64
General n: v
′
Av = (v
1
, . . . , v
n
)
¸
¸
a
11
· · · a
1n
.
.
.
.
.
.
.
.
.
a
n1
· · · a
nn
¸
¸
v
1
.
.
.
v
n
= a
11
(v
1
+· · ·)
2
+
a
11
a
12
a
21
a
22
a
11
(v
2
+· · ·)
2
+
a
11
a
12
a
13
a
21
a
22
a
23
a
31
a
32
a
33
a
11
a
12
a
21
a
22
(v
3
+· · ·)
2
+· · ·+
a
11
· · · a
1n
.
.
.
.
.
. · · ·
a
n1
· · · a
nn
a
11
· · · a
1(n−1)
.
.
.
.
.
.
.
.
.
a
(n−1)1
· · · a
(n−1)(n−1)
v
2
n
.
kth order principle minor of A: A
(k)
≡
¸
¸
a
11
· · · a
1k
.
.
.
.
.
.
.
.
.
a
k1
· · · a
kk
, k = 1, . . . , n.
Negative deﬁnite ⇔ A
(1)
= a
11
< 0,
A
(2)

A
(1)

< 0,
A
(3)

A
(2)

< 0, . . .,
A
(n)

A
(n−1)

< 0.
Positive deﬁnite ⇔ A
(1)
= a
11
> 0,
A
(2)

A
(1)

> 0,
A
(3)

A
(2)

> 0, . . .,
A
(n)

A
(n−1)

> 0.
Negative deﬁnite ⇔ A
(1)
 = a
11
< 0, A
(2)
 =
a
11
a
12
a
21
a
22
> 0, A
(3)
 =
a
11
a
12
a
13
a
21
a
22
a
23
a
31
a
32
a
33
< 0, · · ·, (−1)
n
A
(n)
 = (−1)
n
a
11
· · · a
1n
.
.
.
.
.
.
.
.
.
a
n1
· · · a
nn
> 0.
Positive deﬁnite ⇔ A
(1)
 = a
11
> 0, A
(2)
 =
a
11
a
12
a
21
a
22
> 0, A
(3)
 =
a
11
a
12
a
13
a
21
a
22
a
23
a
31
a
32
a
33
> 0, · · ·, A
(n)
 =
a
11
· · · a
1n
.
.
.
.
.
.
.
.
.
a
n1
· · · a
nn
> 0.
The conditions for semideﬁnite are more complicated and will be discussed in Lecture
13 using the concept of eigenvalues of a square matrix.
8.3 SOC again
From the last two sections, the SOC for a local maximum can be summarized as:
SOC: v
′
H
F
(X
∗
)v is negative deﬁnite ⇒ F
11
(X
∗
) < 0,
F
11
(X
∗
) F
12
(X
∗
)
F
21
(X
∗
) F
22
(X
∗
)
=
F
11
F
22
−F
2
12
> 0,
F
11
(X
∗
) F
12
(X
∗
) F
13
(X
∗
)
F
21
(X
∗
) F
22
(X
∗
) F
23
(X
∗
)
F
31
(X
∗
) F
32
(X
∗
) F
33
(X
∗
)
< 0, . . .
Example: max F(x
1
, x
2
) = 3x
1
+ 3x
1
x
2
− 3x
2
1
− x
3
2
. (A cubic function in 2 vari
ables.)
65
FOC: F
1
= 3 + 3x
2
−6x
1
= 0 and F
2
= 3x
1
−3x
2
2
= 0.
Two critical points:
x
1
x
2
=
¸
¸
1
4
−
1
2
and
1
1
.
Hessian matrix: H(x
1
, x
2
) =
F
11
(X) F
12
(X)
F
21
(X) F
22
(X)
=
−6 3
3 −6x
2
; H
1
(X) =
F
11
(X) = −6 < 0, H
2
(X) =
−6 3
3 −6x
2
= 36x
2
−9.
H
2
(
1
4
, −
1
2
) = −27 < 0 ⇒
¸
¸
1
4
−
1
2
is not a local max.
H
2
(1, 1) = 27 > 0 ⇒
1
1
is a local max. F(1, 1) = 2. It is not a global maximum
because F→∞ when x
2
→−∞.
Remark 1: F
11
< 0 and F
11
F
22
−F
2
12
> 0 together implies F
22
< 0.
Remark 2: If H
k
(X
∗
) = 0 for some 1 ≤ k ≤ n then X
∗
is a degenerate critical point.
Although we can still check whether v
′
Hv is negative semideﬁnite, it is insuﬃcient for
a local maximum. We have to check the third or higher order derivatives to determine
whether X
∗
is a local maximum. It is much more diﬃcult than the single variable
case with F
′′
= 0.
8.4 Joint products
A competitive producer produces two joint products. (eg., gasoline and its by prod
ucts or cars and tructs, etc.)
Cost function: C(q
1
, q
2
).
Proﬁt function: Π(q
1
, q
2
; p
1
, p
2
) = p
1
q
1
+p
2
q
2
−C(q
1
, q
2
).
FOC: Π
1
= p
1
−C
1
= 0, Π
2
= p
2
−C
2
= 0; or p
i
= MC
i
.
SOC: Π
11
= −C
11
< 0,
−C
11
−C
12
−C
21
−C
22
≡ ∆ > 0.
Example: C(q
1
, q
2
) = 2q
2
1
+ 3q
2
2
−q
1
q
2
FOC: p
1
− 4q
1
+ q
2
= 0, p
2
+ q
1
− 6q
2
= 0 ⇒ supply functions q
1
= (6p
1
+ p
2
)/23
and q
2
= (p
1
+ 4p
2
)/23.
SOC: −C
11
= −4 < 0,
−C
11
−C
12
−C
21
−C
22
=
−4 1
1 −6
= 23 > 0.
Comparative statics:
Total diﬀerential of FOC:
C
11
C
12
C
21
C
22
dq
1
dq
2
=
dp
1
dp
2
⇒
¸
¸
∂q
1
∂p
1
∂q
1
∂p
2
∂q
2
∂p
1
∂q
2
∂p
2
=
1
∆
C
22
−C
12
−C
21
C
11
.
66
SOC ⇒C
11
> 0, C
22
> 0, ∆ > 0 ⇒
∂q
1
∂p
1
> 0,
∂q
2
∂p
2
> 0.
∂q
1
∂p
2
and
∂q
2
∂p
1
are positive if the joint products are beneﬁcially to each other in the
production so that C
12
< 0.
8.5 Monopoly price discrimination
A monopoly sells its product in two separable markets.
Cost function: C(Q) = C(q
1
+q
2
)
Inverse market demands: p
1
= f
1
(q
1
) and p
2
= f
2
(q
2
)
Proﬁt function: Π(q
1
, q
2
) = p
1
q
1
+p
2
q
2
−C(q
1
+q
2
) = q
1
f
1
(q
1
) +q
2
f
2
(q
2
) −C(q
1
+q
2
)
FOC: Π
1
= f
1
(q
1
)+q
1
f
′
1
(q
1
)−C
′
(q
1
+q
2
) = 0, Π
2
= f
2
(q
2
)+q
2
f
′
2
(q
2
)−C
′
(q
1
+q
2
) = 0;
or MR
1
= MR
2
= MC.
SOC: Π
11
= 2f
′
1
+q
1
f
′′
1
−C
′′
< 0,
2f
′
1
+q
1
f
′′
1
−C
′′
−C
′′
−C
′′
2f
′
2
+q
2
f
′′
2
−C
′′
≡ ∆ > 0.
Example: f
1
= a −bq
1
, f
2
= α −βq
2
, and C(Q) = 0.5Q
2
= 0.5(q
1
+q
2
)
2
.
f
′
1
= −b, f
′
2
= −β, f
′′
1
= f
′′
2
= 0, C
′
= Q = q
1
+q
2
, and C
′′
= 1.
FOC: a −2bq
1
= q
1
+q
2
= α −2βq
2
⇒
1 + 2b 1
1 1 + 2β
q
1
q
2
=
a
α
⇒
q
1
q
2
=
1
(1 + 2b)(1 + 2β) −1
a(1 + 2β) −α
α(1 + 2b) −a
.
SOC: −2b −1 < 0 and ∆ = (1 + 2b)(1 + 2β) −1 > 0.
q
`
p
MC
MR
1
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
MR
2
MR
1+2
Q
∗
q
∗
2
q
∗
1
MC
∗
MC = MR
1+2
⇒Q
∗
, MC
∗
MC
∗
= MR
1
⇒q
∗
1
MC
∗
= MR
2
⇒q
∗
2
8.6 SR supply vs LR supply  Le Chˆatelier principle
A competitive producer employs a variable input x
1
and a ﬁxed input x
2
. Assume
that the input prices are both equal to 1, w
1
= w
2
= 1.
Production function: q = f(x
1
, x
2
), assume MP
i
= f
i
> 0, f
ii
< 0, f
ij
> 0,
67
f
11
f
22
> f
2
12
.
Proﬁt function: Π(x
1
, x
2
; p) = pf(x
1
, x
2
) −x
1
−x
2
.
Shortrun problem (only x
1
can be adjusted, x
2
is ﬁxed):
SR FOC: Π
1
= pf
1
−1 = 0, or w
1
= VMP
1
. SOC: Π
11
= pf
11
< 0,
Comparative statics: f
1
dp+pf
11
dx
1
= 0 ⇒
dx
1
dp
=
−f
1
pf
11
> 0,
dq
s
dp
=
−f
2
1
pf
11
=
−1
p
3
f
11
> 0.
Longrun problem (both x
1
and x
2
can be adjusted):
LR FOC: Π
1
= pf
1
−1 = 0, Π
2
= pf
2
−1 = 0; or w
i
= VMP
i
.
SOC: Π
11
= pf
11
< 0,
pf
11
pf
12
pf
21
pf
22
≡ ∆ > 0.
Comparative statics:
f
1
dp
f
2
dp
+
pf
11
pf
12
pf
21
pf
22
dx
1
dx
2
=
0
0
⇒
dx
1
/dp
dx
2
/dp
=
−1
p(f
11
f
22
−f
2
12
)
f
1
f
22
−f
2
f
12
f
2
f
11
−f
1
f
21
,
dq
L
dp
=
−(f
11
+f
22
−2f
12
)
p
3
(f
11
f
22
−f
2
12
)
Le Chˆatelier principle:
dq
L
dp
>
dq
s
dp
.
Example: f(x
1
, x
2
) = 3x
1/3
1
x
1/3
2
(homogenous of degree 2/3)
LR FOC: px
−2/3
1
x
1/3
2
= 1 = px
1/3
1
x
−2/3
2
⇒x
1
= x
2
= p
3
q
L
= 3p
2
.
SR FOC (assuming ¯ x
2
= 1): px
−2/3
1
= 1 ⇒x
1
= p
3/2
q
s
= 3p
1/2
.
η
L
(LR supply elasticity) = 2 > η
s
(SR supply elasticity) = 1/2.
q
`
p
S
s
(p) = 3p
0.5
S
L
(p) = 3p
2
p
`
π
>
>
>
>
>
>
>
>
>
>
π
L
(p)
π
s
(p; ¯ x
2
)
h(p)
¯ p
Envelop theorem, Hotelling’s lemma, and Le Chˆatelier principle
From the SR problem we ﬁrst derive the SR variable input demand function x
1
=
x
1
(p, ¯ x
2
).
Then the SR supply function is obtained by substituting into the production function
q
s
= f(x
1
(p, ¯ x
2
), ¯ x
2
) ≡ S
s
(p; ¯ x
2
).
The SR proﬁt function is π
s
(p, ¯ x
2
) = pq
s
−x
1
(p, ¯ x
2
) − ¯ x
2
.
Hotelling’s lemma: By envelop theorem,
∂π
s
∂p
= S
s
(p, ¯ x
2
).
From the LR problem we ﬁrst derive the input demand functions x
1
= x
1
(p) and
x
2
= x
2
(p).
Then the LR supply function is obtained by substituting into the production function
68
q
L
= f(x
1
(p), x
2
(p)) ≡ S
L
(p).
The LR proﬁt function is π
L
(p) = pq
L
−x
1
(p) −x
2
(p).
(Also Hotelling’s lemma) By envelop theorem,
∂π
L
∂p
= S
L
(p).
Notice that π
L
(p) = π
s
(p; x
2
(p)).
Let ¯ x
2
= x
2
(¯ p), deﬁne h(p) ≡ π
L
(p) − π
s
(p, ¯ x
2
). h(p) ≥ 0 because in the LR,
the producer can adjust x
2
to achieve a higher proﬁt level.
Also, h(¯ p) = 0 because π
L
(¯ p) = π
s
(¯ p; x
2
(¯ p)) = π
s
(¯ p; ¯ x
2
).
Therefore, h(p) has a minimum at p = ¯ p and the SOC is h
′′
(¯ p) > 0 =
∂
2
π
L
∂p
2
−
∂
2
π
s
∂p
2
> 0,
which implies
Le Chˆatelier principle:
dq
L
dp
−
dq
s
dp
> 0.
8.7 Concavity and Convexity
Similar to the single variable case, we deﬁne the concepts of concavity and convexity
for 2variable functions F(X) = F(x
1
, x
2
) by deﬁning G
+
F
, G
−
F
⊂ R
2
s follows.
G
+
F
≡ {(x
1
, x
2
, y) y ≥ F(x
1
, x
2
), (x
1
, x
2
) ∈ R
2
}, G
−
F
≡ {(x
1
, x
2
, y) y ≤ F(x
1
, x
2
), (x
1
, x
2
) ∈ R
2
}.
If G
+
F
(G
−
F
) is a convex set, then we say F(X) is a convex function (a concave func
tion). If F(X) is deﬁned only for nonnegative values x
1
, x
2
≥ 0, the deﬁnition is
similar. (The extension to nvariable functions is similar.)
Equivalently, given X
0
=
x
0
1
x
0
2
, X
1
=
x
1
1
x
1
2
, 0 ≤ θ ≤ 1, F
0
≡ F(X
0
),
F
1
≡ F(X
1
), we deﬁne
X
θ
≡ (1 − θ)X
0
+ θX
1
=
(1 −θ)x
0
1
+θx
1
1
(1 −θ)x
0
2
+θx
1
2
≡
x
θ
1
x
θ
2
, F(X
θ
) = F((1 − θ)X
0
+
θX
1
)
F
θ
≡ (1 −θ)F(X
0
) +θF(X
1
) = (1 −θ)F
0
+θF
1
.
x
1
`
x
2
X
0
X
θ
X
1
x
0
1
x
θ
1
x
1
1
x
0
2
x
θ
2
x
1
2
X
θ
is located on the
straight line connecting X
0
and X
1
,
when θ shifts from 0 to 1,
X
θ
shifts from X
0
to X
1
.
On 3dimensional (x
1
–x
2
–F) space, (X
θ
, F
θ
) is located on the straight line connecting
(X
0
, F
0
) and (X
1
, F
1
), when θ shifts from 0 to 1, (X
θ
, F
θ
) shifts from (X
0
, F
0
) to
(X
1
, F
1
). On the other hand, (X
θ
, F(X
θ
)) shifts along the surface representing the
graph of F(X).
69
F(X) is strictly concave ⇒ if for all X
0
, X
1
and θ ∈ (0, 1), F(X
θ
) > F
θ
.
F(X) is concave ⇒ if for all X
0
, X
1
and θ ∈ [0, 1], F(X
θ
) ≥ F
θ
.
F(X) is strictly convex ⇒ if for all X
0
, X
1
and θ ∈ (0, 1), F(X
θ
) < F
θ
.
F(X) is convex ⇒ if for all X
0
, X
1
and θ ∈ [0, 1], F(X
θ
) ≤ F
θ
.
Example: 9x
1/3
1
x
1/3
2
.
Assume that F is twice diﬀerentiable.
Theorem 1: F(X) is concave, ⇔ v
′
H
F
v is negative semideﬁnite for all X.
v
′
H
F
v is negative deﬁnite for all X ⇒ F(X) is strictly concave.
Proof: By Taylor’s theorem, there exist
¯
X
0
∈ [X
0
, X
θ
] and
¯
X
1
∈ [X
θ
, X
1
] such that
F(X
1
) = F(X
θ
) +∇F(X
θ
)(X
1
−X
θ
) +
1
2
(X
1
−X
θ
)
′
H
F
(
¯
X
1
)(X
1
−X
θ
)
F(X
0
) = F(X
θ
) +∇F(X
θ
)(X
0
−X
θ
) +
1
2
(X
0
−X
θ
)
′
H
F
(
¯
X
0
)(X
0
−X
θ
)
⇒ F
θ
= F(X
θ
) +
θ(1 −θ)
2
(X
1
−X
0
)
′
[H
F
(
¯
X
0
) +H
F
(
¯
X
1
)](X
1
−X
0
).
Theorem 2: If F(X) is concave and ∇F(X
0
) = 0, then X
0
is a global maximum.
If F(X) is strictly concave and ∇F(X
0
) = 0, then X
0
is a unique global maximum.
Proof: By theorem 1, X
0
must be a local maximum. If it is not a global maximum,
then there exists X
1
such that F(X
1
) > F(X
0
), which implies that F(X
θ
) > F(X
0
)
for θ close to 0. Therefore, X
0
is not a local maximum, a contradiction.
Remark 1 (boundary or corner solution): The boundary or corner condition F
i
(X) ≤
0, x
i
F
i
(X) = 0 (or F
i
(X) ≥ 0, (x
i
− a
i
)F
i
(X) = 0) becomes suﬃcient for global
maximum.
Remark 2 (minimization problem): For the minimization problem, we replace con
cavity with convexity and negative deﬁnite with positive deﬁnite. If F(X) is convex
and ∇F(X
∗
) = 0, then X
∗
is a global minimum.
If F(X) is strictly convex and ∇F(X
∗
) = 0, then X
∗
is a unique global minimum.
8.8 Learning and utility maximization
Consumer B’s utility function is
U = u(x, k) +m−h(k), x, m, k ≥ 0,
where x is the quantity of commodity X consumed, k is B’s knowledge regard
ing the consumption of X, m is money, u(x, k) is the utility obtained,
∂
2
u
∂x
2
< 0,
∂
2
u
∂x∂k
> 0 (marginal utility of X increases with k), and h(k) is the disutility of acquir
ing/maintaining k, h
′
> 0, h
′′
> 0. Assume that B has 100 dollar to spend and the
price of X is P
x
= 1 so that m = 100 −x and U = u(x, k) + 100 −x −h(k). Assume
70
also that k is ﬁxed in the short run. The short run utility maximization problem
is
max
x
u(x, k) + 100 −x −h(k), ⇒FOC:
∂u
∂x
−1 = 0, SOC:
∂
2
u
∂x
2
< 0.
The short run comparative static
dx
dk
is derived from FOC as
dx
dk
= −
∂
2
u/∂x
2
∂
2
u/∂x∂k
> 0,
that is, consumer B will consume more of X if B’s knowledge of X increases.
In the long run consumer B will change k to attain higher utility level. The long run
utility maximization problem is
max
x,k
u(x, k) + 100 −x −h(k), ⇒FOC:
∂u
∂x
−1 = 0,
∂u
∂k
−h
′
(k) = 0.
The SOC is satisﬁed if we assume that u(x, k) is concave and h(k) is convex (h
′′
(k) >
0), because F(x, k) ≡ u(x, k) + 100 −x −h(k), x, k ≥ 0 is strictly concave then.
Consider now the speciﬁc case when
u(x, k) = 3x
2/3
k
1/3
and h(k) = 0.5k
2
.
1. Calculate consumer B’s short run consumption of X, x
s
= x(k). (In this part,
you may ignore the nonnegative constraint m = 100 −x ≥ 0 and the possibility
of a corner solution.)
x = 1/(8k).
2. Calculate the long run consumption of X, x
L
.
x = 32
3. Derive the short run value function V (k) ≡ u(x(k), k) + 100 −x(k) −h(k).
4. Solve the maximization problem max
k
V (k).
k = 4.
5. Explain how the SOC is satisﬁed and why the solution is the unique global
maximum.
(demand functions) Consider now the general case when P
x
= p so that
U = 3x
2/3
k
1/3
+ 100 −px −0.5k
2
.
1. Calculate consumer B’s short run demand function of X, x
s
= x(p; k). (Warn
ing: There is a nonnegative constraint m = 100 − px ≥ 0 and you have to
consider both interior and corner cases.)
x
s
(p) = 8kp
−3
(100/p) if p ≤
√
2k/5 (p <
√
2k/5).
2. Calculate the long run demand function of X, x
L
(p). and the optimal level of
K, k
L
(p). (Both interior and corner cases should be considered too.)
x
L
(p) = 32p
−5
(100/p) if p > (8/25)
1/4
(p < (8/25)
1/4
), k
L
(p) = [x
L
(p)]
2/5
.
71
8.9 Homogeneous and homothetic functions
A homogeneous function of degree k is f(x
1
, . . . , x
n
) such that
f(hx
1
, . . . , hx
n
) = h
k
f(x
1
, . . . , x
n
) ∀h > 0.
If f is homogeneous of degree k
1
and g homogeneous of degree k
2
, then fg is homoge
neous of degree k
1
+k
2
,
f
g
is homogeneous of degree k
1
−k
2
, and f
m
is homogeneous
of degree mk
1
.
If Q = F(x
1
, x
2
) is homogeneous of degree 1, then Q = x
1
F(1,
x
2
x
1
) ≡ x
1
f
x
2
x
1
. If
m = H(x
1
, x
2
) is homogeneous of degree 0, then m = H(1,
x
2
x
1
) ≡ h
x
2
x
1
.
Euler theorem: If f(x
1
, . . . , x
n
) is homogeneous of degree k, then
x
1
f
1
+. . . +x
n
f
n
= kf
If f is homogeneous of degree k, then f
i
is homogenous of degree k −1.
Examples: 1. CobbDouglas function Q = Ax
α
1
x
β
2
is homogeneous of degree α +β.
2. CES function Q = {ax
ρ
1
+bx
ρ
2
}
k
ρ
is homogenous of degree k.
3. A quadratic form x
′
Ax is homogeneous of degree 2.
4. Leontief function Q = min {ax
1
, bx
2
} is homogeneous of degree 1.
Homothetic functions: If f is homogeneous and g = H(f), H is a monotonic
increasing function, then g is called a homothetic function.
Example: Q = αln x
1
+β ln x
2
= ln
x
α
1
x
β
2
is homothetic.
The MRS of a homothetic function depends only on the ratio
x
2
x
1
.
8.10 Problems
1. Use Newton’s method to ﬁnd the root of the nonlinear equation X
3
+2X+2 = 0
accurate to 2 digits.
2. Find (a) lim
X→0
1 −2
−X
X
, (b) lim
X→0
+
1 −e
−aX
X
, (c) lim
X→0
e
2X
−e
X
X
.
3. Given the total cost function C(Q) = e
aQ+b
, use L’hˆopital’s rule to ﬁnd the
AVC at Q = 0
+
.
4. Let z = x
1
x
2
+x
2
1
+ 3x
2
2
+x
2
x
3
+x
2
3
.
(a) Use matrix multiplication to represent z.
(b) Determine whether z is positive deﬁnite or negative deﬁnite.
(c) Find the extreme value of z. Check whether it is a maximum or a minimum.
72
5. The cost function of a competitive producer is C(Q; K) =
Q
3
3K
+ K, where K
is, say, the plant size (a ﬁxed factor in the short run).
(a) At which output level, the SAC curve attains a minimum?
(b) Suppose that the equilibrium price is p = 100. The proﬁt is Π = 100Q−
Q
3
3K
− K. For a given K, ﬁnd the supply quantity Q(K) such that SR
proﬁt is maximized.
(c) Calculate the SR maximizing proﬁt π(K).
(d) Find the LR optimal K = K
∗
to maximize π(K).
(e) Calculate the LR supply quantity Q
∗
= Q(K
∗
).
(f) Now solve the 2variable maximization problem
max
Q,K≥0
Π(Q, K) = pQ−C(Q) = 100Q−
Q
3
3K
−K.
and show that Π(Q, K) is concave so that the solution is the unique global
maximum.
6. A competitive ﬁrm produces two joint products. The total cost function is
C(q
1
, q
2
) = 2q
2
1
+ 3q
2
2
−4q
1
q
2
.
(a) Use the ﬁrst order conditions for proﬁt maximization to derive the supply
functions.
(b) Check that the second order conditions are satisﬁed.
7. Check whether the function f(x, y) = e
x+y
is concave, convex, or neither.
8. (a) Check whether f(x, y) = 2 lnx+3 ln y−x−2y is concave, convex, or neither.
(Assume that x > 0 and y > 0.)
(b) Find the critical point of f.
(c) Is the critical point a local maximum, a global maximum, or neither?
9. Suppose that a monopoly can produce any level of output at a constant marginal
cost of $c per unit. Assume that the monopoly sells its goods in two diﬀerent
markets which are separated by some distance. The demand curve in the ﬁrst
market is given by Q
1
= exp[−aP
1
] and the curve in the second market is given
by Q
2
= exp[−bP
2
]. If the monopolist wants to maximize its total proﬁts, what
level of output should be produced in each market and what price will prevail in
each market? Check that your answer is the unique global maximum. (Hints: 1.
P
1
= −(1/a) ln Q
1
and P
2
= −(1/b) ln Q
2
. 2. Π = P
1
Q
1
+P
2
Q
2
−(Q
1
+Q
2
)c =
−(1/a)Q
1
ln Q
1
−(1/b)Q
2
ln Q
2
−(Q
1
+Q
2
)c is strictly concave.)
10. The production function of a competitive ﬁrm is given by q = f(x
1
, x
2
) =
3x
1
3
1
x
1
3
2
, where x
1
is a variable input and x
2
is a ﬁxed input. Assume that the
prices of the output and the ﬁxed input are p = w
2
= 1. In the short run,
the amount of the ﬁxed input is given by x
2
= ¯ x
2
. The proﬁt function of the
competitive ﬁrm is given by π = f(x
1
, ¯ x
2
) −w
1
x
1
− ¯ x
2
.
73
(a) State the FOC for the SR proﬁt maximization and calculate the SR input
demand function x
S
1
= x
S
1
(w
1
) and the SR input demand elasticity
w
1
x
S
1
∂x
S
1
∂w
1
.
(b) Now consider the LR situation when x
2
can be adjusted. State the FOC
for LR proﬁt maximization and calculate the LR input demand funtion
x
L
1
= x
L
1
(w
1
) and the LR input demand elasticity
w
1
x
L
1
∂x
L
1
∂w
1
.
(c) Verify the Le Chˆatelier principle:
w
1
x
L
1
∂x
L
1
∂w
1
>
w
1
x
S
1
∂x
S
1
∂w
1
.
(d) Show that the LR proﬁt is a strictly concave function of (x
1
, x
2
) for x
1
, x
2
>
0 and therefore the solution must be the unique global maximum.
11. Let U(x, y) = x
a
y +xy
2
, x, y, a > 0.
(a) For what value(s) of a U(x, y) is homogeneous?.
(b) For what value(s) of a U(x, y) is homothetic?
74
9 Optimization and Equality Constraints and Nonlinear Pro
gramming
In some maximization problems, the agents can choose only values of (x
1
, . . . , x
n
)
that satisfy certain equalities. For example, a consumer has a budget constraint
p
1
x
1
+· · · +p
n
x
n
= m.
max
x
1
,...,xn
U(x) = U(x
1
, . . . , x
n
) subject to p
1
x
1
+· · · +p
n
x
n
= m.
The procedure is the same as before. (1) Use FOC to ﬁnd critical points; (2) use
SOC to check whether a critical point is a local maximum, or show that U(X) is
quasiconcave so that a critical point must be a global maximum.
Deﬁne B = {(x
1
, . . . , x
n
) such that p
1
x
1
+· · · +p
n
x
n
= m}.
A local maximum X
∗
= (x
∗
1
, . . . , x
∗
n
): ∃ǫ > 0 such that U(X
∗
) ≥ U(X) for all X ∈ B
satisfying x
i
∈ (x
∗
i
−ǫ, x
∗
i
+ǫ) ∀i.
A critical point: A X
∗
∈ B satisfying the FOC for a local maximum.
A global maximum X
∗
: F(X
∗
) ≥ F(X) ∀X ∈ B.
A unique global maximum X
∗
: F(X
∗
) > F(X) ∀X ∈ B, X = X
∗
.
To deﬁne the concept of a critical point, we have to know what is the FOC ﬁrst.
9.1 FOC and SOC for a constraint maximization
Consider the 2variable utility maximization problem:
max
x
1
,x
2
U(x
1
, x
2
) subject to p
1
x
1
+p
2
x
2
= m.
Using the budget constraint, x
2
=
m−p
1
x
1
p
2
= h(x
1
),
dx
2
dx
1
= −
p
1
p
2
= h
′
(x
1
), and it
becomes a single variable maximization problem:
max
x
1
U
x
1
,
m−p
1
x
1
p
2
, FOC:
dU
dx
1
= U
1
+U
2
−
p
1
p
2
= 0, SOC:
d
2
U
dx
2
1
< 0.
d
2
U
dx
2
1
= U
11
−2
p
1
p
2
U
12
+
p
1
p
2
2
U
22
=
−1
p
2
2
0 −p
1
−p
2
−p
1
U
11
U
12
−p
2
U
21
U
22
.
By FOC,
U
1
p
1
=
U
2
p
2
≡ λ (MU of $1).
FOC: U
1
= p
1
λ, U
2
= p
2
λ, SOC:
0 U
1
U
2
U
1
U
11
U
12
U
2
U
21
U
22
> 0.
Alternatively, we can deﬁne Lagrangian:
L ≡ U(x
1
, x
2
) +λ(m−p
1
x
1
−p
2
x
2
)
75
FOC: L
1
=
∂L
∂x
1
= U
1
− λp
1
= 0, L
2
=
∂L
∂x
2
= U
2
− λp
2
= 0, L
λ
=
∂L
∂λ
=
m−p
1
x
1
−p
2
x
2
= 0.
SOC:
0 −p
1
−p
2
−p
1
L
11
L
12
−p
2
L
21
L
22
> 0.
general 2variable with 1constraint case:
max
x
1
,x
2
F(x
1
, x
2
) subject to g(x
1
, x
2
) = 0.
Using the constraint, x
2
= h(x
1
),
dx
2
dx
1
= −
g
1
g
2
= h
′
(x
1
), and it becomes a single
variable maximization problem:
max
x
1
F (x
1
, h(x
1
)) FOC:
dF
dx
1
= F
1
+F
2
h
′
(x
1
) = 0, SOC:
d
2
F
dx
2
1
< 0.
d
2
F
dx
2
1
=
d
dx
1
(F
1
+F
2
h
′
) = F
11
+ 2h
′
F
12
+ (h
′
)
2
F
22
+F
2
h
′′
h
′′
=
d
dx
1
−
g
1
g
2
=
−1
g
2
g
11
+ 2g
12
h
′
+g
22
(h
′
)
2
.
⇒
d
2
F
dx
2
1
=
F
11
−
F
2
g
2
g
11
+ 2
F
12
−
F
2
g
2
g
12
h
′
+
F
22
−
F
2
g
2
g
22
(h
′
)
2
= −
0 −h
′
1
−h
′
F
11
−
F
2
g
2
g
11
F
12
−
F
2
g
2
g
12
1 F
21
−
F
2
g
2
g
21
F
22
−
F
2
g
2
g
22
=
−1
g
2
2
0 g
1
g
2
g
1
F
11
−
F
2
g
2
g
11
F
12
−
F
2
g
2
g
12
g
2
F
21
−
F
2
g
2
g
21
F
22
−
F
2
g
2
g
22
By FOC,
F
1
g
1
=
F
2
g
2
≡ λ (Lagrange multiplier).
Alternatively, we can deﬁne Lagrangian:
L ≡ F(x
1
, x
2
) −λg(x
1
, x
2
)
FOC: L
1
=
∂L
∂x
1
= F
1
−λg
1
= 0, L
2
=
∂L
∂x
2
= F
2
−λg
2
= 0, L
λ
=
∂L
∂λ
= −g(x
1
, x
2
) =
0.
SOC:
0 g
1
g
2
g
1
L
11
L
12
g
2
L
21
L
22
> 0.
nvariable 1constraint case:
max
x
1
,...,xn
F(x
1
, . . . , x
n
) subject to g(x
1
, . . . , x
n
) = 0.
L ≡ F(x
1
, . . . , x
n
) −λg(x
1
, . . . , x
n
)
76
FOC: L
i
=
∂L
∂x
i
= F
i
−λg
i
= 0, i = 1, . . . , n, L
λ
=
∂L
∂λ
= −g(x
1
, . . . , x
n
) = 0.
SOC:
0 −g
1
−g
1
L
11
< 0,
0 −g
1
−g
2
−g
1
L
11
L
12
−g
2
L
21
L
22
> 0
0 −g
1
−g
2
−g
3
−g
1
L
11
L
12
L
13
−g
2
L
21
L
22
L
23
−g
3
L
31
L
32
L
33
< 0, etc.
9.2 Examples
Example 1: max F(x
1
, x
2
) = −x
2
1
−x
2
2
subject to x
1
+x
2
= 1.
L = −x
2
1
−x
2
2
+λ(1 −x
1
−x
2
).
FOC: L
1
= −2x
1
−λ = 0, L
2
= −2x
2
−λ = 0 and L
λ
= 1 −2x
1
−2x
2
= 0.
Critical point:
x
1
x
2
=
¸
¸
1
2
1
2
, λ = −1.
SOC:
0 1 1
1 −2 0
1 0 −2
= 4 > 0.
⇒
1/2
1/2
is a local maximum.
x
1
`
x
2
,
x
1
`
x
2
,
,
Example 2: max F(x
1
, x
2
) = x
1
+x
2
subject to x
2
1
+x
2
2
= 1.
L = x
1
+x
2
+λ(1 −x
2
1
−x
2
2
).
FOC: L
1
= 1 −2λx
1
= 0, L
2
= 1 −2λx
2
= 0 and L
λ
= 1 −x
2
1
−x
2
2
= 0.
Two critical points: x
1
= x
2
= λ = 1/
√
2 and x
1
= x
2
= λ = −1/
√
2.
SOC:
0 −2x
1
−2x
2
−2x
1
−2λ 0
−2x
2
0 −2λ
= 8λ(x
2
1
+x
2
2
).
⇒ x
1
= x
2
= λ = 1/
√
2 is a local maximum and x
1
= x
2
= λ = −1/
√
2 is a local
minimum.
77
9.3 Cost minimization and cost function
min C(x
1
, x
2
; w
1
, w
2
) = w
1
x
1
+w
2
x
2
subject to x
a
1
x
1−a
2
= q, 0 < a < 1.
L = w
1
x
1
+w
2
x
2
+λ(q −x
a
1
x
1−a
2
).
FOC: L
1
= w
1
−aλx
a−1
1
x
1−a
2
= 0, L
2
= w
2
−(1−a)λx
a
1
x
−a
2
= 0, L
λ
= q −x
a
1
x
1−a
2
= 0.
⇒
w
1
w
2
=
ax
2
(1 −a)x
1
⇒x
1
= q
¸
aw
2
(1 −a)w
1
1−a
, x
2
= q
¸
(1 −a)w
1
aw
2
a
.
SOC:
0 −ax
a−1
1
x
1−a
2
−(1 −a)x
a
1
x
−a
2
−ax
a−1
1
x
1−a
2
a(1 −a)λx
a−2
1
x
1−a
2
−a(1 −a)λx
a−1
1
x
−a
2
−(1 −a)x
a
1
x
−a
2
−a(1 −a)λx
a−1
1
x
−a
2
a(1 −a)λx
a
1
x
−a−1
2
= −
a(1 −a)q
3
λ
(x
1
x
2
)
2
< 0.
⇒x
1
= q
¸
aw
2
(1 −a)w
1
1−a
, x
2
= q
¸
(1 −a)w
1
aw
2
a
is a local minimum.
The total cost is C(w
1
, w
2
, q) = q
¸
a
1 −a
1−a
+
1 −a
a
a
¸
w
a
1
w
1−a
2
.
9.4 Utility maximization and demand function
max U(x
1
, x
2
) = a ln x
1
+b ln x
2
subject to p
1
x
1
+p
2
x
2
= m.
L = a ln x
1
+b ln x
2
+λ(m−p
1
x
1
−p
2
x
2
).
FOC: L
1
=
a
x
1
−λp
1
= 0, L
2
=
b
x
2
−λp
2
= 0 and L
λ
= m−p
1
x
1
−p
2
x
2
= 0.
⇒
a
b
x
2
x
1
=
p
1
p
2
⇒x
1
=
am
(a +b)p
1
, x
2
=
bm
(a +b)p
2
SOC:
0 −p
1
−p
2
−p
1
−a
x
2
1
0
−p
2
0
−b
x
2
2
=
ap
2
2
x
2
1
+
bp
2
1
x
2
2
> 0.
⇒x
1
=
am
(a +b)p
1
, x
2
=
bm
(a +b)p
2
is a local maximum.
9.5 Quasiconcavity and quasiconvexity
As discussed in the intermediate microeconomics course, if the MRS is strictly de
creasing along an indiﬀerence curve (indiﬀerence curve is convex toward the origin),
then the utility maximization has a unique solution. A utility function U(x
1
, x
2
)
is quasiconcave if MRS (=
U
1
U
2
) is decreasing along every indiﬀerence curve. In
case MRS is strictly decreasing, the utility function is strictly quasiconcave. If
U(x
1
, x
2
) is (strictly) quasiconcave, then a critical point must be a (unique) global
maximum.
Two ways to determine whether U(x
1
, x
2
) is quasiconcave: (1) the set {(x
1
, x
2
) ∈
R
2
U(x
1
, x
2
) ≥
¯
U} is convex for all
¯
U. (Every indiﬀerence curve is convex toward the
78
origin.)
(2)
d
dx
1
−
U
1
U
2
U(x
1
,x
2
)=
¯
U
> 0. (MRS is strictly decreasing along every indiﬀerence
curve.)
In sections 6.14 and 7.7 we used F(X) to deﬁne two sets, G
+
F
and G
−
F
, and we say
that F(X) is concave (convex) if G
−
F
(G
+
F
) is a convex set. Now we say that F(X) is
quasiconcave (quasiconvex) if every ycrosssection of G
−
F
(G
+
F
) is a convex set. A
ycrosssection is formally deﬁned as
G
−
F
(¯ y) ≡ {(x
1
, x
2
) F(x
1
, x
2
) ≥ ¯ y} ⊂ R
2
, G
+
F
(¯ y) ≡ {(x
1
, x
2
) F(x
1
, x
2
) ≤ ¯ y} ⊂ R
2
.
Clearly, G
−
F
is the union of all G
−
F
(¯ y): G
−
F
= ∪
¯ y
G
−
F
(¯ y).
Formal deﬁnition: F(x
1
, . . . , x
n
) is quasiconcave if ∀X
0
, X
1
∈ A and 0 ≤ θ ≤ 1,
F(X
θ
) ≥ min{F(X
0
), F(X
1
)}.
F(x
1
, . . . , x
n
) is strictly quasiconcave if ∀X
0
= X
1
∈ A and 0 < θ < 1, F(X
θ
) >
min{F(X
0
), F(X
1
)}.
x
1
`
x
2
x
1
`
x
2
If F is (strictly) concave, then F must be (strictly) quasiconcave.
Proof: If F(X
θ
) < min{F(X
0
), F(X
1
)}, then F(X
θ
) < (1 − θ)F(X
0
) + θF(X
1
).
Geometrically, if G
−
F
is convex, then G
−
F
(¯ y) must be convex for every ¯ y.
If F is quasiconcave, it is not necessarily that F is concave.
Counterexample: F(x
1
x
2
) = x
1
x
2
, x
1
, x
2
> 0 is strictly quasiconcave but not con
cave. In this case, every G
−
F
(¯ y) is convex but still G
−
F
is not convex.
Bordered Hessian: B
1
 ≡
0 F
1
F
1
F
11
, B
2
 ≡
0 F
1
F
2
F
1
F
11
F
12
F
2
F
21
F
22
, B
3
 ≡
0 F
1
F
2
F
3
F
1
F
11
F
12
F
13
F
2
F
21
F
22
F
23
F
3
F
31
F
32
F
33
,
etc.
Theorem 1: Suppose that F is twice diﬀerentiable. If F is quasiconcave, then
B
2
 ≥ 0, B
3
 ≤ 0, etc. Conversely, if B
1
 < 0, B
2
 > 0, B
3
 < 0, etc., then
F is strictly quasiconcave.
Proof (n = 2):
dMRS
dx
1
=
B
2

F
3
2
and therefore F is quasiconcave if and only if B
2
 ≥ 0.
79
Consider the following maximization problem with a linear constraint.
max
x
1
,...,xn
F(x
1
, . . . , x
n
) subject to a1x
1
+· · · +a
n
x
n
= b.
Theorem 2: If F is (strictly) quasiconcave and X
0
satisﬁes FOC, then X
0
is a
(unique) global maximum.
Proof: By theorem 1, X
0
must be a local maximum. Suppose there exists X
1
satis
fying the linear constraint with U(X
1
) > U(X
0
). Then U(X
θ
) > U(X
0
), a contra
diction.
Theorem 3: A monotonic increasing transformation of a quasiconcave function is
a quasiconcave function. A quasiconcave function is a monotonic increasing trans
formation of a concave function.
Proof: A monotonic increasing transformation does not change the sets {(x
1
, x
2
) ∈
R
2
U(x
1
, x
2
) ≥
¯
U}. To show the opposite, suppose that f(x
1
, x
2
) is quasiconcave.
Deﬁne a monotonic transformation as
g(x
1
, x
2
) = H(f(x
1
, x
2
)) where H
−1
(g) = f(x, x).
9.6 Elasticity of Substitution
Consider a production function Q = F(x
1
, x
2
).
Cost minimization ⇒
w
1
w
2
=
F
1
F
2
≡ θ. Let
x
2
x
1
≡ r. On an isoquant F(x
1
, x
2
) =
¯
Q,
there is a relationship r = φ(θ). The elasticity of substitution is deﬁned as σ ≡
θ
r
dr
dθ
.
If the input price ratio θ =
w
1
w
2
increases by 1%, a competitive producer will increase
its input ratio r =
x
2
x
1
by σ%.
Example: For a CES function Q = {ax
ρ
1
+bx
ρ
2
}
k
ρ
, σ =
1
1 −ρ
.
9.7 Problems
1. Let A =
a
11
a
12
a
21
a
22
, B =
¸
0 c
1
c
2
c
1
a
11
a
12
c
2
a
21
a
22
, and C =
−c
2
c
1
.
(a) Calculate the determinant B.
(b) Calculate the product C
′
AC. Does B have any relationship with C
′
AC?
2. Consider the utility function U(x
1
, x
2
) = x
α
1
x
β
2
.
(a) Calculate the marginal utilities U
1
=
∂U
∂x
1
and U
2
=
∂U
∂x
2
.
(b) Calculate the hessian matrix H =
U
11
U
12
U
21
U
22
.
80
(c) Calculate the determinant of the matrix B =
¸
0 U
1
U
2
U
1
U
11
U
12
U
2
U
21
U
22
.
(d) Let C =
−U
2
U
1
. Calculate the product C’HC.
3. Use the Lagrangian multiplier method to ﬁnd the critical points of f(x, y) =
x+y subject to x
2
+y
2
= 2. Then use the bordered Hessian to determine which
point is a maximum and which is a minimum. (There are two critical points.)
4. Determine whether f(x, y) = xy is concave, convex, quasiconcave, or quasicon
vex. (x > 0 and y > 0.)
5. Let U(x, y), U, x, y > 0, be a homogenous of degree 1 and concave function with
U
xy
= 0.
(a) Show that V (x, y) = [U(x, y)]
a
is strictly concave if 0 < a < 1 and V (x, y)
is strictly quasiconcave for all a > 0. Hint: U
xx
U
yy
= [U
xy
]
2
.
(b) Show that F(x, y) =
x
β
+y
β
a/β
, x, y > 0 and −∞ < β < 1 is homoge
nous of degree 1 and concave if a = 1.
(c) Determine the range of a so that F(x, y) is strictly concave.
(d) Determine the range of a so that F(x, y) is strictly quasiconcave.
81
9.8 Nonlinear Programming
The general nonlinear programming problem is:
max
x
1
,...,xn
F(x
1
, . . . , x
n
) subject to
g
1
(x
1
, . . . , x
n
) ≤ b
1
.
.
.
g
m
(x
1
, . . . , x
n
) ≤ b
m
x
1
, . . . , x
n
≥ 0.
In equality constraint problems, the number of constraints should be less than the
number of policy variables, m < n. For nonlinear programming problems, there is no
such a restriction, m can be greater than or equal to n. In vector notation,
max
x
F(x) subject to g(x) ≤ b, x ≥ 0.
9.9 KuhnTucker condition
Deﬁne the Lagrangian function as
L(x, y) = F(x) +y(b −g(x)) = F(x
1
, . . . , x
n
) +
m
¸
j=1
y
j
(b
j
−g
j
(x
1
, . . . , x
n
)).
KuhnTucker condition: The FOC is given by the KuhnTucker conditions:
∂L
∂x
i
=
∂F
∂x
i
−
m
¸
j=1
y
j
∂g
j
∂x
i
≤ 0, x
i
∂L
∂x
i
= 0 x
i
≥ 0, i = 1, . . . , n
∂L
∂y
j
= b
j
−g
j
(x) ≥ 0, y
j
∂L
∂y
j
= 0 y
j
≥ 0, j = 1, . . . , m
Kuhn Tucker theorem: x
∗
solves the nonlinear programming problem if (x
∗
, y
∗
)
solves the saddle point problem:
L(x, y
∗
) ≤ L(x
∗
, y
∗
) ≤ L(x
∗
, y) for all x ≥ 0, y ≥ 0,
Conversely, suppose that f(x) is a concave function and g
j
(x) are convex func
tions (concave programming) and the constraints satisfy the constraint qualiﬁ
cation condition that there is some point in the opportunity set which satisﬁes
all the inequality constraints as strict inequalities, i.e., there exists a vector
x
0
≥ 0 such that g
j
(x
0
) < b
j
, j = 1, . . . , m, then x
∗
solves the nonlinear pro
gramming problem only if there is a y
∗
for which (x
∗
, y
∗
) solves the saddle point
problem.
If constraint qualiﬁcation is not satisﬁed, it is possible that a solution does not satisfy
the KT condition. If it is satisﬁed, then the KT condition will be necessary. For
the case of concave programming, it is also suﬃcient.
82
In economics applications, however, it is not convenient to use KT condition to
ﬁnd the solution. In stead, we ﬁrst solve the equality constraint version of the problem
and then use KT condition to check or modify the solution when some constraints
are violated.
The KT condition for minimization problems: the inequalities reversed.
9.10 Examples
Example 1. (Joint product proﬁt maximization) The cost function of a competitive
producer producing 2 joint products is c(x
1
, x
2
) = x
2
1
+x
1
x
2
+x
2
2
. The proﬁt function
is given by π(p
1
, p
2
) = p
1
x
1
+p
2
x
2
−(x
2
1
+x
1
x
2
+x
2
2
).
max
x
1
≥0,x
2
≥0
f(x
1
, x
2
) = p
1
x
1
+p
2
x
2
−(x
2
1
+x
1
x
2
−x
2
2
)
KT condition: f
1
= p
1
−2x
1
−x
2
≤ 0, f
2
= p
2
−x
1
−2x
2
≤ 0, x
i
f
i
= 0, i = 1, 2.
Case 1. p
1
/2 < p
2
< 2p
1
.
x
1
= (2p
1
−p
2
)/3, x
2
= (2p
2
−p
1
)/3.
Case 2. 2p
1
< p
2
.
x
1
= 0, x
2
= p
2
/2.
Case 3. 2p
2
< p
1
.
x
1
= p
1
/2, x
2
= 0.
Example 2. The production function of a producer is given by q = (x
1
+1)(x
2
+1)−1.
For q = 8, calculate the cost function c(w
1
, w
2
).
min
x
1
≥0,x
2
≥0
w
1
x
1
+w
2
x
2
subject to −[(x
1
+ 1)(x
2
+ 1) −1] ≥ −8
Lagrangian function: L = w
1
x
1
+w
1
x
2
+λ[(x
1
+ 1)(x
2
+ 1) −9].
KT conditions: L
1
= w
1
−λ(x
2
−1) ≥ 0, L
2
= w
2
−λ(x
2
−1) ≥ 0, x
i
L
i
= 0, i = 1, 2,
and L
λ
= (x
1
+ 1)(x
2
+ 1) −9 ≥ 0, λL
λ
= 0.
Case 1. w
1
/9 < w
2
< 9w
1
.
x
1
=
9w
2
/w
1
−1, x
2
=
9w
1
/w
2
−1 and c(w
1
, w
2
) = 6
√
w
1
w
2
−w
1
−w
2
.
Case 2. 9w
1
< w
2
.
x
1
= 8, x
2
= 0, and c(w
1
, w
2
) = 8w
1
.
Case 3. 9w
2
< w
1
.
x
1
= 0, x
2
= 8, c(w
1
, w
2
) = 8w
2
.
83
Example 3. The utility function of a consumer is U(x
1
, x
2
) = x
1
(x
2
+1). The market
price is p
1
= p
2
= 1 and the consumer has $11. Therefore the budget constraint is
x
1
+ x
2
≤ 11. Suppose that both products are under rationing. Besides the money
price, the consumer has to pay ρ
i
rationing points for each unit of product i consumed.
Assume that ρ
1
= 1 and ρ
2
= 2 and the consumer has q rationing points. The
rationing point constraint is x
1
+2x
2
≤ q. The utility maximization problem is given
by
max
x
1
,x
2
U(x
1
, x
2
) = x
1
(x
2
+1) subject to x
1
+x
2
≤ 11, x
1
+2x
2
≤ q, x
1
, x
2
≥ 0.
Lagrangian function: L = x
1
(x
2
+ 1) +λ
1
(11 −x
1
−x
2
) +λ
2
(q −x
1
−2x
2
).
KT conditions: L
1
= x
2
+1−λ
1
−λ
2
≤ 0, L
2
= x
1
−λ
1
−2λ
2
≤ 0, x
i
L
i
= 0, i = 1, 2,
and L
λ
1
= 11 −x
1
−x
2
≥ 0, L
λ
2
= q −x
1
−2x
2
≥ 0 λ
i
L
λ
i
= 0.
Case 1: q < 2.
x
1
= q, x
2
= 0, λ
1
= 0, and λ
2
= 1.
Case 2: 2 ≤ q ≤ 14.
x
1
= (q + 2)/2, x
2
= (q −2)/4, λ
1
= 0, and λ
2
= (q + 2)/4.
Case 3: 14 < q ≤ 16.
x
1
= 22 −q, x
2
= q −11, λ
1
= 3(q −14), and λ
2
= 2(16 −q).
Case 4: 16 < q.
x
1
= 6, x
2
= 5, λ
1
= 6, and λ
2
= 0.
9.11 Problems
1. Given the individual utility function U(X, Y ) = 2
√
X +Y ,
a) show that U is quasiconcave for X ≥ 0 and Y ≥ 0,
b) state the KuhnTucker conditions of the following problem:
max
X≥0, Y ≥0
2
√
X +Y
s. t. P
X
X +P
Y
Y ≤ I,
c) derive the demand functions X(P
X
, P
Y
, I) and Y (P
X
, P
Y
, I) for the case I ≥
P
2
Y
P
X
,
check that the KT conditions are satisﬁed,
d) and do the same for I <
P
2
Y
P
X
.
e) Given that I = 1 and P
Y
= 1, derive the ordinary demand function X = D(P
X
).
f) Are your answers in (c) and (d) global maximum? Unique global maximum? Why
84
or why not?
2. A farm has a total amount of agricultural land of one acre. It can produce two
crops, corn (C) and lettuce (L), according to the production functions C = N
C
and
L = 2
√
N
L
respectively, where N
C
(N
L
) is land used in corn (lettuce) production.
The prices of corn and lettuce are p and q respectively. Thus, if the farm uses N
C
of land in corn production and N
L
in lettuce production, (N
C
≥ 0, N
L
≥ 0, and
N
C
+N
L
≤ 1) its total revenue is pN
C
+ 2q
√
N
L
.
a) Suppose the farm is interested in maximizing its revenue. State the revenue max
imization problem and the KuhnTucker conditions.
b) Given that q > p > 0, how much of each output will the farm produce? Check
that the KT conditions are satisﬁed.
c) Given that p ≥ q > 0, do the same as (b).
3. Suppose that a ﬁrm has two activities producing two goods “meat” (M) and
“egg” (E) from the same input “chicken” (C) according to the production functions
M = C
M
and E = C
1/2
E
, where C
M
(respectively C
E
) ≥ 0 is the q. Suppose in the
short run, the ﬁrm has
¯
C units of chicken that it must take as given and suppose that
the ﬁrm faces prices p > 0, q > 0 of meat and egg respectively.
a) Show that the proﬁt function π = pC
M
+qC
0.5
E
is quasiconcave in (C
M
, C
E
).
b) Write down the short run proﬁt maximization problem.
c) State the KuhnTucker conditions.
d) Derive the short run supply functions. (There are two cases.)
e) Is your solution a global maximum? Explain.
4. State the KuhnTucker conditions of the following nonlinear programming problem
max U(X, Y ) = 3 lnX + ln Y
s. t. 2X +Y ≤ 24
X + 2Y ≤ 24
X ≥ 0, Y ≥ 0.
Show that X = 9, Y = 6, λ
1
= 1/6, and λ
2
= 0 satisfy the KuhnTucker conditions.
What is the economic interpretations of λ
1
= 1/6 and λ
2
= 0 if the ﬁrst constraint is
interpreted as the income constraint and the second constraint as the rationing point
constraint of a utility maximization problem?
9.12 Linear Programming – A Graphic Approach
Example 1 (A Production Problem). A manufacturer produces tables x
1
and
desks x
2
. Each table requires 2.5 hours for assembling (A), 3 hours for buﬃng (B),
and 1 hour for crating (C). Each desks requires 1 hour for assembling (A), 3 hours
for buﬃng (B), and 2 hours for crating (C). The ﬁrm can use no more than 20 hours
for assembling, 30 hours for buﬃng, and 16 hours for crating each week. Its proﬁt
margin is $3 per table and $4 per desk.
85
max Π = 3x
1
+ 4x
2
(4)
subject to 2.5x
1
+x
2
≤ 20 (5)
3x
1
+ 3x
2
≤ 30 (6)
x
1
+ 2x
2
≤ 16 (7)
x
1
, x
2
≥ 0. (8)
extreme point: The intersection of two constraints.
extreme point theorem: If an optimal feasible value of the objective function exists,
it will be found at one of the extreme points.
In the example, There are 10 extreme points, but only 5 are feasible: (0 ,0), (8,
0), (6
2
3
, 3
1
3
), (4, 6), and (0, 8), called basic feasible solutions. At (4, 6), Π = 36 is the
optimal.
x
1
`
x
2
\
\
\
\
\
\
\
\
\
\
\
\
\
\
\
\
\
\
\
\\
Z
Z
Z
Z
Z
Z
Z
Z
Z
Z
Z
Z
(4,6)
y
`
y
2
`
`
`
`
`
`
`
`
`
`
`
`
`
`
H
H
H
H
H
H
H
H
H
H
H
H
H
H
H
(9,3)
Example 2 (The Diet Problem). A farmer wants to see that her herd gets the
minimum daily requirement of three basic nutrients A, B, and C. Daily requirements
are 14 for A, 12 for B, and 18 for C. Product y
1
has 2 units of A and 1 unit each of
B and C; product y
2
has 1 unit each of A and B and 3 units of C. The cost of y
1
is
$2, and the cost of y
2
is $4.
min c = 2y
1
+ 4y
2
(9)
subject to 2y
1
+y
2
≥ 14 (10)
y
1
+y
2
≥ 12 (11)
y
1
+ 3y
2
≥ 18 (12)
y
1
, y
2
≥ 0. (13)
86
Slack and surplus variables: To ﬁnd basic solutions, equations are needed. This is
done by incorporating a separate slack or surplus variable s
i
into each inequality.
In example 1, the system becomes
2.5x
1
+x
2
+s
1
= 20 3x
1
+ 3x
2
+s
2
= 30 x
1
+ 2x
2
+s
3
= 16.
In matrix for,
2.5 1 1 0 0
3 3 0 1 0
1 2 0 0 1
¸
¸
x
1
x
2
s
1
s
2
s
3
¸
¸
¸
¸
¸
¸
=
20
30
16
¸
¸
.
In example 2, the inequalities are ”≥” and the surplus variables are substracted:
2y
1
+y
2
−s
1
= 14 y
1
+y
2
−s
2
= 12 y
1
+ 3y
2
−s
3
= 18.
In matrix for,
2 1 −1 0 0
1 1 0 −1 0
1 3 0 0 −1
¸
¸
y
1
y
2
s
1
s
2
s
3
¸
¸
¸
¸
¸
¸
=
14
12
18
¸
¸
.
For a system of m equations and n variables, where n > m, a solution in which at
least n − m variables equal to zero is an extreme point. Thus by setting n − mand
solving the m equations for the remaining m variables, an extreme point can be found.
There are n!/m!(n −m)! such solutions.
9.13 Linear programming – The simplex algorithm
The algorithm moves from one basic feasible solution to another, always improving
upon the previous solutions, until the optimal solution is reached. In each step, those
variables set equal to zero are called not in the basis and those not set equal to zero
are called in the basis. Let use example one to illustrate the procedure.
1. The initial Simplex Tableau
x
1
x
2
s
1
s
2
s
3
Constant
2.5 1 1 0 0 20
3 3 0 1 0 30
1 2 0 0 1 16
3 4 0 0 0 0
The ﬁrst basic feasible solution can be read from the tableau as x
1
= 0, x
2
= 0,
s
1
= 20, s
2
= 30, and s
3
= 16. The value of Π is zero.
2. The Pivot Element and a change of Basis
87
(a) The negative indicator with the largest absolute value determines the vari
able to enter the basis. Here it is x
2
. The x
2
column is called the pivot
column (jth column).
(b) The variable to be eliminated is determined by the smallest displacement
ratio. Displacement ratios are found by dividing the elements of the con
stant column by the elements of the pivot column. Here the smallest is
16/2=8 and row 3 is the pivot row (ith row). The pivot element is 2.
3. (Pivoting) and we are going to move to the new basic solution with s
3
= 0
and x
2
> 0. First, divides every element of the pivoting row by the pivoting
element (2 in this example) to make the pivoting element equal to 1. Then
subtracts a
kj
times the pivoting row from kth row to make the jth column a
unit vector. (This procedure is called the Gaussian elimination method, usually
used in solving simultaneous equations.) After pivoting, the Tableau becomes
x
1
x
2
s
1
s
2
s
3
Constant
2 0 1 0 .5 12
1.5 0 0 1 1.5 6
.5 1 0 0 .5 8
1 0 0 0 2 32
The basic solution is x
2
= 8, s
1
= 12, s
2
= 6, and x
1
= s
3
= 0. The value of Π
is 32.
4. (Optimization) Repeat steps 23 until a maximum is reached. In the exam
ple, x
1
column is the new pivoting column. The second row is the pivoting
row and 1.5 is the pivoting element. After pivoting, the tableau becomes
x
1
x
2
s
1
s
2
s
3
Constant
0 0 1 −
4
3
1.5 4
1 0 0
2
3
1 4
0 1 0 −
1
3
1 6
0 0 0
2
3
1 36
The basic solution is x
1
= 4, x
2
= 6, s
1
= 4, and s
2
= s
3
= 0. The value
of Π is 36. Since there is no more negative indicators, the process stops and the
basic solution is the optimal. s
1
= 4 > 0 and s
2
= s
3
= 0 means that the ﬁrst
constraint is not binding but the second and third are binding. The indicators
for s
2
and s
3
, t
2
=
2
3
and t
3
= 1 are called the shadow values, representing the
marginal contributions of increasing one hour for buﬃng or crating.
Because y
1
= y
2
= 0 is not feasible, the simplex algorithm for minimization
problem is more complex. Usually, we solve its dual problem.
88
9.14 Linear programming – The dual problem
To every linear programming problem there corresponds a dual problem. If the orig
inal problem, called the primal problem, is
max
x
F = cx subject to Ax ≤ b, x ≥ 0
then the dual problem is
min
y
G = yb subject to yA ≥ c, y ≥ 0
where
A =
a
11
. . . a
1n
.
.
.
.
.
.
.
.
.
a
m1
. . . a
mn
¸
¸
¸
, x =
x
1
.
.
.
x
n
¸
¸
¸
, b =
b
1
.
.
.
b
m
¸
¸
¸
, c = (c
1
, . . . , c
n
), y = (y
1
, . . . , y
m
).
Existence theorem: A necessary and suﬃcient condition for the existence of a so
lution is that the opportunity sets of both the primal and dual problems are
nonempty.
Proof: Suppose x, y are feasible. Then Ax ≤ b, yA ≥ c. It follows that
F(x) = cx ≤ yAx and G(y) = yb ≥ yAx. Therefore, F(x) ≤ G(y).
Duality theorem: A necessary and suﬃcient condition for a feasible vector to rep
resent a solution is that there exists a feasible vector for the dual problem for
which the values of the objective functions of both problems are equal.
Complementary slackness theorem: A necessary and suﬃcient condition for fea
sible vectors x
∗
, y
∗
to solve the dual problems is that they satisfy the comple
mentary slackness condition:
(c −y
∗
A)x
∗
= 0 y
∗
(b −Ax
∗
) = 0.
Proof: Use KuhnTucker theorem.
Dual of the Diet Problem
max c
∗
= 14x
1
+ 12x
2
+ 18x
3
(14)
subject to 2x
1
+x
2
+x
1
≤ 2 (15)
x
1
+x
2
+ 3x
3
≤ 4 (16)
x
1
, x
2
, x
3
≥ 0. (17)
x
1
, x
2
, x
3
, is interpreted as the imputed value of nutrient A, B, C, respectively.
89
10 General Equilibrium and Game Theory
10.1 Utility maximization and demand function
A consumer wants to maximize his utility function subject to his budget constraint:
max U(x
1
, . . . , x
n
) subj. to p
1
x
1
+· · · +p
n
x
n
= I.
Endogenous variables: x
1
, . . . , x
n
Exogenous variables: p
1
, . . . , p
n
, I (the consumer is a price taker)
Solution is the demand functions x
k
= D
k
(p
1
, . . . , p
n
, I), k = 1, . . . , n
Example: max U(x
1
, x
2
) = a ln x
1
+b ln x
2
subject to p
1
x
1
+p
2
x
2
= m.
L = a ln x
1
+b ln x
2
+λ(m−p
1
x
1
−p
2
x
2
).
FOC: L
1
=
a
x
1
−λp
1
= 0, L
2
=
b
x
2
−λp
2
= 0 and L
λ
= m−p
1
x
1
−p
2
x
2
= 0.
⇒
a
b
x
2
x
1
=
p
1
p
2
⇒x
1
=
am
(a +b)p
1
, x
2
=
bm
(a +b)p
2
SOC:
0 −p
1
−p
2
−p
1
−a
x
2
1
0
−p
2
0
−b
x
2
2
=
ap
2
2
x
2
1
+
bp
2
1
x
2
2
> 0.
⇒x
1
=
am
(a +b)p
1
, x
2
=
bm
(a +b)p
2
is a local maximum.
10.2 Proﬁt maximization and supply function
A producer’s production technology can be represented by a production function
q = f(x
1
, . . . , x
n
). Given the prices, the producer maximizes his proﬁts:
max Π(x
1
, . . . , x
n
; p, p
1
, . . . , p
n
) = pf(x
1
, . . . , x
n
) −p
1
x
1
−· · · −p
n
x
n
Exogenous variables: p, p
1
, . . . , p
n
(the producer is a price taker)
Solution is the supply function q = S(p, p
1
, . . . , p
n
) and the input demand functions,
x
k
= X
k
(p, p
1
, . . . , p
n
) k = 1, . . . , n
Example: q = f(x
1
, x
2
) = 2
√
x
1
+ 2
√
x
2
and Π(x
1
, x
2
; p, p
1
, p
2
) = p(2
√
x
1
+ 2
√
x
2
) −
p
1
x
1
−p
2
x
2
,
max
x
1
.x
2
p(2
√
x
1
+ 2
√
x
2
) −p
1
x
1
−p
2
x
2
FOC:
∂Π
∂x
1
=
p
√
x
1
−p
1
= 0 and
∂Π
∂x
2
=
p
√
x
2
−p
2
= 0.
⇒x
1
= (p/p
1
)
2
, x
2
= (p/p
2
)
2
(input demand functions) and
q = 2(p/p
1
) + 2(p/p
2
) = 2p(
1
p
1
+
1
p
2
) (the supply function)
90
Π = p
2
(
1
p
1
+
1
p
2
)
SOC:
∂
2
Π
∂x
2
1
∂
2
Π
∂x
1
∂x
2
∂
2
Π
∂x
1
∂x
2
∂
2
Π
∂x
2
1
¸
¸
¸
¸
=
−p
2x
3/2
1
0
0
−p
2x
3/2
2
¸
¸
¸
¸
is negative deﬁnite.
10.3 Transformation function and proﬁt maximization
In more general cases, the technology of a producer is represented by a transformation
function: F
j
(y
j
1
, . . . , y
j
n
) = 0, where (y
j
1
, . . . , y
j
n
) is called a production plan, if y
j
k
> 0
(y
j
k
) then k is an output (input) of j.
Example: a producer produces two outputs, y
1
and y
2
, using one input y
3
. Its
technology is given by the transformation function (y
1
)
2
+ (y
2
)
2
+ y
3
= 0. Its proﬁt
is Π = p
1
y
1
+p
2
y
2
+p
3
y
3
. The maximization problem is
max
y
1
,y
2
,y
3
p
1
y
1
+p
2
y
2
+p
3
y
3
subject to (y
1
)
2
+ (y
2
)
2
+y
3
= 0.
To solve the maximization problem, we can eliminate y
3
: x = −y
3
= (y
1
)
2
+(y
2
)
2
> 0
and
max
y
1
,y
2
p
1
y
1
+p
2
y
2
−p
3
[(y
1
)
2
+ (y
2
)
2
].
The solution is: y
1
= p
1
/(2p
3
), y
2
= p
2
/(2p
3
) (the supply functions of y
1
and y
2
), and
x = −y
3
= [p
1
/(2p
3
)]
2
+ [p
2
/(2p
3
)]
2
(the input demand function for y
3
).
10.4 The concept of an abstract economy and a competitive equilibrium
Commodity space: Assume that there are n commodities. The commodity space is
R
n
+
= {(x
1
, . . . , x
n
); x
k
≥ 0}
Economy: There are I consumers, J producers, with initial endowments of com
modities ω = (ω
1
, . . . , ω
n
).
Consumer i has a utility function U
i
(x
i
1
, . . . , x
i
n
), i = 1, . . . , I.
Producer j has a production transformation function F
j
(y
j
1
, . . . , y
j
n
) = 0,
A price system: (p
1
, . . . , p
n
).
A private ownership economy: Endowments and ﬁrms (producers) are owned by
consumers.
Consumer i’s endowment is ω
i
= (ω
i
1
, . . . , ω
i
n
),
¸
I
i=1
ω
i
= ω.
Consumer i’s share of ﬁrm j is θ
ij
≥ 0,
¸
I
i=1
θ
ij
= 1.
An allocation: x
i
= (x
i
1
, . . . , x
i
n
), i = 1, . . . , I, and y
j
= (y
j
1
, . . . , y
j
n
), j = 1, . . . , J.
91
A competitive equilibrium:
A combination of a price system ¯ p = (¯ p
1
, . . . , ¯ p
n
) and an allocation ({¯ x
i
}
i=1,...,I
, {¯ y
j
}
j=1,...,J
)
such that
1.
¸
i
¯ x
i
= ω +
¸
j
¯ y
j
(feasibility condition).
2. ¯ y
j
maximizes Π
j
, j = 1, . . . , J and ¯ x
i
maximizes U
i
, subject to i’s budget con
straint p
1
x
i
1
+. . . +p
n
x
i
n
= p
1
ω
1
1
+. . . +p
n
ω
i
n
+θ
i1
Π
1
+. . . +θ
iJ
Π
J
.
Existence Theorem:
Suppose that the utility functions are all quasiconcave and the production transfor
mation functions satisfy some theoretic conditions, then a competitive equilibrium
exists.
Welfare Theorems: A competitive equilibrium is eﬃcient and an eﬃcient allocation
can be achieved as a competitive equilibrium through certain income transfers.
Constant returns to scale economies and nonsubstitution theorem:
Suppose there is only one nonproduced input, this input is indispensable to produc
tion, there is no joint production, and the production functions exhibits constant
returns to scale. Then the competitive equilibrium price system is determined by the
production side only.
92
10.5 Multiperson Decision Problem and Game Theory
In this chapter, we consider the situation when there are n > 1 persons with diﬀerent
objective (utility) functions; that is, diﬀerent persons have diﬀerent preferences over
possible outcomes. There are two cases:
1. Game theory: The outcome depends on the behavior of all the persons involved.
Each person has some control over the outcome; that is, each person controls certain
strategic variables. Each one’s utility depends on the decisions of all persons. We
want to study how persons make decisions.
2. Public Choice: Persons have to make decision collectively, eg., by voting.
We consider only game theory here.
Game theory: the study of conﬂict and cooperation between persons with diﬀer
ent objective functions.
Example (a 3person game): The accuracy of shooting of A, B, C is 1/3, 2/3, 1,
respectively. Each person wants to kill the other two to become the only survivor.
They shoot in turn starting A.
Question: What is the best strategy for A?
10.6 Ingredients and classiﬁcations of games
A game is a collection of rules known to all players which determine what players
may do and the outcomes and payoﬀs resulting from their choices.
The ingredients of a game:
1. Players: Persons having some inﬂuences upon possible income (decision mak
ers).
2. Moves: decision points in the game at which players must make choices between
alternatives (personal moves) and randomization points (called nature’s moves).
3. A play: A complete record of the choices made at moves by the players and
realizations of randomization.
4. Outcomes and payoﬀs: a play results in an outcome, which in turn determines
the rewords to players.
Classiﬁcations of games:
1. according to number of players:
2person games – conﬂict and cooperation possibilities.
nperson games – coalition formation possibilities in addition.
inﬁniteplayers’ games – corresponding to perfect competition in economics.
2. according to number of strategies:
ﬁnite – strategy (matrix) games, each person has a ﬁnite number of strategies,
93
payoﬀ functions can be represented by matrices.
inﬁnite – strategy (continuous or discontinuous payoﬀ functions) games like
duopoly games.
3. according to sum of payoﬀs:
0sum games – conﬂict is unavoidable.
nonzero sum games – possibilities for cooperation.
4. according to preplay negotiation possibility:
noncooperative games – each person makes unilateral decisions.
cooperative games – players form coalitions and decide the redistribution of
aggregate payoﬀs.
10.7 The extensive form and normal form of a game
Extensive form: The rules of a game can be represented by a game tree.
The ingredients of a game tree are:
1. Players
2. Nodes: they are players’ decision points (personal moves) and randomization
points (nature’s moves).
3. Information sets of player i: each player’s decision points are partitioned into
information sets. An information set consists of decision points that player i can not
distinguish when making decisions.
4. Arcs (choices): Every point in an information set should have the same number of
choices.
5. Randomization probabilities (of arcs following each randomization points).
6. Outcomes (end points)
7. Payoﬀs: The gains to players assigned to each outcome.
A pure strategy of player i: An instruction that assigns a choice for each information
set of player i.
Total number of pure strategies of player i: the product of the numbers of choices of
all information sets of player i.
Once we identify the pure strategy set of each player, we can represent the game
in normal form (also called strategic form).
1. Strategy sets for each player: S
1
= {s
1
, . . . , s
m
}, S
2
= {σ
1
, . . . , σ
n
}.
2. Payoﬀ matrices: π
1
(s
i
, σ
j
) = a
ij
, π
2
(s
i
, σ
j
) = b
ij
. A = [a
ij
], B = [b
ij
].
Normal form:
II
I
σ
1
. . . σ
n
s
1
(a
11
, b
11
) . . . (a
1n
, b
1n
)
.
.
.
.
.
.
.
.
.
.
.
.
s
m
(a
m1
, b
m1
) . . . (a
mn
, b
mn
)
94
10.8 Examples
Example 1: A perfect information game
`
`
`
`
1
L R
`
2
l r
`
2
L R
1
9
9
6
3
7
8
2
S
1
= { L, R }, S
2
= { Ll, Lr, Rl, Rr }.
II
I
Ll Lr Rl Rr
L (1,9) (9,6) (1,9) (9,6)
R (3,7)* (8,2) (3,7) (8,2)
Example 2: Prisoners’ dilemma game
`
`
`
`
1
L R
2
L R L R
4
4
0
5
5
0
1
1
S
1
= { L, R }, S
2
= { L, R }.
II
I
L R
L (4,4) (0,5)
R (5,0) (1,1)*
Example 3: Hijack game
`
`
`
`
1
L R
`
2
L R
−1
2
2
−2
−10
−10
S
1
= { L, R }, S
2
= { L, R }.
II
I
L R
L (1,2) (1,2)*
R (2,2)* (10,10)
Example 4: A simpliﬁed stock price manipulation game
>
>
>
>
`
0
1/2 1/2
`
`
`
1
L R
`
1
l r
`
`
`
`
2
L R L R
4
2
7
5
5
7
4
5
4
2
3
7
S
1
= { Ll, Lr, Rl, Rr }, S
2
= { L, R }.
II
I
L R
Ll (4, 3.5) (4, 2)
Lr (3.5, 4.5) (3.5, 4.5)
Rl (5.5, 5)* (4.5, 4.5)
Rr (5,6) (4,7)
Remark: Each extensive form game corresponds a normal form game. However,
diﬀerent extensive form games may have the same normal form.
95
10.9 Strategy pair and pure strategy Nash equilibrium
1. A Strategy Pair: (s
i
, σ
j
). Given a strategy pair, there corresponds a payoﬀ pair
(a
ij
, b
ij
).
2. A Nash equilibrium: A strategy pair (s
i∗
, σ
j∗
) such that a
i∗j∗
≥ a
ij∗
and b
i∗j∗
≥
b
i∗j
for all (i, j). Therefore, there is no incentives for each player to deviate from
the equilibrium strategy. a
i∗j∗
and b
i∗j∗
are called the equilibrium payoﬀ.
The equilibrium payoﬀs of the examples are marked each with a star in the normal
form.
Remark 1: It is possible that a game does no have a pure strategy Nash equilib
rium. Also, a game can have more than one Nash equilibria.
Remark 2: Notice that the concept of a Nash equilibrium is deﬁned for a normal form
game. For a game in extensive form (a game tree), we have to ﬁnd the normal form
before we can ﬁnd the Nash equilibria.
10.10 Subgames and subgame perfect Nash equilibria
1. Subgame: A subgame in a game tree is a part of the tree consisting of all the
nodes and arcs following a node that form a game by itself.
2. Within an extensive form game, we can identify some subgames.
3. Also, each pure strategy of a player induces a pure strategy for every subgame.
4. Subgame perfect Nash equilibrium: A Nash equilibrium is called subgame
perfect if it induces a Nash equilibrium strategy pair for every subgame.
5. Backward induction: To ﬁnd a subgame perfect equilibrium, usually we work
backward. We ﬁnd Nash equilibria for lowest level (smallest) subgames and
replace the subgames by its Nash equilibrium payoﬀs. In this way, the size of
the game is reduced step by step until we end up with the equilibrium payoﬀs.
All the equilibria, except the equilibrium strategy pair (L,R) in the hijack game, are
subgame perfect.
Remark: The concept of a subgame perfect Nash equilibrium is deﬁned only for an
extensive form game.
10.10.1 Perfect information game and Zemelo’s Theorem
An extensive form game is called perfect information if every information set consists
only one node. Every perfect information game has a pure strategy subgame perfect
Nash Equilibrium.
96
10.10.2 Perfect recall game and Kuhn’s Theorem
A local strategy at an information set u ∈ U
i
: A probability distribution over the
choice set at U
ij
.
A behavior strategy: A function which assigns a local strategy for each u ∈ U
i
.
The set of behavior strategies is a subset of the set of mixed strategies.
Kuhn’s Theorem: In every extensive game with perfect recall, a strategically equiva
lent behavior strategy can be found for every mixed strategy.
However, in a nonperfect recall game, a mixed strategy may do better than be
havior strategies because in a behavior strategy the local strategies are independent
whereas they can be correlated in a mixed strategy.
>
>
>
>
>
>
`
`
`
`
`
`
`
`
`
`
`
`
0
u
11
2
u
12
1/2 1/2
a b
A B
c d c d
¸
1
−1
¸
−1
1
¸
2
−2
¸
0
0
¸
−2
2
¸
0
0
A 2person 0sum nonperfect recall game.
NE is (µ
∗
1
, µ
∗
2
) = (
1
2
ac ⊕
1
2
bd,
1
2
A⊕
1
2
B).
µ
∗
1
is not a behavioral strategy.
10.10.3 Reduction of a game
Redundant strategy: A pure strategy is redundant if it is strategically identical to
another strategy.
Reduced normal form: The normal form without redundant strategies.
Equivalent normal form: Two normal forms are equivalent if they have the same
reduced normal form.
Equivalent extensive form: Two extensive forms are equivalent if their normal forms
are equivalent.
10.11 Continuous games and the duopoly game
In many applications, S
1
and S
2
are inﬁnite subsets of R
m
and R
n
Player 1 controls
m variables and player 2 controls n variables (however, each player has inﬁnite many
strtategies). The normal form of a game is represented by two functions
Π
1
= Π
1
(x; y) and Π
2
= Π
2
(x; y), where x ∈ S
1
⊂ R
m
and y ∈ S
2
⊂ R
n
.
To simplify the presentation, assume that m = n = 1. A strategic pair is (x, y) ∈
S
1
×S
2
. A Nash equilibrium is a pair (x
∗
, y
∗
) such that
Π
1
(x
∗
, y
∗
) ≥ Π
1
(x, y
∗
) and Π
2
(x
∗
, y
∗
) ≥ Π
2
(x
∗
, y) for all x ∈ S
1
y ∈ S
2
.
97
Consider the case when Π
i
are continuously diﬀerentiable and Π
1
is strictly concave
in x and Π
2
strictly concave in y (so that we do not have to worry about the SOC’s).
Reaction functions and Nash equilibrium:
To player 1, x is his endogenous variable and y is his exogenous variable. For each y
chosen by player 2, player 1 will choose a x ∈ S
1
to maximize his objective function
Π
1
. This relationship deﬁnes a behavioral equation x = R
1
(y) which can be obtained
by solving the FOC for player 1, Π
1
x
(x; y) = 0. Similarly, player 2 regards y as en
dogenous and x exogenous and wants to maximize Π
2
for a given x chosen by player
1. Player 2’s reaction function (behavioral equation) y = R
2
(x) is obtained by solving
Π
2
y
(x; y) = 0. A Nash equilibrium is an intersection of the two reaction functions.
The FOC for a Nash equilibrium is given by Π
1
x
(x
∗
; y
∗
) = 0 and Π
2
y
(x
∗
; y
∗
) = 0.
Duopoly game:
There are two sellers (ﬁrm 1 and ﬁrm 2) of a product.
The (inverse) market demand function is P = a −Q.
The marginal production costs are c
1
and c
2
, respectively.
Assume that each ﬁrm regards the other ﬁrm’s output as given (not aﬀected by his
output quantity).
The situation deﬁnes a 2person game as follows: Each ﬁrm i controls his own output
quantity q
i
. (q
1
, q
2
) together determine the market price P = a − (q
1
+ q
2
) which in
turn determines the proﬁt of each ﬁrm:
Π
1
(q
1
, q
2
) = (P−c
1
)q
1
= (a−c
1
−q
1
−q
2
)q
1
and Π
2
(q
1
, q
2
) = (P−c
2
)q
2
= (a−c
2
−q
1
−q
2
)q
2
The FOC are ∂Π
1
/∂q
1
= a −c
1
−q
2
−2q
1
= 0 and ∂Π
2
/∂q
2
= a −c
2
−q
1
−2q
2
= 0.
The reaction functions are q
1
= 0.5(a −c
1
−q
2
) and q
2
= 0.5(a −c
2
−q
1
).
The Cournot Nash equilibrium is (q
∗
1
, q
∗
2
) = ((a − 2c
1
+ c
2
)/3, (a − 2c
2
+ c
1
)/3) with
P
∗
= (a +c
1
+c
2
)/3. (We have to assume that a −2c
1
+c
2
, a −2c
2
+c
1
≥ 0.)
10.11.1 A simple bargaining model
Two players, John and Paul, have $ 1 to divide between them. They agree to spend
at most two days negotiating over the division. The ﬁrst day, John will make an oﬀer,
Paul either accepts or comes back with a counteroﬀer the second day. If they cannot
reach an agreement in two days, both players get zero. John (Paul) discounts payoﬀs
in the future at a rate of α (β) per day.
A subgame perfect equilibrium of this bargaining game can be derived using back
ward induction.
1. On the second day, John would accept any nonnegative counteroﬀer made by
Paul. Therefore, Paul would make proposal of getting the whole $ 1 himself and John
would get $ 0.
2. On the ﬁrst day, John should make an oﬀer such that Paul gets an amount equiv
alent to getting $ 1 the second day, otherwise Paul will reject the oﬀer. Therefore,
John will propose of 1 −β for himself and β for Paul and Paul will accept the oﬀer.
98
An example of a subgame nonperfect Nash equilibrium is that John proposes of
getting 10.5β for himself and 0.5β for Paul and refuses to accept any counteroﬀer
made by Paul. In this equilibrium, Paul is threatened by John’s incredible threat and
accepts only one half of what he should have had in a perfect equilibrium.
10.12 2person 0sum game
1. B = −A so that a
ij
+b
ij
= 0.
2. Maxmin strategy: If player 1 plays s
i
, then the minimum he will have is min
j
a
ij
,
called the security level of strategy s
i
. A possible guideline for player 1 is to
choose a strategy such that the security level is maximized: Player 1 chooses
s
i∗
so that min
j
a
i∗j
≥ min
j
a
ij
for all i. Similarly, since b
ij
= −a
ij
, Player 2
chooses σ
j
∗ so that max
i
a
ij∗
≤ max
i
a
ij
for all j.
3. Saddle point: If a
i∗j∗
= max
i
min
j
a
ij
= min
j
max
i
a
ij
, then (s
i∗
, σ
j∗
) is called a
saddle point. If a saddle point exists, then it is a Nash equilibrium.
A
1
=
2 1 4
−1 0 6
A
2
=
1 0
0 1
In example A
1
, max
i
min
j
a
ij
= min
j
max
i
a
ij
= 1 (s
1
, σ
2
) is a saddle point and
hence a Nash equilibrium. In A
2
, max
i
min
j
a
ij
= 0 = min
j
max
i
a
ij
= 1 and no
saddle point exists. If there is no saddle points, then there is no pure strategy
equilibrium.
4. Mixed strategy for player i: A probability distribution over S
i
. p = (p
1
, . . . , p
m
),
q = (q
1
, . . . , q
n
)
′
. (p, q) is a mixed strategy pair. Given (p, q), the expected
payoﬀ of player 1 is pAq. A mixed strategy Nash equilibrium (p
∗
, q
∗
) is such
that p
∗
Aq
∗
≥ pAq
∗
and p
∗
Aq
∗
≤ p
∗
Aq for all p and all q.
5. Security level of a mixed strategy: Given player 1’s strategy p, there is a pure
strategy of player 2 so that the expected payoﬀ to player 1 is minimized, just
as in the case of a pure strategy of player 1.
t(p) ≡ min
j
{
¸
i
p
i
a
i1
, . . . ,
¸
i
p
i
a
in
}.
The problem of ﬁnding the maxmin mixed strategy (to ﬁnd p
∗
to maximize
t(p)) can be stated as
max
p
t subj. to
¸
i
p
i
a
i1
≥ t, . . . ,
¸
i
p
i
a
in
≥ t,
¸
i
p
i
= 1.
6. Linear programming problem: The above problem can be transformed into a
linear programming problem as follows: (a) Add a positive constant to each
element of A to insure that t(p) > 0 for all p. (b) Deﬁne y
i
≡ p
i
/t(p) and
99
replace the problem of max t(p) with the problem of min 1/t(p) =
¸
i
y
i
. The
constraints become
¸
i
y
i
a
i1
≥ 1, . . . ,
¸
i
y
i
a
in
≥ 1.
min
y
1
,...,ym≥0
y
1
+. . . +y
m
subj. to
¸
i
y
i
a
i1
≥ 1, . . . ,
¸
i
y
i
a
in
≥ 1
7. Duality: It turns out that player 2’s minmax problem can be transformed sim
ilarly and becomes the dual of player 1’s linear programming problem. The
existence of a mixed strategy Nash equilibrium is then proved by using the
duality theorem in linear programming.
Example (tossing coin game): A =
1 0
0 1
.
To ﬁnd player 2’s equilibrium mixed strategy, we solve the linear programming prob
lem:
max
x
1
,x
2
≥0
x
1
+x
2
subj. to x
1
≤ 1 x
2
≤ 1.
The solution is x
1
= x
2
= 1 and therefore the equilibrium strategy for player 2 is
q
∗
1
= q
∗
2
= 0.5.
x
1
`
x
2
1
1
,
y
1
`
y
2
1
1
,
Player 1’s equilibrium mixed strategy is obtained by solving the dual to the linear
programming problem:
min
y
1
,y
2
≥0
y
1
+y
2
subj. to y
1
≥ 1 y
2
≥ 1.
The solution is p
∗
1
= p
∗
2
= 0.5.
10.13 Mixed strategy equilibria for nonzero sum games
The idea of a mixed strategy equilibrium is also applicable to a nonzero sum game.
Similar to the simplex algorism for the 0sum games, there is a Lemke algorism.
Example (Game of Chicken)
100
`
`
`
`
1
S N
2
S N S N
0
0
−3
3
3
−3
−9
−9
S
1
= { S, N }, S
2
= { S, N }.
II
I
Swerve Don’t
Swerve (0,0) (3,3)*
Don’t (3,3)* (9,9)
There are two pure strategy NE: (S, N) and (N, S).
There is also a mixed strategy NE. Suppose player 2 plays a mixed strategy (q, 1−q).
If player 1 plays S, his expected payoﬀ is Π
1
(S) = 0q + (−3)(1 − q). If he plays
N, his expected payoﬀ is Π
1
(N) = 3q + (−9)(1 − q). For a mixed strategy NE,
Π
1
(S) = Π
1
(N), therefore, q =
2
3
.
The mixed strategy is symmetrical: (p
∗
1
, p
∗
2
) = (q
∗
1
, q
∗
2
) = (
2
3
,
1
3
).
Example (Battle of sex Game)
`
`
`
`
1
B O
2
B O B O
5
4
0
0
0
0
4
5
S
1
= { B, O }, S
2
= { B, O }.
II
I
Ball game Opera
Ball game (5,4)* (0,0)
Opera (0,0) (4,5)*
There are two pure strategy NE: (B, B) and (O, O).
There is also a mixed strategy NE. Suppose player 2 plays a mixed strategy (q, 1−q).
If player 1 plays B, his expected payoﬀ is Π
1
(B) = 5q +(0)(1 −q). If he plays O, his
expected payoﬀ is Π
1
(O) = 0q+(4)(1−q). For a mixed strategy NE, Π
1
(B) = Π
1
(O),
therefore, q =
4
9
.
The mixed strategy is: (p
∗
1
, p
∗
2
) = (
5
9
,
4
9
) and (q
∗
1
, q
∗
2
) = (
4
9
,
5
9
).
10.14 Cooperative Game and Characteristic form
2person 0sum games are strictly competitive. If player 1 gains $ 1, player 2 will loss
$ 1 and therefore no cooperation is possible. For other games, usually some coopera
tion is possible. The concept of a Nash equilibrium is deﬁned for the situation when
no explicit cooperation is allowed. In general, a Nash equilibrium is not eﬃcient (not
Pareto optimal). When binding agreements on strategies chosen can be contracted
before the play of the game and transfers of payoﬀs among players after a play of the
game is possible, players will negotiate to coordinate their strategies and redistribute
the payoﬀs to achieve better results. In such a situation, the determination of strate
gies is not the key issue. The problem becomes the formation of coalitions and the
distribution of payoﬀs.
101
Characteristic form of a game:
The player set: N = {1, 2, . . . , n}.
A coalition is a subset of N: S ⊂ N.
A characteristic function v speciﬁes the maximum total payoﬀ of each coalition.
Consider the case of a 3person game. There are 8 subsets of N = {1, 2, 3}, namely,
φ, (1), (2), (3), (12), (13), (23), (123).
Therefore, a characteristic form game is determined by 8 values v(φ), v(1), v(2), v(3), v(12), v(13), v(23
Superadditivity: If A∩ B = φ, then v(A∪ B) ≥ v(A) +v(B).
An imputation is a payoﬀ distribution (x
1
, x
2
, x
3
).
Individual rationality: x
i
≥ v(i).
Group rationality:
¸
i∈S
x
i
≥ v(S).
Core C: the set of imputations that satisfy individual rationality and group rational
ity for all S.
Marginal contribution of player i in a coalition S ∪ i: v(S ∪ i) −v(S)
Shapley value of player i is an weighted average of all marginal contributions
π
i
=
¸
S⊂N
S!(n −S −1)!
n!
[v(S ∪ i) −v(S)].
Example: v(φ) = v(1) = v(2) = v(3) = 0, v(12) = v(13) = v(23) = 0.5, v(123) = 1.
C = {(x
1
, x
2
, x
3
), x
i
≥ 0, x
i
+ x
j
≥ 0.5, x
1
+ x
2
+ x
3
= 1}. Both (0.3, 0.3, 0.4) and
(0.2, 0.4, 0.4) are in C.
The Shapley values are (π
1
, π
2
, π
3
) = (
1
3
,
1
3
,
1
3
).
Remark 1: The core of a game can be empty. However, the Shapley values are
uniquely determined.
Remark 2: Another related concept is the vonNeumann Morgenstern solution. See
CH 6 of Intriligator’s Mathematical Optimization and Economic Theory for the mo
tivations of these concepts.
10.15 The Nash bargaining solution for a nontransferable 2person coop
erative game
In a nontransferable cooperative game, afterplay redistributions of payoﬀs are im
possible and therefore the concepts of core and Shapley values are not suitable. For
the case of 2person games, the concept of Nash bargaining solutions are useful.
Let F ⊂ R
2
be the feasible set of payoﬀs if the two players can reach an agreement
and T
i
the payoﬀ of player i if the negotiation breaks down. T
i
is called the threat
point of player i. The Nash bargaining solution (x
∗
1
, x
∗
2
) is deﬁned to be the solution
to the following problem:
102
x
1
`
x
2
T
1
T
2
x
∗
1
x
∗
2
max
(x
1
,x
2
)∈F
(x
1
−T
1
)(x
2
−T
2
)
See CH 6 of Intriligator’s book for the motivations of the solution concept.
10.16 Problems
1. Consider the following twoperson 0sum game:
I \ II σ
1
σ
2
σ
3
s
1
4 3 2
s
2
3 4 10
s
3
7 6 8
(a) Find the max min strategy of player I s
max min
and the min max strategy
of player II σ
min max
.
(b) Is the strategy pair (s
max min
, σ
min max
) a Nash equilibrium of the game?
(c) What are the equilibrium payoﬀs?
2. Find the maxmin strategy (s
max min
) and the minmax strategy (σ
min max
) of the
following twoperson 0sum game:
I \ II σ
1
σ
2
s
1
3 6
s
2
8 2
s
3
6 3
Is the strategy pair (s
max min
, σ
min max
) a Nash equilibrium? If not, use simplex
method to ﬁnd the mixed strategy Nash equilibrium.
3. Find the (mixed strategy) Nash Equilibrium of the following twoperson game:
I \ II H T
H (2, 2) (2, 1)
T (2, 2) (1,2)
4. Suppose that two ﬁrms producing a homogenous product face a linear demand
curve P = a−bQ = a−b(q
1
+q
2
) and that both have the same constant marginal
costs c. For a given quantity pair (q
1
, q
2
), the proﬁts are Π
i
= q
i
(P − c) =
q
i
(a − bq
1
− bq
2
− c), i = 1, 2. Find the Cournot Nash equilibrium output of
each ﬁrm.
5. Suppose that in a twoperson cooperative game without side payments, if the
two players reach an agreement, they can get (Π
1
, Π
2
) such that Π
2
1
+ Π
2
= 47
103
and if no agreement is reached, player 1 will get T
1
= 3 and player 2 will get
T
2
= 2.
(a) Find the Nash solution of the game.
(b) Do the same for the case when side payments are possible. Also answer
how the side payments should be done?
6. A singer (player 1), a pianist (player 2), and a drummer (player 3) are oﬀered
$ 1,000 to play together by a night club owner. The owner would alternatively
pay $ 800 the singerpiano duo, $ 650 the piano drums duo, and $ 300 the piano
alone. The night club is not interested in any other combination. Howeover,
the singerdrums duo makes $ 500 and the singer alone gets $ 200 a night in a
restaurant. The drums alone can make no proﬁt.
(a) Write down the characteristic form of the cooperative game with side pay
ments.
(b) Find the Shapley values of the game.
(c) Characterize the core.
Contents
Lecture 1 Lecture 2 Lecture 3 Lecture 4 Lecture 5 Lecture 6 Lecture 7 Lecture 8 Lecture 9 Lecture 10 Static Economic Models and The Concept of Equilibrium Matrix Algebra Vector Space and Linear Transformation Determinant, Inverse Matrix, and Cramer’s rule Diﬀerential Calculus and Comparative Statics Comparative Statics – Economic applications Optimization Optimization–multivariate case Optimization with equality constraints and Nonlinear Programming General Equilibrium and Game Theory 1 5 10 16 25 36 44 61 74 89
1
1
Static Economic Models and The Concept of Equilibrium
Here we use three elementary examples to illustrate the general structure of an economic model. 1.1 Partial market equilibrium model
A partial market equilibrium model is constructed to explain the determination of the price of a certain commodity. The abstract form of the model is as follows. Qd = D(P ; a) Qd : Qs : P: a: Qs = S(P ; a) Qd = Qs ,
quantity demanded of the commodity quantity supplied to the market market price of the commodity a factor that aﬀects demand and supply
D(P ; a): demand function S(P ; a): supply function
Equilibrium: A particular state that can be maintained. Equilibrium conditions: Balance of forces prevailing in the model. Substituting the demand and supply functions, we have D(P ; a) = S(P ; a). For a given a, we can solve this last equation to obtain the equilibrium price P ∗ as a function of a. Then we can study how a aﬀects the market equilibrium price by inspecting the function. Example: D(P ; a) = a2 /P , S(P ) = 0.25P . a2 /P ∗ = 0.25P ∗ ⇒ P ∗ = 2a, Q∗ = Q∗ = s d 0.5a. 1.2 General equilibrium model
Usually, markets for diﬀerent commodities are interrelated. For example, the price of personal computers is strongly inﬂuenced by the situation in the market of microprocessors, the price of chicken meat is related to the supply of pork, etc. Therefore, we have to analyze interrelated markets within the same model to be able to capture such interrelationship and the partial equilibrium model is extended to the general equilibrium model. In microeconomics, we even attempt to include every commodity (including money) in a general equilibrium model. Qd1 = D1 (P1 , . . . , Pn ; a) Qd2 = D2 (P1 , . . . , Pn ; a) . . . Qdn = Dn (P1 , . . . , Pn ; a) Qsn = Sn (P1 , . . . , Pn ; a) Qs2 = S2 (P1 , . . . , Pn ; a) Qs1 = S1 (P1 , . . . , Pn ; a) Qdn = Qsn Qd2 = Qs2 Qd1 = Qs1 Qdi : quantity demanded of commodity i Qsi : quantity supplied of commodity i Pi : market price of commodity i a: a factor that aﬀects the economy Di (P1 , . . . , Pn ; a): demand function of commodity i Si (P1 , . . . , Pn ; a): supply function of commodity i We have three variables and three equations for each commodity/market.
a) = 0.4 The ingredients of a model We set up economic models to study economic phenomena (causeeﬀect relationships). . . (a0 − b0 ) + (a1 − b1 )P1 + (a2 − b2 )P2 = 0 (α0 − β0 ) + (α1 − β1 )P1 + (α2 − β2 )P2 = 0. Pn .2 Substituting the demand and supply functions. . . which are relationships among variables. a) = 0 . Causes (Changes in exogenous var. . . . . .) ⇒ Eﬀects (Changes in endogenous var. . . . A 2market linear model: D1 = a0 + a1 P1 + a2 P2 . or how certain economic variables aﬀect other variables. There are n equations and n unknown. we substitute the ﬁrst two equations into the third to obtain ¯ ¯ Y = a + bY + I0 + G ⇒ Y ∗ = (a + I(r) + G)/(1 − b). a) − S1 (P1 . . Pn . . . . S1 = b0 + b1 P1 + b2 P2 . Exogenous variables: variables assumed to aﬀect the endogenous variables but are not aﬀected by them. a) ≡ En (P1 . A model consists of equations. . . Pn . . . . Variables can be divided into three categories: Endogenous variables: variables we choose to represent diﬀerent states of a model.3 National income model The most fundamental issue in macroeconomics is the determination of the national income of a country. . .) Parameters: Coeﬃcients of the equations. . Pn . . 1. Pn . . a) ≡ E2 (P1 . Pn ). . a) ≡ E1 (P1 . 1. Dn (P1 . . b: coeﬃcients of the consumption function. . . . a) − Sn (P1 . . . . Pn . Pn . Pn . In principle. Pn ). Pn . we have D1 (P1 . . . C: Consumption Y : National income I: Investment S: Savings ¯ G: government expenditure r: interest rate a. . . . a) − S2 (P1 . . we can solve the simultaneous equation to ﬁnd the ∗ ∗ equilibrium prices (P1 . it is a simultaneous equation in (P1 . . C I Y S = = = = a + bY (a > 0. S2 = β0 + β1 P1 + β2 P2 . D2 = α0 + α1 P1 + α2 P2 . . For a given a. 0 < b < 1) I(r) ¯ C +I +G Y − C. a) = 0 D2 (P1 . . . To solve the model. . . . . . .
. . a) a. y1. . Qs = S(P . . . y2 . Pn ) ¯ Income model: C = a + bY . Qsi C. . . In general. . . G Parameters Coeﬃcients of D(P . . Behavioral equations Equilibrium cond. I = I(r) Y =C +I +G S =Y −C 1. a a ¯ r. some are equilibrium conditions. . it is usually very diﬃcult to solve the model. . . Some of the equations are behavioral. ym ). y2 . . . . Qdi . S Ex. Deﬁnitions Partial equilibrium model: Qd = D(P . Var. In principle. ym ) x2 = x1 (y1 . However. if some equations are nonlinear. . xn = x1 (y1 . P . x2 . Qd . x2 . . . ym. . . y2 . we use comparative statics method (to be discussed in part 3) to ﬁnd the diﬀerential relationships between ∂x xi and yj : ∂yji . xn ). . . S(P . There should be n equations so that the model can be solved. . .5 The general economic model Assume that there are n endogenous variables and m exogenous variables. Deﬁnitions: to introduce new variables into the model. Fn (x1 . . . . ym ) . a) Qd = Qs General equilibrium model: Qdi = Di (P1 . . . . x2 . ym) = 0. . . . xn . Pn ). Qs Pi .3 End. x2 . Equilibrium conditions: the condition such that the state can be maintained (when diﬀerent forces/motivations are in balance). y2 . . . . I. . y2 . If the equations are all linear in (x1 . Qdi = Qsi Qsi = Si (P1 . . y1 . ym ) = 0 . . . y1. Var. . . . . a). . . y2 . . ym ) = 0 F2 (x1 . . a). xn Exogenous variables: y1 . given the values of the exogenous variables. . we solve to ﬁnd the endogenous variables as functions of the exogenous variables: x1 = x1 (y1 . . F1 (x1 . . y2 . . . . . . Endogenous variables: x1 . . xn . . . b Partial equilibrium model: General equilibrium model: Income model: Equations can be divided into three types: Behavioral equations: representing the decisions of economic agents in the model. Y . then we can use Cramer’s rule (to be discussed in the next part) to solve the equations. xn . and some are deﬁnitions. x2 . .
The partial market model becomes Qd = D(P1 + t). Find the equilibrium national income. P2 ) and Q2 = D 2 (P1 . The demand and supply functions of a twocommodity model are as follows: Qd1 = 18 − 3P1 + P2 Qs1 = −2 + 4P1 Qd2 = 12 + P1 − 2P2 Qs2 = −2 + 3P2 Find the equilibrium of the model. 5.4 1. Calculate P1 and Q1 both as function of t. 2. What restriction(s) on the parameters is needed for an economically reasonable solution to exist? C = 25 + 6Y 1/2 . I0 = 16. 3. (The eﬀect of a sales tax) Suppose that the government imposes a sales tax of t dollars per unit on product 1. P2 ). ∗ ∗ Find the equilibrium prices (P1 . Let the nationalincome Y = C = G = (a) (b) (c) (d) model be: C + I0 + G a + b(Y − T0 ) gY (a > 0. P2 ) and quantities (Q∗ . G0 = 14. Q∗ ) as functions 1 2 of a. In a 2good market equilibrium model. (c) If t increases. Qs = Qd . (a) Find the demand functions Q1 = D 1 (P1 . Q2 = P2 . the equilibrium price is determined by D(P1 + t) = 1 1 S(P1 ). Give the economic meaning of the parameter g. Find the equilibrium solution of the following model: Qs = 6P − 4. (b) Suppose that the supply functions are Q1 = a−1 P1 . Find the equilibrium Y and C from the following: Y = C + I0 + G0 . P2 = Q1 Q23 . will P1 and Q1 increase or decrease? 4. 6. 1 1 Eliminating Qd and Qs . the inverse demand functions are given by −2 −2 1 1 3 3 P1 = Q13 Q2 . (a) Identify the endogenous variables and exogenous variable(s).6 Problems Qd = 3 − P 2 . (b) Let D(p) = 120 − P and S(p) = 2P . 1 Qs = S(P1 ). 1 Qd = Qs . 0 < b < 1) (0 < g < 1) Identify the endogenous variables. 1. .
A−B = 0 2 3 3 . . . In that case. c2p A= . B = . . . a2n b21 b22 . . A= 1 2 3 4 . . . . . . Addition/subtraction: A + B and A − B can be deﬁned only when size(A) = size(B). then A is a square matrix. . ⇒ 4A = . am1 am2 . . (1) it is not always deﬁned.5 2 Matrix Algebra There are n columns each with m elements or m rows each with n elements. We say that the size of A is m × n. . . . . . . . . For 1 2 4 8 example. For example. . . . . . b1p c11 c12 . . A = . . . called a scalar number. .1 Matrix operations A matrix is a two dimensional rectangular array of numbers: a11 a12 . in that case. A 1 × 1 matrix is just a number. . A m × 1 matrix is called a column vector and a 1 × n matrix is called a row vector. size (AB) = size (C) = m × p. . . . 2. . . is more complicated. 3 4 12 16 Multiplication of two matrices: Let size(A) = m × n and size(B) = o × p. a1n a21 a22 . denoted by αA. a11 a12 . ⇒C = . . . (A − B)ij = aij − bij . .. . amn Equality: A = B ⇒ (1) size(A) = size(B). . If m = n. . is always deﬁned with size (αA) = size(A) and (αA)ij = αaij . a1n b11 b12 . 0 5 4 0 + 20 20 5 6 5 + 14 6 + 16 19 22 = = 7 8 15 + 28 18 + 32 43 50 . c1p a21 a22 . . The condition for AB to be meaningful is that the number of columns of A should be equal to the number of rows of B. a2n A≡ . . . . Scalar multiplication: The multiplication of a scalar number α and a matrix A. . . . . . . B= 1 0 0 1 ⇒A+B = 2 2 3 5 . . anp cm1 cm2 . am1 am2 . . ... . .. i. . cmp where cij = Examples: 1 2 3 4 n k=1 aik bkj . . . C = AB. size(A + B) = size(A − B) = size(A) = size(B) and (A + B)ij = aij + bij . 1 2 3 3+8 11 = = . (2) aij = bij for all ij. . a2p c21 c22 . amn bn1 bn2 . . . the multiplication of A and B.e.. . . . n = o. . . (2) AB = BA even when both are deﬁned.
a11 . + and × are binary operations of real numbers R. Another important example: addition and multiplication are binary operations of matrices. .6 1 2 = 3 4 1 2 Notice that 3 4 2. −2. an1 . Deﬁne A ≡ . . . . . an1 x1 + . In matrix algebra. ann xn bn the simultaneous equation system. m R = real numbers Q = rational numbers n Rn = ndimensional column vectors M(m. . . . 3. . the solution to the equation ax = b is x = a−1 b. .3 Commutative. For example. . . . association.. . . −1. n) = m × n matrices M(n) = n × n matrices A binary operation is a law of composition of two elements from a set to form a third element of the same set. + ann xn = bn a1n x1 b1 . . 3 4 = . x ≡ .2 5 6 7 8 5 + 18 10 + 24 7 + 24 14 + 32 5 6 5 6 = 7 8 7 8 23 34 31 46 1 2 . 1. Then the equation . . . . 2. . . . . Income determination model: In the algebra of real numbers. . . I = integers . . 2. . we wish to deﬁne a concept of A−1 for a n × n matrix A so that x = A−1 b is the solution to the equation Ax = b. . + a1n xn = b1 . 2. . . Matrix representation of a linear simultaneous equation system A linear simultaneous equation system: a11 x1 + . N = nature numbers 1. Ax = b is equivalent to Linear 2market model: E1 = (a1 − b1 )p1 + (a2 − b2 )p2 + (a0 − b0 ) = 0 E2 = (α1 − β1 )p1 + (α2 − β2 )p2 + (α0 − β0 ) = 0 ⇒ a1 − b1 a2 − b2 α1 − β1 α2 − β2 p1 p2 + a0 − b0 α0 − β0 = 0 0 . 0. Y =C +I 1 1 −1 Y 0 The notations for some important sets are given by the following table. . . and distributive laws C = a + bY 1 0 −b C a I = I(r) ⇒ 0 1 0 I = I(r) . . and b ≡ . .
In an example. (2) (A + B)′ = A′ + B ′ . .. we already showed that the matrix multiplication does not satisfy the commutative law AB = BA even when both are meaningful. O+A = A+O = A.. 0 0 . we deﬁne zero matrices and identity matrices as 1 0 . b ∈ R. 0 is the unit element of + and 1 is the unit element of ×: 0 + a = a + 0 = a.5 Both On.. 0 . 0. Another example is A = . 2. Association law of + and × in R: (a + b) + c = a + (b + c) and (a × b) × c = a × (b × c) for all a. . b = 0 then a×b = 0. b.2 . .7 Commutative law of + and × in R: a + b = b + a and a × b = b × a for all a. 0 0 1 . .. then A is an idempotent matrix. However. The proof is trivial. However. . this deserves a proof! It is also true that matrix addition and multiplication satisfy the distributive law: (A + B)C = AC + BC and C(A + B) = CA + CB. In matrix algebra. 0 × a = a × 0 = 0. and IA = AI = A. . . 1 4 1 2 3 A= then A′ = 2 5 . C ∈ M(m.. we deﬁne its transpose A′ as a matrix with size n × m such that the ijth element of A′ is equal to the jith element of A. . Distributive law of + and × in R: (a+b)×c = a×c+b×c and c×(a+b) = c×a+c×b for all a. 0.. .. b. . (3) (AB)′ = B ′ A′ . a′ij = aji . n). In the multiplication 1 0 0 0 0 0 of real numbers if a. and 1 × a = a × 1 = a. 0 0 0 . .n ≡ . = = 0 0 0 1 0 0 O2. In ≡ . OA = AO = O.5 0. Idempotent matrix: If AA = A (A must be square). The . You should try to prove these statements as exercises. . c ∈ R. then A is symmetrical.. . Symmetrical matrix: If A = A′ (A must be square).. B. Nevertheless the matrix multiplication satisﬁes the associative law (AB)C = A(BC) when the sizes are such that the multiplications are meaningful.5 Transpose of a matrix: For a matrix A with size m × n. The addition of matrices satisﬁes both commutative and associative laws: A + B = B + A and (A + B) + C = A + (B + C) for all A. .. . 0 . .4 Special matrices In the space of real numbers. . 0 and 1 are very special. Om.5 0. 1 Clearly. .. c ∈ R. 4 5 6 3 6 Properties of matrix transposition: (1) (A′ )′ = A.n and In are idempotent.
a2 = 0. V = (1. (2) On.5 0. Suppose A is a n × n nonsingular matrix and P is a n × n idempotent matrix. Deﬁnition of A−1 : If there exists a B ∈ M(n) such that AB = BA = In . Show that AP A−1 is idempotent. and W = aU + bV . BV . a1 . Both λ1 0 0 In and On. and BW . x2 ) 3 2 2 5 x1 x2 . U = (1. Singular matrix: A square matrix whose inverse matrix does not exist. (1) (A−1 )−1 = A.5 0.5 a and b are real numbers. Nonsingular matrix: A is nonsingular if A−1 exists.n ⇒ On. Both On. −1 Examples: (1) Since II = I. 2.n B = On. (2) (AB)−1 = B −1 A−1 .n are diagonal. Suppose that A and B are n × n symmetric idempotent matrices and AB = B.n and In are symmetrical. Is B idempotent? 4. −1)′ . 0 a2 0 a−1 2 then A−1 does not exist. Diagonal matrix: A symmetrical matrix A is diagonal if aij = 0 for all i = j. (4) If a1 = 0 or a2 = 0. 1 2 Another example is A = . 2 3 Projection matrix: A symmetrical idempotent matrix is a projection matrix.n does not exist. 5. (b) Show that matrix A is idempotent. . 2. Scalar: aa−1 = a−1 a = 1 ⇒ if b satisﬁes ab = ba = 1 then b = a−1 . Properties of matrix inversion: Let A. I −1 = I. (3) (A′ )−1 = (A−1 )′ . (a) (A + B)3 = ? (b) (A + I)3 = ? 3. where 0.8 condition for A to be symmetrical is that aij = aji . 6. Calculate (x1 . B ∈ M(n). (a) If the dimension of X is m × n. (3) a1 0 a−1 0 1 If A = .5 Inverse of a square matrix We are going to deﬁne the inverse of a square matrix A ∈ M(n). then we deﬁne A−1 = B. Let A = I − X(X ′X)−1 X ′ . Show that A − B is idempotent. what must be the dimension of I and A.6 Problems 1. 1)′. Let A and B be n × n matrices and I be the identity matrix.5 . Let B = 0. then A−1 = . Find BU. Another example is A = 0 λ2 0 0 0 λ3 2.
Let I = and J = . (c) Show that (aI + bJ)(cI + dJ) = (ac − bd)I + (ad + bc)J. and J −1 . (a) Show that J 2 = −I. .9 1 0 0 1 0 1 −1 0 7. J 4 . (b) Make use of the above result to calculate J 3 . 1 (d) Show that (aI + bJ)−1 = 2 (aI − bJ) and [(cos θ)I + (sin θ)J]−1 = a + b2 (cos θ)I − (sin θ)J.
1 Vector space. . we say that the mdimensional Euclidean space is a vector space. . v1 w1 v1 + w1 . . Two points . . Vector addition: . . vm αvm x2 T v2 0 v 1 Q 1 2 v +v 0 x2 T E x1 v E x1 2v With such a structure. = . (1) A matrix as a collection of column (row) vectors (2) A matrix as a linear transformation from a vector space to another vector space 3. and linear independence Each point in the mdimensional Euclidean space can be represented as a mdimensional v1 . vm w vm + wm m v1 αv1 . . vm in the mdimensional Euclidean space can be added according to the rule of matrix addition.10 3 Vector Space and Linear Transformation In the last section. . Scalar multiplication: α . we regard a matrix simply as an array of numbers. . A point can be multiplied by a scalar according to the rule of scalar multiplication. + . linear combination. = . . . Now we are going to provide some geometrical meanings to a matrix. where each vi represents the ith coordinate. column vector v = . . .
{v 1 . . 0). = . 0. where i=1 Linear dependence of {v 1 . . . . Therefore. then the collection is said to be linear dependent. . . . i . 6 3 0 α1 + 4α2 α1 v 1 + α2 v 2 = 2α1 + 5α2 = 0 ⇒ α1 = α2 = 0. a2n a21 a22 a2n . v 2 = a2 . . amn am2 amn m1 Linear combination of a collection of vectors {v . . . v n }: If one of the vectors is a linear combination of others. . A m × n matrix: a collection of n mdimensional column vectors: a11 a11 a12 . . . 0). . . . . 1 4 Example 3: v 1 = 2 . Alternatively. If a vector is a linear combination of other vectors. v 2 } is 0 3α1 + 6α2 linear independent. 1 n n αi v i . . v 2 = 5 . . . . . v v } linear dependent. . . . Therefore. a1 0 0 Example 1: v 1 = 0 . . a am1 am2 . Example 2: v = 3 6 9 2 1 3 1 2 3 . 0) is a linear combination of it. v 3 } must be linear independent. . {v . . v 3 = 8 . a1n a12 a1n a21 a22 . . v ∈ R . is 2v = v + v . {v 1 . . . a1 a2 a3 = 0. then it is linear independent. α3 ) = (0. . . . . .. . 0 0 a3 If α1 v 1 + α2 v 2 + α3 v 3 = 0 then (α1 . . . . . . v n }: If the collection is not linear dependent. v 3 = 0 . . . . . . Therefore. . α2 . . αn ) = (0. . then it can be removed without changing the span. . v }: w = (α1 . v 2 . v 2 = 5 .11 m mdimensional column vector space: R = v1 . . . . . . the collection is linearly dependent if (0. . 1 4 7 1 2 . . Span of {v 1 . . . vm We use superscripts to represent individual vectors. Linear independence of {v 1 . . v n }: The space of linear combinations.
. Example 1: A = Example 2: B = Rank(B) = 1. . . . . ami a11 . a1n . N(A) = 2 1 . . 0 0 1 0 1 0 Standard basis vectors of Rn : e1 ≡ . . . . v1n . ⇒ f (g(x)) = A(By). i i . . 1 0 0 a1i . Given x ∈ Rn . Rank(A) ≡ dim(R(A)).12 v11 . . . . The multiplication of two matrices can be interpreted as the composition of two linear transformations. en ≡ . . R(B) = k k . . i i Let v be the ith column of A. ym am1 . Rank . ⇒ v 1 = f (e1 ) = Ae1 . vm1 . . . Therefore. . . . . Ax ∈ Rm .k ∈ R . .. v n }) = Maximum # of indepen.. . 1 0 1 1 0 . . . . . . .. . = . v = f (e ): . Null space of f (x) = Ax: N(A) ≡ {x ∈ Rn . . . . g : Rp →Rn . g(y) = By. f ◦g : Rp →Rm . . . f : Rn →Rm . . . . .. we can deﬁne a linear transformation from Rn to Rm as f (x) = Ax or y1 a11 . f (x) = Ax. v n } = Range space of f (x) = Ax ≡ R(A). . v i is the image of the ith standard basis vector ei under f . am1 0 am1 . . . . a1n 1 a11 . Rank(A) = 2. . k ∈ R .. dim(R(A)) + dim(N(A)) = n. . . . f : Rn →Rm . f (x) = . . ≡ Dimension(Span{v 1 . e2 ≡ . etc. . . N(B) = 1 0 0 k −k . . . xn It is linear because f (αx + βw) = A(αx + βw) = αAx + βAw = αf (x) + βf (w). = . R(A) = R2 . f (x) = Ax = 0}. . . v = .2 Linear transformation Consider a m × n matrix A. . . Span{v 1 . . amn Therefore. . amn x1 . . . vmn dent vectors. 3. .
then A is nonsingular.n represents the mapping that maps every point to the origin.5 . f −1 f (x) = x ⇒ A−1 Ax = x.5 2 k . If dim(N(A)) = 0. then f (x) is not onto.13 The composition is meaningful only when the dimension of the range space of g(y) is equal to the dimension of the domain of f (x). Each A ∈ M(n) represents a linear transformation f : Rn →Rn . 0 0 a3 3 6 9 x2 x= Ax = x′ = 1 2 . which is the same condition for the validity of the matrix multiplication.5 1. then R(A) = Rn and f (x) is onto.. A projection matrix repre0. ⇒ if Rank(A) = n.5 sents a mapping that projects points onto a linear subspace of Rn .5 1 .3 Inverse transformation and inverse of a square matrix Remark: On.5 .5 0. then f (x) is one to one and onto and there exits an inverse mapping f −1 : Rn →Rn represented by a n×n square matrix A−1 . then A−1 is the inverse mapping of A. eg. d x d d d Ax d d d d d d E x1 s d d d d d d d d T Consider now the special case of square matrices. 0. if Rank(A) < n.5 . The deﬁnition of the inverse matrix A−1 is such that AA−1 = A−1 A = I. Ax′ = −k = 0 0 1. If dim(R(A)) = n. and A is singular. In represents the identity mapping that maps a point to itself.5 0. a1 0 0 1 4 7 Examples: Rank 0 a2 0 = 3 and Rank 2 5 8 = 2. 3. then f (x) is one to one. no inverse mapping exists. ⇒ if Rank(A) = n. If we regard A as a linear transformation from Rn →Rn and I as the identity transformation that maps every vector (point) into itself.5 projects points onto the 45 degree line. Every linear transformation f : Rn →Rm can be represented by f (x) = Ax for some m × n matrix.
Show that y must be in the null space of I − A. Show that x must be in the range space of I − A. x3 a (a) Is v 1 = 0 . (Hint: Try y = x.4 Problems 0 1 0 1. A = 1 1 1 . i. 2. Let B = 0 0 1 0 0 0 and TB the corresponding linear transformation TB : R3 → R3 . What should be the values of a and b? (f) What is the dimension of the null space of B? .) (c) Suppose that y is in the range space of A. show that there exists a vector y such that (I − A)y = x. in the null space of TB ? Why or why not? 0 0 (b) Is v 2 = 0 .e. x = a . a = 0.. Ax = 0. (d) Suppose that A is n × n and Rank[A] = n − k. x1 where x = x2 ∈ R3 . Let A be an idempotent matrix. What is the rank of I − A? 1 1 1 1 0 0 1 1 3 3 3 0 1 0 . what are the values of α and β? (c) What is the dimension of the range space of A? (d) Determine the rank of A.. i. Let I = 3 3 3 1 1 1 0 0 1 b β 3 3 3 B = I − A. and 3. TB (x) = Bx. (a) Calculate AA and BB. b = 0. (b) Suppose that x = 0 is in the null space of A. n > k > 0. in the range space of TB ? Why or why not? How b c 3 d ? about v = 0 (c) Find Rank(B). (e) Suppose now that x is in the null space of B.e. (b) If y is in the range space of A.14 3. y = α . (a) Show that I − A is also idempotent.
Let A = 1/5 2/5 2/5 4/5 1 1 1 and B = 0 1 1 . (c) Determine whether they are idempotent. .15 (g) Determine the rank of B? 4. 0 0 1 (a) Determine the ranks of A and B. (b) Determine the null space and range space of each of A and B and explain why.
3) a11 a22 a33 + (−1)N (1.1. 2 is mapped to i2 . 2. . .in ) a1i1 a2i2 . . . + .2) a11 a23 a32 + (−1)N (2. N(i1 . Using the properties of determinants.16 4 Determinant. a1n . 2.. . 3. . n} is a 11 mapping of {1. Permutation set of {1. 1) = 3.1) a12 a23 a31 + (−1)N (3.. anin . Then we derive a general procedure to solve a simultaneous equation. We also write π = (i1 . 1)}. P2 = {(1. First. . in ): If there exist k and l such that k < l and ik > il . . . n}: Pn ≡ {π = (i1 . i2 .2 Determinant a11 a12 a21 a22 . then we say that an inversion occurs. 2) = 1. 2. : . . a2n . . 1)}. . . . . ın ): Total number of inversions in (i1 . N(3. . P4 : 4! = 24 permutations. . 1.3.3) a12 a21 a33 (−1) (−1)N (2. . 2. and Cramer’s rule In this section we are going to derive a general method to calculate the inverse of a square matrix. . a11 a12 a13 n = 3: a21 a22 a23 = a31 a32 a33 N (1. (2. (3. i2 .i2 . i2 . 1. P3 = {(1. 2. i2 . Inversions in a permutation π = (i1 .. . . .. in ).2) a11 a22 + (−1)N (2. .. 2) = 0.1) a13 a22 a31 = a11 a22 a33 − a11 a23 a32 − a12 a21 a33 + a12 a23 a31 + a13 a21 a32 − a13 a22 a31 . . we deﬁne the determinant of a square matrix. in n is mapped to in . and i1 i2 . 3). 3). (2.in )∈Pn Determinant of A = A ≡ a11 a12 a21 a22 n = 2: = (−1)N (1. N(1. ..1 Permutation group A permutation of {1. . . . (i1 .2.i2 .2. . . . Examples: 1. . n π= meaning that 1 is mapped to i1 . .1. written as 1 2 . N(1. 1) = 1. ann (−1)N (i1 .. 2. ın ) when no confusing. (1. . i2 . . 3. 2). 2. .. .1) a12 a21 = a11 a22 − a12 a21 . . N(2.. 2. . .3. 1) = 2. . . 3) = 1. 2). 2). . N(2. . 4. 3) = 0. Inverse Matrix. 2) = 2. 1).. .. . we ﬁnd a procedure to compute the inverse matrix. N(1. .2) a13 a21 a32 + (−1)N (3. in ) : π is a permutation}. 3. 4. (2. N(3. 1. . an1 an2 . 1. (3. . 3. . . . n} onto itself. . . N(2..
Example: 4 3 3 4 Property 3: If two column vectors are identical. we can replace “column vectors” in the properties below by “row vectors”. the determinant is equal to the negative of itself. other vectors ) = αD(v i . .17 a11 a11 n = 2: a21 a12 d d a12 d d a13 d d a11 d d d d a12 d d n = 3: a21 a31 a22 a32 a23 a33 a21 a31 a22 a32 a22 d d 4. other vectors ) + βD(w i. v n }. Proof: Each term of the new determinant corresponds to a term of A of opposite sign because the number of inversion increases or decreases by 1. then the determinant is 0. . . Proof: By property 2. v n ). By property 1. . which is possible only when the determinant is 0. Minors: The determinant Aij  is called the minor of the element aij . other vectors ). 3 4 Submatrix: We denote by Aij the submatrix of A obtained by deleting the ith row and jth column from A. D : Rn×n →R. we can regard determinants as a function of n column vectors A = D(v 1 . . . . v 2 . = 1 × 4 − 2 × 3 = −2. . we can replace “column vectors” in the properties below by “row vectors”. Proof: Given other column vectors. . v 2. 2 1 1 2 = 2 × 3 − 1 × 4 = 2. Proof: Each term of A′  corresponds to a term of A of the same sign. 1+5×2 2 1 2 5×2 2 1 2 2 2 Example: = + = +5× = 3+5×4 4 3 4 5×4 4 3 4 4 4 1 2 + 5 × 0. the determinant function is a linear function of v i : D(αv i + βw i. the determinant changes sign. Cofactors: Cij ≡ (−1)i+j Aij  is called the cofactor of aij . By property 1.3 Properties of determinant Property 1: A′  = A. Since a n × n matrix can be regarded as n column vectors A = {v 1 . the determinant does not change. Property 4: If you add a linear combination of other column vectors to a column vector. Property 2: If two column vectors are interchanged.
⇒ AC ′ = C ′ A = . . . . . ij Given j ′ = ¯ = n aij ′ Ci¯ = 0. 8 0 7 0 7 8 A−1 = . Cn1 Cn2 .18 n Property 5 (Laplace theorem): Given i = ¯ A = i. Cnn A 0 . 0 1 ′ −1 3. j i=1 Therefore. 0 0 A . Cnn C11 C21 . . . . . ij ij j=1 Given j = ¯ A = n ai¯Ci¯. .. . ⇒ if A = 0 then A C = A . 5 6 4 6 4 5 Example: 0 = 4 × −5× +6× . we C11 C12 . . j. . . C1n C2n . . 8 0 7 0 7 8 7 8 0 n Property 6: Given i = ¯ i.. . . . . . j. . if A = a11 a22 − a12 a21 = 0. . 8 0 7 0 7 8 4. . . . . C1n C21 C22 . . . . . . 1 2 3 5 6 4 6 4 5 Example: 4 5 6 = 1 × −2× +3× . you get 0 instead of the determinant. . Cofactor matrix: C ≡ . . A a11 a22 − a12 a21 −a21 a11 1 2 3 Example 2: A = 4 5 6 ⇒ A = 27 = 0 and 7 8 0 5 6 4 6 4 5 C11 = = −48. . C12 = − = 42. . . . . ′ j=1 ai′ j C¯ = 0. a21 a22 −a12 a11 1 ′ 1 a22 −a12 C = .. can calculate the inverse of A as follows. C13 = = −3. . . Adjoint of A: Adj A ≡ C = . . a¯ C¯ . . Cn1 C12 C22 . . .4 Computation of the inverse matrix Using properties 5 and 6. . C2n 1. A a11 a12 a22 −a21 Example 1: A = then C = . j i=1 j Proof: In the deﬁnition of the determinant of A. . Proof: The sum becomes the determinant of a matrix with two identical rows (columns). 0 0 . all terms with aij can be put togethere to become aij Cij . . Cn2 ′ 2. if you multiply cofactors by the elements from a diﬀerent row or column. . . . .
6 4 5 −48 24 −3 −48 24 −3 1 = 42 −21 6 . C33 = = −3. b. where Ai is a matrix obtained by replacing Cramer’s rule: xi = A A the ith column of A by b. . . 27 −3 6 −3 −3 6 −3 1 7 1 4 If A = 0. a11 a1n 0 . . 1 0 −b 0 1 0 1 1 −1 a1 − b1 b0 − a0 α1 − β1 β0 − α0 .6 Economic applications a1 − b1 a2 − b2 α1 − β1 α2 − β2 p1 p2 = b0 − a0 β0 − α0 . The solution to the simultaneous A C ′b . a1 − b1 a2 − b2 α1 − β1 α2 − β2 0 −b C a 1 0 I = I(r) . 0 −b 1 0 1 −1 .5 Cramer’s rule C′ . equation Ax = b is x = A−1 b = 4. the linear transformation TA is not onto and therefore an inverse transformation does not exist. C22 = 8 0 2 3 C31 = = −3. an1 ann 0 of A are linear dependent. . C ′ −3 6 −3 C21 = − 3 1 2 = −21. A−1 = 42 −21 6 . 1 0 −b 0 1 0 1 1 −1 C= I= 1 a −b 0 I(r) 0 1 0 −1 . then A is singular and A−1 does not The reason is that A = exist. . 4. = . p2 = . The column vectors . Ai = {v 1 . + · · · + Cn1 . A n Ai  j=1 Cj bj = . v n }. . 0 7 8 3 1 2 = 6. . If A = 0 then A is nonsingular and A−1 = Linear 2market model: b0 − a0 a2 − b2 β0 − α0 α2 − β2 p1 = a1 − b1 a2 − b2 α1 − β1 α2 − β2 1 Income determination model: 0 1 a 0 −b I(r) 1 0 0 1 −1 . . . C23 = − = 6. . .19 2 3 = 24. . 1 −1 Y 0 1 0 1 1 0 1 Y = 0 a 1 I(r) 1 0 . 0 ⇒ AC ′ = 0n×n ⇒ C11 . . C32 = − 5 6 −48 42 −3 C = 24 −21 6 . v i+1 . v i−1.
′ M = M0 + mY M ′ = M0 + m′ Y ′ X (export). ′ ′ ′ (1 − b + m)Y − m′ Y ′ = a + I0 + M0 − M0 (1 − b′ + m′ )Y ′ − mY = a′ + I0 + M0 − M0 . Y. R (interest rate). Y′ = ′ 1 − b + m a + I0 + M0 − M0 ′ ′ ′ −m a + I0 + M0 − M0 . that is. L(demand for money). I ′. k. I.7 Inputoutput table Assumption: Technologies are all ﬁxed proportional. 1−b i k −l R= Twocountry income determination model: Another extension of the income determination model is to consider the interaction between domestic country and the rest of the world (foreign country). 1−b i k −l Y = ¯ a + I0 + G i ¯ M −l . R 1−b i k −l Y R = ¯ a + I0 + G ¯ M ¯ 1 − b a + I0 + G ¯ k M . Y ′. b. M (money supply) ¯ ¯ ex. we regard interest rate as given and consider only the product market. ′ ′ I = I0 I = I0 M (import). Now we enlarge the model to include the money market and regard interest rate as the price (an endogenous variable) determined in the money market. var: G. Substitute into equilibrium conditions: good market: money market endogenous variables: ¯ ¯ a + bY + I0 − iR + G = Y . M (quantity of money). Y . domestic good market: foreign good market: endogenous variables: C = a + bY C ′ = a′ + b′ Y ′ C. good market: money market: C = a + bY L = kY − lR ¯ I = I0 − iR M =M ¯=Y C +I +G M =L end. var: C. . Y . Substituting into the equilibrium conditions. l.20 ISLM model: In the income determination model. X = M ′ and X ′ = M. X ′. 1−b+m −m′ −m 1 − b′ + m′ 4. you need aji units of Xj . By deﬁnition. kY − lR = M. to produce one unit of product Xi . ′ ′ ′ ′ ′ ′ C +I +X −M =Y C +I +X −M =Y C . i. parameters: a. I. 1−b+m −m′ −m 1 − b′ + m′ Y = ′ a + I0 + M0 − M0 −m′ ′ ′ ′ a + I0 + M0 − M0 1 − b′ + m′ 1−b+m −m′ −m 1 − b′ + m′ Y Y′ = ′ a + I0 + M0 − M0 ′ ′ a′ + I0 + M0 − M0 . M ′.
then the volume is 0 and the vectors are linearly dependent. . . A−1 does not exist. It is equal to the volume of the parallelepiped formed by the vectors {v 1 . . For n = 2. . . xn a11 x1 + a12 x2 + . . . . . 4. . . v n ) is called an alternative linear nform of Rn . . . . a2n IO table: A = . . . . . . . . . Therefore. .21 a11 a12 . .8 A geometric interpretation of determinants Because of properties 2 and 4. + a2n xn Ax = . . + a1n xn a21 x2 + a22 x2 + . . A is the area of the parallelogram formed by a11 a21 .. . The net output is x − Ax = (I − A)x. an1 an2 . one of them must be a linear combination of others. . ann Column i represents the coeﬃcients of inputs needed to produce one unit of Xi . an1 x1 + an2 x2 + . . dn −1 x = (I − A) d. . . . then since d = (I − A)x. . . and A is singular. . we will need a list of inputs . v n }. . . an inverse mapping does not exist. . a1n a21 a22 . . + ann xn d1 d2 If we want to produce a net amount of d = . . x1 x2 Suppose we want to produce a list of outputs x = . See the diagram: a12 a22 x2 T A = Area of D ¢ ¢ ¢ v 2 ¢ ¢ ¢ ¢ D ¢ ¢ ¢ ¢ ¢ ¢ 1 ¢ ¢ v I ¢ ¢ ¢ E x1 If the determinant is 0. . the determinant function D(v 1 .
In this case. 2 4 7 Multiplying the ﬁrst equation by 2. There are two cases. On x1 x2 space. there exists an equation whose LHS is a linear combination of the LHS of other equations but whose RHS is diﬀerent from the same linear combination of the RHS of other equations. the two equations are x2 k represented by the same line and every point on the line is a solution. . First. 1 2 3 1 2 1 2 Examples: Rank = 2. Rank(B)}. . Case 2: Rank(M) = Rank(A) + 1. There will be more variables than equations after removing these equations. On x1 x2 space.9 Rank of a matrix and solutions of Ax = d when A = 0 Rank(A) = the maximum # of independent vectors in A = {v 1 . 1 2 x1 3 1 2 Example: = . there will be inﬁnite number of solutions.22 4. and calculate . . Case 1: Rank(M) = Rank (A). When A = 0. The sox1 3 − 2k lutions are = for any k. some equations are linear combinations of others (the equations are dependent) and can be removed without changing the solution space. Rank(A) = Rank = 1 = Rank(M) = 2 4 x2 6 2 4 1 2 3 Rank . Property 1: Rank(AB) ≤ min{Rank(A). Property 2: dim(Null Space of TA ) + dim(Range Space of TA ) = n. In this case. 2 4 6 The second equation is just twice the ﬁrst equation and can be discarded. the equation system is contraditory and there will be no solutions. Rank(A) = Rank = 1 < Rank(M) = x2 2 4 7 2 4 1 2 3 Rank = 2. form the augmented matrix M ≡ [A. the two equations are represented by two parallel lines and cannot have any intersection points. and Rank(A) < n. Therefore. Hence. there exists a row of A that is a linear combination of other rows .d] the rank of M. whereas the second equation x1 says 2x1 + 4x2 = 7. Therefore. . Rank(A) = the size of the largest nonsingular square submatrices of A. Consider the simultaneous equation Ax = d. . 2x1 + 4x2 = 6. v n } = dim(Range Space of TA ). it is impossible to have any satisfying both x2 equations simultaneously. 1 2 x1 3 1 2 Example: = . Rank 4 5 6 = 2 because is non3 4 4 5 7 8 9 singular.
0 0 0 (a) Find AA and AAA. 8 9 7 (a) calculate the cofactors C11 . A = . 4)′. and Rank[AAA]. (c) Find Rank[X(X ′ X)−1 X ′ ] and Rank[I − X(X ′X)−1 X ′ ]. 2. B.23 4. 3. v3 } linearly independent? Why or why not? 6 5 .. and AAAx. Given the 3 × 3 matrix A = 5 3 4 . (c) and use Cramer’s rule to ﬁnd X1 of the following equation system: 2 1 6 X1 1 5 3 4 X2 = 2 . Is {v1 . and v3 = (3. C31 . Let A = 0 0 1 . v2 . 4. Use Cramer’s rule to solve the nationalincome model C = a + b(Y − T ) T = −t0 + t1 Y Y = C + I0 + G (1) (2) (3) 0 1 0 5. 3 6 3 6 1 3 6 12 (a) Find the ranks of A. 2. 1 −1 6. C21 . Find the inverse of A = 8 7 2 1 6 3. and C = . v2 = (2. and C. Rank[AA]. (b) Let x = (1. 2. 3)′.B= . (c) Do the same for the following system: 1 2 3 6 X1 X2 = 4 12 . . (b) use Laplace expansion theorem to ﬁnd A. Let X = 1 0 . 1 1 ′ (a) Find X X and (X ′ X)−1 . 8 9 7 X3 3 (Hint: Make use of the results of (a). compute Ax. 3)′. 5)′.) 4. Suppose v1 = (1. 1 2 1 2 1 1 2 4 7. (c) Find Rank[A]. (b) Use the results of (a) to determine whether the following system has any solution: 1 2 3 6 X1 X2 = 1 1 . (b) Compute X(X ′ X)−1 X ′ and I − X(X ′ X)−1 X ′ .10 Problems 1. AAx.
24 3 2 , I the 2 × 2 identity matrix, and λ a scalar number. 1 2 (a) Find A − λI. (Hint: It is a quadratic function of λ.) (b) Determine Rank(A − I) and Rank(A − 4I). (Remark: λ = 1 and λ = 4 are the eigenvalues of A, that is, they are the roots of the equation A − λI = 0, called the characteristic equation of A.) (c) Solve the simultaneous equation system (A − I)x = 0 assuming that x1 = 1. (Remark: The solution is called an eigenvector of A associated with the eigenvalue λ = 1.) (d) Solve the simultaneous equation system (A − 4I)y = 0 assuming that y1 = 1. (e) Determine whether the solutions x and y are linearly independent. 8. Let A =
25
5
Diﬀerential Calculus and Comparative Statics
As seen in the last chapter, a linear economic model can be represented by a matrix equation Ax = d(y) and solved using Cramer’s rule, x = A−1 d(y). On the other hand, a closed form solution x = x(y) for a nonlinear economic model is, in most applications, impossible to obtain. For general nonlinear economic models, we use diﬀerential calculus (implicit function theorem) to obtain the derivatives of endogenous variables ∂xi with respect to exogenous variables : ∂yj f1 (x1 , . . . , xn ; y1 , . . . , ym ) = 0 . . . fn (x1 , . . . , xn ; y1, . . . , ym ) = 0 ∂x1 ∂x1 ∂f1 ∂f1 ∂f1 ∂f1 ∂x1 . . . ∂xn ∂y1 . . . ∂ym ∂y1 . . . ∂ym . . . . . = − . . . .. .. .. . . . . . ⇒ . . . . . . . . . ∂fn ∂xn ∂fn ∂fn ∂xn ∂fn ... ... ... ∂x1 ∂xn ∂y1 ∂ym ∂y1 ∂ym −1 ∂x1 ∂x1 ∂f1 ∂f1 ∂f1 ∂f1 ∂x1 . . . ∂xn ∂y1 . . . ∂ym ∂y1 . . . ∂ym . . = − . . . . . .. .. .. . . . . . ⇒ . . . . . . . . . . ∂fn ∂xn ∂xn ∂fn ∂fn ∂fn ... ... ... ∂y1 ∂ym ∂x1 ∂xn ∂y1 ∂ym ∂xi ∂xi represents a causeeﬀect relationship. If > 0 (< 0), then xi will increase Each ∂yj ∂yj (decrease) when yj increases. Therefore, instead of computing xi = xi (y), we want to ∂xi determine the sign of for each ij pair. In the following, we will explain how it ∂yj works. 5.1 Diﬀerential Calculus x = f (y) ⇒ f ′ (y ∗) = dx dy ≡ lim f (y ∗ + ∆y) − f (y ∗) . ∆y→0 ∆y
y=y ∗
On yx space, x = f (y) is represented by a curve and f ′ (y ∗ ) represents the slope of the tangent line of the curve at the point (y, x) = (y ∗, f (y ∗)). Basic rules: dx 1. x = f (y) = k, = f ′ (y) = 0. dy dx 2. x = f (y) = y n , = f ′ (y) = ny n−1 . dy dx 3. x = cf (y), = cf ′ (y). dy
26 4. x = f (y) + g(y), 5. x = f (y)g(y), dx = f ′ (y) + g ′ (y). dy
dx = f ′ (y)g(y) + f (y)g ′(y). dy dx f ′ (y)g(y) − f (y)g ′(y) 6. x = f (y)/g(y), = . dy (g(y))2 dx dx 1 7. x = eay , = aeay . x = ln y, = . dy dy y dx dx 8. x = sin y, = cos y. x = cos y, = − sin y. dy dy Higher order derivatives: d d d2 d d2 f ′′ (y) ≡ f (y) = 2 f (y), f ′′′ (y) ≡ f (y) dy dy dy dy dy 2 5.2 Partial derivatives
=
d3 f (y). dy 3
In many cases, x is a function of several y’s: x = f (y1 , y2 , . . . , yn ). The partial ∗ ∗ ∗ derivative of x with respect to yi evaluated at (y1 , y2 , . . . , yn ) = (y1 , y2 , . . . , yn ) is ∂x ∂yi
∗ ∗ ∗ (y1 ,y2 ,...,yn )
∗ ∗ ∗ ∗ ∗ ∗ f (y1 , . . . , yi + ∆yi , . . . , yn ) − f (y1 , . . . , yi , . . . , yn ) ≡ lim , ∆yi →0 ∆yi
that is, we regard all other independent variables as constant (f as a function of yi only) and take derivative. ∂xn xm 1 2 9. = nxn−1 xm . 1 2 ∂x1 Higher order derivatives: We can deﬁne higher order derivatives as before. For the case with two independent variables, there are 4 second order derivatives: ∂ ∂x ∂2x ∂ ∂x ∂2x ∂ ∂x ∂2x ∂ ∂x ∂2x = 2, = , = , = 2. ∂y1 ∂y1 ∂y1 ∂y2 ∂y1 ∂y2 ∂y1 ∂y1 ∂y2 ∂y1 ∂y2 ∂y2 ∂y2 ∂y2 Notations: f1 , f2 , f11 , f12 , f21 , f22 . f1 . ∇f ≡ . : Gradient vector of f . . f n f11 . . . f1n . . H(f ) ≡ . . . . . : second order derivative matrix, called Hessian of f . . . fn1 . . . fnn Equality of crossderivatives: If f is twice continously diﬀerentiable, then fij = fji and H(f ) is symmetric. 5.3 Economic concepts similar to derivatives
Elasticity of Xi w.r.t. Yj : EXi ,Yj ≡
Yj ∂Xi , the percentage change of Xi when Yj Xi ∂Yj P dQd increases by 1 %. Example: Qd = D(P ), EQd ,P = Qd dP
5 f (k) (a) f (k+1) (c) f ′′ (a) (y − a)2 + . dx f (y) dx 1 1 Example: x = ln y ⇒ y = ex ⇒ = x = .4 Mean value and Taylor’s Theorems Continuity theorem: If f (y) is continuous on the interval [a. b] such that f (c) = 0. 5.Y . . b] and f (a) = f (b) = 0. Diﬀerential of x = f (y): dx = df = f ′ (y)dy. the percentage change of X per unit of X dt time. there exists a c ∈ (a. dx = f ′ (y)g ′(z) = f ′ (g(z))g ′(z). dy > 0 (so that we can divid something by dx or by dy) but dx. dx. then there exists a c ∈ (a. ∆y ∆y ∆y dx. b].+ (y − a)k + (y − a)k+1 . y = g(z) ⇒ x = f (g(z)). 2! k! (k + 1)! Concepts of diﬀerentials and applications Let x = f (y). 2. dy: Inﬁnitesimal changes of x and y. y) such that f (y) = f (a) + f ′(a)(y − a) + 5. b−a Taylor’s Theorem: If f (y) is k + 1 times continously diﬀerentiable on [a.Y = EX1 . y = z 2 + 1 ⇒ dz Inverse function rule: x = f (y). then for each y ∈ [a.Y − EX2 . ⇒ y = f −1 (x) ≡ g(x).X = 1/EX.Y = EX1 . Example: x = (z 2 + 1)3 ⇒ x = y 3. 3. dy < a for any positive real number a (so that ∆y→dy). ∆x f (y + ∆y) − f (y) ∆x Finite quotient: = ⇒ ∆x = ∆y. b) such that f (b) − f (a) = f ′ (c)(b − a) or f (b) − f (a) = f ′ (c). dz dx = 3y 2 2z = 6z(z 2 + 1)2 . Mean value theorem: If f (y) is continously diﬀerentiable on [a.Y . EX1 /X2 . b]. dy 1 dx = f ′ (y)dy. Chain rule: x = f (y). then there exists a c ∈ (a. EY. . EX1 X2 .27 Basic rules: 1. f (b) ≥ 0. b] and f (a) ≤ 0. . b) such that f ′ (c) = 0. dy e y dx = f ′ (y)dy. dy = g ′(x)dx ⇒ dx = f ′ (y)g ′(x)dx. 1 dX Growth rate of X = X(t): GX ≡ . b]. Rolle’s theorem: If f (y) is continuous on the interval [a. dy = g ′(z)dz ⇒ dx = f ′ (y)g ′(z)dz.Y + EX2 . called the ﬁnite diﬀerence of x.Y . = g ′(x) = ′ . then there exists a c ∈ [a. Deﬁne ∆x ≡ f (y + ∆y) − f (y).
+fn (y1 . . y2 + ∆y2 ) − f (y1. dyn Total diﬀerential of x = f (y1 . ⇒ x = f (g 1(z1 . . dy1 . x2 . − dx1 U2 1 x 3 2 −2 Multivariate chain rule: x = f (y1. we use implicit function rule. . z2 ). . dy2 to ﬁnd the derivative dx = (f1 . x2 ) = U deﬁnes a relationship between x1 and dx2 . the relationship between two variables are deﬁned implicitly. : Inﬁnitesimal changes of x (endogenous). fn ) . y2 )dy1 + f2 (y1 . x2 ) = 3x1 + 3x2 = 6 deﬁnes an indiﬀerence curve passing through the point (x1 . called the ﬁnite diﬀerence of x. . . . yn )dy1 +. dy = .28 5. . Deﬁne ∆x ≡ f (y1 + ∆y1 . . ∂z1 dz1 dz2 . x2 )dx2 = dU = 0 ⇒. z2 ). y2 ). x2 )dx1 + U2 (x1 . dU = U1 (x1 . g 2(z1 . f1 g2 +f2 g2 ) . y2). . 1) can be calculated using implicit function rule. y2 )dy2. yn ): dy1 . y2 ) = ∆y1 + ∆y2 ∆y1 ∆y2 dx = f1 (y1 . dy1. 1). . We can use the total diﬀerentials dx. . y2 + ∆y2 ) f (y1. y1 . dz1 dz2 1 2 1 2 = (f1 g1 +f2 g1 . dx1 U2 (x1 . f2 ) dy1 dy2 = (f1 . To ﬁnd the slope of the curve dx1 U1 (x1 . y2 ). x2 ) 1 1 3 3 Example: U(x1 . y1 = g 1(z1 . . yn )dyn = (f1 . y2 + ∆y2 ) − f (y1 . y2 + ∆y2 ) − f (y1. yn (exogenous). . . y2 ) = f (y1 + ∆y1 . z2 ).6 Concepts of total diﬀerentials and applications Let x = f (y1 . y2 + ∆y2 ) − f (y1. y2 + ∆y2 ) + f (y1 . y2) f (y1 + ∆y1 . dx. the indiﬀerence curve U(x1 . . . . For ¯ example. z2 )) ≡ H(z1 . z2 ). The slope (Marginal Rate of Substitution) at (1. dyn Implicit function rule: In many cases. . x2 ) = (1. . = (∇f )′ dy. f2 ) 1 1 g1 g2 2 2 g1 g2 ∂x . x2 ) dx2 ¯ =− . y2 = g 2 (z1 . . . dx2 U1 x 3 1 =− = − 1 2 = − = −1. dx = df = f1 (y1 . y2 + ∆y2 ) − f (y1 . . y2 + ∆y2 ) − f (y1. ∆x = f (y1 + ∆y1 . .
yn ) . For example. x= . yn xn represented by a vector function x1 f 1 (y1 . y2 =h(y1 ) ⇒ dx = f1 dy1 + f2 dy2 = f1 dy1 + f2 h′ dy1 = (f1 + f2 h′ )dy1. for a linear mapping with A = 0. ∂z1 6 7 Example: x = y1 y2 . y2 ). . an 11 inverse mapping exists and the partial derivatives are given by the inverse matrix of A. dx dy1 = f1 + f2 h′ . . ∂x1 /∂y1 = a11 where∂y1 /∂x1 = a22 etc. . A general nonlinear mapping from Rn to Rn . x1 x2 = a11 a12 a21 a22 y1 y2 y1 y2 = a22 A −a21 A −a12 A a11 A x1 x2 . Total derivative: Partial derivative (direct eﬀect of y1 on x): Indirect eﬀect through y2 : Example: (x1 . . . . x1 along the indiﬀerence curve passing through (1. = = f1 g2 + f2 g2 . ∂y2 dy1 ∂x ∂f = = f1 (y1 . The idea can be generalized to nonlinear mappings. ∂z1 ∂z1 ∂z2 ∂z2 ∂x 5 7 6 6 = 6y1 y2 (2) + 7y1 y2 (4). . is . x2 ) x 3 m(x1 . → x = . y1 = 2z1 + 3z2 . = ≡ F (y). x2 ) x2 3 + the MRS at a point The rate of change of MRS 3x1 1/3 +3x2 =6 1/3 3x1 1/3 +3x2 =6 1/3 −2 x2 3 . . . n xn f (y1 . . w. . x2 ) = = − = − 12. . x2 ) = 3x1 −2 dx2 U1 (x1 . we discussed a linear mapping x = Ay and its inverse mapping y = A−1 x when A = 0. y2 ). Total derivative: x = f (y1 . y2 = h(y1 ). ∂y1 ∂y1 1 3 3x2 . yn ) . . x = f (y1 . 1) is a total derivative dm dx1 5.t.7 d2 x2 = dx2 1 ∂m ∂m dx2 ∂m ∂m = + = + ∂x1 ∂x2 dx1 ∂x1 ∂x2 − x1 −2 3 1 3 Given the utility function U(x1 . y2 = 4z1 + 5z2 . . Therefore.r. y = . − dx1 U2 (x1 . .29 ⇒ ∂H ∂x ∂H ∂x 1 2 1 2 = = f1 g1 + f2 g1 . Inverse function theorem In Lecture 3. A y1 x1 . h(y1 )) ≡ g(y1). x2 ) is ∂x dy2 = f2 h′ .
. . ∂y1 ∂yn Jacobian: Jacobian matrix: JF (y) ≡ ∂(x1 . . n n f1 . .. . . − . fn dx1 .30 ∂x1 ∂x1 . .. . θ = tan−1 1 2 1 1 Notice that g1 = 1/(f1 ) in general. . . . . . . . .. . JF (r. . .. . then the solution is given by x = −A−1 (By + c). . θ) = cos θ −r sin θ sin θ r cos θ . When r = 0. n n n n dyn g1 . . . In that case.. ∂y1 ∂yn . . . . . Linear model: If all the equations are linear. . = . . J = 0 and the mapping is degenerate. . . yn ) = 1 1 f1 . . + . . ∂(y1 . . xn ) ∗ ∗ −1 x = F (y). . . . . . . gn dx1 f1 . . xn ) ≡ JF (y). . . . . . .e. . . there exists a function G(x) ≡ such that y = G(x) if . fn Inverse function theorem: If x∗ = F (y ∗ ) and JF (y ∗ ) = 0 (JF (y ∗ ) is nonsingular). . = . . . . 5.8 Implicit function theorem and comparative statics If A = 0. .. = .. . gn dxn f1 . ⇒ r = x2 + x2 . J = JF  = r(cos2 θ + sin2 θ) = r > 0. 0). . . n g (x1 . . fn dy1 . . then F (y) is invertible nearby x∗ . . .. . . . . . . . the whole set {r = 0. just like the case in Lecture 3 when the Null space is a line. . . . . −π ≤ θ < π} is mapped to the origin (0. . . . i. Reasoning: 1 1 dx1 f1 . . g 1(x1 . . . JG (x ) = (JF (y )) . . ∂xn ∂xn . . .. . x1 x2 = F (r. . n n dxn f1 . . Ax+By = c ⇔ . . . ⇒ . . . . . . . xn ) . the model can be represtned in matrix form as a11 · · · a1n x1 b11 · · · b1m y1 c1 0 . we can derive a similar derivative matrix for general nonlinear cases. Using total diﬀerentials of the equations. . . = . . . . . . . . fn dyn −1 1 1 1 1 dy1 g1 . fn dxn Example: x2 and x1 JG = (JF )−1 . . . The derivative matrix [∂xi /∂yj ]ij = A−1 B. an1 · · · ann xn bn1 · · · anm ym cn 0 . . . . θ) = r cos θ r sin θ . . . fn . that is.
. . xn . . . . . . ∂y1 ∂ym 1 1 −1 ∂f . . . .. . = − . . (x1 . ym ) F (x(y). Jacobian matrix: Jx ≡ . . y1 . . .. . .. y ∗) = 0 and Jx (x∗ . .. . . . .. . . . xn xn (y1 . .. . In that case. . . . ∂f 1 1 ∂y1 dx1 f1 . . = x(y) = such that x∗ = x(y ∗ ) and . . . . . 1) dx1 dx2 is an equilibrium. + . = 0 . . ∂y1 ∂ym ∂y1 ∂ym Reasoning: ∂f 1 ∂f 1 1 1 ∂y1 . fn Implicit function theorem: If F (x∗ . .. y ∗) is nonsingular). xn .. . . . . . ∂f n ∂xn n n ∂xn ∂f n f1 . . y) = 0. y1. . that is. y) = 0. . . x2 ) = (1. .. . . n n ∂f n n ∂f dxn f1 . When y = 1. . fn ∂ym ∂ym . fn dx1 . . . . . . . . 1 1 ∂y1 ∂y1 f1 . . . . ⇒ . . . . . . . fn dym . ∂f n n n ∂f n dym df n 0 f1 . . . . . ⇒ . . y ∗ ) = 0 (Jx (x∗ . . . . ⇔ F (x. . . . fn . ∂f . . ym ) = 0 1 1 f1 . .. . . .. ym ) = 0 . ∂ym dy1 df 1 0 f1 . . function x = . . . . . . To calculate and at the equilibrium we use the implicit dy dy function theorem: dx1 ∂f 1 −1 1 1 ∂y f1 f2 dy dx = − 2 2 ∂f 2 2 f1 f2 dy ∂y = − 2x1 x2 x2 1 2 −1 −1 −1 0 =− 2 1 2 −1 −1 −1 0 = 1/4 1/2 . there exists a x1 x1 (y1 . . . . . . . . . . . .. fn (x1 . . . = − . . . . fn ∂ym dy1 . . . n n f1 . 1 1 ∂x1 ∂x1 −1 ∂f . . . fn dxn . . . . . . . .. y) = 0 is solvable nearby (x∗ . ∂y1 ∂ym 1 2 2 Example: f = x1 x2 − y = 0. . . . . . . . . . . . then F (x. .. . = . fn . f = 2x1 − x2 − 1 = 0. . . y ∗). ym ) .31 We can regard the LHS of a nonlinear economic model as a mapping from Rn+m to Rn : f1 (x1 .
In a 2good market equilibrium model. Q2 ) and P2 = P2 (Q1 . In a Cournot quantity competition duopoly model with heterogeneous products. ∂ 2 Y /∂X1 . use chain rule to derive ∂Z/∂X1 and ∂Z/∂X2 . ∂ 2 Y /∂X1 . X2 )+f (X1 )+g(X2). use implicit function rule to derive ∂Y /∂X1 and ∂Y /∂X2 . 13. ∂Q2 /∂P1 ∂Q2 /∂P2 . ∂P2 /∂Q1 ∂P2 /∂Q2 ∂(Q1 . .9 Problems 1. use chain rule to derive ∂Z/∂X1 and ∂Z/∂X2 . Given Y = F (X1 . F (Y1 . (a) For what value of c can we invert the demand functions to obtain P1 and P2 as functions of Q1 and Q2 ? (b) Calculate the inverse demand functions P1 = P1 (Q1 . the demand functions are given by Q1 = a − P1 − cP2 . X2 ) = X1 X2 + X1 X2 and X2 = 2X1 + 1. 4. Q2 )Q1 and T R2 (Q1 . 1 ≥ c > 0. P2 = A2 Qα Qβ−1 . 11. and ∂ 2 Y /∂X1 ∂X2 . 7. X) = Y12 + Y22 + X 2 − 1 = 0. Q2 )Q2 . ﬁnd dQ/dP . 5. (c) Derive the total revenue functions T R1 (Q1 . Q2 ) What condition(s) should the parameters satisfy so that we can invert the functions to obtain the demand functions? (b) Derive the Jacobian matrix of the derivatives of (Q1 . 1 2 1 2 (a) Calculate the Jacobian matrix ∂(P1 . Given Y = X1 X2 + 2X1 X2 . α. 2 6. 9. ﬁnd ∂Y /∂X1 . and ∂ 2 Y /∂X1 ∂X2 . 2 3. Y = X1 X2 . Given that Q = D(q ∗ e/P ). Given the consumption function C = C(Y − T (Y )). 2 2 2 2. X 2 + Y 3 = 1. and Z = Y1 Y2 . β > 0. Q2 ) = P2 (Q1 . Q2 ) with respect to ∂Q1 /∂P1 ∂Q1 /∂P2 (P1 . Y1 = X1 + 2X2 . ﬁnd the partial derivative ∂U/∂X1 and the total derivative dU/dX1 . Q2 ) = P1 (Q1 . Y2. use implicit function rule to ﬁnd dY /dX. ﬁnd dC/dY . Let U(X1 . the inverse demand functions are given by P1 = A1 Qα−1 Qβ . P2 ) ∂P1 /∂Q1 ∂P1 /∂Q2 and Jacobian . Q2 ). 2 2 8. Given the demand function Qd = (100/P ) − 10. 2 2 10. Y2 = 2X1 + X2 . Z = Y 2 + 2Y − 2. and the total diﬀerential DY . P2 ). X) = Y1 − Y2 + X − 1 = 0 and G(Y1 .32 5. 12. X1 + 2X2 + Y 2 = 1. Y2. use implicit function theorem to derive dY1 /dX and dY2 /dX. ﬁnd the demand elasticity η. Q2 = a − cP1 − P2 . ﬁnd ∂Y /∂X1 .
b). . It is easy ′ ′ to see that f− (c) ≥ 0 and f+ (c) ≤ 0. b−a b−a r (r+1) Taylor’s Theorem: If f (x) ∈ C [a. b) such that Q. b).10 Proofs of important theorems of diﬀerentiation Rolle’s theorem: If f (x) ∈ C[a. Proof: Case 1: f (x) ≡ 0 ∀x ∈ [a. (x) exists for all x ∈ (a.E. Then there exists a c ∈ (a.E. Q./ Case 2: f (x) ≡ 0 ∈ [a. r! (r + 1)! . b] ⇒ f ′ (x) = 0 ∀x ∈ (a. Assume that M = 0 (otherwise m = 0 and the proof is similar).33 5. f (b) = f (a)+f ′ (a)(b−a)+ f ′′ (a)(b−a)2 +. Then 1 1 1 f (r+1) (c)(b−a)r+1 . or φ′ (c) = f ′ (c) − f (b) − f (a) f (b) − f (a) = 0 ⇒ f ′ (c) = = 0. . and f (a) = f (b) = 0. . + f (r) (a)(b − a)r . Hence.D. there exists a c ∈ (a. b−a It is clear that φ(x) ∈ C[a. b] ⇒ ∃e. (r + 1)! 2 r! Consider the function 1 φ(x) ≡ f (b) − f (x) + f ′ (x)(b − x) + f ′′ (x)(b − x)2 + . . c such that f (e) = m ≤ f (x) ≤ M = f (c) and M > m. . b]. + 2 1 (r) ξ f (x)(b − x)r + (b − x)r+1 . b] and f ′ (x) exists for all x ∈ (a. b) such that f (b) − f (a) = f ′ (c)(b − a). b). b) such that f ′ (c) = 0. φ(a) = φ(b) = 0 so that the conditions of Rolle’s Theorem are satisﬁed for φ(x). b) such that φ′ (c) = 0. Mean Value theorem: If f (x) ∈ C[a. then there exists a c ∈ (a. . b] and φ′ (x) exists for all x ∈ (a. b] and f there exists a c ∈ (a. f ′ (c) = 0. b). Therefore. Proof: Consider the function φ(x) ≡ f (x) − f (a) + f (b) − f (a) (x − a) . Also. b).+ f (r) (a)(b−a)r + 2 r! (r + 1)! Proof: Deﬁne ξ ∈ R (b − a)r+1 1 1 ξ ≡ f (b)− f (a) + f ′ (a)(b − a) + f ′′ (a)(b − a)2 + . f ′ (x) exists for all x ∈ (a.D.
b] and φ′ (x) exists for all x ∈ (a. there exist x1 . Find U. So ﬁx y such that y − y0 < λρ. Show that f (x) is one to one in U. b).34 It is clear that φ(x) ∈ C[a. Choose an open ball B about x0 with radius ρ such that the closure [B] ⊆ U. (2.) Choose λ ≡ A/2. Since f ∈ C 1 (E).D. φy (x) ∈ [B]. Proof: (1. With φy deﬁned above. it is enough to show that y ∈ V whenever y − y0 < λρ.) Choose y1 . A = 0. Therefore. and f −1 : V → U is C 1 (U). φy (x) is a contraction of the complete space [B] into itself. f is one to one in U. Also. Since J(φy (x)) = I − A−1 J(f (x)) = A−1 [A − J(f (x))] ⇒ J(φy (x)) < on U. 2 If x ∈ [B] ⊆ U. y2 ∈ V . (3. f (U) = V . φ(a) = φ(b) = 0 so that the conditions of Rolle’s Theorem are satisﬁed for φ(x). Suppose f : E → Rn is C 1 (E). b) such that φ′ (c) = 0. Inverse Function Theorem: Let E ⊆ R be an open set.) Let V = f (U) and y0 = f (x0 ) ∈ V for x0 ∈ U. a ∈ E. and A = J(f (a)). there exists a c ∈ (a. V = f (U) is open so that f −1 is continuous. 2 2 That is. r! n Q.E.) For each y ∈ Rn deﬁne φy on E by φy (x) ≡ x + A−1 (y − f (x)). Hence. Thus. (4. f (x2 ) = y2 . ρ φy (x0 ) − x0 = A−1 (y − y0 ) < A−1 λρ = . then φy (x) − x0 ≤ φy (x) − φy (x0 ) + φy (x0 ) − x0 < 1 ρ x − x0 + ≤ ρ. f (a) = b. b ∈ V . φy (x) has a unique ﬁxed point x ∈ [B] and y = f (x) ∈ f ([B]) ⊂ f (U) = V . f is one to one on U. . there exists a neighborhood U ⊆ E with a ∈ U such that J(f (x)) − A < λ. x2 ∈ U such that f (x1 ) = y1 . or φ′ (c) = ξ − f (r+1) (c) = 0 ⇒ f (r+1) (c) = ξ. ⇒ (x2 −x1 )−A−1 (y2 −y1 ) ≤ or x2 − x1 ≤ 1 1 1 x2 −x1 ⇒ x2 −x1 ≤ A−1 (y2 −y1 ) ≤ y2 −y1 2 2 2λ 1 y2 − y1 . 2 Therefore φy (x) is a contraction mapping and there exists at most one ﬁxed point in U. φy (x2 ) − φy (x1 ) = x2 − x1 + A−1 (f (x1 ) − f (x2 )) = (x2 − x1 ) − A−1 (y2 − y1 ). Hence. Notice that f (x) = y if and only if x is a ﬁxed point of 1 φy . It follows that (f ′ )−1 exists locally about a. V ⊂ Rn such that a ∈ U. To prove that V is open. Since λ f −1 (y2 ) − f −1 (y1 ) − (f ′ )−1 (y1 )(y2 − y1 ) = (x2 − x1 ) − (f ′ )−1 (y1 )(y2 − y1 ) = −(f ′ )−1 (y1 )[−f ′ (x1 )(x2 − x1 ) + f (x2 ) − f (x1 )]. f −1 ∈ C −1 . Then there exist open sets U.
φ(x. y) = π ◦ (F ◦ F −1 )(x. 0).1≤j≤m = ∂fi ∂xn+j 1≤i. is continuous. where it exists. Since f −1 is diﬀerentiable. such that F : A × B→V has a C 1 inverse F −1 : V →A × B. φ(x. Note that since J(F (a. By the Inverse Function Theorem there exists an open set V ⊆ Rn+m containing F (a. and J(f (a. b) = 0. such that for each x ∈ B. Then π ◦ F (x.D. y) = (x. f (x. Then there exist open sets A ⊂ Rn and B ⊂ Rm a ∈ A and b ∈ B. f ′ is continuous and its inversion. y) = π(x. y) ≡ (x. b). x) = 0 and g : B → A is C 1 (B). F −1 is of the form F −1 (x. Also.1≤j≤n ∂xi ∂xn+j 1≤i≤n. there exists a unique g(x) ∈ A such that f (g(x). y)) = f ◦ F −1 (x. 0) and an open set of the form A × B ⊆ E containing (a. y)) for some C 1 function φ.E. y) = y and f (x. b)) = 0. (a.D.E. y) = (π ◦ F ) ◦ F −1 (x. Q. b)) = M = 0.j≤n ∂fi ∂xj 1≤i≤m. it is continuous. Therefore (f −1 )′ is continuous or f −1 ∈ C 1 (V ). So. deﬁne g : A→B by g(x) = φ(x.35 We have f −1 (y2 ) − f −1 (y1) − (f ′ )−1 (y1 )(y2 − y1 ) (f ′ )−1 ≤ y2 − y1 λ f (x2 ) − f (x1 ) − f ′ (x1 )(x2 − x1 ) . x2 − x1 As y2 →y1 . b ∈ Rm . Q. x2 →x1 . Proof: Deﬁning F : Rn+m →Rn+m by F (x. b) = (a. Deﬁne the projection π : Rn+m →Rm by π(x. y) = y. . φ(x. Therefore f (x. b)) = ∂xi ∂xj 1≤i. Suppose f : E → Rn is C 1 (E) and f (a. 0)) = 0. y). y)). b) ∈ E. J(F (a.j≤m I O N M . Implicit Function Theorem: Let E ⊆ R(n+m) be an open set and a ∈ Rn . y) = f (x. Hence (f 1− )′ (y) = {f ′[f −1 (y)]} for y ∈ V .
¯ I ′ (r)dr + dG ¯ ¯ Y = C(Y ) + I(r) + G ⇒ dY = C ′ (Y )dY + I ′ (r)dr + dG ⇒ dY = . Q. and export functions into the equilibrium condition. e) My > 0.2..2 Income determination model C = C(Y ) I = I(r) ¯ Y =C +I +G dQ ∂D dP ∂D df 1 = − − =0 dY2 dY ∂P dY ∂Y . Y ) = Q − D(P. e)−M(Y. var: Y. J = dP = ∂P ∂Y 1 −S ′ (P ) 0 dY ∂D ∂D − ∂D ′ ∂Y ∂P S (P ) − 0 −S ′ (P ) dQ = = ∂Y > 0. I ′ (r) < 0 end. . var: Q. e). both are assumed here as exogenous variables. ¯ ⇒F (Y. ∂r 1−C ∂G 6. e. M = M(Y. ex. I ¯ ex. 1 − C ′ (Y ) ∂Y I ′ (r) ∂Y 1 = < 0. Xe > 0 where import M is a function of domestic income and exchange rate e and export X is a function of exchange rate and foreign income Y ∗ . r. Y ) Q = S(P ) ∂D ∂D < 0. Substituting consumption. e)−Y = 0. Y. >0 ∂P ∂Y ′ S (P ) > 0 end. ′ (Y ) ¯ = 1 − C ′ (Y ) > 0. Q. var. ∂P ∂D ∂Y 1 0 J ∂D = ∂Y > 0. Y ) = 0 f 2 (P.1 Income determination and trade Consider an income determination model with import and export: C = C(Y ) 1 > Cy > 0. X = X(Y ∗ . P. Y ∗ ) ≡ C(Y )+I+X(Y ∗ . import. Me < 0 C + I + X − M = Y. we have ¯ C(Y )+I+X(Y ∗ .36 6 6. f 1 (P. var.1 Comparative Statics – Economic applications Partial equilibrium model Q = D(P. e)−M(Y. Xy∗ > 0. df dQ dP = − S ′ (P ) =0 dY dY dY ∂D ∂P 1 −S ′ (P ) 1 − 1 dP = dY = −S ′ (P ) + ∂D < 0. G. dY J J 6. e) = Y. ¯ I = I. Y ) = Q − S(P ) = 0 dQ ∂D ∂D dY 1 − . C. J − 0 < C ′ (Y ) < 1..
37 Use implicit function rule to derive ∂Y ¯ and determine its sign: ∂I 1 ∂Y FI =− = . ∗ with a similar assumption on the foreigner’s consumption function: 1 > Cy∗ > 0. e) = Y ∗ . ¯ (Xe − Me )de + dI ¯ (Me − Xe )de + dI ∗ Calculate the total diﬀerential of the system (Now Y ∗ becomes endogenous): 1 − C ′ + My −Xy∗ −My 1 − C ∗ ′ + Xy∗ J = 1 − C ′ + My −Xy∗ −My 1 − C ∗ ′ + Xy∗ dY dY ∗ = . ¯ Fy 1 − C ′ (Y ) + My ∂I Use implicit function rule to derive ∂Y and determine its sign: ∂e Fe Xe − Me ∂Y =− = . e) = X(Y ∗ . Use Cramer’s rule to derive ∂Y ∂Y ∗ and and determine their signs: ∂e ∂e dY = ¯ (Xe − Me )de + dI −Xy∗ ¯∗ 1 − C ∗ ′ + Xy∗ (Me − Xe )de + dI J ¯ ¯ (Xe − Me )(1 − C ∗ ′ + Xy∗ − Xy∗ )de + (1 − C ∗ ′ + Xy∗ )dI + Xy∗ dI ∗ = . e) = Y ¯ C ∗ (Y ∗ ) + I ∗ + M(Y.2. ∗ ′ (Y ) + M ∂Y Fy 1−C y 6. e) and M ∗ (Y ∗ . e) = M(Y. X ∗ (Y. e) − X(Y ∗ . = (1 − C ′ + My )(1 − C ∗ ′ + My∗ ) − My Xy∗ > 0.2 Interdependence of domestic and foreign income Now extend the above income determination model to analyze the joint dependence of domestic income and foreign income: ¯ C(Y ) + I + X(Y ∗ . e) and the system becomes: ¯ C(Y ) + I + X(Y ∗ . e) = Y ∗ . Since domestic import is the same as foreigner’s export and domestic export is foreigner’s import. ∂e Fy 1 − C ′ (Y ) + My Use implicit function rule to derive ∂Y and determine its sign: ∂Y ∗ Fy∗ Xy ∗ ∂Y =− = . e) − M(Y. e) − M ∗ (Y ∗ . J . e) − M(Y. e) = Y ¯ C ∗ (Y ∗ ) + I ∗ + X ∗ (Y.
C. J ∂I ∂I 6. ¯ ¯ = J < 0. var: Y . ∂e J ¯ ¯ −(Xe − Me )(1 − C ′ + My − My )de + (1 − C ′ + My )dI ∗ + Xy dI . P2 ) = S1 2 D (P1 . ¯ D 1 (P1 . Q2d = D 2 (P1 .4 ¯ ¯ Lr dG + I ′ dM . ′ ¯ Q2s = S2 (P2 ) S2 (P2 ) > 0 Q1s = S1 Q1d = Q1s Q2d = Q2s end. J ∂Y ∗ −(Xe − Me )(1 − C ′ ) = < 0. P2 ) D1 < 0. var: G. r) > 0. ¯ ¯ −LY dG + (1 − C ′ )dM J dY = = ∂Y Lr ∂Y I′ = > 0. ¯ ¯ J J ∂G ∂M ¯ 1 − C ′ dG ¯ LY dM J = Twomarket general equilibrium model 1 2 2 Q1d = D 1 (P1 . ¯ Y − C(Y ) − I(r) = G ¯ L(Y. var: Q1 . D1 > 0. ∂e J ∂Y ∗ ∂Y 1 − C ∗ ′ + My∗ My =− > 0. var: S1 . ⇒ 1 1 ¯ D1 dP1 + D2 dP2 = dS1 2 2 ′ D1 dP1 + D2 dP2 − S2 dP2 = 0. Q2 . ¯ ¯ J J ∂G ∂M 6. P2 . M s . r) = M 1 − C ′ −I ′ LY Lr ¯ dG −I ′ ¯ dM Lr J dY dr = ¯ dG ¯ dM ¯ ¯ ex. M. P2 ) − S2 (P2 ) = 0 . r. I = I(r) I ′ (r) < 0 ¯ Y =C +I +G end. ¯ (1 − C ′ (Y ))dY − I ′ (r)dr = dG ∂L ∂L ¯ dY + dr = dM ∂Y ∂r J = 1 − C ′ −I ′ LY Lr = (1−C ′ )Lr +I ′ LY < 0. P2 ) D2 < 0. M d . dr = J ∂r LY ∂r (1 − C ′ ) =− > 0.38 ¯ 1 − C ′ + My (Xe − Me )de + dI ¯∗ (Me − Xe )de + dI −My J dY ∗ = = Derive ∂Y ∗ ∂Y and ¯ and determine their signs: ¯ ∂I ∂I ∂Y (Xe − Me )(1 − C ∗ ′ ) = > 0. I. P1 .3 ISLM model C = C(Y ) 0 < C ′ (Y ) < 1 ∂L ∂L M d = L(Y. <0 ∂Y ∂r ¯ Ms = M d M = M s. ⇒ . = < 0. = > 0. ¯ ex.
n ex. To calculate ¯ . Pn ). var: Pn . Pn ). ⇒ ∂Pn ∂Pu D (Pn . dPu S (Pn ) − ∂D n /∂Pn dPu dPu S (Pn ) − ∂D n /∂Pn The markets for used cars and for new cars are actually interrelated. = 0 ¯ dQu . J = Qu = D u (Pu . Qu . In each period. ∂D n ∂D n dPn + dPu = S ′ (Pn )dPn . ∂D n /∂Pn < 0 ∂D n /∂Pu > 0. ∂Q2 ¯ ∂Q1 From Q1s = S1 . denoted by Qu . The supply function of new cars is Qn = S(Pn ). ∂D n /∂Pu < 0. we have to use chain rule: ∂ S1 ∂ S1 ∂Q2 ′ dP2 = S2 ¯ < 0. Pu ). ¯ = 1. n n Dn − S ′ Du u u Dn Du ¯ Qu = Qu n n Dn dPn +Du dPu = S ′ dPn . Instead of analyzing the eﬀects of a change in Pu on the new car market. where Qn is the quantity of new cars and Pn (Pu ) the price of a new (used) car. Pu . Qn . ∂D n /∂Pn > 0. ¯ D u (Pu . dQn . = <0 ¯ ¯ J J d S1 d S1 1 1 D1 D2 2 2 ′ D1 D2 − S 2 1 2 ′ 1 2 = D1 (D2 −S2 )−D2 D1 . n dPn −Du dPu Dn − S ′ = < 0. Pn ) = Qu . we have to use chain rule: ¯ = ∂ Qu ∂ Qu ∂ Qu dPn S ′ ¯ < 0. Qn = S(Pn ). u u ¯ Dn dPn +Du dPu = Q dPn dPu n u n u = (Dn −S ′ )Du −Du Dn . Pu ). S ′ (Pn ) > 0. ⇒ J > 0. var: Pu . n n Dn − S ′ Du u u Dn Du ¯ ex. Qn . ⇒ J > 0. J = 1 1 2 2 Assumption: D1  > D2  and D2  > D1  (owneﬀects dominate). = n < 0. = S ′ (Pn ) = ′ > 0.1 Car market Suppose we want to analyze the eﬀect of the price of used cars on the market of new cars. n n u u Assumption: Dn  > Du  and Du  > Dn  (owneﬀects dominate). var: Pn . ¯ ¯ J J dQu dQu ∂Qn ∂Qn ¯ ∂Qu From Qu = Qu . The demand for new cars is given by Qn = D n (Pn . Pu ) = S(Pn ). ⇒ D n (Pn . Qn = D n (Pn . end. To calculate ¯ . ¯ = 1. The demand for used cars is Qu = D u (Pu . ¯ the quantity of used cars supplied is ﬁxed. ¯ ∂ S1 d S1 6.4. end. Pu ) = S(Pn ). var: Qu . we want to know how a change in ¯ Qu aﬀects both markets.39 1 1 D1 D2 2 2 ′ D1 D2 − S 2 dP1 dP2 = ¯ d S1 0 . dPn ∂D n /∂Pu dQn dPn S ′ (Pn )∂D n /∂Pu = ′ > 0. ′ 2 dP1 dP2 D 2 − S2 −D1 = 2 < 0.
= J dK 1 0 −FLL FLK J Problem Q = D(P ) D ′ (P ) < 0 Q = S(P. W ) 0 < CY < 1 CW > 0 I = I(r) I ′ (r) < 0 M D = L(Y. 1. The ISLM model becomes C = C(Y. well as on income Y . derive ∂Y /∂W and ∂r/∂W . L) FLK > 0. W ) − I(r) = 0. J = 1 −h′ (w) −FLL 1 = = 1−h′ (w)FLL > 0. r) = M. Let the demand and supply functions for a commodity be 2. ∂S/∂t < 0. L(Y. FLK > 0. (c) Determine the sign of dQ/dt and dP/dt. as MS = M Y =C +I MS = MD (a) Which variables are endogenous? Which are exogenous? (b) Which equations are behavioral/institutional. w. L − h(w) = 0 w − FL = 0 1 −h′ (w) −FLL 1 0 dL = dK 6. (d) Use the Q − P diagram to explain your results. (b) Use Cramer’s rule to compute dQ/dt and dP/dt. (d) Use Cramer’s rule to derive the eﬀects of an increase in W on Y and r. ∂L endogenous variables: L. t) ∂S/∂P > 0.6 FLK dL dw −h′ (w) 1 J = ′ ⇒ dL − h′ (w)dw = 0 . (c) Derive the total diﬀerential for each of the two equations. where t is the tax rate on the commodity.5 Classic labor market model L = h(w) h′ > 0 labor supply function ∂Q w = MPPL = = FL (K. ie.. r) LY > 0 Lr < 0 which is predetermind. exogenous variable: K. . (a) Derive the total diﬀerentail of each equation.40 6. J 0 FLK dK = h FLK dw > 0. FLL < 0 labor demand function. which are equilibrium conditions? The model can be reduced to Y − C(Y. dw − FLL dL − FLK dK = 0 . (e) Determine the signs of ∂Y /∂W and ∂r/∂W . Suppose consumption C depends on total wealth W .
S2 < 0. 2 The supply of the ﬁrst commodity is given exogenously. Therefore. ie. Both D 1 and D 2 depend only on their own prices.. P2 ) S 2 (P1 . Consider a 2industry (e. P2 ) = D1 (P1 ) + G S 2 (P1 . The model can be reduced to S 1 (P1 . the supply of each commodity depends on the price of the other commodity as well as on its own price. The demand functions of a 2commodity market model are: Qd = D 1 (P1 . Because each industry requires in its production process outputs of the other. (d) Use Cramer’s rule to derive ∂P1 /∂G and ∂P2 /∂G. 2 2 (a) Which variables are endogenous? Which are exogenous? (b) Which equations are behavioral? Which are deﬁnitional? Which are equilibrium conditions? The model above can be reduced to : D 1 (P1 . (e) Determine the signs of ∂P1 /∂G and ∂P2 /∂G.g. 5. 1 1 1 1 S1 > 0. P2 ) 1 Qd = D 2 (P1 . manufacturing and agriculture) general equilibrium model. 2 ′ D1 (P1 ) < 0. (Hint: Use chain rule. 1 1 Qd = Qs . The 1 supply of the second commodity depends on its own price. The 2 equilibrium conditions are: Qd = Qs . (f) Compute ∂Q1 /∂G and ∂Q2 /∂G. P2 ) Qs . S2 > S1 . 2 2 2 2 S1 < 0. (a) Which variables are endogenous? Which are exogenous? (b) Which equations are behavioral? Which are deﬁnitional? Which are equilibrium conditions? 4. P2 ) = D2 (P2 ) (c) Compute the total diﬀerential of each equation. the model may be written as follows: Qd 1 Qd 2 Qs 1 Qs 2 Qd 1 Qd 2 = = = = = = D1 (P1 ) + G D2 (P2 ) S 1 (P1 . The agricultural products have only private demand D 2 .) (g) Give an economic interpretation of the results. Qs = S2 (P2 ). P2 ). The demand for manufacturing product consists of two components: private demand D 1 and goverment demand G. ′ D2 (P2 ) < 0.41 3. 1 Qs . P2 ) − S1 = 0 D 2 (P1 . Qs = S1 . S2 > 0. P2 ) − S2 (P2 ) = 0 . S1 > S2 .
6. (a) Calculate the Jacobian matrix ∂(P1 . PC are price of ﬁsh and price of chicken respectively. Q2 ) and U2 (Q1 . N). (b) Use Cramer’s rule to ﬁnd dPF /dN and dPC /dN. P2 = U2 (Q1 . i that each one is a gross substitute for the other (ie. Q2 ). P2 ). Q2 = P2 . C ′ > 0. Di < 0. (d) Use Cramer’s rule to derive ∂P1 /∂S1 and ∂P2 /∂S1 . Q2 ) What condition(s) should the parameters satisfy so that we can invert the functions to obtain the demand functions? (b) Derive the Jacobian matrix of the derivatives of (Q1 . Q2 ). where U1 (Q1 . and Q∗ and Q∗ are market equilibrium quantities. FN > 0. (c) Calculate the total diﬀerential of each equation of the reduced model.) da da . N) DC (PC − PF ) = C(PC ) (a) Find the total diﬀerential of the reduced system. i = 1. The demand functions for ﬁsh and chicken are as follows: ′ Qd = DF (PF − PC ). (f) Compute ∂Q1 /∂S1 and ∂Q2 /∂S1 and determine their signs. (e) Determine the signs of ∂P1 /∂S1 and ∂P2 /∂S1 . 7. ∂P2 /∂Q1 ∂P2 /∂Q2 ∂(Q1 . In a 2good market equilibrium model.42 i Suppose that both commodities are not Giﬀen good (hence. DC < 0 C where PF . Q2 ) are the partial derivatives of a utility function U(Q1 . (Hint: Eliminate P1 and P2 . and that S2 (P2 ) > 0. Find the comparative 1 2 dQ∗ dQ∗ 1 2 statics and . 2). Dj > 0. the inverse demand functions are given by P1 = U1 (Q1 . that i i ′ Di  > Dj . P2 ) ∂P1 /∂Q1 ∂P1 /∂Q2 and Jacobian . ∂Q2 /∂P1 ∂Q2 /∂P2 (c) Suppose that the supply functions are Q1 = a−1 P1 . The supply of chicken depends only on its price PC : Qs = C(PC ). What is the economic meaning of your results? (d) Find dQC /dN. DF < 0 F ′ Qd = DC (PC − PF ). Q2 ) with respect to Q1 and Q2 . . The supply of ﬁsh depends on the number of ﬁshermen (N) as well as its price PF : Qs = F F (PF . FPF > 0. (c) Determine the signs of dPF /dN and dPC /dN. respectively. i = j).. Q2 ) with respect to ∂Q1 /∂P1 ∂Q1 /∂P2 (P1 . The model can be reduced to C DF (PF − PC ) = F (PF .
Determine the signs of dP2 /dt and dQ2 /dt. i i i Suppose that Di < 0 and Si′ > 0. 2. In a 2good market equilibrium model with a sales tax of t dollars per unit on product 1. P2 ) = Q1 = S1 (P1 ). (b) Calculate dQ2 /dt. (e) Explain your results in economics. Determine the signs of dP2 /dt and dQ2 /dt. i. the model becomes D 1 (P1 + t. i (d) Suppose that Dj < 0. i (c) Suppose that Dj > 0. P2 ) = Q2 = S2 (P2 ). i = j.43 8. (a) Calculate dP2 /dt. j = 1. . Di  > Dj . D 2 (P1 + t.
when doing comparative static analysis. y1 .xn k g (x . xn ) such that (x1 . . . xn . x . and Lr < 0 in the ISLM model. . max f (x1 . . . ym : variables given to the agent (exogenous variables). . ym ) ≥ 0 . sometimes more appropriate. . . y ) ≥ 0. The agent will choose a feasible decision such that the objective function is maximized. . . . . . .1 Neoclassic methodology Neoclassic assumption: An agent. g 1 (x1 . . Remarks: (1) Biological behavior is an alternative assumption. ..44 7 Optimization A behavioral equation is a summary of the decisions of a group of economic agents in a model. . xn . . . . . p2 . A producer will choose to supply the quantity such that his proﬁt is maximizaed. . . . In the last chapter. q2 ) subject to the budget constraint p1 q1 + p2 q2 = m ⇒ demand functions q1 = D 1 (p1 . ym). . . Maximization ⇒ Behavioral equations Game equilibrium ⇒ Equilibrium conditions x1 . . ym ). . . m). . ym ) subject to . y1 . A demand (supply) function summarizes the consumption (production) decisions of consumers (producers) under diﬀerent market prices. . (2) Sometimes an agent is actually a group of people with diﬀerent personalities like a company and we have to use game theoretic equilibrium concepts to characterize the collective behavior. i = 1.. . . y1 . . y . . n (derived from FOC). Opportunity set: the agent can choose only (x1 . Example 1: A consumer maximizes his utility function U(q1 . y1 . . . . etc. y1 . we are going to provide a theoretical foundation for the determination of these signs. . has an objective function in mind (or has well deﬁned preferences). xn . . .. . In this chapter. . D ′ (P ) < 0 and S ′ (P ) > 0 in the partial market equilibrium model. A is usually deﬁned by inequality constriants. . A consumer will choose the quantity of each commodity within his/her budget constraints such that his/her utility function is maximized. 1 n 1 m Solution (behavioral equations): xi = xi (y1 . m) and q2 = D 2 (p1 . . Ly > 0.. when making decisions. C ′ (Y ) > 0. . . I ′ (r) < 0. . . 7. . . . xn : variables determined by the agent (endogenous variables). . . x1 . ∂xi /∂yj : derived by the comparative static method (sometimes its sign can be determined from SOC). . . Objective function: f (x1 . xn . . ym) ∈ A ⊂ Rn+m . . . p2 . . we always assumed that the signs of derivatives of a behavioral equation in a model are known. . The derivative of a behavioral function represents how agents react when an independent variable changes. For example. .
Qu : unique global max not global max but not unique Ql EQ Qg EQ Qu EQ The agent will choose only a global maximum as the quantity supplied to the market.) We will ﬁnd conditions (eg. A local maximum Ql : there exists ǫ > 0 such that F (Ql ) ≥ F (Q) for all Q ∈ (Ql − ǫ. the supply quantity is not unique. In that case. One of them must be a global maximum. otherwise the problem does not have a solution (the maximum occurs at ∞. ⇒ to ﬁnd a global maximum.2 Diﬀerent concepts of maximization Suppose that a producer has to choose a Q to maximize its proﬁt π = F (Q): maxQ F (Q). it is possible that there are more than one global maximum. A global maximum Qg : F (Qg ) ≥ F (Q) for all Q. 7. Therefore. A local maximum must be a critical point but a critical point does not have to be a local maximum. This is called the ﬁrst order condition (FOC) for a local maximum.45 Example 2: A producer maximizes its proﬁt Π(Q. Assume that F ′ (Q) and F ′′ (Q) exist.3 FOC and SOC for a local maximum At a local maximum Ql . However. n endogenous variables without constraints: next n endogenous variables with equality constraints: Nonlinear programming: n endogenous variables with inequality constraints Linear programming: Linear objective function with linear inequality constraints Game theory: more than one agents with diﬀerent objective functions 7. we prefer that the maximization problem has a unique global maximum. P ) = P Q − C(Q) where C(Q) is the cost of producing Q units of output ⇒ the supply function Q = S(P ). . A critical point Qc : F ′ (Qc ) = 0. we ﬁrst ﬁnd all the local maxima. A unique global maximum must be a global maximum and a global maximum must be a local maximum. F F F T l : local max.. increasing MC or decreasing MRS) so that there is only one local maximum which is also the unique global maximum. one endogenous variable: this chapter. Ql + ǫ). A unique global maximum Qu : F (Qu ) > F (Q) for all Q = Qu . the slope of the graph of F (Q) must be horizontal F ′ (Ql ) = 0. Q T T Qg : a global max.
F (4) (5) = −24 < 0 and Q = 5 is a local maximum. T T F ′′ (Qc ) = 0 degenerate F ′′ (Qc ) > 0 local min. If the lowest order nonzero derivative is of odd order. Remark 2 (Unboundedness): If limQ→∞ F (Q) = ∞. Remember that Q is endogenous and P is exogenous here. Remark 1 (Degeneracy): For a degenerate critical point. F ′ (5) = F ′′ (5) = F ′′′ (5) = 0. Therefore. Remark 3 (Nondiﬀerentiability): If F (Q) is not diﬀerentiable. it is possible that the solution occurs at Q = 0 (or at Q = a). Qc EQ Qc EQ Qc EQ A degenerate critical point: F ′ (Q) = F ′′ (Q) = 0. 5. FOC is modiﬁed to become F ′ (Q) ≤ 0. FOC: F ′ (Ql ) = 0 SOC: F ′′ (Ql ) < 0 Example: F (Q) = −15Q + 9Q2 − Q3 . eg. F ′ (5) = F ′′ (5) = 0 and F ′′′ (5) = 6 = 0 and Q = 5 is not a local maximum. F ′ (Q) = −15 + 18Q − 3Q2 = −3(Q − 1)(Q − 5). F (Q) = −(Q − 5)4 . 7. we have to check higher order derivatives. F ′′ (1) = 12 > 0. There are two critical points: Q = 1. then it is a local maximum (minimum). ∂Q The FOC is the inverse supply function (a behavioral equation) of the producer: P = C ′ (Q) = MC. F (Q) = (Q − 5)3 . eg. A nondegenerate critical point: F ′ (Q) = 0. F ′′ (Q) = 0. To take care of such possibilities.. Q = 5 is a local maximum. then it is a reﬂect point. A nondegenerate critical point is a local maximum (minimum) if F ′′ (Q) < 0 (F ′′ (Q) > 0).4 Supply function of a competitive producer Consider ﬁrst the proﬁt maximization problem of a competitive producer: max Π = P Q − C(Q). and F ′′ (5) = −12 < 0. F ′′ (Q) = 18 − 6Q. To ﬁnd . If the lowest order nonzero derivative is of even order and negative (positive).. (a − Q)F ′ (Q) = 0). then we have to use other methods to ﬁnd a global maximum. then a global maximum does not exist. Remark 4 (Boundary or corner solution): When there is nonnegative restriction Q ≥ 0 (or an upper limit Q ≤ a). It is a global maximum for 0 ≤ Q < ∞. QF ′ (Q) = 0 (or F ′ (Q) ≥ 0. Q FOC ⇒ ∂Π = P − C ′ (Q) = 0.46 F F F T F ′′ (Qc ) < 0 local max.
T MC Pc S(Pc ) = {Q1 . P T P MC MC is the supply curve. SOC: FXX < 0 dY FXX dX FXY Case 1: FXY > 0 ⇒ =− > 0. but the supply curve is only part of the increasing sections of the MC curve and can be discontinuous. ∂U FOC: = u′ (Q) − P = 0. the sign of depends only on the sign of FXY . TheredP ∂Q2 Remark: The result is true no matter what the cost function C(Q) is. dY FXX dX FXY Case 2: FXY < 0 ⇒ =− < 0. dQs > 0.47 dQ the comparative statics . Y ). FOC: FX (X ∗ . dY FXX dX Therefore. EQ Q1 Q2 EQ 7. we use the total diﬀerential method discussed in the last dP chapter: dQ 1 dP = C ′′ (Q)dQ. dY 7. MC = C ′ (Q) can be nonmonotonic. supply curve has two component. a behavioral equation) . which is = −C ′′ (Q) < 0. we need the SOC. ⇒ X ∗ = X(Y ) · · · · · · Behavioral Equation dX FXY Comparative statics: FXX dX + FXY dY = 0 ⇒ =− . ⇒ = ′′ .5 Maximization and comparative statics: general procedure X Maximization problem of an agent: max F (X. dP C (Q) To determine the sign of fore. dP dQ ∂2Π .6 Utility Function A consumer wants to maximize his/her utility function U = u(Q) + M = u(Q) + (Y − P Q). Y ) = 0. Q2 }. ∂Q ⇒ u′ (Qd ) = P (inverse demand function) ⇒ Qd = D(P ) (demand function.
π(w. K). a simpliﬁed version of Hotelling’s lemma. dP Qd =D(P ) In the input demand function problem. K(Q))/∂Q =SMC(Q. Its proﬁt is Π = pf (x) − wx. a behavioral equation) ∂2Π dx = −1 ⇒ = h′ (w/p) < 0. FOC implies dY that the indirect eﬀect of Y on M is zero. 7.8 Envelope theorem Deﬁne the maximum function M(Y ) ≡ maxX F (X. ⇒ K = K(Q) is the optimal ﬁrm size. where x is the quantity of an input employed. By envelope theorem. In the consumer utility maximization problem. Let p = 1 and still write π(w) ≡ f (h(w)) − wh(w). That is. LTC(Q) = minK C(Q. Each K corresponds to a STC. Each STC tangents to the LTC (STC = LTC) at the quantity Q such that K = K(Q).48 dQd ∂2U = UP Q = −1 ⇒ = D ′ (P ) < 0. The envelope theorem dΠ implies that π ′ (w) = = −h(w). LTC is the envelope of STC’s. the input demand is a decreasing func∂x∂(w/p) d(w/p) w tion of the real input price . K(Q)). p) ≡ pf (h(w/p)) − wh(w/p) is the proﬁt function. Y ) = F (X(Y ). K: ﬁrm size. .7 Input Demand Function The production function of a producer is given by Q = f (x). V (P ) ≡ U(D(P )) + Y − P D(P ) is called the indirect utility function. dw x=h(w) Example: The relationships between LR and SR cost curves STC(Q. where p (w) is the price of the output (input). K) = C(Q. K). At the maximization point X = X(Y ). LMC(Q) = dLTC(Q)/dQ = ∂C(Q. Notice that the endogenous variable is K and the exogenous is Q here. when K = K(Q) is optimal for producing Q. dY ∂Y X=X(Y ) dX Proof: M ′ (Y ) = FX + FY . Y ) = M ′ (Y ) = . The FOC of proﬁt maximization problem is pf ′ (x) − w = 0 ⇒ f ′ (x) = w/p (inverse input demand function) ⇒ x = h(w/p) (input demand function. Y ) then the total derivative dM ∂F (X. p 7. The envelope theorem implies that V ′ (P ) = dU ≡ −D(P ). SMC=LMC. this is a simpliﬁed version of Roy’s identity. the demand function is a decreasing ∂Q∂P dP function of price.
LAC is the envelope of SAC’s and each SAC tangents to the LAC (SAC = LAC) at the quantity Q such that K = K(Q). Therefore. Case 1: MR is downward sloping and MC is upward sloping: When t increases. C = C(q).9 Eﬀect of a Unit Tax on Monopoly Output (Samuelson) Assumptions: a monopoly ﬁrm. Case 3: Both MR and MC are downward sloping. FOC ⇐⇒ MR = MC + t. The diagram shows that dq ∗ /dt > 0. t = unit tax max π(q) = P q − C(q) − tq = qf (q) − C(q) − tq q: endogenous. P P T T MC+t MC MRE MC+t MC < q < MRE q Case 2: Both MR and MC are downward sloping and MR is steeper. Why? MR = MC + t violates SOC < 0. 7. an equilibrium is determined by the intersection point of MR and MC + t curves. q = D(P ) ⇐⇒ P = f (q). . t: exogenous FOC: ∂π/∂q = f (q) + qf ′ (q) − C ′ (q) − t = 0. MR decreasing more ⇒ t ↑ q ↓. MC decreasing. but MC is steeper. The derivative dq ∗ /dt can be determined by the sign of the cross derivative: ∂2π = −1 < 0 ∂q∂t Therefore. on q–p space. therefore. It is opposite to our comparative statics result. q ∗ decreases as seen from the left diagram below. (inverse functions).49 Since LAC(Q) =LTC(Q)/Q and SAC(Q) =STC(Q)/Q. The FOC deﬁnes a relationship between the monopoly output q ∗ and the tax rate t as q ∗ = q(t) (a behavioral equation). the intersection of MR and MC + t is not a proﬁt maximizing point. The result can be obtained using the q–pdiagram. we have dq ∗ /dt < 0.
the input market equilibrium is X = 2r2 +1 2b and W = b + X = ar+2r . where W is the price of X. The inverse demand function for Q is P = a − Q.10 MCE q A Price taker vs a price setter A producer employs an input X to produce an output Q. 2 −br Equating the demand and supply of Q. Q r r The output supply function is Q = Q(P ) = r P2−br . Case 3: The ﬁrm is a monopolist in Q and a monopsony in X. Case 2: The ﬁrm is a price taker in Q and a monopsony in X. the equilibrium values of W . Case 1: The ﬁrm is a monopolist in Q and a price taker in X. Given W .. The production function is Q = rX. 2r 2 +1 Substituting back into the production and output demand functions. The producer is the only seller of Q and only buyer of X. and P . x 2 . The inverse supply function of X is W = b + X. Substituting back into the production and output demand functions. its object is b 2Q max π = P Q − (b + X)X = P Q − (b + (Q/r))(Q/r). the output market equilibrium is Q = ar2 −2 r 2a+br and P = a − Q = 2r2 +1 . x ⇒ FOC ar−2r 2 X−W = 0. P is endogenous and W is exogenous. X. To the producer.e. i. It depends on the producer’s power in each market. We want to ﬁnd the 2market equilibrium. Q (output) and X (input). Q. Given P . P is exogenous and W is endogenous. its object is max π = P Q−W X = (a−Q)Q−W X = (a−rX)(rX)−W X. ⇒ FOC P − − 2 = 0. the output market equilibrium is Q = rX = f ar 2 − br2r 2 + 1 and P = a − Q = f a + ar 2 + b2r 2 + 1. To the producer. the output market equilibrium is X = Q/r = f ar − br 2 − 2 and W = b + X = f ar + br 2 − 3br 2 − 2. its object is max π = (a − Q)Q − (b + X)X = (a − (rX))(rX) − (b + X)X. To the producer. 2r 2 ar−b Equating the demand and supply of X. There are two markets. The input demand function is X = X(W ) = ar−W . both P and W are endogenous.50 P T MR MC+t > 7.
r.) (Show that π is strictly concave in X.t. 7. dN 168 − N C L wN .) Find the proﬁt maximizing X as a function of a and b. ∂X ∂X Determine the sign of the comparative statics and and explain your results ∂a ∂b in economics. Derive the price and the wage rate set by the ﬁrm P and W and compare the results with that of cases 1 and 2. X = X(a. and the budget constraint which equates his consumption (C) to his wage income (wN). The slope the supply function N ′ (w) is determined by the sign of the crossderivative fN w C ← N ւ U.51 (We can also eliminate X instead of Q. N. L) = U(wN. The FOC requires that the total derivative of U w. dU w 1 w 1 = √ −√ = √ −√ = 0. b). Numerical Examples: √ √ Example 1: U = 2 C + 2 L elasticity of substitution σ > 1 √ √ √ √ U = 2 C+2 L = 2 wN+2 168 − N . N be equal to 0.11 Labor Supply Function Consider a consumer/worker trying to maximize his utility function subject to the time constraint that he has only 168 hours a week to spend between work (N) and leisure (L). therefore. w) Here N is endogenous and w is exogenous. and the wage rate w. N + L = 168. The two procedures are the same. C = wN. FOC: dU = fN = UC w + UL (−1) = 0. Uc . as follows: max U = U(C. dN FOC deﬁnes a relationship between the labor supply of the consumer/worker. UL ւ տ տ L w fN w = UC + wNUCC − NULC The sign of fN w is indeterminate. which is exactly the labor supply function of the individual N ∗ = N(w). 168 − N) ≡ f (N. the slope of N ∗ = N(w) is also indeterminate.
which is negatively sloped. Some Rules: d x (e ) = ex . the inverse labor supply function is w = N/(168 − N). dx . Example 3: U = CL CobbDouglas( σ = 1) U = CL = wN(168 − N) dU = w(168 − 2N) = 0 dN The labor supply function is a constant N = 84 and the curve is vertical. C +L wN + 168 − N Therefore. 7. dx d ax e = aeax . . which is positively sloped. The Taylor expansion of the exponential functionat x = 0 becomes ex = 1 + x + x2 /2! + · · · + xn /n! + . = f (n) (x).12 Exponential and logarithmic functions and interest compounding Exponential function f (x) = ex is characterized by f (0) = 1 and f (x) = f ′ (x) = f ′′ (x) = . the inverse labor supply function is w = [(168 − N)/N]2 . . . . dx d f (x) e = f ′ (x)ef (x) . dN (C + L)2 wN(168 − N) CL U= = . w w T T σ=1 EN EN Backwardbending Labor Supply Curve: It is believed that the labor supply curve can be backwardbending.52 Therefore. w w T σ>1 T EN σ<1 EN Example 2: U = CL C+L elasticity of substitution σ < 1 dU wL2 − C 2 = = 0.
this is the formula for instantaneous compounding. a > 0. then the interest rate each time becomes r/m. r = interest rate. ln x More generally. therefore.53 The inverse of ex is the logarithmic function: ln x. if x = ay . n = number of periods V = A(1 + r)n If we compound interest m times per period. V → Aern . then we deﬁne y ≡ loga x = . Interest compounding A: principal (PV). . y dt f dt Example: f (t) = g(t)h(t). ln f (t) = ln g(t) + ln h(t). dx e y y = ex T y = ln x Ex x ax = eln a = e(ln a)x ⇒ d lna x = dx 1 ln a 1 x d x a = (ln a)e(ln a)x = (ln a)ax . the number of times of compounding becomes mn. ln x = ln x = 1/x. and V = A[(1 + r/m)m ]n m→∞ lim (1 + r/m)m = 1 + r + r 2 /2! + · · · + r n /n! + . . . growth rate of f (t) is equal to the sum of the growth rates of g(t) and h(t). = er Therefore. dx growth rate of a function of time y = f (t): growth rate ≡ 1 dy f′ d = = [ln f (t)]. V = Future Value. then we deﬁne y ≡ ln x. If x = ey . ln a 1 d Using inverse function rule: dx = ey dy.
Then the utility function become U(X) + (100 − XP ). can be derived by maximizing the utility function U(C. ⇒ at t = 25. The consumption function of a consumer.13 Timing: (when to cut a tree) t: number √ years to wait. then to wait 25 years before cutting the tree is the optimum. (b) What is the behavior equation implied by the ﬁrst order condition? (c) Derive dX/dP and determine its sign. . where u′1 (C) > 0. X = X(W ). can be derived by maximizing the proﬁt function Π = F (X) − W X with respect to X. 7. Suppose that the total income of the consumer is $ 100 and that the price of X is P . at t = 26. where X is the quantity of input X and W is the price of X. 3. Derive the supply function. where X is the quantity of commodity X consumed and M is money. Derive the comparative statics dX/dW and determine its sign. Suppose the total cost function of a competitive ﬁrm is C(Q) = eaQ+b . 2. Y ) = u1 (C) + u2 (Y − C). C = C(Y ). What is the economic meaning of your result? 4.54 7. max A(t) = Ke t e−rt . growth rate = 10% = r. (a) Find the ﬁrst order condition of the utility maximization problem. Suppose that A(t) = ef (t) FOC becomes: A′ (t)/A(t) = f ′ (t) = r. The utility function of a consumer is given by U = U(X) + M. u2 ”(Y − C) < 0.14 Problems 1. t∗ = 1/(4r 2) 2 t For example. The input demand function of a competitive producer. SOC: It can be shown that A′′ < 0. growth rate < 10%. t √ The FOC is A′ = Because A(t) = 0. FOC implies: 1 √ − r A(t) = 0 2 t 1 √ − r = 0. A(t) present value after t years of V (t) = Ke√ t : the market value of the tree after t years A(t) = Ke t e−rt is the present value We want to ﬁnd the optimal t such that the present value is maximized. ⇒ f ′ (t) is the instantaneous growth rate (or the marginal growth rate) at t. Derive the comparative statics dC/dY and determine its sign. t = 25. if r = 10%. u′2 (Y − C) > 0 and u1 ”(C) < 0. it is better to cut and sell the tree at t = 25 and put the proceed in the bank than waiting longer. Therefore.
given the quantity of ﬁxed input K. (a) State the FOC and the SOC of the proﬁt maximization problem. QB . w is wage rate. Show that if u increases. K).there is a L which maximizes Π. C ′ > 0. x = x(b). the optimal output is a function of oil price. you are asked to derive the relation between L and p. (a) State the FOC and SOC of the proﬁt maximization problem. and r is the rental rate on capital. b is the price of oil. for the production of a commodity (Q). The proﬁt of ﬁrm A is ΠA = P QA − T CA = QA f (QA + QB ) − C(QA ). dQB (b) When QB increases. b) is the total cost function. The labor supply function is L = L(w). D(w) is labor demand function. Hence. Consider a duopoly market with two ﬁrms. The production function of a competitive ﬁrm is given by Q = F (L. and D ′ (w) < 0. where x is output. b). will ﬁrm A’s output QA increase or decrease? 6. which he sells in a perfect competition market. Assume that Cbx = ∂ 2 C/∂b∂x > 0. The short run proﬁt function is given by Π = pQ−wL−rK. where L is variable input and K is ﬁxed input. there is an optimal labor input which maximizes the monopsonist’s total proﬁt Π = pf (L) − W (L)L. Consider a monopsony who uses a single input. there is a QA which maximizes ﬁrm A’s proﬁt. His production function is Q = F (L). that is. In the short run. and that the union wishes to maximize the sum of the wage bill plus the unemployment compensation S = wD(w) + u(L − D(w)). For a given output of ﬁrm B. (Hint: w is endogenous and u is exogenous. where w is wage per worker. an increase in oil price will increase marginal cost. (f ′ (L) > 0). A and B. or more convenient for this problem. that unemployed workers are paid unemployment insurance at a rate of $u per worker. Assume that FLK > 0. is dx/db > 0? 7. Suppose that a union has a ﬁxed supply of labor (L) to sell. f ” < 0. where p is output price. the short run demand for L can be regarded as a function of K. For any given oil price b. The proﬁt of a monopolistic ﬁrm is given by Π = R(x) − C(x. (b) Derive the comparative statics dL/dK and determine its sign. w = L−1 (L) = W (L). that is. f ′ < 0. 8. (b) Derive the comparative statics dL/dp and determine its sign. QA = RA (QB ).55 5. labor (L). C” > 0. .) 9. and C(x. The cost function of ﬁrm A is T CA = C(QA ). R(x) is total revenue. Will an increase in oil price increase output. ′ (a) Find the slope of the reaction function RA = dQA . Given the commodity price p. This relationship between QB and QA is called the reaction function of ﬁrm A. In this problem. then the union should set a higher w. there is an optimal output which maximizes proﬁt. The inverse demand function is P = f (QA + QB ). that is.
X θ ≡ (1 − θ)X 0 + θX 1 ∈ A. Also. Convex sets A is a convex set if ∀X 0 . F G− ≡ {(x. F − G+ . GF ⊂ R 2 . T T G+ F F (X) + GF F (X) G− F EX G− F EX G− is a convex set ⇒ F (X) is concave F G+ is a convex set ⇒ F (X) is convex F Equivalent Deﬁnition: Given X 0 < X 1 .15 Concavity and Convexity The derivation of a behavioral equation X = X(Y ) from the maximization problem maxx F (X. x ∈ R}.) x2 x2 T 1( 0) d d T 52 43 E x1 some nonconvex sets some convex sets E x1 (1) If A1 and A2 are convex. (2) The convex hull of A is the smallest convex set that contains A. y) y ≤ F (x). Here we are going to discuss a condition on F (X. Convex and concave functions Given a function F (X). Y ) is valid only if there exists a unique global maximum for every Y . then X = X(Y ) has multiple values and the comparative static analysis is not valid. X 1 ∈ A and 0 ≤ θ ≤ 1. X 1 ∈ A then the whole line connecting X 0 and X 1 is in A. If there are multiple global maximum. the empty set itself is a convex set.56 7. (If X 0 . 0 ≤ θ ≤ 1. we deﬁne the sets G+ ≡ {(x. F 1 = F (X 1 ). the deﬁnition is similar. Also . denote F 0 = F (X 0 ). For example. then A1 ∩ A2 is convex but A1 ∪ A2 is not necessarily convex. F If G+ (G− ) is a convex set. then we say F (X) is a convex function (a concave funcF F tion). F (X θ ) = F ((1 − θ)X 0 + θX 1 ). Deﬁne X θ ≡ (1 − θ)X 0 + θX 1 . x ∈ R}. If F (X) is deﬁned only for nonnegative values X ≥ 0. the convex hull of {X 0 } ∪ {X 1 } is the straight line connecting X 0 and X 1 . Y ) so that a critical point is always a unique global maximum and the comparative static analysis is always valid. y) y ≥ F (x).
F (X θ ) < F θ . 1 − Xθ 1 − X 0) X (1 − θ)(X 1−θ Fθ − F0 θ(F 1 − F 0 ) θ = = . F 1 ) and when θ shifts from 0 to 1. F 0 ) and (X 1 .57 deﬁne F θ ≡ (1 − θ)F (X 0 ) + θF (X 1 ) = (1 − θ)F 0 + θF 1 . F (X) is strictly concave . F 0 ) to (X 1 . 1 − Fθ 1 − F 0) F (1 − θ)(F 1−θ T F (X) X0 Xθ X1 EX F (X) is strictly concave ⇒ if for all X 0 . For a strictly concave function. 1). (They have something to do with the whole graph of F . 1). 1]. F 1 = F (X 1 ) T F (X θ ) F (X) Fθ F0 F1 T F 0 = F (X 0 ) X 0 X θ X 1 EX X 0 X θ X 1 EX Therefore. (X θ . F (X) is convex ⇒ if for all X 0 . F θ ) is located on the straight line connecting (X 0 . (X θ . X 1 and θ ∈ [0. F (X θ ) > F θ . not just the behavior of F nearby a point. F (X θ ) ≤ F θ . X 1 and θ ∈ (0. F θ ) shifts from (X 0 .) The graph of a concave function can have a ﬂat part. F (X) is concave ⇒ if for all X 0 . F (X θ )) shifts along the curve representing the graph of F (X) (the left ﬁgure). X 1 and θ ∈ [0. (X θ . the graph should be curved everywhere except at kink points. Put the two ﬁgures together: F (X ) θ F θ Xθ − X0 θ(X 1 − X 0 ) θ = = . F 1 ) (the right ﬁgure). X 1 and θ ∈ (0. F (X θ ) ≥ F θ . F (X) is strictly convex ⇒ if for all X 0 . 1]. T T F (X) F (X) is concave F (X) (not strictly) E EX X Notice that these concepts are global concepts. On the other hand.
there exist X 0 ∈ [X 0 . the income tax schedule T = f (Y ) is a piecewiselinear function. Remark 1 (boundary/corner solution): The boundary or corner condition F ′ (X) ≤ 0. X θ ] and X 1 ∈ [X θ . Theorem 1: F (X) is concave. F F F T T $$ $$ T ¨ ¨¨ ¨¨ ¨ ¨ ¨¨ EX ¢ ¢ ¢ ¢ ¢ ¢ a convex ¢ piecewiselinear ¢ function ¢ ¢ ¢ ¢ a concave piecewiselinear function EX EX Remark 2: A piecewiselinear function consists of linear components. then there exists X 1 such that F (X 1 ) > F (X 0 ). then X 0 is a unique global maximum. X 0 is not a local maximum. . ⇔ F ′′ (X) ≤ 0 for all X. for example. If F (X) is strictly concave and F ′ (X 0 ) = 0. ¯ ¯ Proof: By Taylor’s theorem. we assume that F ′′ (X) exists for all X. Therefore. a contradiction. then X 0 is a global maximum. F ′′ (X) < 0 for all X ⇒ F (X) is strictly concave.58 F F T T F (X) is strictly convex EX F (X) is convex (not strictly) H EX Remark 1: A linear function is both concave and convex since F θ ≡ F (X θ ). If it is not a global maximum. Other examples are concave F (X) = 2X X≤1 1+X X >1 convex F (X) = X X ≤1 2X − 1 X > 1 In the following theorems. 2 Theorem 2: If F (X) is concave and F ′ (X 0 ) = 0. which implies that F (X θ ) > F (X 0 ) for θ close to 0. X 0 must be a local maximum. X 1 ] such that 1 ¯ F (X 1 ) = F (X θ ) + F ′ (X θ )(X 1 − X θ ) + F ′′ (X 1 )(X 1 − X θ )2 2 1 ¯ F (X 0 ) = F (X θ ) + F ′ (X θ )(X 0 − X θ ) + F ′′ (X 0 )(X 0 − X θ )2 2 1 ¯ ¯ ⇒ F θ = F (X θ ) + θ(1 − θ)(X 1 − X 0 )2 [F ′′ (X 0 ) + F ′′ (X 1 )]. Proof: By theorem 1.
then X ∗ is a global minimum. If F (X) is strictly convex and F ′ (X ∗ ) = 0. However. but it must be continuous on the interior points. a > 1 T F X a. X a is strictly concave if a < 1. (X − a)F ′ (X) = 0) becomes suﬃcient for global maximum. a < 0 EX EX EX F T F eaX T F T −aX 2 + bX + c EX ln X EX EX Remark 4: A concave function does not have to be diﬀerentiable. then X ∗ is a unique global minimum. 0 < a < 1 T X a. x→a x→a h(x) x→a h(x) − h(a) ∆x→0 h(a + ∆x) − h(a) h (a) The same procedure also works for the case with g(a) = h(a) = ∞. If F (X) is convex and F ′ (X ∗ ) = 0. Remark 3: The sum of two concave functions is concave. we replace concavity with convexity and F ′′ (X) < 0 with F ′′ (X) > 0. The product of two concave function is not necesarily concave. h(x) x .59 XF ′ (X) = 0 (or F ′ (X) ≥ 0. 7. g(x) ln[(ax + 1)/2] Example 1: f (x) = = . Remark 2 (minimization problem): For the minimization problem.16 Indeterminate forms and L’Hˆpital’s rule o g(x) . strictly convex if a > 1. limx→a f (x) can be calculated. 0 Let f (x) = g(x) g(x) − g(a) g(a + ∆x) − g(a) g ′ (a) lim f (x) = lim = lim = lim = ′ . F T F X a. g(a) = h(a) = 0 and g(x) and h(x) be continuous at x = a. eX is strictly convex and ln X is strctly concave with X > 0. f (a) is h(x) 0 not deﬁned because it is .
ax + 1 ln a ln a . x1 = 1 − + 1 = ≈ 1. xn+1 = xn − = xn − + 2 .47. h′ (x) = 1. h(x) e 1 ⇒ h′ (∞) = ∞ and g ′ (∞) = 1.44225. 2 3x 3 x 1 5 5 5 9 331 Starting x0 = 1.60 g(0) = h(0) = 0. . h′ (x) = ex . two or three steps would be good enough. g ′ (x) = . ∞ g(x) ln x ′ 1 Example 3: f (x) = = . g ′(x) = ⇒ h′ (0) = 1 and g ′ (0) = (ln a)ax . h(x) x x ⇒ limx→0+ h′ (x) = 1 and limx→0+ g ′(x) = ∞.17 Newton’s method 1 = 0. f ′ (xn ) f (xn ) Recursive formula: xn+1 = xn − Ex xn xn+1 x∗ If f (x) is not too strange. ⇒ limx→0 f (x) = .666. h (x) = 1. 2 2 x g(x) Example 2: f (x) = = x . ⇒ limx→0+ f (x) = 7. x3 − 3 x 1 Example: f (x) = x3 − 3. ⇒ limx→∞ f (x) = = 0. g ′ (x) = 1. Usually. The 3 3 3 9 25 225 √ true value is x∗ = 3 3 ≈ 1. ∞ We can approximate a root x∗ of a nonlinear equation f (x) = 0 using an algorithm called Newton’s method. T f ′ (xn ) = f (xn ) xn − xn+1 f (xn ) . x2 = − + = ≈ 1. we will have limn→∞ xn = x∗ . f ′ (x) = 3x2 .
. x2 ) = −x2 + 4x1 x2 − x2 . xc ): = 0 ∀i. xn ) = F (X).xn Suppose that the objective function has n variables: max F (x1 . F2 = 4x1 − 2x2 = 0.. x∗ ): ∃ǫ > 0 such that F (X ∗ ) ≥ F (X) for all X 1 n satisfying xi ∈ (x∗ − ǫ. . The procedure is the same as that of the single variable case. . . i = 1. . or show that F (X) is concave so that a critical point must be a global maximum.61 8 Optimization–Multivariate Case x1 . A local maximum X ∗ = (x∗ . A unique global maximum X ∗ : F (X ∗ ) > F (X) ∀X = X ∗ . we have a reason to suspect that the SOC is not suﬃcient. n.1 SOC SOC of variablewise maximization is wrong Example: max F (x1 . $$$ $$$ $ $ £ $$ E x1 $$ $$ £ $$ 1 $$ 1 £ $$ 4 £ 4 £ £ £ £ £ £ £ 1 £ E xi . . If we regard variables other than xi as ﬁxed. .. xi = 0 is a maximum. 1 n ∂xi A global maximum X ∗ : F (X ∗ ) ≥ F (X) ∀X.x2 FOC: F1 = −2x1 + 4x2 = 0. x2 T (0.. ⇒ x1 = x2 = 0. . .0) is a saddle point.. since there are n×n second order derivatives (the Hessian matrix H(F )) and the SOC above does not consider the crossderivatives. therefore. . . 8. . x∗ + ǫ) ∀i. 4 £ £ £ £ £ £ £ £ F T 4 1 When xj = 0. . we have the necessary conditions FOC: ∂F = 0 and ∂xi SOC: ∂2F < 0. . ∂x2 i However. . 1 2 x1 . i i ∂F (X c ) c A critical point X = (xc . then it is a single variable maximization problem and. (2) use SOC to check whether a critical point is a local maximum. (1) Use FOC to ﬁnd critical points. In the next section we will provide a counterexample and give a true SOC. SOC? F11 = F22 = −2 < 0. .
The solution depends on x2 . 2 In the next section. The FOC is M ′ (x2 ) = F1 h′ + F2 = F2 = 0 (since F1 = 0). we can assume that aij = aji so that A is symmetrical. 0) = 0 < F (k.2 Quadratic forms and their signs Since vi vj = vj vi . F11 (X ∗ ) F12 (X ∗ ) F21 (X ∗ ) F22 (X ∗ ) x1 − x∗ 1 x2 − x∗ 2 1 ≡ v ′ H ∗ v. . x1 = h(x2 ). SOC of sequential maximization: We can solve the maximization problem sequentially. . x2 − x∗ ) 1 2 2 F11 (X ∗ ) F12 (X ∗ ) F21 (X ∗ ) F22 (X ∗ ) x1 − x∗ 1 x2 − x∗ 2 + higher order terms.. the true SOC is: F11 < 0 and F11 F22 − F12 > 0. f (v1 . . we can ﬁnd a x1 such that F (x1 . F1 (X ∗ ) = F2 (X ∗ ) = 0 and we have the approximation for X close to X ∗ : 1 F (X) − F (X ∗ ) ≈ (x1 − x∗ . The SOC is 2 M ′′ (x2 ) = F1 h′′ + F11 (h′ )2 + 2F12 h′ + F22 = (−F12 + F11 F22 )/F11 < 0. we will derive a systematic method to test the true SOC. x2 − x∗ ) 1 2 2 True SOC: v ′ H ∗ v < 0 for all v = 0. x2 ). For the nvariable case. . . . That is. 2 Therefore. . . 0) is not a local maximum! The true SOC should take into consideration the possibility that when x1 and x2 increase simultaneously F may increase even if individual changes cannot increase F . i. Since X ∗ is a critical point. vn ) = aij vi vj = (v1 . . . 1 −1 v1 2 2 Example 1: (v1 − v2 )2 = v1 − 2v1 v2 + v2 = (v1 . we regard x1 as endogenous variable and x2 as an exogenous variable in the ﬁrst stage. dx1 /dx2 = h′ (x2 ) = −F12 /F11 . ≡ v ′ Av. Using implicit function rule. x2 ) is maximized. we can also use Taylor’s theorem to expand F (X) around a critical point X ∗ : F (X) = F (X ∗ ) + F1 (X ∗ )(x1 − x∗ ) + F2 (X ∗ )(x2 − x∗ ) 1 2 1 + (x1 − x∗ . In the second stage we maximize M(x2 ) ≡ F (h(x2 ). a11 · · · a1n v1 n n . vn ) . k) = 2k 2 for all k = 0. .e. . −1 1 v2 A quadratic form in n variables is a second degree homogenous function. . . . v2 ) . 8. . for a given x2 . i=1 j=1 an1 · · · ann vn . ⇒ (0. SOC using Taylor’s expansion: To ﬁnd the true SOC. . the sequential argument is more complicated and we use Taylor’s expansion. The FOC and SOC are F1 = 0 and F11 < 0. .62 F (0.
2 2 Example 1: F (v1 . a11 = . (v1 + v2 )2 ≥ 0. x2 . The matrix is 4 −2 the Hessian of F of the counterexample in section 1. −2 < 0 but −2 4 4 −2 v1 v2 . Therefore. negative deﬁnite if v ′ Av < 0 for all v = 0. The proofs for them are more diﬃcult and discussed in Lecture 13. positive deﬁnite if v ′ Av > 0 for all v = 0. v2 ) = a11 v1 + 2a12 v1 v2 + a22 v2 = a21 a22 v2 a11 a2 2 a2 a12 a12 2 2 v2 v1 + 2 v1 v2 + 12 v2 + − 12 + a22 v2 = a11 v1 + a11 a2 a11 a11 11 2 a11 a12 a21 a22 2 + v2 . 12 a22 For negative or positive semideﬁnite. = 0 if v1 = −v2 ⇒ positive semideﬁnite. 2 2 v1 + v2 > 0. < = Notice that if v ′ Av is negative (positive) deﬁnite then it must be negative (positive) semideﬁnite. if v1 = 0 or v2 = 0 ⇒ positive. 12 a22 a12 = a11 a22 − a2 > 0. v2 ) −2 4 = −12 < 0 ⇒ neither positive nor negative.63 0 1 1 0 v1 . = 0 if v1 = v2 ⇒ negative semideﬁnite. 2 2 −v1 − 2v2 < 0. x3 ) 0 2 0 x2 . < < 2 > = (6) v1 v2 0 if sign(v1 ) sign(v2 ) ⇒ neither positive nor negative deﬁnite. v2 Example 2: 2v1 v2 = (v1 . positive semideﬁnite if v ′ Av ≥ 0 for all v ∈ Rn . −(v1 − v2 )2 ≤ 0. if v1 = 0 or v2 = 0 ⇒ negative. Testing the sign of a quadratic form: a11 a12 v1 2 2 n = 2: v ′ Av = (v1 . replace strict inequalities with semiinequalities. v2 ) v ′ Av v ′ Av v ′ Av v ′ Av (1) (2) (3) (4) 1 0 3 x1 Example 3: x2 + 2x2 + 6x1 x3 = (x1 . the counterexample violates the SOC. v2 ) = −2v1 + 8v1 v2 − 2v2 = (v1 . a11 Negative deﬁnite ⇔ a11 < 0 and Positive deﬁnite ⇔ a11 > 0 and a11 a21 a11 a21 a12 = a11 a22 − a2 > 0. 1 2 3 0 0 x3 is is is is called called called called negative semideﬁnite if v ′ Av ≤ 0 for all v ∈ Rn . > 2 2 > (5) v1 − v2 0 if v1  v  ⇒ neither positive nor negative deﬁnite.
A  A  A  a11 a12 Negative deﬁnite ⇔ A(1)  = a11 < 0. (n−1) > 0. (−1) A  = (−1) > 0. . (2) < 0. . . .) . . = a11 (v1 + · · ·)2 . (A cubic function in 2 vari2 ables. . . n. A  = > 0. a11 · · · a1n . (1) > 0. . . ak1 · · · akk (2) A  A(3)  A(n)  Negative deﬁnite ⇔ A(1) = a11 < 0.. . . . . General n: v ′ Av = (v1 . a11 · · · a1k . a21 a22 a23 > 0. .. . a31 a32 a33 an1 · · · ann a11 a12 Positive deﬁnite ⇔ A(1)  = a11 > 0. . . . A  A  A  (2) (3) (n) A  A  A  Positive deﬁnite ⇔ A(1) = a11 > 0. · · · a(n−1)(n−1) 2 vn . A(3)  = a21 a22 a11 · · · a1n a11 a12 a13 . ··· . .64 a11 · · · a1n . . . .. . F31 (X ∗ ) F32 (X ∗ ) F33 (X ∗ ) 2 Example: max F (x1 . · · ·. . . (n) . . . (1) < 0. an1 · · · ann a(n−1)1 ··· a1(n−1) .. (2) > 0. . a31 a32 a33 an1 · · · ann The conditions for semideﬁnite are more complicated and will be discussed in Lecture 13 using the concept of eigenvalues of a square matrix.3 SOC again From the last two sections. (n−1) < 0. . . . F21 (X ∗ ) F22 (X ∗ ) F23 (X ∗ ) < 0. x2 ) = 3x1 + 3x1 x2 − 3x1 − x3 . k = 1. . .. . . a21 a22 a23 < 0. . . = F21 (X ∗ ) F22 (X ∗ ) F11 (X ∗ ) F12 (X ∗ ) F13 (X ∗ ) 2 F11 F22 − F12 > 0. . vn ) . kth order principle minor of A: A(k) ≡ . .. n (n) n . . . . an1 · · · ann a11 a12 a21 a22 + a11 v1 . · · ·. . 8. the SOC for a local maximum can be summarized as: F11 (X ∗ ) F12 (X ∗ ) SOC: v ′ HF (X ∗ )v is negative deﬁnite ⇒ F11 (X ∗ ) < 0. A(2)  = > 0. . . A(3)  = a21 a22 a11 · · · a1n a11 a12 a13 . . vn a11 a12 a13 a21 a22 a23 a31 a32 a33 (v3 +· · ·)2 +· · ·+ (v2 +· · ·)2 + a11 a12 a21 a22 a11 . . . . . A(2)  = > 0. .. . .
q2 . Proﬁt function: Π(q1 . (eg. p2 + q1 − 6q2 = 0 ⇒ supply functions q1 = (6p1 + p2 )/23 and q2 = (p1 + 4p2 )/23. Π2 = p2 − C2 = 0. q2 ) = 2q1 + 3q2 − q1 q2 FOC: p1 − 4q1 + q2 = 0. 2 1 x1 1 Two critical points: = 41 and . Although we can still check whether v ′ Hv is negative semideﬁnite. F (1. 8. q2 ). It is not a global maximum 1 because F →∞ when x2 → − ∞. H 2(X) = = 36x2 − 9.65 FOC: F1 = 3 + 3x2 − 6x1 = 0 and F2 = 3x1 − 3x2 = 0. −C11 −C12 SOC: Π11 = −C11 < 0. gasoline and its by products or cars and tructs. x2 1 − 2 F11 (X) F12 (X) −6 3 Hessian matrix: H(x1 . p1 . or pi = MCi . p2 ) = p1 q1 + p2 q2 − C(q1 . H 1 (X) = F21 (X) F22 (X) 3 −6x2 −6 3 F11 (X) = −6 < 0. it is insuﬃcient for a local maximum. − ) = −27 < 0 ⇒ 1 is not a local max. −C11 −C12 −4 1 SOC: −C11 = −4 < 0. 1) = 2. Remark 2: If H k (X ∗ ) = 0 for some 1 ≤ k ≤ n then X ∗ is a degenerate critical point..) Cost function: C(q1 . ≡ ∆ > 0. 4 2 − 2 1 H 2 (1. 3 −6x2 1 1 1 4 H 2 ( . = 23 > 0. It is much more diﬃcult than the single variable case with F ′′ = 0.4 Joint products 2 2 Example: C(q1 . . = −C21 −C22 1 −6 A competitive producer produces two joint products. 1) = 27 > 0 ⇒ is a local max. etc. x2 ) = = . 2 Remark 1: F11 < 0 and F11 F22 − F12 > 0 together implies F22 < 0. We have to check the third or higher order derivatives to determine whether X ∗ is a local maximum. −C21 −C22 Comparative statics: Total diﬀerential of FOC: C11 C12 C21 C22 dq1 dq2 = dp1 dp2 ∂q1 ⇒ ∂p1 ∂q2 ∂p1 ∂q1 ∂p2 = 1 ∂q2 ∆ ∂p2 C22 −C12 −C21 C11 . FOC: Π1 = p1 − C1 = 0. q2 ).
66 SOC ⇒ C11 > 0, C22 > 0, ∆ > 0 ⇒ ∂q2 ∂q1 > 0, > 0. ∂p1 ∂p2
∂q1 ∂q2 and are positive if the joint products are beneﬁcially to each other in the ∂p2 ∂p1 production so that C12 < 0. 8.5 Monopoly price discrimination
Example: f1 = a − bq1 , f2 = α − βq2 , and C(Q) = 0.5Q2 = 0.5(q1 + q2 )2 . ′ ′ ′′ ′′ f1 = −b, f2 = −β, f1 = f2 = 0, C ′ = Q = q1 + q2 , and C ′′ = 1. 1 + 2b 1 q1 a FOC: a − 2bq1 = q1 + q2 = α − 2βq2 ⇒ = 1 1 + 2β q2 α 1 q1 a(1 + 2β) − α ⇒ = . q2 (1 + 2b)(1 + 2β) − 1 α(1 + 2b) − a SOC: −2b − 1 < 0 and ∆ = (1 + 2b)(1 + 2β) − 1 > 0. p
T
A monopoly sells its product in two separable markets. Cost function: C(Q) = C(q1 + q2 ) Inverse market demands: p1 = f1 (q1 ) and p2 = f2 (q2 ) Proﬁt function: Π(q1 , q2 ) = p1 q1 + p2 q2 − C(q1 + q2 ) = q1 f1 (q1 ) + q2 f2 (q2 ) − C(q1 + q2 ) ′ ′ FOC: Π1 = f1 (q1 ) + q1 f1 (q1 ) −C ′ (q1 + q2 ) = 0, Π2 = f2 (q2 ) + q2 f2 (q2 ) −C ′ (q1 + q2 ) = 0; or MR1 = MR2 = MC. ′ ′′ 2f1 + q1 f1 − C ′′ −C ′′ ′ ′′ SOC: Π11 = 2f1 + q1 f1 − C ′′ < 0, ≡ ∆ > 0. ′′ ′ ′′ −C 2f2 + q2 f2 − C ′′
MC∗
d d d d d d MR1+2 d d MR d 2 d MR1
∗ q1 ∗ q2
MC
MC = MR1+2 ⇒ Q∗ , MC∗
∗ MC∗ = MR1 ⇒ q1 ∗ MC∗ = MR2 ⇒ q2
Q
∗
Eq
8.6
SR supply vs LR supply  Le Chˆtelier principle a
A competitive producer employs a variable input x1 and a ﬁxed input x2 . Assume that the input prices are both equal to 1, w1 = w2 = 1. Production function: q = f (x1 , x2 ), assume MPi = fi > 0, fii < 0, fij > 0,
67
2 f11 f22 > f12 . Proﬁt function: Π(x1 , x2 ; p) = pf (x1 , x2 ) − x1 − x2 .
Shortrun problem (only x1 can be adjusted, x2 is ﬁxed): SR FOC: Π1 = pf1 − 1 = 0, or w1 = VMP1 . SOC: Π11 = pf11 < 0, 2 dx1 −f1 dq s −f1 −1 Comparative statics: f1 dp+pf11 dx1 = 0 ⇒ = > 0, = = 3 > 0. dp pf11 dp pf11 p f11 Longrun problem (both x1 and x2 can be adjusted): LR FOC: Π1 = pf1 − 1 = 0, Π2 = pf2 − 1 = 0; or wi = VMPi . pf11 pf12 SOC: Π11 = pf11 < 0, ≡ ∆ > 0. pf21 pf22 f1 dp pf11 pf12 dx1 0 Comparative statics: + = f2 dp pf21 pf22 dx2 0 −1 dq L −(f11 + f22 − 2f12 ) dx1 /dp f1 f22 − f2 f12 ⇒ = , = 2 2 dx2 /dp f2 f11 − f1 f21 p(f11 f22 − f12 ) dp p3 (f11 f22 − f12 ) dq s dq L > . Le Chˆtelier principle: a dp dp Example: f (x1 , x2 ) = 3x1 x2 (homogenous of degree 2/3) −2/3 1/3 1/3 −2/3 LR FOC: px1 x2 = 1 = px1 x2 ⇒ x1 = x2 = p3 q L = 3p2 . −2/3 SR FOC (assuming x2 = 1): px1 ¯ = 1 ⇒ x1 = p3/2 q s = 3p1/2 . L s η (LR supply elasticity) = 2 > η (SR supply elasticity) = 1/2. L p π ¨ π (p) s 0.5 T T ¨ S (p) = 3p ¨
¨ ¨¨ ¨¨ ¨¨
1/3 1/3
π s (p; x2 ) ¯ h(p)
S L (p) = 3p2
Eq Ep
p ¯
Envelop theorem, Hotelling’s lemma, and Le Chˆtelier principle a From the SR problem we ﬁrst derive the SR variable input demand function x1 = x1 (p, x2 ). ¯ Then the SR supply function is obtained by substituting into the production function q s = f (x1 (p, x2 ), x2 ) ≡ S s (p; x2 ). ¯ ¯ ¯ The SR proﬁt function is π s (p, x2 ) = pq s − x1 (p, x2 ) − x2 . ¯ ¯ ¯ ∂π s Hotelling’s lemma: By envelop theorem, = S s (p, x2 ). ¯ ∂p From the LR problem we ﬁrst derive the input demand functions x1 = x1 (p) and x2 = x2 (p). Then the LR supply function is obtained by substituting into the production function
68 q L = f (x1 (p), x2 (p)) ≡ S L (p). The LR proﬁt function is π L (p) = pq L − x1 (p) − x2 (p). ∂π L (Also Hotelling’s lemma) By envelop theorem, = S L (p). ∂p Notice that π L (p) = π s (p; x2 (p)). Let x2 = x2 (¯), deﬁne h(p) ≡ π L (p) − π s (p, x2 ). h(p) ≥ 0 because in the LR, ¯ p ¯ the producer can adjust x2 to achieve a higher proﬁt level. Also, h(¯) = 0 because π L (¯) = π s (¯; x2 (¯)) = π s (¯; x2 ). p p p p p ¯ ∂2 πL ∂2πs ′′ Therefore, h(p) has a minimum at p = p and the SOC is h (¯) > 0 = ¯ p − 2 > 0, ∂p2 ∂p which implies dq L dq s Le Chˆtelier principle: a − > 0. dp dp 8.7 Concavity and Convexity
Similar to the single variable case, we deﬁne the concepts of concavity and convexity for 2variable functions F (X) = F (x1 , x2 ) by deﬁning G+ , G− ⊂ R2 s follows. F F G+ ≡ {(x1 , x2 , y) y ≥ F (x1 , x2 ), (x1 , x2 ) ∈ R2 }, F G− ≡ {(x1 , x2 , y) y ≤ F (x1 , x2 ), (x1 , x2 ) ∈ R2 }. F If G+ (G− ) is a convex set, then we say F (X) is a convex function (a concave funcF F tion). If F (X) is deﬁned only for nonnegative values x1 , x2 ≥ 0, the deﬁnition is similar. (The extension to nvariable functions is similar.) Equivalently, given X 0 = F 1 ≡ F (X 1 ), we deﬁne x0 1 x0 2 , X1 = x1 1 x1 2 ≡ , 0 ≤ θ ≤ 1, F 0 ≡ F (X 0 ), xθ 1 xθ 2 , F (X θ ) = F ((1 − θ)X 0 +
X θ ≡ (1 − θ)X 0 + θX 1 =
θX 1 ) F θ ≡ (1 − θ)F (X 0 ) + θF (X 1 ) = (1 − θ)F 0 + θF 1 . x2
T
(1 − θ)x0 + θx1 1 1 (1 − θ)x0 + θx1 2 2
x0 2 xθ 2 x1 2
d d Xθ d d d d X1
X0
X θ is located on the straight line connecting X 0 and X 1 , when θ shifts from 0 to 1,
E x1
x0 1
xθ 1
x1 1
X θ shifts from X 0 to X 1 .
On 3dimensional (x1 –x2 –F ) space, (X θ , F θ ) is located on the straight line connecting (X 0 , F 0 ) and (X 1 , F 1), when θ shifts from 0 to 1, (X θ , F θ ) shifts from (X 0 , F 0 ) to (X 1 , F 1 ). On the other hand, (X θ , F (X θ )) shifts along the surface representing the graph of F (X).
X 1 and θ ∈ [0. k is B’s knowledge regard2 ing the consumption of X. concave ⇒ if for all X 0 . 1). convex ⇒ if for all X 0 . If it is not a global maximum. there exist X 0 ∈ [X 0 . F (X θ ) ≥ F θ . v ′ HF v is negative deﬁnite for all X ⇒ F (X) is strictly concave. Remark 1 (boundary or corner solution): The boundary or corner condition Fi (X) ≤ 0. X 1 ] such that 1 ¯ F (X 1) = F (X θ ) + ∇F (X θ )(X 1 − X θ ) + (X 1 − X θ )′ HF (X 1 )(X 1 − X θ ) 2 1 ¯ F (X 0) = F (X θ ) + ∇F (X θ )(X 0 − X θ ) + (X 0 − X θ )′ HF (X 0 )(X 0 − X θ ) 2 θ(1 − θ) 1 ¯ ¯ ⇒ F θ = F (X θ ) + (X − X 0 )′ [HF (X 0 ) + HF (X 1 )](X 1 − X 0 ). 1].8 Learning and utility maximization Consumer B’s utility function is U = u(x. Assume that F is twice diﬀerentiable. 1). Therefore. h′ > 0. we replace concavity with convexity and negative deﬁnite with positive deﬁnite. Assume that B has 100 dollar to spend and the price of X is Px = 1 so that m = 100 − x and U = u(x. m is money. xi Fi (X) = 0 (or Fi (X) ≥ 0. and h(k) is the disutility of acquir∂x∂k ing/maintaining k. then X ∗ is a unique global minimum. X 0 is not a local maximum. F (X θ ) > F θ . F (X θ ) < F θ . k) is the utility obtained. If F (X) is strictly concave and ∇F (X 0 ) = 0. Assume . 1/3 1/3 Example: 9x1 x2 . then there exists X 1 such that F (X 1 ) > F (X 0 ). (xi − ai )Fi (X) = 0) becomes suﬃcient for global maximum. x. Theorem 1: F (X) is concave. a contradiction. ∂ u < 0. where x is the quantity of commodity X consumed. u(x. X 1 and θ ∈ (0. which implies that F (X θ ) > F (X 0 ) for θ close to 0. h′′ > 0. X 0 must be a local maximum. ¯ ¯ Proof: By Taylor’s theorem. m. X θ ] and X 1 ∈ [X θ . If F (X) is strictly convex and ∇F (X ∗ ) = 0. strictly convex ⇒ if for all X 0 . then X 0 is a unique global maximum.69 F (X) F (X) F (X) F (X) is is is is strictly concave ⇒ if for all X 0 . If F (X) is convex and ∇F (X ∗ ) = 0. 2 Theorem 2: If F (X) is concave and ∇F (X 0 ) = 0. X 1 and θ ∈ (0. Proof: By theorem 1. then X 0 is a global maximum. 8. k) + 100 − x − h(k). then X ∗ is a global minimum. 1]. k) + m − h(k). F (X θ ) ≤ F θ . X 1 and θ ∈ [0. ⇔ v ′ HF v is negative semideﬁnite for all X. Remark 2 (minimization problem): For the minimization problem. ∂x2 ∂2u > 0 (marginal utility of X increases with k). k ≥ 0.
k) + 100 − x(k) − h(k). x = 32 3.70 also that k is ﬁxed in the short run. Solve the maximization problem maxk V (k). 2. k) ≡ u(x.) xL (p) = 32p−5 (100/p) if p > (8/25)1/4 (p < (8/25)1/4). kL (p) = [xL (p)]2/5 . k) + 100 − x − h(k). xs = x(k). k) + 100 − x − h(k). Calculate consumer B’s short run consumption of X.) √ √ xs (p) = 8kp−3 (100/p) if p ≤ 2k/5 (p < 2k/5). (Warning: There is a nonnegative constraint m = 100 − px ≥ 0 and you have to consider both interior and corner cases. x. (In this part. ⇒ FOC: − 1 = 0. ∂x ∂u − h′ (k) = 0. you may ignore the nonnegative constraint m = 100 − x ≥ 0 and the possibility of a corner solution. x ∂x ∂x2 The short run comparative static dx is derived from FOC as dk ∂ 2 u/∂x2 dx =− 2 > 0. 4. 2. consumer B will consume more of X if B’s knowledge of X increases. k) = 3x2/3 k 1/3 and h(k) = 0. Derive the short run value function V (k) ≡ u(x(k). xs = x(p. k) is concave and h(k) is convex (h′′ (k) > 0). Calculate the long run consumption of X. The short run utility maximization problem is ∂u ∂2u max u(x. xL (p). . In the long run consumer B will change k to attain higher utility level. because F (x. Explain how the SOC is satisﬁed and why the solution is the unique global maximum.5k 2 . kL (p). SOC: < 0. (demand functions) Consider now the general case when Px = p so that U = 3x2/3 k 1/3 + 100 − px − 0. The long run utility maximization problem is max u(x. Calculate the long run demand function of X. Consider now the speciﬁc case when u(x. and the optimal level of K. k = 4. k). k) + 100 − x − h(k).) x = 1/(8k). x. xL . Calculate consumer B’s short run demand function of X.5k 2 . ∂k The SOC is satisﬁed if we assume that u(x. 5. 1. 1. dk ∂ u/∂x∂k that is. (Both interior and corner cases should be considered too.k ⇒ FOC: ∂u − 1 = 0. k ≥ 0 is strictly concave then.
. x2 ) is homogeneous of degree 0. . . Let z = x1 x2 + x2 + 3x2 + x2 x3 + x2 . H is a monotonic increasing function. . (c) lim . 1 2 3 (a) Use matrix multiplication to represent z. Example: Q = α ln x1 + β ln x2 = ln xα xβ 1 2 is homothetic. X→0 X→0 X→0 X X X 3. then g is called a homothetic function. x1 x1 Euler theorem: If f (x1 . . If x1 x1 x2 x2 m = H(x1 . + xn fn = kf If f is homogeneous of degree k. x2 x2 If Q = F (x1 . . bx2 } is homogeneous of degree 1. Find (a) lim . 4. CobbDouglas function Q = Axα xβ is homogeneous of degree α + β. and f m is homogeneous g of degree mk1 . then x1 f1 + . 1 2 3. . (b) lim+ . use L’hˆpital’s rule to ﬁnd the o + AVC at Q = 0 . .71 8. x2 .9 Homogeneous and homothetic functions A homogeneous function of degree k is f (x1 . Given the total cost function C(Q) = eaQ+b . . xn ) is homogeneous of degree k. . Homothetic functions: If f is homogeneous and g = H(f ). then Q = x1 F (1. . . Leontief function Q = min {ax1 . . . Check whether it is a maximum or a minimum. hxn ) = hk f (x1 . (c) Find the extreme value of z. (b) Determine whether z is positive deﬁnite or negative deﬁnite. ) ≡ h . CES function Q = {axρ + bxρ } ρ is homogenous of degree k. then m = H(1. . A quadratic form x′ Ax is homogeneous of degree 2.10 Problems 1. If f is homogeneous of degree k1 and g homogeneous of degree k2 . then fi is homogenous of degree k − 1. 1 2 k 2. is homogeneous of degree k1 − k2 . . 4. xn ) such that f (hx1 . x1 The MRS of a homothetic function depends only on the ratio 8. xn ) ∀h > 0. . Examples: 1. then f g is homogef neous of degree k1 + k2 . Use Newton’s method to ﬁnd the root of the nonlinear equation X 3 +2X +2 = 0 accurate to 2 digits. x2 ) is homogeneous of degree 1. . . 1 − e−aX e2X − eX 1 − 2−X 2. ) ≡ x1 f .
the plant size (a ﬁxed factor in the short run). (c) Calculate the SR maximizing proﬁt π(K). (a) At which output level. x2 ) − w1 x1 − x2 . P1 = −(1/a) ln Q1 and P2 = −(1/b) ln Q2 . The proﬁt function of the ¯ competitive ﬁrm is given by π = f (x1 . (Hints: 1. or neither.) 10. 8. say. the amount of the ﬁxed input is given by x2 = x2 . The cost function of a competitive producer is C(Q. convex. Assume that the monopoly sells its goods in two diﬀerent markets which are separated by some distance. x2 ) = 1 1 3 3 3x1 x2 . where x1 is a variable input and x2 is a ﬁxed input. Check whether the function f (x. what level of output should be produced in each market and what price will prevail in each market? Check that your answer is the unique global maximum. The production function of a competitive ﬁrm is given by q = f (x1 . K) is concave so that the solution is the unique global maximum. 7. A competitive ﬁrm produces two joint products. (c) Is the critical point a local maximum. the SAC curve attains a minimum? (b) Suppose that the equilibrium price is p = 100. The total cost function is 2 2 C(q1 . For a given K. ¯ ¯ . a global maximum. (f) Now solve the 2variable maximization problem Q3 max Π(Q. (a) Use the ﬁrst order conditions for proﬁt maximization to derive the supply functions. Assume that the prices of the output and the ﬁxed input are p = w2 = 1. K) = is. where K 3K 5. 2. The proﬁt is Π = 100Q − Q3 − K. (e) Calculate the LR supply quantity Q∗ = Q(K ∗ ). In the short run. or neither.) (b) Find the critical point of f . (a) Check whether f (x. Suppose that a monopoly can produce any level of output at a constant marginal cost of $c per unit.72 Q3 + K.K≥0 3K and show that Π(Q. q2 ) = 2q1 + 3q2 − 4q1 q2 . y) = ex+y is concave. The demand curve in the ﬁrst market is given by Q1 = exp[−aP1 ] and the curve in the second market is given by Q2 = exp[−bP2 ]. or neither? 9. (d) Find the LR optimal K = K ∗ to maximize π(K). 6. convex. Q. y) = 2 ln x+3 ln y −x−2y is concave. Π = P1 Q1 + P2 Q2 − (Q1 + Q2 )c = −(1/a)Q1 ln Q1 − (1/b)Q2 ln Q2 − (Q1 + Q2 )c is strictly concave. (b) Check that the second order conditions are satisﬁed. (Assume that x > 0 and y > 0. K) = pQ − C(Q) = 100Q − − K. ﬁnd the supply quantity Q(K) such that SR 3K proﬁt is maximized. If the monopolist wants to maximize its total proﬁts.
(d) Show that the LR proﬁt is a strictly concave function of (x1 . Let U(x. x. (b) For what value(s) of a U(x. a > 0. 1 1 1 (c) Verify the Le Chˆtelier principle: a L w1 ∂x1 xL ∂w1 1 > S w1 ∂x1 xS ∂w1 1 . x2 ) for x1 . (a) For what value(s) of a U(x. State the FOC for LR proﬁt maximization and calculate the LR input demand funtion L w1 ∂x xL = xL (w1 ) and the LR input demand elasticity xL ∂w11 . 1 x 1 (b) Now consider the LR situation when x2 can be adjusted. y) = xa y + xy 2 . 11. y. x2 > 0 and therefore the solution must be the unique global maximum. y) is homothetic? . y) is homogeneous?.73 (a) State the FOC for the SR proﬁt maximization and calculate the SR input S S 1 ∂x demand function xS = x1 (w1 ) and the SR input demand elasticity wS ∂w11 .
U2 = p2 λ. we have to know what is the FOC ﬁrst. x1 . . or show that U(X) is quasiconcave so that a critical point must be a global maximum. . A unique global maximum X ∗ : F (X ∗ ) > F (X) ∀X ∈ B. (1) Use FOC to ﬁnd critical points.74 9 Optimization and Equality Constraints and Nonlinear Programming In some maximization problems. x∗ ): ∃ǫ > 0 such that U(X ∗ ) ≥ U(X) for all X ∈ B 1 n satisfying xi ∈ (x∗ − ǫ. . xn ) that satisfy certain equalities. dx2 1 d2 U p1 = U11 − 2 U12 + 2 dx1 p2 By FOC. .. x1 . (2) use SOC to check whether a critical point is a local maximum. m − p1 x1 p2 . SOC: d2 U < 0. the agents can choose only values of (x1 . . SOC: Alternatively. FOC: dU p1 = U1 + U2 − dx1 p2 p1 p2 2 = 0. xn ) subject to p1 x1 + · · · + pn xn = m. To deﬁne the concept of a critical point.1 FOC and SOC for a constraint maximization Consider the 2variable utility maximization problem: max U(x1 . . For example. x2 ) + λ(m − p1 x1 − p2 x2 ) . p1 p2 U22 = −1 p2 2 0 −p1 −p2 −p1 U11 U12 . i i A critical point: A X ∗ ∈ B satisfying the FOC for a local maximum. X = X ∗ . we can deﬁne Lagrangian: 0 U1 U2 U1 U11 U12 U2 U21 U22 > 0. . The procedure is the same as before. and it p2 dx1 p2 becomes a single variable maximization problem: Using the budget constraint. x2 = max U x1 x1 . x∗ + ǫ) ∀i. .x2 m − p1 x1 dx2 p1 = h(x1 ). = − = h′ (x1 ). . . A global maximum X ∗ : F (X ∗ ) ≥ F (X) ∀X ∈ B. a consumer has a budget constraint p1 x1 + · · · + pn xn = m. .. A local maximum X ∗ = (x∗ .. −p2 U21 U22 FOC: U1 = p1 λ.xn max U(x) = U(x1 . Deﬁne B = {(x1 . L ≡ U(x1 . . . x2 ) subject to p1 x1 + p2 x2 = m. . xn ) such that p1 x1 + · · · + pn xn = m}. . 9.. U1 U2 = ≡ λ (MU of $1). .
. . xn ) = 0. and it becomes a single dx1 g2 dF d2 F = F1 + F2 h′ (x1 ) = 0. dx1 dx2 1 d2 F d 2 = (F1 + F2 h′ ) = F11 + 2h′ F12 + (h′ ) F22 + F2 h′′ 2 dx1 dx1 h′′ = ⇒ d2 F = dx2 1 0 d dx1 − g1 g2 = −1 g11 + 2g12 h′ + g22 (h′ )2 .xn 0 g1 g2 g1 L11 L12 g2 L21 L22 max F (x1 . . x1 . . h(x1 )) FOC: x1 dx2 g1 = − = h′ (x1 ). . variable maximization problem: max F (x1 . nvariable 1constraint case: x1 . x2 ) = ∂x1 ∂x2 ∂λ > 0. . xn ) − λg(x1 .. . . .. . . . g2 F2 F2 g12 h′ + F22 − g22 (h′ )2 g2 g2 0 g1 g2 g1 g2 F2 F2 F11 − g11 F12 − g12 g2 g2 F2 F2 F21 − g21 F22 − g22 g2 g2 F11 − F2 g11 g2 + 2 F12 − −h′ 1 F2 F2 −h′ F11 − g11 F12 − g12 −1 g2 g2 =− = 2 g2 F2 F2 1 F21 − g21 F22 − g22 g2 g2 F2 F1 By FOC.75 ∂L ∂L ∂L = U1 − λp1 = 0. Lλ = = −g(x1 .. . . x2 = h(x1 ). x2 ) = 0. . xn ) . 0 −p1 −p2 SOC: −p1 L11 L12 > 0. SOC: ∂L ∂L ∂L = F1 − λg1 = 0. L2 = = U2 − λp2 = 0. g1 g2 Alternatively. x2 ) − λg(x1 . SOC: < 0. L2 = = F2 − λg2 = 0.. = ≡ λ (Lagrange multiplier). x2 ) FOC: L1 = 0. xn ) subject to g(x1 . −p2 L21 L22 FOC: L1 = max F (x1 . we can deﬁne Lagrangian: L ≡ F (x1 . Lλ = = ∂x1 ∂x2 ∂λ m − p1 x1 − p2 x2 = 0. L ≡ F (x1 .x2 general 2variable with 1constraint case: Using the constraint. . x2 ) subject to g(x1 .
. i = 1. L2 = 1 − 2λx2 = 0 and Lλ = 1 − x2 − x2 = 0. ∂xi ∂λ 0 −g1 −g2 −g3 0 −g1 −g2 0 −g1 −g1 L11 L12 L13 < 0. . . Lλ = = −g(x1 . . x2 ) = −x1 − x2 subject to x1 + x2 = 1.76 FOC: Li = SOC: ∂L ∂L = Fi − λgi = 0. n. . 1 2 FOC: L1 = −2x1 − λ = 0. 1 2 L = x1 + x2 + λ(1 − x2 − x2 ). x2 ) = x1 + x2 subject to x2 + x2 = 1. x2 2 0 1 1 SOC: 1 −2 0 = 4 > 0. . etc. 1 0 −2 1/2 ⇒ is a local maximum. 2 = 2 − λ = 0 and Lλ = 1 − 2x1 − 2x2 = 0. 1 2 √ √ Two critical points: x1 = x2 = λ = 1/ 2 and x1 = x2 = λ = −1/ 2.2 Examples 2 Example 1: max F (x1 . 0 −2x1 −2x2 0 SOC: −2x1 −2λ = 8λ(x2 + x2 ). 1 2 FOC: L1 = 1 − 2λx1 = 0. . 1 2 −2x2 0 −2λ √ √ ⇒ x1 = x2 = λ = 1/ 2 is a local maximum and x1 = x2 = λ = −1/ 2 is a local minimum. λ = −1. 2 L = −x2 − x2 + λ(1 − x1 − x2 ). −g1 L11 −g2 L21 L22 L23 −g2 L21 L22 −g3 L31 L32 L33 9. xn ) = 0. . −g1 L11 L12 > 0 < 0. L −2x 1 x1 2 Critical point: = 1 . . 1/2 x2 T d d dr d d d x2 T E x1 d d d dr d d d d d d d d d d d d d d d d d d r d d d d d d d d E x1 Example 2: max F (x1 .
L2 = w2 − (1 − a)λxa x−a = 0. L2 = − λp2 = 0 and Lλ = m − p1 x1 − p2 x2 = 0. x2 . Lλ = q − xa x1−a = 0. x2 ) ∈ ¯ ¯ R2 U(x1 . (x1 x2 )2 (1 − a)w1 . x2 = q is a local minimum. x2 ) U1 is quasiconcave if MRS (= ) is decreasing along every indiﬀerence curve. x2 ) is quasiconcave: (1) the set {(x1 . x2 = b x1 p2 (a + b)p1 (a + b)p2 0 −p1 −p2 −a ap2 bp2 −p1 0 SOC: = 22 + 21 > 0. If U(x1 . x2 1 x1 x2 −b −p2 0 x2 2 am bm ⇒ x1 = . x2 = q . w1. 1 L = w1 x1 + w2 x2 + λ(q − xa x1−a ). Two ways to determine whether U(x1 . w2 (1 − a)x1 (1 − a)w1 aw2 SOC: 0 1−a −axa−1 x2 1 −(1 − a)xa x−a 1 2 1−a a −axa−1 x2 −(1 − a)x1 x−a 1 2 a−2 1−a a(1 − a)λx1 x2 −a(1 − a)λxa−1 x−a 1 2 a−1 −a −a(1 − a)λx1 x2 a(1 − a)λxa x−a−1 1 2 1−a a a(1 − a)q 3 λ =− < 0.4 Utility maximization and demand function aw2 ⇒ x1 = q (1 − a)w1 max U(x1 . A utility function U(x1 . 0 < a < 1. then a critical point must be a (unique) global maximum. x2 ) = a ln x1 + b ln x2 subject to p1 x1 + p2 x2 = m. FOC: L1 = x1 x2 a x2 p1 am bm ⇒ = ⇒ x1 = . w2 . aw2 1−a a a 1−a a 1−a The total cost is C(w1 . then the utility maximization has a unique solution. (Every indiﬀerence curve is convex toward the . (a + b)p1 (a + b)p2 9. the utility function is strictly quasiconcave. 1 2 1 2 1 2 1−a a ax2 aw2 (1 − a)w1 w1 ⇒ = ⇒ x1 = q . x2 ) ≥ U } is convex for all U . a b − λp1 = 0. if the MRS is strictly decreasing along an indiﬀerence curve (indiﬀerence curve is convex toward the origin). x2 = is a local maximum. 1−a a 9. w2 ) = w1 x1 + w2 x2 subject to xa x2 = q. q) = q + w1 w2 .5 Quasiconcavity and quasiconvexity As discussed in the intermediate microeconomics course.3 Cost minimization and cost function 1−a min C(x1 . In U2 case MRS is strictly decreasing. 1 2 FOC: L1 = w1 − aλxa−1 x1−a = 0. x2 ) is (strictly) quasiconcave. L = a ln x1 + b ln x2 + λ(m − p1 x1 − p2 x2 ).77 9.
F (X 1 )}. then G− (¯) must be convex for every y . and we say F − that F (X) is concave (convex) if GF (G+ ) is a convex set. (MRS is strictly decreasing along every indiﬀerence ¯ U (x1 . G− is the union of all G− (¯): G− = ∪y G− (¯). . Geometrically. Now we say that F (X) is F + quasiconcave (quasiconvex) if every ycrosssection of G− (GF ) is a convex set. etc.78 origin. x2 ) F (x1 . then F must be (strictly) quasiconcave. B 3  ≡ F2 F21 F22 0 F1 F2 F3 F1 F11 F12 F13 . Conversely. etc. x2 ) F (x1 . F y F 0 F1 . then B 2  ≥ 0. F (X θ ) ≥ min{F (X 0).7 we used F (X) to deﬁne two sets. xn ) is quasiconcave if ∀X 0 . . x2 ) ≤ y } ⊂ R2 . then F is strictly quasiconcave. . X 1 ∈ A and 0 ≤ θ ≤ 1. F2 F21 F22 F23 F3 F31 F32 F33 Bordered Hessian: B 1  ≡ etc. .) − U1 U2 > 0. GF and G− . x2 > 0 is strictly quasiconcave but not concave. Proof: If F (X θ ) < min{F (X 0). Theorem 1: Suppose that F is twice diﬀerentiable. if G− is convex. A F ycrosssection is formally deﬁned as G− (¯) ≡ {(x1 . . ¯ F y Clearly. B 2  > 0. x2 x2 T T E x1 d d d d d d E x1 If F is (strictly) concave. ¯ F y G+ (¯) ≡ {(x1 .) d (2) dx1 curve. If F is quasiconcave. F (x1 . .x2 )=U + In sections 6. F (X θ ) > min{F (X 0). dx1 F2 . B 3  < 0. ¯ F y F F y F Formal deﬁnition: F (x1 . . if B 1  < 0. F (X 1 )}. B 2  dMRS Proof (n = 2): = 3 and therefore F is quasiconcave if and only if B 2  ≥ 0. B 2  ≡ F1 F11 0 F1 F2 F1 F11 F12 . In this case. .. x2 ) ≥ y } ⊂ R2 . F (X 1 )}. Counterexample: F (x1 x2 ) = x1 x2 . xn ) is strictly quasiconcave if ∀X 0 = X 1 ∈ A and 0 < θ < 1. every G− (¯) is convex but still G− is not convex. it is not necessarily that F is concave. B 3  ≤ 0. x1 .14 and 7. ¯ F F y If F is quasiconcave. then F (X θ ) < (1 − θ)F (X 0 ) + θF (X 1 ).
The elasticity of substitution is deﬁned as σ ≡ r dθ w1 If the input price ratio θ = increases by 1%.xn max F (x1 . U21 U22 (a) Calculate the determinant B. x2 ) = xα xβ . ∂x1 ∂x2 U11 U12 .6 Elasticity of Substitution Consider a production function Q = F (x1 . Then U(X θ ) > U(X 0 ). Does B have any relationship with C ′ AC? 2.. Theorem 3: A monotonic increasing transformation of a quasiconcave function is a quasiconcave function. 9. To show the opposite. x2 ) ∈ ¯ R2 U(x1 . w1 F1 x2 ¯ Cost minimization ⇒ = ≡ θ. On an isoquant F (x1 . w2 F2 x1 θ dr . B = c1 a11 a12 . then X 0 is a (unique) global maximum. X 0 must be a local maximum. 1 2 (a) Calculate the marginal utilities U1 = (b) Calculate the hessian matrix H = ∂U ∂U and U2 = . Proof: A monotonic increasing transformation does not change the sets {(x1 . Consider the utility function U(x1 . x1 . Let A = (b) Calculate the product C ′ AC. Suppose there exists X 1 satisfying the linear constraint with U(X 1 ) > U(X 0 ). Example: For a CES function Q = {axρ + bxρ } ρ . a contradiction. 1. . a competitive producer will increase w2 x2 its input ratio r = by σ%.. .7 Problems a11 a12 a21 a22 0 c1 c2 . suppose that f (x1 . . Let ≡ r.. there is a relationship r = φ(θ). x2 ) = Q. x). x2 ) = H(f (x1 . and C = c2 a21 a22 −c2 c1 . σ = 1 2 1−ρ 9. x1 k 1 .. Theorem 2: If F is (strictly) quasiconcave and X 0 satisﬁes FOC. x2 ) ≥ U}.79 Consider the following maximization problem with a linear constraint. . x2 )) where H −1(g) = f (x. x2 ) is quasiconcave. . Proof: By theorem 1. A quasiconcave function is a monotonic increasing transformation of a concave function. xn ) subject to a1x1 + · · · + an xn = b. x2 ). Deﬁne a monotonic transformation as g(x1 .
y > 0 and −∞ < β < 1 is homogenous of degree 1 and concave if a = 1. U2 U21 U22 (d) Let C = −U2 U1 . 3. y) = xy is concave. (a) Show that V (x. x. Determine whether f (x. y). Let U(x. x. be a homogenous of degree 1 and concave function with Uxy = 0. y) = [U(x. y) is strictly quasiconcave for all a > 0. y > 0.) 5. (x > 0 and y > 0. (d) Determine the range of a so that F (x. (There are two critical points. y) is strictly concave. (b) Show that F (x. Hint: Uxx Uyy = [Uxy ]2 .80 0 U1 U2 (c) Calculate the determinant of the matrix B = U1 U11 U12 . Calculate the product C’HC. a/β . y)]a is strictly concave if 0 < a < 1 and V (x. y) = x + y subject to x2 + y 2 = 2. quasiconcave. Then use the bordered Hessian to determine which point is a maximum and which is a minimum. y) = xβ + y β . or quasiconvex. U. y) is strictly quasiconcave. (c) Determine the range of a so that F (x. convex. Use the Lagrangian multiplier method to ﬁnd the critical points of f (x.) 4.
xn ) subject to x1 . . suppose that f (x) is a concave function and g j (x) are convex functions (concave programming) and the constraints satisfy the constraint qualiﬁcation condition that there is some point in the opportunity set which satisﬁes all the inequality constraints as strict inequalities. . ∂xi j=1 yj m xi ∂L = 0 xi ≥ 0. . m ∂yj Kuhn Tucker theorem: x∗ solves the nonlinear programming problem if (x∗ .. there exists a vector x0 ≥ 0 such that g j (x0 ) < bj . m. xn ≥ 0. . x ≥ 0. max F (x) x subject to g(x) ≤ b. In vector notation. For nonlinear programming problems. . .. . . . m can be greater than or equal to n. In equality constraint problems. the number of constraints should be less than the number of policy variables. there is no such a restriction. Conversely. . m < n. .. . . . .xn g m (x1 . . . max F (x1 .e. .. . . . . .81 9. it is possible that a solution does not satisfy the KT condition.9 KuhnTucker condition Deﬁne the Lagrangian function as m L(x. . . ∂xi i = 1. xn )). y) for all x ≥ 0. y ≥ 0. . . then x∗ solves the nonlinear programming problem only if there is a y ∗ for which (x∗ . . . j = 1. ∂yj ∂L = 0 yj ≥ 0. y ∗) solves the saddle point problem: L(x. If it is satisﬁed.8 Nonlinear Programming The general nonlinear programming problem is: g 1 (x1 . xn ) ≤ b1 . y ∗ ) ≤ L(x∗ . . xn ) + j=1 yj (bj − g j (x1 . then the KT condition will be necessary. xn ) ≤ bm x1 . . . . .. y ∗) ≤ L(x∗ . n ∂L = bj − g j (x) ≥ 0. . y) = F (x) + y(b − g(x)) = F (x1 . 9. For the case of concave programming. it is also suﬃcient. . If constraint qualiﬁcation is not satisﬁed. i. . . KuhnTucker condition: The FOC is given by the KuhnTucker conditions: ∂L ∂F = − ∂xi ∂xi ∂g j yj ≤ 0. y ∗ ) solves the saddle point problem. j = 1. .
x2 ≥0 max f (x1 . w2 ) = 8w1 . x2 ) = x2 + x1 x2 + x2 . p1 /2 < p2 < 2p1 . x2 = 0. In stead. L2 = w2 − λ(x2 − 1) ≥ 0. xi Li = 0. (Joint product proﬁt maximization) The cost function of a competitive producer producing 2 joint products is c(x1 . 2. λLλ = 0. calculate the cost function c(w1 . x2 = 8. Case 1. x1 = 0. 9w2 < w1 . i = 1. Example 2. x1 = 8. p2 ) = p1 x1 + p2 x2 − (x2 + x1 x2 + x2 ). Case 3. 9w1 < w2 . w2 ) = 8w2 . xi fi = 0. it is not convenient to use KT condition to ﬁnd the solution. x2 = (2p2 − p1 )/3. x1 = p1 /2. Case 2. The KT condition for minimization problems: the inequalities reversed. w1 /9 < w2 < 9w1 . x1 = (2p1 − p2 )/3. c(w1 .82 In economics applications. however. x2 = p2 /2. Case 1. KT conditions: L1 = w1 − λ(x2 − 1) ≥ 0. x1 ≥0. and Lλ = (x1 + 1)(x2 + 1) − 9 ≥ 0. . i = 1. Case 2. 9. 1 2 x1 ≥0. w2 ) = 6 w1 w2 − w1 − w2 . x2 ) = p1 x1 + p2 x2 − (x2 + x1 x2 − x2 ) 1 2 KT condition: f1 = p1 − 2x1 − x2 ≤ 0. x1 = 0. The proﬁt function 1 2 is given by π(p1 . we ﬁrst solve the equality constraint version of the problem and then use KT condition to check or modify the solution when some constraints are violated. x2 = 0. Case 3. √ x1 = 9w2 /w1 − 1. f2 = p2 − x1 − 2x2 ≤ 0. x2 = 9w1 /w2 − 1 and c(w1 .x2 ≥0 min w1 x1 + w2 x2 subject to − [(x1 + 1)(x2 + 1) − 1] ≥ −8 Lagrangian function: L = w1 x1 + w1 x2 + λ[(x1 + 1)(x2 + 1) − 9]. and c(w1 . w2 ).10 Examples Example 1. The production function of a producer is given by q = (x1 +1)(x2 +1)−1. 2. 2p2 < p1 . 2p1 < p2 . For q = 8.
x1 + 2x2 ≤ q. and λ2 = 2(16 − q). Besides the money price. x2 ) = x1 (x2 +1). and λ2 = 0. derive the ordinary demand function X = D(PX ). Suppose that both products are under rationing. Lagrangian function: L = x1 (x2 + 1) + λ1 (11 − x1 − x2 ) + λ2 (q − x1 − 2x2 ). b) state the KuhnTucker conditions of the following problem: √ max 2 X + Y X≥0. KT conditions: L1 = x2 + 1 − λ1 − λ2 ≤ 0.83 Example 3. Case 4: 16 < q. λ1 = 6. PY . and Lλ1 = 11 − x1 − x2 ≥ 0.x2 subject to x1 + x2 ≤ 11. PX X + PY Y ≤ I. f) Are your answers in (c) and (d) global maximum? Unique global maximum? Why P2 . λ1 = 0. x2 = q − 11. x2 = (q − 2)/4. Y c) derive the demand functions X(PX . 2. t. I) and Y (PX . λ1 = 0. I) for the case I ≥ PX . 2 PY d) and do the same for I < PX . x2 ≥ 0. e) Given that I = 1 and PY = 1. Lλ2 = q − x1 − 2x2 ≥ 0 λi Lλi = 0. xi Li = 0. Assume that ρ1 = 1 and ρ2 = 2 and the consumer has q rationing points.11 Problems √ 1. x2 ) = x1 (x2 + 1) x1 . the consumer has to pay ρi rationing points for each unit of product i consumed. 9. Y ≥0 s. x2 = 0. x2 = 5. The market price is p1 = p2 = 1 and the consumer has $11. Therefore the budget constraint is x1 + x2 ≤ 11. Case 1: q < 2. x1 = 22 − q. The rationing point constraint is x1 + 2x2 ≤ q. check that the KT conditions are satisﬁed. L2 = x1 − λ1 − 2λ2 ≤ 0. PY . The utility maximization problem is given by max U(x1 . Case 2: 2 ≤ q ≤ 14. The utility function of a consumer is U(x1 . x1 = q. i = 1. Y ) = 2 X + Y . Case 3: 14 < q ≤ 16. Given the individual utility function U(X. and λ2 = 1. x1 = (q + 2)/2. and λ2 = (q + 2)/4. x1 = 6. x1 . a) show that U is quasiconcave for X ≥ 0 and Y ≥ 0. λ1 = 3(q − 14).
where NC (NL ) is land used in corn (lettuce) production. do the same as (b). λ1 = 1/6. according to the production functions C = NC and corn L = 2 NL respectively. t. Suppose in the ¯ short run. 3 hours for buﬃng (B).) e) Is your solution a global maximum? Explain. Each table requires 2. a) Suppose the farm is interested in maximizing its revenue. and 2 hours for crating (C). q > 0 of meat and egg respectively. Y = 6. Y ≥ 0. Its proﬁt margin is $3 per table and $4 per desk. Y ) = 3 ln X + ln Y 2X + Y ≤ 24 X + 2Y ≤ 24 X ≥ 0. State the KuhnTucker conditions of the following nonlinear programming problem max s.5 hours for assembling (A). and 1 hour for crating (C). It can produce two crops. √ (C) and lettuce (L). b) Given that q > p > 0. the ﬁrm has C units of chicken that it must take as given and suppose that the ﬁrm faces prices p > 0. 3 hours for buﬃng (B). NL ≥ 0. (NC ≥ 0. Show that X = 9. The ﬁrm can use no more than 20 hours for assembling.12 Linear Programming – A Graphic Approach Example 1 (A Production Problem). 0. 30 hours for buﬃng. if the farm uses NC of land in corn production and NL in lettuce production. State the revenue maximization problem and the KuhnTucker conditions. how much of each output will the farm produce? Check that the KT conditions are satisﬁed.84 or why not? 2. where CM (respectively CE ) ≥ 0 is the q. c) State the KuhnTucker conditions. . What is the economic interpretations of λ1 = 1/6 and λ2 = 0 if the ﬁrst constraint is interpreted as the income constraint and the second constraint as the rationing point constraint of a utility maximization problem? 9. (There are two cases. and 16 hours for crating each week. Thus.5 a) Show that the proﬁt function π = pCM + qCE is quasiconcave in (CM . U(X. Each desks requires 1 hour for assembling (A). c) Given that p ≥ q > 0. d) Derive the short run supply functions. and λ2 = 0 satisfy the KuhnTucker conditions. b) Write down the short run proﬁt maximization problem. Suppose that a ﬁrm has two activities producing two goods “meat” (M) and “egg” (E) from the same input “chicken” (C) according to the production functions 1/2 M = CM and E = CE . 4. The prices of corn and lettuce are p and q respectively. CE ). 3. A farm has a total amount of agricultural land of one acre. and √ NC + NL ≤ 1) its total revenue is pNC + 2q NL . A manufacturer produces tables x1 and desks x2 .
6) rr dv rr Z v dZ rr vdZZ rr v d Z rr r v d Z v v v v v E x1 e e de e d e d ed HH e d e HH d e Hd H Hd e H (9.85 max Π subject to 2. At (4. There are 10 extreme points. In the example. and the cost of y2 is $4. 3 1 ). it will be found at one of the extreme points. and C.5x1 + x2 3x1 + 3x2 x1 + 2x2 x1 . A farmer wants to see that her herd gets the minimum daily requirement of three basic nutrients A. The cost of y1 is $2. (8. 12 for B. B. (6 2 . and (0. y2 = ≥ ≥ ≥ ≥ 2y1 + 4y2 14 12 18 0. and 18 for C. Π = 36 is the 3 3 optimal. x2 = ≤ ≤ ≤ ≥ 3x1 + 4x2 20 30 16 0.0). Daily requirements are 14 for A. called basic feasible solutions. (9) (10) (11) (12) (13) . 6). (4. but only 5 are feasible: (0 . (4) (5) (6) (7) (8) extreme point: The intersection of two constraints. min c subject to 2y1 + y2 y1 + y2 y1 + 3y2 y1 . x2 v T v v v v y2 T v d Zd v rZ v rd Z r Z dr v rq Z v dZ (4. Product y1 has 2 units of A and 1 unit each of B and C. extreme point theorem: If an optimal feasible value of the objective function exists. 0). 6). product y2 has 1 unit each of A and B and 3 units of C.3) e d q HH e d HH e d H H e d H Ey Example 2 (The Diet Problem). 8).
2. the inequalities are ”≥” and the surplus variables are substracted: 2y1 + y2 − s1 = 14 In matrix for. In example 1. a solution in which at least n − m variables equal to zero is an extreme point. always improving upon the previous solutions.13 Linear programming – The simplex algorithm The algorithm moves from one basic feasible solution to another. an extreme point can be found. For a system of m equations and n variables. those variables set equal to zero are called not in the basis and those not set equal to zero are called in the basis. This is done by incorporating a separate slack or surplus variable si into each inequality. and s3 = 16.86 Slack and surplus variables: To ﬁnd basic solutions. There are n!/m!(n − m)! such solutions. The Pivot Element and a change of Basis . 2 1 −1 0 0 1 1 0 −1 0 1 3 0 0 −1 y1 y2 s1 s2 s3 14 = 12 . In example 2. s2 = 30. 16 3x1 + 3x2 + s2 = 30 x1 + 2x2 + s3 = 16. 18 y1 + y2 − s2 = 12 y1 + 3y2 − s3 = 18. until the optimal solution is reached. 1. x2 = 0.5 1 1 0 0 3 3 0 1 0 1 2 0 0 1 x1 x2 s1 s2 s3 20 = 30 . 9. Let use example one to illustrate the procedure. The value of Π is zero. where n > m.5x1 + x2 + s1 = 20 In matrix for. equations are needed. The initial Simplex Tableau x1 x2 2. the system becomes 2.5 1 3 3 1 2 3 4 s1 1 0 0 0 s2 0 1 0 0 s3 0 0 1 0 Constant 20 30 16 0 The ﬁrst basic feasible solution can be read from the tableau as x1 = 0. 2. s1 = 20. In each step. Thus by setting n − mand solving the m equations for the remaining m variables.
s1 = 4. Here the smallest is 16/2=8 and row 3 is the pivot row (ith row). In the example. The x2 column is called the pivot column (jth column).5 12 1.5 4 3 2 1 0 0 1 4 3 1 0 1 0 −3 1 6 2 0 0 0 1 36 3 The basic solution is x1 = 4. x2 = 6.5 is the pivoting element. the tableau becomes x1 x2 s1 s2 s3 Constant 0 0 1 − 4 1. (Optimization) Repeat steps 23 until a maximum is reached. The indicators for s2 and s3 . the Tableau becomes x1 x2 s1 s2 s3 Constant 2 0 1 0 . Since there is no more negative indicators. Usually.) After pivoting.5 1 0 0 . the simplex algorithm for minimization problem is more complex. representing the 3 marginal contributions of increasing one hour for buﬃng or crating. t2 = 2 and t3 = 1 are called the shadow values. After pivoting. The second row is the pivoting row and 1. we solve its dual problem. First. The pivot element is 2. and s2 = s3 = 0. x1 column is the new pivoting column. Displacement ratios are found by dividing the elements of the constant column by the elements of the pivot column. s2 = 6. .5 6 . and x1 = s3 = 0. (b) The variable to be eliminated is determined by the smallest displacement ratio. usually used in solving simultaneous equations. 3. Then subtracts akj times the pivoting row from kth row to make the jth column a unit vector.87 (a) The negative indicator with the largest absolute value determines the variable to enter the basis. the process stops and the basic solution is the optimal.5 8 1 0 0 0 2 32 The basic solution is x2 = 8.5 0 0 1 1. The value of Π is 32. Because y1 = y2 = 0 is not feasible. divides every element of the pivoting row by the pivoting element (2 in this example) to make the pivoting element equal to 1. (Pivoting) and we are going to move to the new basic solution with s3 = 0 and x2 > 0. s1 = 4 > 0 and s2 = s3 = 0 means that the ﬁrst constraint is not binding but the second and third are binding. s1 = 12. The value of Π is 36. (This procedure is called the Gaussian elimination method. 4. Here it is x2 .
x2 . . cn ). x3 = ≤ ≤ ≥ 14x1 + 12x2 + 18x3 2 4 0. . y = (y1 . am1 . F (x) ≤ G(y). x≥0 then the dual problem is min G = yb y subject to yA ≥ c. x = . . . A= . . C. respectively. . x1 . . . . . bm c = (c1 . x1 . is max F = cx x subject to Ax ≤ b. y are feasible. . a1n . (14) (15) (16) (17) y ∗(b − Ax∗ ) = 0. amn Existence theorem: A necessary and suﬃcient condition for the existence of a solution is that the opportunity sets of both the primal and dual problems are nonempty. It follows that F (x) = cx ≤ yAx and G(y) = yb ≥ yAx. . y ∗ to solve the dual problems is that they satisfy the complementary slackness condition: (c − y ∗ A)x∗ = 0 Proof: Use KuhnTucker theorem. If the original problem. called the primal problem. xn b1 . B. . b = . Then Ax ≤ b. Duality theorem: A necessary and suﬃcient condition for a feasible vector to represent a solution is that there exists a feasible vector for the dual problem for which the values of the objective functions of both problems are equal.14 Linear programming – The dual problem To every linear programming problem there corresponds a dual problem. y≥0 where a11 . Therefore. Complementary slackness theorem: A necessary and suﬃcient condition for feasible vectors x∗ . x3 . ym ). . . . . . Proof: Suppose x. x2 . . . .88 9.. . yA ≥ c. . is interpreted as the imputed value of nutrient A. . Dual of the Diet Problem max c∗ subject to 2x1 + x2 + x1 x1 + x2 + 3x3 x1 .
to p1 x1 + · · · + pn xn = I. . . . . . . I). . . k = 1. . x2 . n √ √ √ √ Example: q = f (x1 . p. . xn Exogenous variables: p1 . . . . ∂x1 x1 ∂x2 x2 ⇒ x1 = (p/p1 )2 . p1 . pn ) k = 1.1 General Equilibrium and Game Theory Utility maximization and demand function A consumer wants to maximize his utility function subject to his budget constraint: max U(x1 . x2 = ⇒ b x1 p2 (a + b)p1 (a + b)p2 0 −p1 −p2 −a ap2 bp2 0 −p1 SOC: = 22 + 21 > 0.2 Proﬁt maximization and supply function A producer’s production technology can be represented by a production function q = f (x1 . . xn ) − p1 x1 − · · · − pn xn Exogenous variables: p. . xk = Xk (p. . . p1 . p1 . . x2 ) = 2 x1 + 2 x2 and Π(x1 . . pn ) and the input demand functions. . x2 ) = a ln x1 + b ln x2 subject to p1 x1 + p2 x2 = m. . .89 10 10. . x2 1 x1 x2 −b −p2 0 x2 2 am bm ⇒ x1 = . . . L2 = − λp2 = 0 and Lλ = m − p1 x1 − p2 x2 = 0. . . pn . . (a + b)p1 (a + b)p2 10. . p2 ) = p(2 x1 + 2 x2 ) − p1 x1 − p2 x2 . x2 = (p/p2 )2 (input demand functions) and q = 2(p/p1 ) + 2(p/p2 ) = 2p( p11 + p12 ) (the supply function) . . . . pn (the producer is a price taker) Solution is the supply function q = S(p. xn ). . . . . . . a b FOC: L1 = − λp1 = 0. . . . xn . pn . xn ) subj. . √ √ max p(2 x1 + 2 x2 ) − p1 x1 − p2 x2 x1 .x2 FOC: ∂Π p ∂Π p = √ − p1 = 0 and = √ − p2 = 0. . . x2 = is a local maximum. . p1 . the producer maximizes his proﬁts: max Π(x1 . pn ) = pf (x1 . p. . . L = a ln x1 + b ln x2 + λ(m − p1 x1 − p2 x2 ). x1 x2 p1 am bm a x2 = ⇒ x1 = . . . n Example: max U(x1 . p1 . . . Endogenous variables: x1 . . Given the prices. I (the consumer is a price taker) Solution is the demand functions xk = Dk (p1 .
. yn ) = 0.3 ∂2Π ∂x2 1 ∂2Π ∂x1 ∂x2 −p ∂2Π ∂x1 ∂x2 2x3/2 = 1 ∂2Π 0 ∂x2 1 0 2x2 3/2 −p Transformation function and proﬁt maximization In more general cases.4 The concept of an abstract economy and a competitive equilibrium Commodity space: Assume that there are n commodities. J producers. J. A price system: (p1 . . . . . . To solve the maximization problem. . . . . . . . xi ). where (y1 . . 10. . . ωn ). if yk > 0 j (yk ) then k is an output (input) of j. yn ) is called a production plan. . . . . i i Consumer i’s endowment is ω i = (ω1 . i Consumer i has a utility function U i (x1 . . . . y1 . j = 1. . The commodity space is n R+ = {(x1 . with initial endowments of commodities ω = (ω1 . . Example: a producer produces two outputs. . i=1 θij = 1. yn ). . . i = 1. Its technology is given by the transformation function (y1 )2 + (y2 )2 + y3 = 0. . I. i=1 I Consumer i’s share of ﬁrm j is θij ≥ 0.y2 The solution is: y1 = p1 /(2p3 ). . using one input y3 . . . y2 = p2 /(2p3 ) (the supply functions of y1 and y2 ). . . . and y j = (y1 . 1 n . i = 1. the technology of a producer is represented by a transformation j j j j j function: F j (y1 . .y2 . yn ) = 0. y1 and y2 . .y3 max p1 y1 + p2 y2 + p3 y3 subject to (y1 )2 + (y2 )2 + y3 = 0. pn ). .90 Π = p2 ( p11 + SOC: 1 ) p2 is negative deﬁnite. . . . . . The maximization problem is y1 . j j An allocation: xi = (xi . . ωn ). A private ownership economy: Endowments and ﬁrms (producers) are owned by consumers. . . I ω i = ω. 10. xn ). n j j Producer j has a production transformation function F j (y1 . . . xi ). and x = −y3 = [p1 /(2p3 )]2 + [p2 /(2p3 )]2 (the input demand function for y3 ). . I. Its proﬁt is Π = p1 y1 + p2 y2 + p3 y3 . . we can eliminate y3 : x = −y3 = (y1 )2 + (y2 )2 > 0 and max p1 y1 + p2 y2 − p3 [(y1 )2 + (y2 )2 ]. xk ≥ 0} Economy: There are I consumers. . .
Existence Theorem: Suppose that the utility functions are all quasiconcave and the production transformation functions satisfy some theoretic conditions. this input is indispensable to production. . ... Then the competitive equilibrium price system is determined by the production side only. . . {¯j }j=1.. J and xi maximizes U i . subject to i’s budget con¯ ¯ i i 1 i straint p1 x1 + . i xi = ω + j y j (feasibility condition).I . . + θiJ ΠJ .. pn ) and an allocation ({¯i }i=1.. . + pn ωn + θi1 Π1 + . Welfare Theorems: A competitive equilibrium is eﬃcient and an eﬃcient allocation can be achieved as a competitive equilibrium through certain income transfers.J ) ¯ p ¯ x y such that 1. and the production functions exhibits constant returns to scale. . . . + pn xn = p1 ω1 + .. . y maximizes Π . .. . . .91 A competitive equilibrium: A combination of a price system p = (¯1 . Constant returns to scale economies and nonsubstitution theorem: Suppose there is only one nonproduced input. . ¯ ¯ j j 2.. there is no joint production. j = 1. then a competitive equilibrium exists.
C is 1/3. Example (a 3person game): The accuracy of shooting of A. according to number of players: 2person games – conﬂict and cooperation possibilities. Players: Persons having some inﬂuences upon possible income (decision makers). Each person has some control over the outcome. respectively. nperson games – coalition formation possibilities in addition. .5 Multiperson Decision Problem and Game Theory In this chapter. Game theory: the study of conﬂict and cooperation between persons with diﬀerent objective functions. inﬁniteplayers’ games – corresponding to perfect competition in economics. which in turn determines the rewords to players. We want to study how persons make decisions. 2. 2. Game theory: The outcome depends on the behavior of all the persons involved.92 10. A play: A complete record of the choices made at moves by the players and realizations of randomization. B. 4. Classiﬁcations of games: 1. diﬀerent persons have diﬀerent preferences over possible outcomes. Outcomes and payoﬀs: a play results in an outcome. Moves: decision points in the game at which players must make choices between alternatives (personal moves) and randomization points (called nature’s moves).6 Ingredients and classiﬁcations of games A game is a collection of rules known to all players which determine what players may do and the outcomes and payoﬀs resulting from their choices. Public Choice: Persons have to make decision collectively. by voting. 2. Each person wants to kill the other two to become the only survivor. There are two cases: 1. They shoot in turn starting A. we consider the situation when there are n > 1 persons with diﬀerent objective (utility) functions. Each one’s utility depends on the decisions of all persons. eg. 2/3. 1. The ingredients of a game: 1. Question: What is the best strategy for A? 10. 3. according to number of strategies: ﬁnite – strategy (matrix) games. We consider only game theory here. each person controls certain strategic variables. that is. that is.. each person has a ﬁnite number of strategies.
7 The extensive form and normal form of a game Extensive form: The rules of a game can be represented by a game tree. cooperative games – players form coalitions and decide the redistribution of aggregate payoﬀs. . The ingredients of a game tree are: 1.. Randomization probabilities (of arcs following each randomization points). . S2 = {σ1 . (am1 . .. according to sum of payoﬀs: 0sum games – conﬂict is unavoidable. bm1 ) . An information set consists of decision points that player i can not distinguish when making decisions. bmn ) . Normal form: d II d d I s1 . . .. sm . 5. 2. . σn (a1n . π2 (si . Arcs (choices): Every point in an information set should have the same number of choices. 6. . . . Once we identify the pure strategy set of each player. . b11 ) .. σj ) = aij .. 1. σj ) = bij . 3. Nodes: they are players’ decision points (personal moves) and randomization points (nature’s moves). Information sets of player i: each player’s decision points are partitioned into information sets. 10. . sm }. . σ1 (a11 . 4. Total number of pure strategies of player i: the product of the numbers of choices of all information sets of player i. A pure strategy of player i: An instruction that assigns a choice for each information set of player i. nonzero sum games – possibilities for cooperation. Payoﬀs: The gains to players assigned to each outcome. B = [bij ]. . . inﬁnite – strategy (continuous or discontinuous payoﬀ functions) games like duopoly games. Outcomes (end points) 7. we can represent the game in normal form (also called strategic form). according to preplay negotiation possibility: noncooperative games – each person makes unilateral decisions. . (amn . Strategy sets for each player: S1 = {s1 . 4. Payoﬀ matrices: π1 (si . Players 2. . . A = [aij ]. .93 payoﬀ functions can be represented by matrices. b1n ) . . σn }. 3.
diﬀerent extensive form games may have the same normal form. R }. Example 3: Hijack game 1 R L 2 L d R d −1 2 2 −10 −2 −10 S1 = { L.2) d II d d S1 = { L. 4.94 10.6) (1. R }.7) Remark: Each extensive form game corresponds a normal form game.2)* (10. Rl. Lr. R }. Ll Lr Rl Rr L R (4.5.5.9) (9. R }. S2 = { Ll.6) (4. Rl. S2 = { L. S2 = { L. I d Ll Lr Rl Rr L (1.5) (5.7)* (8.6) R (3. Lr.7) (8.0) (1. 5)* (4.9) (9. However. 2) (3.5) (5.2) (1.1)* d II d d I L R L R (1.2) (3. Rr }. .5) (4.10) Example 4: A simpliﬁed stock price manipulation game 0 d II d I d 2 7 5 5 7 4 5 4 2 3 7 S1 = { Ll. 3.5) (5. Rr }. 4.5. 1/2 ¨rr1/2 ¨ rr 1 1¨¨ e R ¡d r L l e ¡ ¨d 2 ¡e ¡e © d L¡ e R L¡ e R 4 I L R L R (4.5) (3.8 Examples Example 1: A perfect information game 1 L R 2 2 d r l L d R d d 1 9 9 6 3 7 8 2 d II d Example 2: Prisoners’ dilemma game 1 R L ¨ 2 © L dd L dd R R 4 4 0 5 5 0 1 1 S1 = { L.2)* (2. R }.5. R }. S2 = { L.4) (0. 4.
σj ). we can identify some subgames.R) in the hijack game. Within an extensive form game. A Strategy Pair: (si .10 Subgames and subgame perfect Nash equilibria 1. each pure strategy of a player induces a pure strategy for every subgame. The equilibrium payoﬀs of the examples are marked each with a star in the normal form. All the equilibria. there corresponds a payoﬀ pair (aij . 2. Every perfect information game has a pure strategy subgame perfect Nash Equilibrium. Backward induction: To ﬁnd a subgame perfect equilibrium.95 10. In this way. σj∗) such that ai∗j∗ ≥ aij∗ and bi∗j∗ ≥ bi∗j for all (i. 2. Remark: The concept of a subgame perfect Nash equilibrium is deﬁned only for an extensive form game. Also. . Subgame perfect Nash equilibrium: A Nash equilibrium is called subgame perfect if it induces a Nash equilibrium strategy pair for every subgame. Remark 2: Notice that the concept of a Nash equilibrium is deﬁned for a normal form game. j). Subgame: A subgame in a game tree is a part of the tree consisting of all the nodes and arcs following a node that form a game by itself. For a game in extensive form (a game tree). are subgame perfect.9 Strategy pair and pure strategy Nash equilibrium 1. Given a strategy pair.1 Perfect information game and Zemelo’s Theorem An extensive form game is called perfect information if every information set consists only one node. 10. 4. bij ). Remark 1: It is possible that a game does no have a pure strategy Nash equilibrium. we have to ﬁnd the normal form before we can ﬁnd the Nash equilibria. the size of the game is reduced step by step until we end up with the equilibrium payoﬀs. there is no incentives for each player to deviate from the equilibrium strategy.10. 5. usually we work backward. a game can have more than one Nash equilibria. except the equilibrium strategy pair (L. A Nash equilibrium: A strategy pair (si∗ . 3. Therefore. Also. We ﬁnd Nash equilibria for lowest level (smallest) subgames and replace the subgames by its Nash equilibrium payoﬀs. ai∗j∗ and bi∗j∗ are called the equilibrium payoﬀ. 10.
A ⊕ B). A behavior strategy: A function which assigns a local strategy for each u ∈ Ui . in a nonperfect recall game. S1 and S2 are inﬁnite subsets of Rm and Rn Player 1 controls m variables and player 2 controls n variables (however. y ∗) ≥ Π2 (x∗ . Reduced normal form: The normal form without redundant strategies. Equivalent normal form: Two normal forms are equivalent if they have the same reduced normal form. y). each player has inﬁnite many strtategies). Kuhn’s Theorem: In every extensive game with perfect recall. y ∗ ) and Π2 (x∗ . A strategic pair is (x. y) for all x ∈ S1 y ∈ S2 . The set of behavior strategies is a subset of the set of mixed strategies. y ∗) such that Π1 (x∗ . d 1 ¨ d d u12 © d d 1 −1 d d c d c d d d −1 1 ¡e ¡e ¡e 0 ¡ e −2 0¡ e ¡ e2 e ¡ e ¡ e ¡ −2 0 2 0 0 10. The normal form of a game is represented by two functions Π1 = Π1 (x. where x ∈ S1 ⊂ Rm and y ∈ S2 ⊂ Rn .96 10. 10. 1/2 ¨ rr 1/2 ¨ ¨ 1 1 1 1 rr u¨¨ 2 11 NE is (µ∗ . y) and Π2 = Π2 (x. ¨rr A 2person 0sum nonperfect recall game.2 Perfect recall game and Kuhn’s Theorem A local strategy at an information set u ∈ Ui : A probability distribution over the choice set at Uij . Equivalent extensive form: Two extensive forms are equivalent if their normal forms are equivalent. y) ∈ S1 × S2 .10.11 Continuous games and the duopoly game In many applications. To simplify the presentation. a strategically equivalent behavior strategy can be found for every mixed strategy. a mixed strategy may do better than behavior strategies because in a behavior strategy the local strategies are independent whereas they can be correlated in a mixed strategy. µ∗ ) = ( ac ⊕ bd. assume that m = n = 1. y ∗ ) ≥ Π1 (x. However. A Nash equilibrium is a pair (x∗ .3 Reduction of a game Redundant strategy: A pure strategy is redundant if it is strategically identical to another strategy. 1 2 2 2 2 2 a d b A dd B µ∗ is not a behavioral strategy. .10.
5(a − c1 − q2 ) and q2 = 0. If they cannot reach an agreement in two days. otherwise Paul will reject the oﬀer.1 A simple bargaining model Two players. 1. 2.11. both players get zero. q2 ) together determine the market price P = a − (q1 + q2 ) which in turn determines the proﬁt of each ﬁrm: Π1 (q1 . x is his endogenous variable and y is his exogenous variable. Player 2’s reaction function (behavioral equation) y = R2 (x) is obtained by solving 2 Πy (x. They agree to spend at most two days negotiating over the division. y) = 0. John will propose of 1 − β for himself and β for Paul and Paul will accept the oﬀer. a − 2c2 + c1 ≥ 0.) 10. On the ﬁrst day. Paul either accepts or comes back with a counteroﬀer the second day. Therefore. Π1 (x. The ﬁrst day. y ∗ ) = 0 and Π2 (x∗ . The situation deﬁnes a 2person game as follows: Each ﬁrm i controls his own output quantity qi . Therefore. A subgame perfect equilibrium of this bargaining game can be derived using backward induction. (a − 2c2 + c1 )/3) with P ∗ = (a + c1 + c2 )/3. respectively. (We have to assume that a − 2c1 + c2 . John will make an oﬀer. x y Duopoly game: There are two sellers (ﬁrm 1 and ﬁrm 2) of a product. The (inverse) market demand function is P = a − Q. (q1 . John (Paul) discounts payoﬀs in the future at a rate of α (β) per day. John would accept any nonnegative counteroﬀer made by Paul. ∗ ∗ The Cournot Nash equilibrium is (q1 . This relationship deﬁnes a behavioral equation x = R1 (y) which can be obtained by solving the FOC for player 1. A Nash equilibrium is an intersection of the two reaction functions. Assume that each ﬁrm regards the other ﬁrm’s output as given (not aﬀected by his output quantity). Reaction functions and Nash equilibrium: To player 1. The marginal production costs are c1 and c2 . The reaction functions are q1 = 0. John and Paul. . player 1 will choose a x ∈ S1 to maximize his objective function Π1 . q2 ) = (P −c1 )q1 = (a−c1 −q1 −q2 )q1 and Π2 (q1 . q2 ) = ((a − 2c1 + c2 )/3. Paul would make proposal of getting the whole $ 1 himself and John would get $ 0. player 2 regards y as enx dogenous and x exogenous and wants to maximize Π2 for a given x chosen by player 1. John should make an oﬀer such that Paul gets an amount equivalent to getting $ 1 the second day. The FOC for a Nash equilibrium is given by Π1 (x∗ . have $ 1 to divide between them. y ∗) = 0. Similarly. q2 ) = (P −c2 )q2 = (a−c2 −q1 −q2 )q2 The FOC are ∂Π1 /∂q1 = a − c1 − q2 − 2q1 = 0 and ∂Π2 /∂q2 = a − c2 − q1 − 2q2 = 0. For each y chosen by player 2.97 Consider the case when Πi are continuously diﬀerentiable and Π1 is strictly concave in x and Π2 strictly concave in y (so that we do not have to worry about the SOC’s). y) = 0.5(a − c2 − q1 ). On the second day.
. In A2 . i i pi ain }. i 6. . maxi minj aij = minj maxi aij = 1 (s1 . 3. Player 2 chooses σj ∗ so that maxi aij∗ ≤ maxi aij for all j.5β for himself and 0. then the minimum he will have is minj aij . . 5. . i pi ain ≥ t.5β for Paul and refuses to accept any counteroﬀer made by Paul. the expected payoﬀ of player 1 is pAq. q ∗ ) is such that p∗ Aq ∗ ≥ pAq ∗ and p∗ Aq ∗ ≤ p∗ Aq for all p and all q. qn )′ . q = (q1 . σ2 ) is a saddle point and hence a Nash equilibrium. A mixed strategy Nash equilibrium (p∗ .12 2person 0sum game 1. . In this equilibrium. . q) is a mixed strategy pair. pi = 1. The problem of ﬁnding the maxmin mixed strategy (to ﬁnd p∗ to maximize t(p)) can be stated as max t subj. . . since bij = −aij . Paul is threatened by John’s incredible threat and accepts only one half of what he should have had in a perfect equilibrium. . then it is a Nash equilibrium. Similarly. Given (p. Linear programming problem: The above problem can be transformed into a linear programming problem as follows: (a) Add a positive constant to each element of A to insure that t(p) > 0 for all p. just as in the case of a pure strategy of player 1. . 4. 2. . to p i pi ai1 ≥ t. 10. B = −A so that aij + bij = 0. σj∗ ) is called a saddle point. pm ). A1 = 2 1 4 −1 0 6 A2 = 1 0 0 1 In example A1 . t(p) ≡ min{ j pi ai1 . Security level of a mixed strategy: Given player 1’s strategy p. p = (p1 . If a saddle point exists. then there is no pure strategy equilibrium. If there is no saddle points. Saddle point: If ai∗j∗ = maxi minj aij = minj maxi aij . . Mixed strategy for player i: A probability distribution over Si .98 An example of a subgame nonperfect Nash equilibrium is that John proposes of getting 10. . then (si∗ . called the security level of strategy si . (b) Deﬁne yi ≡ pi /t(p) and . Maxmin strategy: If player 1 plays si . maxi minj aij = 0 = minj maxi aij = 1 and no saddle point exists. A possible guideline for player 1 is to choose a strategy such that the security level is maximized: Player 1 chooses si∗ so that minj ai∗j ≥ minj aij for all i. there is a pure strategy of player 2 so that the expected payoﬀ to player 1 is minimized. q). . . . (p.
The min y1 + . . to x1 ≤ 1 x2 ≤ 1.99 replace the problem of max t(p) with the problem of min 1/t(p) = constraints become i yi ai1 ≥ 1. . .13 Mixed strategy equilibria for nonzero sum games The idea of a mixed strategy equilibrium is also applicable to a nonzero sum game. . The solution is x1 = x2 = 1 and therefore the equilibrium strategy for player 2 is ∗ ∗ q1 = q2 = 0. . The existence of a mixed strategy Nash equilibrium is then proved by using the duality theorem in linear programming. + ym subj.y2 ≥0 min y1 + y2 subj. 1 2 10.5.x2 ≥0 Example (tossing coin game): A = max x1 + x2 subj. 1 0 .ym ≥0 i yi . i yi ain ≥ 1 7.. . to i yi ai1 ≥ 1. there is a Lemke algorism. to y1 ≥ 1 y2 ≥ 1. 0 1 To ﬁnd player 2’s equilibrium mixed strategy. ... . x2 y2 T T d d d dr d d d d d d d dr d d d d 1 1 E x1 1 1 E y1 Player 1’s equilibrium mixed strategy is obtained by solving the dual to the linear programming problem: y1 . Example (Game of Chicken) . i yi ain ≥ 1. ..5. y1 . Duality: It turns out that player 2’s minmax problem can be transformed similarly and becomes the dual of player 1’s linear programming problem. . we solve the linear programming problem: x1 . Similar to the simplex algorism for the 0sum games. The solution is p∗ = p∗ = 0.
In general.0) (3. Π1 (S) = Π1 (N).0) Opera (0. 9 ) and (q1 . therefore. S2 = { B. his expected payoﬀ is Π1 (B) = 5q + (0)(1 − q). 1 − q).14 Cooperative Game and Characteristic form 2person 0sum games are strictly competitive. For other games.9) S1 = { S. his expected payoﬀ is Π1 (S) = 0q + (−3)(1 − q). 4 ∗ ∗ The mixed strategy is: (p∗ . 1 ). p∗ ) = ( 5 . player 2 will loss $ 1 and therefore no cooperation is possible. The problem becomes the formation of coalitions and the distribution of payoﬀs. the determination of strategies is not the key issue. q = 2 .100 1 N S ¨ 2 © S dd S dd N N 0 0 −3 3 3 −3 −9 −9 d II d I d Swerve Don’t Swerve (0. If he plays N. 5 ). 1 − q). 1 2 9 9 9 10. his expected payoﬀ is Π1 (N) = 3q + (−9)(1 − q). O }. a Nash equilibrium is not eﬃcient (not Pareto optimal). 3 ∗ ∗ The mixed strategy is symmetrical: (p∗ . N }. For a mixed strategy NE.0) (4. If he plays O. If player 1 gains $ 1. q2 ) = ( 4 . When binding agreements on strategies chosen can be contracted before the play of the game and transfers of payoﬀs among players after a play of the game is possible. In such a situation. . his expected payoﬀ is Π1 (O) = 0q +(4)(1−q). The concept of a Nash equilibrium is deﬁned for the situation when no explicit cooperation is allowed. If player 1 plays S. 4 therefore.4)* (0.3)* (9. There are two pure strategy NE: (S. B) and (O. p∗ ) = (q1 . q = 9 . There is also a mixed strategy NE. N }. Suppose player 2 plays a mixed strategy (q. There is also a mixed strategy NE. There are two pure strategy NE: (B. S2 = { S. Suppose player 2 plays a mixed strategy (q. usually some cooperation is possible. N) and (N. For a mixed strategy NE. O }. S).5)* S1 = { B. If player 1 plays B.3)* Don’t (3. 1 2 3 3 Example (Battle of sex Game) 1 O B ¨ 2 © B dd B dd O O 5 4 0 0 0 0 4 5 d II d I d Ball game Opera Ball game (5. Π1 (B) = Π1 (O). q2 ) = ( 2 . players will negotiate to coordinate their strategies and redistribute the payoﬀs to achieve better results. O).
3. the Shapley values are uniquely determined. v(123) = 1. .4. then v(A ∪ B) ≥ v(A) + v(B). Let F ⊂ R2 be the feasible set of payoﬀs if the two players can reach an agreement and Ti the payoﬀ of player i if the negotiation breaks down. Marginal contribution of player i in a coalition S ∪ i: v(S ∪ i) − v(S) Shapley value of player i is an weighted average of all marginal contributions S!(n − S − 1)! [v(S ∪ i) − v(S)]. π2 . v(12). 0.101 Characteristic form of a game: The player set: N = {1.2. . (3). n}. v(2). 10. namely. x2 . Individual rationality: xi ≥ v(i).4) are in C. For the case of 2person games. Consider the case of a 3person game. the concept of Nash bargaining solutions are useful. The Nash bargaining solution (x∗ . 0. x3 ). . The Shapley values are (π1 . x2 . 2.4) and (0. Core C: the set of imputations that satisfy individual rationality and group rationality for all S. v(3). 2. (12). φ. v(1). There are 8 subsets of N = {1. 1 ). A coalition is a subset of N: S ⊂ N. xi + xj ≥ 0. A characteristic function v speciﬁes the maximum total payoﬀ of each coalition.15 The Nash bargaining solution for a nontransferable 2person cooperative game In a nontransferable cooperative game. Ti is called the threat point of player i. . C = {(x1 . x1 + x2 + x3 = 1}. v(13). Group rationality: i∈S xi ≥ v(S). xi ≥ 0. 0. π3 ) = ( 1 .3. See CH 6 of Intriligator’s Mathematical Optimization and Economic Theory for the motivations of these concepts. 1 .5. afterplay redistributions of payoﬀs are impossible and therefore the concepts of core and Shapley values are not suitable. Therefore. (23). (1). Both (0. However. 0. Remark 2: Another related concept is the vonNeumann Morgenstern solution. (2).5. n! πi = S⊂N Example: v(φ) = v(1) = v(2) = v(3) = 0. 3}. 3 3 3 Remark 1: The core of a game can be empty. An imputation is a payoﬀ distribution (x1 . a characteristic form game is determined by 8 values v(φ). x3 ). v(2 Superadditivity: If A ∩ B = φ. (123). (13). v(12) = v(13) = v(23) = 0. x∗ ) is deﬁned to be the solution 1 2 to the following problem: .
i = 1. use simplex method to ﬁnd the mixed strategy Nash equilibrium. Π2 ) such that Π2 + Π2 = 47 1 . they can get (Π1 .2) 4. 3. Suppose that two ﬁrms producing a homogenous product face a linear demand curve P = a−bQ = a−b(q1 +q2 ) and that both have the same constant marginal costs c. Consider the following I \ II s1 s2 s3 (a) Find the max min strategy of player I smax min and the min max strategy of player II σmin max . 1) T (2. Find the maxmin strategy (smax min ) and the minmax strategy (σmin max ) of the following twoperson 0sum game: I \ II σ1 σ2 s1 3 6 s2 8 2 s3 6 3 Is the strategy pair (smax min . 10. Suppose that in a twoperson cooperative game without side payments. 2. (b) Is the strategy pair (smax min .102 x2 T (x1 . the proﬁts are Πi = qi (P − c) = qi (a − bq1 − bq2 − c). Find the (mixed strategy) Nash Equilibrium of the following twoperson game: I \ II H T H (2. σmin max ) a Nash equilibrium? If not. For a given quantity pair (q1 .x2 )∈F max (x1 − T1 )(x2 − T2 ) x∗ 2 T2 T1 x∗ 1 E x1 See CH 6 of Intriligator’s book for the motivations of the solution concept. 2) (2. 2) (1. Find the Cournot Nash equilibrium output of each ﬁrm.16 Problems twoperson 0sum game: σ1 σ2 σ3 4 3 2 3 4 10 7 6 8 1. if the two players reach an agreement. 5. σmin max ) a Nash equilibrium of the game? (c) What are the equilibrium payoﬀs? 2. q2 ).
103 and if no agreement is reached.000 to play together by a night club owner. (b) Do the same for the case when side payments are possible. (a) Find the Nash solution of the game. (a) Write down the characteristic form of the cooperative game with side payments. Also answer how the side payments should be done? 6. a pianist (player 2). Howeover. the singerdrums duo makes $ 500 and the singer alone gets $ 200 a night in a restaurant. (b) Find the Shapley values of the game. The night club is not interested in any other combination. (c) Characterize the core. The drums alone can make no proﬁt. and a drummer (player 3) are oﬀered $ 1. and $ 300 the piano alone. $ 650 the piano drums duo. player 1 will get T1 = 3 and player 2 will get T2 = 2. A singer (player 1). The owner would alternatively pay $ 800 the singerpiano duo. .
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue listening from where you left off, or restart the preview.