1

Lecture Notes on

Industrial Organization (I)

Chien-Fu CHOU

January 2004
 2

Contents

Lecture 1 Introduction 1

Lecture 2 Two Sides of a Market 3

Lecture 3 Competitive Market 8

Lecture 4 Monopoly 11

Lecture 5 Basis of Game Theory 20

Lecture 6 Duopoly and Oligopoly – Homogeneous products 32

Lecture 7 Differentiated Products Markets 46

Lecture 8 Concentration, Mergers, and Entry Barriers 62

Lecture 9 Research and Development (R&D) 81

Lecture 10 Network Effects, Compatibility, and Standards 93

Lecture 11 Advertising 102

Lecture 12 Quality 109

Lecture 13 Pricing Tactics 112

Lecture 14 Marketing Tactics: Bundling, Upgrading, and Dealerships 114

 1

1 Introduction
1.1 Classification of industries and products
2M¬Å¼¹™Ä}é, 2M¬ÅW“™Ä}é;
«%Íß%’eé.

1.2 A model of industrial organization analysis:

(FS Ch1)

Structuralist:

1. The inclusion of conduct variables is not essential to the development of an

operational theory of industrial organization.

2. a priori theory based upon structure-conduct and conduct-performance links

yields ambiguous predictions.

3. Even if a priori stucture-conduct-performance hypotheses could be formulated,

attempting to test those hypotheses would encounter serious obstacles.

Behaviorist: We can do still better with a richer model that includes intermediate
behavioral links.

1.3 Law and Economics

Antitrust law, t>q¶
Patent and Intellectual Property protection ù‚DN‹ßž\ˆ
Cyber law or Internet Law 昶

1.4 Industrial Organization and International Trade

 2

Basic Conditions
Supply Demand
Raw materials Price elasticity
Technology Substitutes
Unionization Rate of growth
Product durability Cyclical and
Value/weight seasonal character
Business attitudes Purchase method
Public polices Marketing type

?
Market Structure
Number of sellers and buyers
Product differentiation
Barriers to entry
Cost structures
Vertical integration
Conglomerateness

?
Conduct
Pricing behavior
Product strategy and advertising
Research and innovation
Plant investment
Legal tactics

?
Performance
Production and allocative efficiency
Progress
Full employment
Equity
 3

2 Two Sides of a Market

2.1 Comparative Static Analysis
Assume that there are n endogenous variables and m exogenous variables.
Endogenous variables: x1 , x2 , . . . , xn
Exogenous variables: y1 , y2 , . . . , ym .
There should be n equations so that the model can be solved.

F1 (x1 , x2 , . . . , xn ; y1 , y2 , . . . , ym ) = 0
F2 (x1 , x2 , . . . , xn ; y1 , y2 , . . . , ym ) = 0
..
.
Fn (x1 , x2 , . . . , xn ; y1 , y2 , . . . , ym ) = 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:

x1 = x1 (y1 , y2 , . . . , ym )
x2 = x2 (y1 , y2 , . . . , ym )
..
.
xn = xn (y1 , y2 , . . . , ym ).

We use comparative statics method to find the differential relationships between

xi and yj : ∂xi /∂yj . Then we check the sign of ∂xi /∂yj to investigate the causality
relationship between xi and yj .
 4

2.2 Utility Maximization and Demand Function

2.2.1 Single product case
A consumer wants to maximize his/her utility function U = u(Q) + M = u(Q) +
(Y − P Q).
∂U
FOC: = u0 (Q) − P = 0,
∂Q
⇒ u0 (Qd ) = P (inverse demand function)
⇒ Qd = D(P ) (demand function, a behavioral equation)
∂2U dQd
= UP Q = −1 ⇒ = D 0 (P ) < 0, the demand function is a decreasing
∂Q∂P dP
function of price.

2.2.2 Multi-product case

A consumer wants to maximize his utility function subject to his budget constraint:

max U (x1 , . . . , xn ) subj. to p1 x1 + · · · + pn xn = I.

Endogenous variables: x1 , . . . , xn
Exogenous variables: p1 , . . . , pn , I (the consumer is a price taker)
Solution is the demand functions xk = Dk (p1 , . . . , pn , I), k = 1, . . . , n

Example: max U (x1 , x2 ) = a ln x1 + b ln x2 subject to p1 x1 + p2 x2 = m.

L = a ln x1 + b ln x2 + λ(m − p1 x1 − p2 x2 ).
a b
FOC: L1 = − λp1 = 0, L2 = − λp2 = 0 and Lλ = m − p1 x1 − p2 x2 = 0.
x1 x2
a x2 p1 am bm
⇒ = ⇒ x1 = , x2 =
b x1 p2 (a + b)p1 (a + b)p2
 0 −p1 −p2 
 −a 
 ap2 bp2
 −p1 0

SOC:  x12
 = 22 + 21 > 0.

 −b  x1 x2
 −p 0
2
x22
 
am bm
⇒ x1 = , x2 = is a local maximum.
(a + b)p1 (a + b)p2

2.3 Indivisibility, Reservation Price, and Demand Function

In many applications the product is indivisible and every consumer needs at most
one unit.
Reservation price: the value of one unit to a consumer.
If we rank consumers according to their reservation prices, we can derive the market
demand function.

Example: Ui = 31 − i, i = 1, 2, · · · , 30.
 5

Ui , P
rr
rr
6
rr
rr
The trace of Ui ’s becomes
rr the demand curve.
rr
rr
rr
r rr
rr
rr
rr
rr
rr
rr
rr
- i, Q

2.4 Demand Function and Consumer surplus

Demand Function: Q = D(p). Inverse demand function: p = P (Q).
p dQ pD 0 (p) P (Q)
Demand elasticity: ηD ≡ = = .
Q dp D(p) QP (Q)
Total Revenue: T R(Q) = QP (Q) = pD(p).
T R(Q) pD(p)
Average Revenue: AR(Q) = = P (Q) = .
Q D(p)
dT R(Q)
Marginal Revenue: M R(Q) = or
dQ

P 0 (Q)Q
   
0 1
M R(Q) = P (Q) + QP (Q) = P (Q) 1 + = P (Q) 1 + .
P (Q) η
R∞
Consumer surplus: CS(p) ≡ p D(p)dp.

A 1
2.4.1 Linear demand function: Q = D(p) = − p or P (Q) = A − bQ
b b
a
T R = AQ − bQ2 , AR = A − bQ, M R = A − 2bQ, η = 1 − ,
bQ
RA (A − p)p
CS(p) = p D(p)dp = .
2b

2.4.2 Const. elast. demand function: Q = D(p) = apη or P (Q) = AQ1/η

1 1+η
T R = AQ1+ η , AR = AQ1/η ,and M R = AQ1/η .
η

2.4.3 Quasi-linear utility function: U (Q) = f (Q) + m ⇒ P (Q) = f 0 (Q)

T R = Qf 0 (Q), AR = f 0 (Q), M R = f 0 (Q) + Qf 00 (Q),
CS(p) = f (Q) − pQ = f (Q) − Qf 0 (Q).
 6

2.5 Profit maximization and supply function

2.5.1 From cost function to supply function
Consider first the profit maximization problem of a competitive producer:
∂Π
max Π = P Q − C(Q), FOC ⇒ = P − C 0 (Q) = 0.
Q ∂Q
The FOC is the inverse supply function (a behavioral equation) of the producer: P
= C 0 (Q) = MC. Remember that Q is endogenous and P is exogenous here. To find
dQ
the comparative statics , we use the total differential method discussed in the last
dP
chapter:
dQ 1
dP = C 00 (Q)dQ, ⇒ = 00 .
dP C (Q)
dQ ∂2Π
To determine the sign of , we need the SOC, which is 2
= −C 00 (Q) < 0.
dP ∂Q
dQs
Therefore, > 0.
dP

2.5.2 From production function to cost function

A producer’s production technology can be represented by a production function
q = f (x1 , . . . , xn ). Given the prices, the producer maximizes his profits:

max Π(x1 , . . . , xn ; p, p1 , . . . , pn ) = pf (x1 , . . . , xn ) − p1 x1 − · · · − pn xn

Exogenous variables: p, p1 , . . . , pn (the producer is a price taker)

Solution is the supply function q = S(p, p1 , . . . , pn ) and the input demand functions,
xk = Xk (p, p1 , . . . , pn ) k = 1, . . . , n

√ √ √ √
Example: q = f (x1 , x2 ) = 2 x1 + 2 x2 and Π(x1 , x2 ; p, p1 , p2 ) = p(2 x1 + 2 x2 ) −
p 1 x1 − p 2 x2 ,
√ √
max p(2 x1 + 2 x2 ) − p1 x1 − p2 x2
x1 .x2

∂Π p ∂Π p
FOC: = √ − p1 = 0 and = √ − p2 = 0.
∂x1 x1 ∂x2 x2
2 2
⇒ x1 = (p/p1 ) , x2 = (p/p2 ) (input demand functions) and
q = 2(p/p1 ) + 2(p/p2 ) = 2p( p11 + p12 ) (the supply function)
Π = p2 ( p11 + p12 )
SOC:
∂2Π ∂2Π −p
   
0
 ∂x2 ∂x1 ∂x2   2x−3/2 
1 = 1
 ∂2Π ∂2Π   −p
   
0

−3/2
∂x1 ∂x2 ∂x21 2x2
is negative definite.
 7

2.5.3 Joint products, transformation function, and profit maximization

In more general cases, the technology of a producer is represented by a transformation
function: F j (y1j , . . . , ynj ) = 0, where (y1j , . . . , ynj ) is called a production plan, if ykj > 0
(ykj ) then k is an output (input) of j.

Example: a producer produces two outputs, y1 and y2 , using one input y3 . Its
technology is given by the transformation function (y1 )2 + (y2 )2 + y3 = 0. Its profit
is Π = p1 y1 + p2 y2 + p3 y3 . The maximization problem is

max p1 y1 + p2 y2 + p3 y3 subject to (y1 )2 + (y2 )2 + y3 = 0.

y1 ,y2 ,y3

To solve the maximization problem, we can eliminate y3 : x = −y3 = (y1 )2 + (y2 )2 > 0
and
max p1 y1 + p2 y2 − p3 [(y1 )2 + (y2 )2 ].
y1 ,y2

The solution is: y1 = p1 /(2p3 ), y2 = p2 /(2p3 ) (the supply functions of y1 and y2 ), and
x = −y3 = [p1 /(2p3 )]2 + [p1 /(2p3 )]2 (the input demand function for y3 ).

2.6 Production function and returns to scale

∂Q ∂Q
Production function: Q = f (L, K). M PK = M PK =
∂L ∂K
IRTS: f (hL, hK) > hf (L, K). CRTS: f (hL, hK) = hf (L, K).
DRTS: f (hL, hK) < hf (L, K).
∂2Q ∂2Q
Supporting factors: > 0. Substituting factors: < 0.
∂L∂K ∂L∂K
Example 1: Cobb-Douglas case F (L, K) = ALa K b .

Example 2: CES case F (L, K) = A[aLρ + (1 − a)K ρ ]1/ρ .

2.7 Cost function: C(Q)

C(Q)
Total cost T C = C(Q) Average cost AC = Marginal cost M C = C 0 (Q).
Q
Example 1: C(Q) = F + cQ

Example 2: C(Q) = F + cQ + bQ2

Example 3: C(Q) = cQa .

 8

3 Competitive Market
Industry (Market) structure:
Short Run: Number of firms, distribution of market shares, competition decision vari-
ables, reactions to other firms.
Long Run: R&D, entry and exit barriers.

Competition: In the SR, firms and consumers are price takers.

In the LR, there is no barriers to entry and exit ⇒ 0-profit.

3.1 SR market equilibrium

3.1.1 An individual firm’s supply function
A producer i in a competitive market is a price taker. It chooses its quantity to
maximize its profit:

max pQi − Ci (Qi ) ⇒ p = Ci0 (Qi ) ⇒ Qi = Si (p).

Qi

3.1.2 Market supply function

P
Market supply is the sum of individual supply function S(p) = i Si (p).
On the Q-p diagram, it is the horizontal sum of individual supply curves.

p p
6 S1 S2 S 6 S1 S2 S
@
@
@
@
@
@
p∗ @
@
-Q @D - Q
Q∗1 Q∗2 Q ∗

3.2 Market equilibrium

Market equilibrium is determined by the intersection of the supply and demand as in
the diagram.

Formally, suppose there are n firms. A state of the market is a vector (p, Q1 , Q2 , . . . , Qn ).
An equilibrium is a state (p∗ , Q∗1 , Q∗2 , . . . , Q∗n ) such that:
1. D(p∗ ) = S(p∗ ).
2. Each Q∗i maximizes Πi (Qi ) = p∗ Qi − Ci (Qi ), i = 1, . . . , n.
3. Πi (Q∗i ) = p∗ Q∗i − ci (Q∗i ) ≥ 0.
 9

p
3.2.1 Example 1: C1 (Q1 ) = Q21 , C2 (Q2 ) = 2Q22 , D = 12 −
4

p p 3p
p = C10 (Q1 ) = 2Q1 , p = C20 (Q2 ) = 4Q2 , ⇒ S 1 = , S2 = , S(P ) = S1 +S2 = .
2 4 4
p∗ 3p∗
D(p∗ ) = S(p∗ ) ⇒ 12− = ⇒ p∗ = 12, Q∗ = S(p∗ ) = 9, Q∗1 = S1 (p∗ ) = 6, Q∗2 = S2 (p∗ ) = 3.
4 4
p p
6 S1  S2
S 6
@ @
@
@
@
@
 @
@
12 
@ c @ S

@ @

@ @

@ D @ D-

@ -Q @ Q
9 Q∗

3.2.2 Example 2: C(Q) = cQ (CRTS) and D(p) = max{A − bp, 0}

If production technology is CRTS, then the equilibrium market price is determined
by the AC and the equilibrium quantity is determined by the market demand.

p∗ = c, Q∗ = D(p∗ ) = max{A − bp∗ , 0}.

A A
If ≤ c then Q∗ = 0. If > c then Q∗ = A − bc > 0.
b b

3.2.3 Example 3: 2 firms, C1 (Q1 ) = c1 Q1 , C2 (Q2 ) = c2 Q2 , c1 < c2

p∗ = c 1 , Q∗ = Q∗1 = D(p∗ ) = D(c1 ), Q∗2 = 0.

3.2.4 Example 4: C(Q) = F + cQ or C 00 (Q) > 0 (IRTS), no equilibrium

If C 00 (Q) > 0, then the profit maximization problem has no solution.
If C(Q) = F + cQ, then p∗ = c cannot be and equilibrium because
Π(Q) = cQ − (F + cQ) = −F < 0.
 10

3.3 General competitive equilibrium

Commodity space: Assume that there are n commodities. The commodity space is
n
R+ = {(x1 , . . . , xn ); xk ≥ 0}

Economy: There are I consumers, J producers, with initial endowments of com-

modities ω = (ω1 , . . . , ωn ).
Consumer i has a utility function U i (xi1 , . . . , xin ), i = 1, . . . , I.
Producer j has a production transformation function F j (y1j , . . . , ynj ) = 0,

A price system: (p1 , . . . , pn ).

A private ownership economy: Endowments and firms (producers) are owned by

consumers.
Consumer i’s endowment is ω i = (ω1i , . . . , ωni ), Ii=1 ω i = ω.
P

Consumer i’s share of firm j is θ ij ≥ 0, Ii=1 θ ij = 1.

P

An allocation: xi = (xi1 , . . . , xin ), i = 1, . . . , I, and y j = (y1j , . . . , ynj ), j = 1, . . . , J.

A competitive equilibrium:
A combination of a price system p̄ = (p̄1 , . . . , p̄n ) and an allocation ({x̄i }i=1,...,I , {ȳ j }j=1,...,J )
such
Pthat
1. i x̄i = ω + j ȳ j (feasibility condition).
P

2. ȳ j maximizes Πj , j = 1, . . . , J and x̄i maximizes U i , subject to i’s budget con-

straint p1 xi1 + . . . + pn xin = p1 ω11 + . . . + pn ωni + θi1 Π1 + . . . + θiJ ΠJ .

Existence Theorem:
Suppose that the utility functions are all quasi-concave 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 non-substitution 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.
 11

4 Monopoly
A monopoly industry consists of one single producer who is a price setter (aware of
its monopoly power to control market price).

4.1 Monopoly profit maximization

Let the market demand of a monopoly be Q = D(P ) with inverse function P = f (Q).
Its total cost is TC = C(Q). The profit maximization problem is

max π(Q) = P Q−T C = f (Q)Q−C(Q) ⇒ f 0 (Q)Q+f (Q) = MR(Q) = MC(Q) = C 0 (Q) ⇒ QM .

Q≥0

d2 π
The SOC is = MR0 (Q) − MC0 (Q) < 0.
dQ2
Long-run existence condition: π(Qm ) ≥ 0.

Example: TC(Q) = F + cQ2 , f (Q) = a − bQ, ⇒ MC = 2cQ, MR = a − 2bQ.

a a(b + 2c) a2
⇒ Qm = , Pm = , ⇒ π(Qm ) = − F.
2(b + c) 2(b + c) 4(b + c)
a2
When < F , the true solution is Qm = 0 and the market does not exist.
4(b + c)
P
6
@
A
A@ MC
Pm AA@@
A @
MC A @
A @
A @
AMR @D - Q
Qm

4.1.1 Lerner index

The maximization can be solved using P as independent variable:

max π(P ) = P Q − T C = P D(P ) − C(D(P ))

P ≥0

Pm − C 0 D(P ) 1
⇒ D(P ) + P D 0 (P ) = C 0 (D(P ))D 0 (P ) ⇒ =− 0 = .
Pm D (P )P ||
Pm − C 0
Lerner index: . It can be calculated from real data for a firm (not necessarily
Pm
monopoly) or an industry. It measures the profit per dollar sale of a firm (or an
industry).
 12

4.1.2 Monopoly and social welfare

P P P
6 6 6
@
A @ @
A
A@ MC @ MC=S A@ MC
Pm AA@@ @
@
A @
A @
A @ P∗ @ A @
A @ @ A @
MC A @ @ A @
A @ @ A @
AMR @D - Q @D - Q AMR @D - Q
Qm Q∗ Qm Q∗

4.1.3 Rent seeking (¥) activities

R&D, Bribes, Persuasive advertising, Excess capacity to discourage entry, Lobby
expense, Over doing R&D, etc are means taken by firms to secure and/or maintain
their monopoly profits. They are called rent seeking activities because monopoly
profit is similar to land rent. They are in many cases regarded as wastes because they
don’t contribute to improving productivities.

4.2 Monopoly price discrimination

Indiscriminate Pricing: The same price is charged for every unit of a product sold to
any consumer.
Third degree price discrimination: Different prices are set for different consumers, but
the same price is charged for every unit sold to the same consumer (linear pricing).
Second degree price discrimination: Different price is charged for different units sold
to the same consumer (nonlinear pricing). But the same price schedule is set for
different consumers.
First degree price discrimination: Different price is charged for different units sold to
the same consumer (nonlinear pricing). In addition, different price schedules are set
for different consumers.

4.2.1 Third degree price discrimination

Assume that 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 f10 (q1 ) − C 0 (q1 + q2 ) = 0, Π2 = f2 (q2 ) + q2 f20 (q2 ) − C 0 (q1 + q2 ) = 0;
or MR1 = MR2 = MC.  
0 00 00
 2f10 + q1 f100 − C 00 −C 00 
SOC: Π11 = 2f1 + q1 f1 − C < 0,    ≡ ∆ > 0.
−C 00 2f20 + q2 f200 − C 00 
Example: f1 = a − bq1 , f2 = α − βq2 , and C(Q) = 0.5Q2 = 0.5(q1 + q2 )2 .
f10 = −b, f20 = −β, f100 = f200 = 0, C 0 = Q = q1 + q2 , and C 00 = 1.
 13

    
1 + 2b 1 q1 a
FOC: a − 2bq1 = q1 + q2 = α − 2βq2 ⇒ =
   1 1+ 2β q 2 α
q1 1 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
6 MC
MC = MR1+2 ⇒ Qm , MC∗

aa
Q MC∗ = MR1 ⇒ q1∗
@ Qa
@Qaaa
MC∗ @QQ aa MC∗ = MR2 ⇒ q2∗
@ Q aa
Q aa
@ Q aa
@ Q a MR1+2
@ Q
Q
@ Q
@ Q
Q MR2
@ MR1 Q- q
q1∗ q2∗ Qm

4.2.2 First Degree

Each consumer is charged according to his total utility, i.e., T R = P Q = U (Q). The
total profit to the monopoly is Π(Q) = U (Q) − C(Q). The FOC is U 0 (Q) = C 0 (Q),
i.e., the monopoly regards a consumer’s MU (U 0 (Q)) curve as its MR curve and
maximizes its profit.

max Π = U (Q) − C(Q) ⇒ U 0 (Q) = C 0 (Q).

Q

The profit maximizing quantity is the same as the competition case, Qm1 = Q∗ . How-
U (Q∗ )
ever, the price is much higher, Pm1 = = AU > P ∗ = U 0 (Q∗ ). There is no
Q∗
inefficiency. But there is social justice problem.

P
6
H
@
A HH
r
A@ HH
Pm1 AA@@ HH HH AU
A @
P∗ A @ C 0 (Q)
A @
AMR @ D = M U
A @ -Q
Qm Q∗ = Qm1

4.2.3 Second degree discrimination

See Varian Ch14 or Ch25.3 (under).
 14

P P P
6 6 6
D2 D2 D2
Q Q Q
Q Q Q
Q Q Q
D1 Q D1 Q D1 Q
@ Q @ Q @ Q
@ B Q @ Q @ Q
Q Q Q
@ Q @ Q @ Q
A @ C Q @ Q @ Q
@ Q
Q- Q @ QQ- @ QQ
-Q
Q
By self selection principle, P1 Q1 = A, P2 Q2 = A + C, Π = 2A + C is maximized when
Q1 is such that the hight of D2 is twice that of D1 .

4.3 Multiplant Monopoly and Cartel

Now consider the case that a monopoly has two plants.
Cost functions: TC1 = C1 (q1 ) and TC2 = C2 (q2 )
Inverse market demand: P = D(Q) = D(q1 + q2 )
Profit function: Π(q1 , q2 ) = P (q1 + q2 ) − C1 (q1 ) − C2 (q2 ) = D(q1 + q2 )(q1 + q2 ) −
C1 (q1 ) − C2 (q2 )
FOC: Π1 = D 0 (Q)Q + D(Q) − C10 (q1 ) = 0, Π2 = D 0 (Q)Q + D(Q) − C20 (q2 ) = 0;
orMR = MC1 = MC2 .
0
SOC:0 Π11 = 2D (Q) + D 00 (Q)Q − C100 < 0, 
 2D (Q) + D 00 (Q)Q − C100 2D 0
(Q) + D 00
(Q)Q 
00  ≡ ∆ > 0.
 0 00 0 00

 2D (Q) + D (Q)Q 2D (Q) + D (Q)Q − C1
Example: D(Q) = A − Q, C1 (q1 ) = q12 , and C2 (q2 ) = 2q22 .     
4 2 q1 A
FOC: M R = A − 2(q1 + q2 ) = M C1 = 2q1 = M C2 = 4q2 . =
2 6 q2 A
⇒ q1 = 0.2A, q2 = 0.1A, Pm = 0.7A
p
6
MR = MC1+2 ⇒ Qm , MR∗
MC2 MC1 MC1+2
@
A 

MR∗ = MC1 ⇒ q1∗
A@

pm A @
 MR∗ = MC2 ⇒ q2∗
A
@ 
 A
@
MR∗ 

A A
@
@

 A @

 A @


A MR @D -q
q2∗ q1∗ Qm
 15

4.4 Multiproduct monopoly

Consider a producer who is monopoly (the only seller) in two joint products.

Q1 = D1 (P1 , P2 ), Q2 = D2 (P1 , P2 ), TC = C(Q1 , Q2 ).

The profit as a function of (P1 , P2 ) is

Π(P1 , P2 ) = P1 D1 (P1 , P2 ) + P2 D2 (P1 , P2 ) − C(D1 (P1 , P2 ), D2 (P1 , P2 )).

Maximizing Π(P1 , P2 ) w. r. t. P1 , we have

∂D1 ∂D2 ∂C ∂D1 ∂C ∂D2

D1 (P1 , P2 ) + P1 + P2 − − = 0,
∂P1 ∂P1 ∂Q1 ∂P1 ∂Q2 ∂P1
P1 − MC1 1 P2 − MC2 TR2 21
⇒ = +
P1 |11 | P2 TR1 |11 |
P1 − MC1 1
Case 1: 12 > 0, goods 1 and 2 are substitutes, > .
P1 |11 |
P1 − MC1 1
Case 2: 12 < 0, goods 1 and 2 are complements, < .
P1 |11 |
Actually, both P1 and P2 are endogenous and have to be solved simultaneously.
 R2 21  P1 − C10   1 
1 −
R1 |11 |    |11 |
 P P 1
0  = 
 
 −R1 12
2 − C 2
1 
1
R2 |22 | P2 |22 |

P1 − C10 R2
   
1 |22 | + 21
⇒ P P 1 = R 1
.
   
0 
2 − C2 11 22 − 12 21 | | + R1 

11 12
P2 R2

4.4.1 2-period model with goodwill (Tirole EX 1.5)

∂D2
Assume that Q2 = D2 (p2 ; p1 ) and < 0, ie., if p1 is cheap, the monopoy gains
∂p1
goodwill in t = 2.

max p1 D1 (p1 ) − C1 (D1 (p1 )) + δ[p2 D2 (p2 ; p1 ) − C2 (D2 (p2 ; p1 ))].

p1 ,p2

4.4.2 2-period model with learning by doing (Tirole EX 1.6)

∂C2
Assume that TC2 = C2 (Q2 ; Q1 ) and < 0, ie., if Q1 is higher, the monopoy gains
∂Q1
more experience in t = 2.

max p1 D1 (p1 ) − C1 (D1 (p1 )) + δ[p2 D2 (p2 ) − C2 (D2 (p2 ), D1 (p1 ))].
p1 ,p2
 16

Continuous time (Tirole EX 1.7):

Z ∞ Z t
max [R(qt ) − C(wt )qt ]e−rt dt, wt = qτ dτ,
qt ,wt 0 0

where R(qt ) is the revenue at t,R0 > 0, R00 < 0, r is the interest rate, Ct = C(wt ) is
the unit production cost at t, C 0 < 0, and wt is the experience accumulated by t.
√ 1
Example: R(q) = q and C(w) = a + .
w

4.5 Durable good monopoly

Flow (perishable) goods: ‡
Durable goods: ˝‹

Coase (1972:) A durable good monopoly is essentially different from a perishable

good monopoly.

Perishable goods: .°v‚ ÒÖ , ©‚·b½h˛.

Durable goods: .°v‚ Ò.Ö , ¥‚˛-‚ÿ..y, ÝBªJž“.

4.5.1 A two-period model

There are 100 potential buyers of a durable good, say cars. The value of the service
of a car to consumer i each period is Ui = 101 − i, i = 1, . . . , 100.
Assume that MC = 0 and 0 < δ < 1 is the discount rate.
Ui , P
rr
rr
6
rr
rr
The trace of Ui ’s becomes
rr the demand curve.
rr
rr
rr
r rr
rr
rr
rr
rr
rr
rr
rr
- i, Q

ù»j¶: 1. É•.“. P1R , P2R are the rents in periods 1 and 2.

2. “i. P1s , P2s are the prices of a car in periods 1 and 2.

4.5.2 É•.“
The monopoly faces the same demand function P = 100 − Q in each period. The
monopoly profit maximization implies that MR = 100 − 2Q = 0. Therefore,

P1R = P2R = 50, π1R = π2R = 2500, ΠR = π1R + δπ2R = 2500(1 + δ),

where δ < 1 is the discounting factor.

 17

4.5.3 “i
We use backward induction method to find the solution to the profit maximization
problem. We first assume that those consumers who buy in period t = 1 do not resale
their used cars to other consumers.

Suppose that q1s = q̄1 , ⇒ the demand in period t = 2 becomes

q2s = 100 − q̄1 − P2s , ⇒ P2s = 100 − q̄1 − q2 and
MR2 = 100 − q̄1 − 2q2 = 0, q2s = 50 − 0.5q̄1 = P2s , π2s = (100 − q̄1 )2 /4.
Now we are going to calculate the location of the marginal consumer q̄1 who is indif-
ferent between buying in t = 1 and buying in t = 2.
(1 + δ)(100 − q̄1 ) − P1s = δ[(100 − q̄1 ) − P2s ] ⇒ (100 − q̄1 ) − P1s = −δP2s ,
⇒ P1s = 100 − q̄1 + δP2s = (1 + 0.5δ)(100 − q̄1 ).
max Πs = π1s + δπ2s = q1 P1s + δ(50 − 0.5q1 )2 = (1 + 0.5δ)q1 (100 − q1 ) + 0.25(100 − q1 )2 ,
q1

FOC⇒ q1s = 200/(4+δ), P1s = 50(2+δ)2 /(4+δ), Πs = 2500(2+δ)2 /(4+δ) < ΠR = (1+δ)2500.
When a monopoly firm sells a durable good in t = 1 instead of leasing it, the monopoly
loses some of its monopoly power, that is why Πs < ΠR .

4.5.4 Coase problem

Sales in t will reduce monopoly power in the future. Therefore, a rational expectation
consumer will wait.

Coase conjecture (1972): In the ∞ horizon case, if δ→1 or ∆t→0, then the monopoly
profit Πs →0.

The conjecture was proved in different versions by Stokey (1981), Bulow (1982), Gul,
Sonnenschein, and Wilson (1986).

Tirole EX 1.8.
1. A monopoly is the only producer of a durable good in t = 1, 2, 3, . . .. If (q1 , q2 , q3 , . . .)
and (p1 , p2 , p3 , . . .) are the quantity and price sequences for the monopoly product,
the profit is
∞
X
Π= δ t p t qt .
t=1

2. There is a continuum of consumers indexed by α ∈ [0, 1], each needs 1 unit of the
durable good.
α
vα = α + δα + δ 2 α + . . . = : The utility of the durable good to consumer α.
1−δ
If consumer α purchases the good at t, his consumer surplus is
α
δ t (vα − pt ) = δ t ( − pt ).
1−δ
 18

3. A linear stationary equilibrium is a pair (λ, µ), 0 < λ, µ < 1, such that
(a) If vα > λpt , then consumer α will buy in t if he does not buy before t.
(b) If at t, all consumers with vα > v (vα < v) have purchased (not purchased) the
durable good, then the monopoly charges pt = µv.
(c) The purchasing strategy of (a) maximizes consumer α’s consumer surplus, given
the pricing strategy (b).
(d) The pricing strategy of (b) maximizes the monopoly profits, given the purchacing
strategy (a).

The equilibrium is derived in Tirole as

1 √
λ= √ , µ = [ 1 − δ − (1 − δ)]/δ, lim λ = ∞, lim µ = 0.
1−δ δ→1 δ→1

One way a monopoy of a durable good can avoid Coase problem is price commitment.
By convincing the consumers that the price is not going to be reduced in the future,
it can make the same amount of profit as in the rent case. However, the commitment
equilibrium is not subgame perfect. Another way is to make the product less durable.

4.6 Product Selection, Quality, and Advertising

Tirole, CH2.
Product space, Vertical differentiation, Horizontal differentiation.
Goods-Characteristics Approach, Hedonic prices.
Traditional Consumer-Theory Approach.

4.6.1 Product quality selection, Tirole 2.2.1, pp.100-4.

Inverse Demand: p = P (q, s), where s is the quality of the product.
Total cost: TC = C(q, s), Cq > 0, Cs > 0.

Social planner’s problem:

Z q
max W (q, s) = P (x, s)ds − C(q, s),
q,s 0
Z q
FOC: (1) Wq = P (q, s) − Cq = 0, (2) Ws = Ps (x, s)dx − Cs = 0.
0

(1) P = MC,
1Rq
(2) Ps dx = Cs /q: Average marginal valuation of quality should be equal to the
q 0
marginal cost of quality per unit.

Monopoly profit maximization:

max Π(q, s) = qP (x, s) − C(q, s), FOC Πq = MR − Cq = 0, Πs = qPs (x, s) − Cs = 0.

q,s
 19

Ps = Cs /q: Marginal consumer’s marginal valuation of quality should be equal to the

marginal cost of quality per unit.

Example 1: P (q, s) = f (q) + s, C(q, s) = sq, ⇒ Ps = 1, Cs = q, no distortion.

Example 2: There is one unit of consumers indexed by x ∈ [0, x̄]. U = xs − P , F (x)

1Rq
is the distribution function of x. ⇒ P (q, s) = sF −1 (1 − q) ⇒ Ps dx ≥ Ps (q, s),
q 0
monopoly underprovides quality.

Example 3: U = x + (α − x)s − P , x ∈ [0, α], F (x) is the distribution function

1 Rq
of x ⇒ P (q, s) = αs + (1 − s)F −1 (1 − q) ⇒ Ps dx ≤ Ps (q, s), monopoly overpro-
q 0
vides quality.
 20

5 Basis of Game Theory

In this part, 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 3-person 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?

5.1 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:
2-person games – conflict and cooperation possibilities.
n-person games – coalition formation (¯ó©d) possibilities in addition.
infinite-players’ 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,
 21

payoff functions can be represented by matrices.

infinite – strategy (continuous or discontinuous payoff functions) games like
duopoly games.

3. according to sum of payoffs:

0-sum games – conflict is unavoidable.
non-zero sum games – possibilities for cooperation.

4. according to preplay negotiation possibility:

non-cooperative games – each person makes unilateral decisions.
cooperative games – players form coalitions and decide the redistribution of
aggregate payoffs.

5.2 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 point).
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: S1 = {s1 , . . . , sm }, S2 = {σ1 , . . . , σn }.

2. Payoff matrices: π1 (si , σj ) = aij , π2 (si , σj ) = bij . A = [aij ], B = [bij ].

Normal form:
@ II
@
I @ σ1 ... σn
s1 (a11 , b11 ) ... (a1n , b1n )
.. .. .. ..
. . . .
sm (am1 , bm1 ) . . . (amn , bmn )
 22

5.3 Examples


Example 1: A perfect information game

 Q R
L 
1
2 Q
 L 
Q2
l @ r @ R
@ @

II
    
1 9 3 8 @
9 6 7 2 @
I @ Ll Rl Lr Rr
L (1,9) (1,9) (9,6) (9,6)
S1 = { L, R }, S2 = { Ll, Lr, Rl, Rr }. R (3,7)* (8,2) (3,7) (8,2)

Example 2: Prisoners’ dilemma game
L 
1
  
Q R
Q

@ 2 Q

@
L @R L @R

II
    
4 0 5 1 @
4 5 0 1 @
I @ L R
L (4,4) (0,5)
S1 = { L, R }, S2 = { L, R }. R (5,0) (1,1)*


Example 3: Hijack game
L 
Q R
1
 Q
 L 
 Q2
@ R
@

−1
2 
2
 
−10

II
@
−2 −10 @
I @ L R
L (-1,2) (-1,2)*
S1 = { L, R }, S2 = { L, R }. R (2,-2)* (-10,-10)

stock price manipulation game

Example 4: A simplified
 1/2 
0
HH1/2
1 H 1
L  l
H

@
AR
@ r II
@
 L

A 
A

A 2 @ @
R L R I @ L R

A 
A 
 
4 3 Ll (4, 3.5) (4, 2)
2 7 5 4 4 7 Lr (3.5, 4.5) (3.5, 4.5)
5 7 5 2
Rl (5.5, 5)* (4.5, 4.5)
S1 = { Ll, Lr, Rl, Rr }, S2 = { L, R }. 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.
 23

5.4 Strategy pair and pure strategy Nash equilibrium

1. A Strategy Pair: (si , σj ). Given a strategy pair, there corresponds a payoff pair
(aij , bij ).

2. A Nash equilibrium: A strategy pair (si∗ , σj∗ ) such that ai∗j∗ ≥ aij∗ and bi∗j∗ ≥
bi∗j for all (i, j). Therefore, there is no incentives for each player to deviate from
the equilibrium strategy. ai∗j∗ and bi∗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.

5.5 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.

5.5.1 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.
 24

5.5.2 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 .
A behavior strategy: A function which assigns a local strategy for each u ∈ Ui .
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 non-perfect 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
 HH
 1/2  HH1/2  A 2-person 0-sum non-perfect recall game.
1 1 1 1
 HH2
 
u ∗ ∗
11 NE is (µ 1 , µ 2 ) = ( ac ⊕ bd, A ⊕ B).
a @ b @ 2 2 2 2
@ A
 @B µ∗1 is not a behavioral strategy.


A
@ u12 @
 
A  
1 c
Ad c
Ad −1
−1
A
A 1
     
2 0 −2 0
−2 0 2 0
5.5.3 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.

Equivalent transformation:
(1) Inflation-Deflation;
 

   
1 1
 
 
A A
r r
 A  A
 
2 A 2 A
A
@ A @ A
 @   @ A 


A 

A
1 @ A1 @ 1 A

A
A
A
A

A
A
A
A
A
A

A
A
A
A
A
A
 25

(2) Addition of superfluous move;

 
 
1 1
  HH

A
  HH 
  HH
r  r
A 

@
2 A 2
A
@ A @
 @   @   @ 


A 

A
2 @
1 @ A2 1 @

A
A
A
A
A

A
A
A
A
A
A
A

A
A
A
A
A
A
A

(3) Coalesing of moves;

 
 
1 1
 
 
A EA
 
r  r 
A EA
 
2 A 2 E A
A
@ A @ E A
 @   @  E A


A 

A
1 @ A1 1 @ E A

A
A
A E A

A
A
A
A
A E A

A
A
A
A
A E A

(4) Interchange of moves.

 

   
1 1
 
 
 A  A
r r
 A  A
 
2 A 1 A
A A
@ A @ A
 @   @ 


A 

A 
1 @ A1 2 @ A1

A
A
A
A

A
A
A
A
A
A

A
A
A
A
A
A

5.6 Continuous games and the duopoly game

In many applications, S1 and S2 are infinite subsets of Rm and Rn 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
Π1 = Π1 (x; y) and Π2 = Π2 (x; y), where x ∈ S1 ⊂ Rm and y ∈ S2 ⊂ Rn .
To simplify the presentation, assume that m = n = 1. A strategic pair is (x, y) ∈
S1 × S2 . A Nash equilibrium is a pair (x∗ , y ∗ ) such that
Π1 (x∗ , y ∗ ) ≥ Π1 (x, y ∗ ) and Π2 (x∗ , y ∗ ) ≥ Π2 (x∗ , y) for all x ∈ S1 y ∈ S2 .
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).
 26

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 ∈ S1 to maximize his objective function
Π1 . This relationship defines a behavioral equation x = R1 (y) which can be obtained
by solving the FOC for player 1, Π1x (x; y) = 0. Similarly, player 2 regards y as en-
dogenous and x exogenous and wants to maximize Π2 for a given x chosen by player
1. Player 2’s reaction function (behavioral equation) y = R 2 (x) is obtained by solving
Π2y (x; y) = 0. A Nash equilibrium is an intersection of the two reaction functions.
The FOC for a Nash equilibrium is given by Π1x (x∗ ; y ∗ ) = 0 and Π2y (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 c1 and c2 , respectively.
Assume that each firm regards the other firm’s output as given (not affected by his
output quantity).
The situation defines a 2-person game as follows: Each firm i controls his own output
quantity qi . (q1 , q2 ) together determine the market price P = a − (q1 + q2 ) which in
turn determines the profit of each firm:
Π1 (q1 , q2 ) = (P −c1 )q1 = (a−c1 −q1 −q2 )q1 and Π2 (q1 , 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.
The reaction functions are q1 = 0.5(a − c1 − q2 ) and q2 = 0.5(a − c2 − q1 ).
The Cournot Nash equilibrium is (q1∗ , q2∗ ) = ((a − 2c1 + c2 )/3, (a − 2c2 + c1 )/3) with
P ∗ = (a + c1 + c2 )/3. (We have to assume that a − 2c1 + c2 , a − 2c2 + c1 ≥ 0.)

5.7 2-person 0-sum game

1. B = −A so that aij + bij = 0.
2. Maxmin strategy: If player 1 plays si , then the minimum he will have is minj aij ,
called the security level of strategy si . 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. Similarly, since bij = −aij , Player 2
chooses σj ∗ so that maxi aij∗ ≤ maxi aij for all j.
3. Saddle point: If ai∗j∗ = maxi minj aij = minj maxi aij , then (si∗ , σj∗ ) is called a
saddle point. If a saddle point exists, then it is a Nash equilibrium.
   
2 1 4 1 0
A1 = A2 =
−1 0 6 0 1
In example A1 , maxi minj aij = minj maxi aij = 1 (s1 , σ2 ) is a saddle point and
hence a Nash equilibrium. In A2 , maxi minj aij = 0 6= minj maxi aij = 1 and no
saddle point exists. If there is no saddle points, then there is no pure strategy
equilibrium.
 27

4. Mixed strategy for player i: A probability distribution over Si . p = (p1 , . . . , pm ),

q = (q1 , . . . , qn )0 . (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.
X X
t(p) ≡ min{ pi ai1 , . . . , pi ain }.
j
i i

The problem of finding the maxmin mixed strategy (to find p∗ to maximize
t(p)) can be stated as
X X X
max t subj. to pi ai1 ≥ t, . . . , pi ain ≥ t, pi = 1.
p
i i i

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 yi ≡ pP i /t(p) and
replace the problemPof max t(p) withP the problem of min 1/t(p) = i yi . The
constraints become i yi ai1 ≥ 1, . . . , i yi ain ≥ 1.
X X
min y1 + . . . + ym subj. to yi ai1 ≥ 1, . . . , yi ain ≥ 1
y1 ,...,ym ≥0
i i

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.
 
1 0
Example (tossing coin game): A = .
0 1
To find player 2’s equilibrium mixed strategy, we solve the linear programming prob-
lem:
max x1 + x2 subj. to x1 ≤ 1 x2 ≤ 1.
x1 ,x2 ≥0

The solution is x1 = x2 = 1 and therefore the equilibrium strategy for player 2 is

q1∗ = q2∗ = 0.5.
x2 y2
6 6

@ @
@ @
@r @r
@ @
1 1
@ @
@ @
@ @
@ - x1 @ - y1
1 1
 28

Player 1’s equilibrium mixed strategy is obtained by solving the dual to the linear
programming problem:

min y1 + y2 subj. to y1 ≥ 1 y2 ≥ 1.
y1 ,y2 ≥0

The solution is p∗1 = p∗2 = 0.5.

mixed strategy equilibria for non-zero sum games

The idea of a mixed strategy equilibrium is also applicable to a non-zero sum game.
Similar to the simplex algorism for the 0-sum games, there is a Lemke algorism.

Example (Game of Chicken)

S 
1
  
Q N
Q

@ 2 Q
 @ II
S @
N S @@N I @
@
Swerve Don’t
Swerve (0,0) (-3,3)*
    
0 −3 3 −9
0 3 −3 −9 Don’t (3,-3)* (-9,-9)

S1 = { S, N }, S2 = { S, N }.

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 Π1 (S) = 0q + (−3)(1 − q). If he plays
N , his expected payoff is Π1 (N ) = 3q + (−9)(1 − q). For a mixed strategy NE,
Π1 (S) = Π1 (N ), therefore, q = 32 .
The mixed strategy is symmetrical: (p∗1 , p∗2 ) = (q1∗ , q2∗ ) = ( 23 , 13 ).

5.8 Cooperative Game and Characteristic form

2-person 0-sum 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.

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.
 29

Consider the case of a 3-person 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 deter-
mined by 8 values v(φ), v(1), v(2), v(3), v(12), v(13), v(23), v(123).
Super-additivity: If A ∩ B = φ, then v(A ∪ B) ≥ v(A) + v(B).
An imputation is a payoff distribution (x1 , x2 , x3 ).
Individual rationality: P xi ≥ v(i).
Group rationality: i∈S xi ≥ 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
X |S|!(n − |S| − 1)!
πi = [v(S ∪ i) − v(S)].
S⊂N
n!

Example: v(φ) = v(1) = v(2) = v(3) = 0, v(12) = v(13) = v(23) = 0.5, v(123) = 1.
C = {(x1 , x2 , x3 ), xi ≥ 0, xi + xj ≥ 0.5, x1 + x2 + x3 = 1}. Both (0.3, 0.3, 0.4) and
(0.2, 0.4, 0.4) are in C.
The Shapley values are (π1 , π2 , π3 ) = ( 13 , 13 , 13 ).

Remark 1: The core of a game can be empty. However, the Shapley values are
uniquely determined.
Remark 2: Another related concept is the von-Neumann Morgenstern solution. See
CH 6 of Intriligator’s Mathematical Optimization and Economic Theory for the mo-
tivations of these concepts.

5.9 The Nash bargaining solution for a nontransferable 2-person cooper-

ative game
In a nontransferable cooperative game, after-play redistributions of payoffs are im-
possible and therefore the concepts of core and Shapley values are not suitable. For
the case of 2-person games, the concept of Nash bargaining solutions are useful.
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. Ti 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:
x2
6

max (x1 − T1 )(x2 − T2 )

(x1 ,x2 )∈F

x∗2

T2
- x1
T1 x∗1
 30

See CH 6 of Intriligator’s book for the motivations of the solution concept.

5.10 Problems
1. Consider the following two-person 0-sum game:
I \ II σ 1 σ2 σ3
s1 4 3 -2
s2 3 4 10
s3 7 6 8

(a) Find the max min strategy of player I smax min and the min max strategy
of player II σmin max .
(b) Is the strategy pair (smax min , σmin max ) a Nash equilibrium of the game?
(c) What are the equilibrium payoffs?

2. Find the maxmin strategy (smax min ) and the minmax strategy (σmin max ) of the
following two-person 0-sum game:
I \ II σ1 σ2
s1 -3 6
s2 8 -2
s3 6 3
Is the strategy pair (smax min , σmin 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 two-person 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(q1 +q2 ) and that both have the same constant marginal
costs c. For a given quantity pair (q1 , q2 ), the profits are Πi = qi (P − c) =
qi (a − bq1 − bq2 − c), i = 1, 2. Find the Cournot Nash equilibrium output of
each firm.

5. Suppose that in a two-person cooperative game without side payments, if the

two players reach an agreement, they can get (Π1 , Π2 ) such that Π21 + Π2 = 47
and if no agreement is reached, player 1 will get T1 = 3 and player 2 will get
T2 = 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?
 31

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 singer-piano duo, $ 650 the piano drums duo, and $ 300 the piano
alone. The night club is not interested in any other combination. Howeover,
the singer-drums 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.
 32

6 Duopoly and Oligopoly–Homogeneous products

6.1 Cournot Market Structure
2 Sellers producing a homogenous product.
TCi (qi ) = ci qi , i = 1, 2.
P (Q) = a − bQ, a, b > 0, a > maxi ci , Q = q1 + q2 .

Simultaneous move: both firms choose (q1 , q2 ) simultaneously.

π1 (q1 , q2 ) = P (Q)q1 − c1 q1 = (a − bq1 − bq2 )q1 − c1 q1 ,
π2 (q1 , q2 ) = P (Q)q2 − c2 q2 = (a − bq1 − bq2 )q2 − c2 q2 .

Definition of a Cournot equilibrium: {P c , q1c , q2c } such that P c = P (Qc ) = a−b(q1c +q2c )
and
π1 (q1c , q2c ) ≥ π1 (q1 , q2c ), π2 (q1c , q2c ) ≥ π2 (q1c , q2 ), ∀(q1 , q2 ).
The first order conditions (FOC) are

∂π1 ∂π2
= a − 2bq1 − bq2 − c1 = 0, = a − bq1 − 2bq2 − c2 = 0.
∂q1 ∂q2
In matrix form,
      c   
2b b q1 a − c1 q1 1 a − 2c1 + c2
= , ⇒ = .
b 2b q2 a − c2 q2c 3b a − 2c2 + c1
 c 
(a − 2c1 + c2 )2
 
c 2a − c1 − c2 c a + c 1 + c2 π1 1
Q = , P = , = .
3b 3 π2c 9b (a − 2c2 + c1 )2
If c1 ↓ (say, due to R&D), then q1c ↑, q2c ↓, Qc ↑, P c ↓, π1c ↑, π2c ↓.

6.1.1 Reaction function and diagrammatic solution

From FOC, we can derive the reaction functions:
a − c1 a − c2
q1 = − 0.5q2 ≡ R1 (q2 ), q2 = − 0.5q1 ≡ R2 (q1 ).
2b 2b
q2
6
A
A R1 (q2 )
A
A
A
HH A
HA
q2c HAH
A HH R2 (q1 )
A HH -
q1
q1c
 33

6.1.2 N -seller case

N sellers, MCi = ci , i = 1, . . . , N , P = P (Q) = a − bQ = a − b N
P
j=1 qj .
N
!
X
πi (q1 , . . . , qN ) = P (Q)qi − ci qi = a − b qj qi − c i qi .
j=1

FOC is
∂πi X X X
= a−b qj − bqi − ci = P − bqi − ci = 0, ⇒ N a − (N + 1)b qj − cj = 0,
∂qi j j j
P P P
X N a − j cj a + j cj c P − c i a + j cj − (N + 1)ci
⇒ Qc = qjc = , Pc = , qi = = .
j
(N + 1)b N +1 b b(N + 1)
Symmetric case ci = c:
a−c a + Nc N a−c
qic = , Pc = , Qc = .
(N + 1)b N +1 N +1 b
When N = 1, it is the monopoly case.
a−c
As N → ∞, (P c , Qc ) → (c, ), the competition case.
b

6.1.3 Welfare analysis for the symmetric case

(a − P )Q P
Consumer surplus CS = , Social welfare W = CS + j πj .
2
For the symmetric case, CS as functions of N are
     2
c 1 a + Nc N a−c 1 N
CS (N ) = a− = (a − c)2 .
2 N +1 N +1 b 2b N + 1
The sum of profits is
X X a−c N a−c N (a − c)2
πj = (P c − c)qj = = 2
.
j j
N + 1 N + 1 b (N + 1) b

c c
X (a − c)2 N + 0.5N 2
W (N ) = CS (N ) + πj = .
j
b (N + 1)2
(a − c)2 X
lim W c (N ) = lim CSc (N ) = , lim πj = 0.
N →∞ N →∞ 2b N →∞
j
P
6
@
@
@
CS @@
Pc @
π1 π2 π3 @
c @
@
@- Q, qi
q1 Qc
 34

6.2 Sequential moves – Stackelberg equilibrium

Consider now that the two firms move sequentially. At t = 1 firm 1 chooses q1 . At
t = 2 firm 2 chooses q2 .

Firm 1 – leader, Firm 2 – follower.

The consequence is that when choosing q2 , firm 2 already knows what q1 is. On
the other hand, in deciding the quantity q1 , firm 1 takes into consideration firm 2’s
possible reaction, i.e., firm 1 assumes that q2 = R2 (q1 ). This is the idea of backward
induction and the equilibrium derived is a subgame perfect Nash equilibrium.
At t = 1, firm 1 chooses q1 to maximize the (expected) profit π1 = π1 (q1 , R2 (q1 )):
a − c2
max[a−b(q1 +R2 (q1 ))]q1 −c1 q1 = [a−b(q1 + −0.5q1 )]q1 −c1 q1 = 0.5(a+c2 −bq1 )q1 −c1 q1 .
q1 2b
The FOC (interior solution) is

dπ1 ∂π1 ∂π1 dR2

= + = 0.5(a − 2c1 + c2 − 2bq1 ) = 0,
dq1 ∂q1 ∂q2 dq1
a − 2c1 + c2 a + 2c1 − 3c2
⇒ q1s = > q1c , q2s = < q2c .
2b 4b
3a − 2c1 − c2 a + 2c1 + c2
Qs = > Qc , P s = < P c , ((a + c1 + c2 )/3 = P c > c1 ).
4b 4
(a − 2c1 + c2 ) (a + 2c1 − 3c2 )
π1s = , π2s = .
8b 16b

1. π1s + π2s ≷ π1c + π2c depending on c1 , c2 .

2. π1s > π1c because π1s = maxq1 π1 (q1 , R2 (q1 )) > π1 (q1c , R2 (q1c )) = π1c .
3. π2s < π2c since q2s < q2c and P s < P c .

q2
6
A
A R1 (q2 )
A
A
A
H A
HHA
q2c HAH
q2s A HH R2 (q1 )
H
A H - q1
q1c q1s

6.2.1 Subgame non-perfect equilibrium

In the above, we derived a subgame perfect equilibrium. On the other hand, the game
has many non-perfect equilibria. Firm 2 can threaten firm 1 that if q1 is too large, he
 35

will chooses a large enough q2 to make market price zero. This is an incredible threat
because firm 2 will hurt himself too. However, if firm 1 believes that the threat will
be executed, there can be all kind of equilibria.

6.2.2 Extension
The model can be extended in many ways. For example, when there are three firms
choosing output quantities sequentially. Or firms 1 and 2 move simultaneously and
then firm 3 follows, etc.

6.3 Conjecture Variation

In Cournot equilibrium, firms move simultaneously and, when making decision, expect
that other firms will not change their quantities. In more general case, firms will form
conjectures about other firms behaviors.

π1e (q1 , q2e ) = (a − bq1 − bq2e )q1 − c1 q1 , π2e (q1e , q2 ) = (a − bq1e − bq2 )q2 − c2 q2 .

The FOC are

dπ1e ∂π e ∂π e dq e dπ2e ∂π e ∂π e dq e
= 1 + e1 2 = 0, = 2 + e2 1 = 0.
dq1 ∂q1 ∂q2 dq1 dq2 ∂q2 ∂q1 dq2
Assume that the conjectures are
dq2e dq1e
= λ1 , = λ2 .
dq1 dq2
In equilibrium, q1e = q1 and q2e = q2 . The FOCs become

a − 2bq1 − bq2 − bλ1 q1 − c1 = 0, a − 2bq2 − bq1 − bλ2 q2 − c2 = 0.

In matrix form,
    
(2 + λ1 )b b q1 a − c1
= ,
b (2 + λ2 )bq2 a − c2
   
q1 1 (a − c1 )(2 + λ2 ) − (a − c2 )
⇒ = .
q2 [3 + 2(λ1 + λ2 ) + λ1 λ2 ]b (a − c2 )(2 + λ1 ) − (a − c1 )
If λ1 = λ2 = 0, then it becomes the Cournot equilibrium.

6.3.1 Stackelberg Case: λ2 = 0, λ1 = R20 (q1 ) = 0.5

Assuming that c1 = c2 = 0.
   
q1 1 a
= .
q2 2b a/2
It is the Stackelberg leadership equilibrium with firm 1 as the leader.
Similarly, if λ1 = 0, λ2 = R10 (q2 ) = 0.5, then firm 2 becomes the leader.
 36

6.3.2 Collusion case: λ1 = q2 /q1 , λ2 = q1 /q2

Assume that MC1 = c1 q1 and M C2 = c2 q2 . The FOCs become
a − 2b(q1 + q2 ) − c1 q1 = 0, a − 2b(q1 + q2 ) − c2 q2 = 0.
That is, MR = MC1 = M C2 , the collusion solution.
The idea can be generalized to N -firm case.

6.4 Bertrand Price Competition

It is easier to change prices than to change quantities. Therefore, firms’ strategic
variables are more likely to be prices.

Bertrand model: firms determine prices simultaneously.

Question: In a homogeneous product market, given (p1 , p2 ), how market demand

is going to divided between firm 1 and firm 2?

Assumption: Consumers always choose to buy from the firm charging lower price.
When two firms charge the same price, the market demand divided equally between
them. Let Q = D(P ) be the market demand.
 
 0 p1 > p2  D(p2 ) p1 > p2
q1 = D1 (p1 , p2 ) = 0.5D(p1 ) p1 = p2 q2 = D2 (p1 , p2 ) = 0.5D(p2 ) p1 = p2
D(p1 ) p1 < p2 , 0 p1 < p2 .
 

P p1 p2
6 6 6
D(P ) D1 (p1 , p2 ) D2 (p1 , p2 )
@
@
r r
@
@ p2 p1
@ @ @
@ @ @
@ @ @
@ -Q @ - q1 @ - q2

Notice that individual firms’ demand functions are discontinuous.

Bertrand game:
π1 (p1 , p2 ) = (p1 − c1 )D1 (p1 , p2 ), π2 (p1 , p2 ) = (p2 − c2 )D2 (p1 , p2 )
Bertrand equilibrium: {pb1 , pb2 , q1b , q2b } such that q1b = D1 (pb1 , pb2 ), q2b = D2 (pb1 , pb2 ), and
π1 (pb1 , pb2 ) ≥ π1 (p1 , pb2 ), π2 (pb1 , pb2 ) ≥ π2 (pb1 , p2 ) ∀ (p1 , p2 ).
We cannot use FOCs to find the reaction functions and the equilibrium as in the
Cournot quantity competition case because the profit functions are not continuous.
 37

6.4.1 If c1 = c2 = c, then pb1 = pb2 = c, q1b = q2b = 0.5D(c)

Proof: 1. pbi ≥ c.
2. Both p1 > p2 > c and p2 > p1 > c cannot be equilibrium since the firm with a
higher price will reduce its price.
3. p1 = p2 > c cannot be equilibrium since every firm will reduce its price to gain the
whole market.
4. p1 > p2 = c and p2 > p1 = c cannot be equilibrium because the firm with p = c
will raise its price to earn positive profit.
5. What left is p1 = p2 = c, where none has an incentive to change.

Bertrand paradox: 1. π1b = π2b = 0.

2. Why firms bother to enter the market if they know that π1b = π2b = 0.

6.4.2 If ci < cj and cj ≤ Pm (ci ), then no equilibrium exists

However, if cj > Pm (ci ), where Pm (ci ) is the monopoly price corresponding to marginal
cost ci , then firm i becomes a monopoly.
Or if there is a smallest money unit e (say, e is one cent), cj − ci > e, then pbi = c2 − e,
pbj = cj , qi = D(c2 − e), q2b = 0 is an equilibrium because none has an incentive to
change. Approximately, we say that pbi = pbj = cj and qib = D(cj ), qjb = 0.

6.5 Price competition, capacity constraint, and Edgeworth Cycle

One way to resolve Bertrand paradox is to consider DRTS or increasing MC. In
Bertrand price competition, if the marginal cost is increasing (TC 00i (qi ) > 0), then we
have to consider the possibility of mixed strategy equilibria. See Tirole’s Supplement
to Chapter 5. Here we discuss capacity constraint and Edgeworth cycle.

6.5.1 Edgeworth model

Assume that both firms has a capacity constraint qi ≤ q̄i = 1 and that the product is
indivisible. c1 = c2 = 0. There are 3 consumers. Consumer i is willing to pay 4 − i
dollars for one unit of the product, i = 1, 2, 3.

P
6
t
D(P )
3
2 t

1 t
-Q
1 2 3
Edgeworth Cycle of (p1 , p2 ): (2, 2)→(3, 2)→(3, 2.9)→(2.8, 2.9)→(2.8, 2.7)→ · · · →(2, 2)
Therefore, the is no equilibrium but repetitions of similar cycles.
 38

6.6 A 2-period model

At t = 1, both firms determine quantities q1 , q2 .
At t = 2, both firms determine prices p1 , p2 after seeing q1 , q2 .
Demand function: P = 10 − Q, MCs: c1 = c2 = 1.

Proposition: If 2qi + qj ≤ 9, then p1 = p2 = 10 − (q1 + q2 ) = P (Q) is the equi-

librium at t = 2.
Proof: Suppose p2 = 10 − (q1 + q2 ), then p1 = p2 maximizes firm 1’s profit. The
reasons are as follows.
1. If p1 < p2 , π1 (p1 ) = q1 (p1 − 1) < π1 (p2 ) = q1 (p2 − 1).
2. If p1 > p2 , π1 (p1 ) = (10−q2 −p1 )(p1 −1), π10 (p1 ) = 10−q2 −2p1 +1 = 2q2 +q1 −9 ≤ 0.

It follows that at t = 1, firms expect that P = 10 − q1 − q2 , the reduced profit

functions are
π1 (q1 , q2 ) = q1 (P (Q)−c1 ) = q1 (9−q1 −q2 ), π2 (q1 , q2 ) = q2 (P (Q)−c2 ) = q2 (9−q1 −q2 ).
Therefore, it becomes an authentic Cournot quantity competition game.

6.7 Infinite Repeated Game and Self-enforcing Collusion

Another way to resolve Bertrand paradox is to consider the infinite repeated version
of the Bertrand game. To simplify the issue, assume that c1 = c2 = 0 and P =
min{1 − Q, 0} = min{1 − q1 − q2 , 0} in each period t. The profit function of firm i at
time t is
πi (t) = πi (q1 (t), q2 (t)) = qi (t)[1 − q1 (t) − q2 (t)].
A pure strategy of the repeated game of firm i is a sequence of functions σi,t of
outcome history Ht−1 :
σi ≡ (σi,0 , σi,1 (H0 ), . . . , σi,t (Ht−1 ), . . .) .
where Ht−1 is the history of a play of the repeated game up to time t − 1:
Ht−1 = ((q1 (0), q2 (0)), (q1 (1), q2 (1)), . . . , (q1 (t − 1), q2 (t − 1))) ,
and σi,t maps from the space of histories Ht−1 to the space of quantities {qi : 0 ≤
qi < ∞}. Given a pair (σ1 , σ2 ), the payoff function of firm i is
∞
X ∞
X
t
Πi (σ1 , σ2 ) = δ πi (σ1,t (Ht−1 ), σ2,t (Ht−1 )) = δ t πi (q1 (t), q2 (t)).
t=0 t=0

1 1
Given the Cournot equilibrium (q1c , q2c ) = ( , ), we can define a Cournot strategy for
3 3
the repeated game as follows:
c 1
σi,t (Ht−1 ) = qic =
∀Ht−1 , σic ≡ (σi,0c c
, σi,1 c
, . . . , σi,t , . . .).
3
It is straightforward to show that the pair (σ1c , σ2c ) is a Nash equilibrium for the
repeated game.
 39

6.7.1 Trigger strategy and tacit collusive equilibrium

A trigger strategy σ T for firm i has a cooperative phase and a non-cooperative phase:
Cooperative phase ¯T
Þ– If both firms cooperate (choose collusion quantity
qi = 0.5Qm = 0.25) up to period t − 1, then firm i will cooperate at period t.
Non-cooperative phase .¯T
Þ– Once the cooperation phase breaks down
(someone has chosen a different quantity), then firm i will choose the Cournot equi-
librium quantity qic .
If both firms choose the same strigger strategy, then they will cooperate forever. If
neither one could benefit from changing to a different strategy, then (σ T , σ T ) is a
subgame-perfect Nash equilbrium.

Formally, a trigger strategy for firm i is

σ T = (σ0T , σ1T , . . . , σtT , . . .), σ0T = 0.5Qm ,


0.5Qm if qj (τ ) = 0.5Qm ∀ 0 ≤ τ < t, j = 1, 2
σtT =
qic otherwise.

9
6.7.2 (σ T , σ T ) is a SPNE for δ >
17
Proof:
1. At every period t in the cooperative phase, if the opponent does not violate the
cooperation, then firm i’s gain to continue cooperation is
1
Π∗ = 0.5πm + δ0.5πm + δ 2 0.5πm + · · · = 0.5πm (1 + δ + δ 2 + · · ·) = .
8(1 − δ)
3
If firm i chooses to stop the cooperative phase, he will set qt = (the profit max-
8
imization output when qj = 0.25) and then trigger the non-cooperative phase and
gains the Cournot profit of 1/9 per period. Firm i’s gain will be

9 9 9 δ
Πv = + δπic + δ 2 0.5πic + · · · = + πic (δ + δ 2 + · · ·) = + .
64 64 64 9(1 − δ)
1
Π∗ − Π v =
(17δ − 9) > 0.
576
Therefore, during the cooperative phase, the best strategy is to continue cooperation.
2. In the non-cooperative phase, the Cournot quantity is the Nash equilibrium quan-
tity in each period.

The cooperative phase is the equilibrium realization path. The non-cooperative phase
is called off-equilibrium subgames.
 40

6.7.3 Retaliation trigger strategy

A retaliation trigger strategy σ RT for firm i has a cooperative phase and a retaliation
phase:
Cooperative phase– the same as a trigger strategy.
Retaliation phase– Once the cooperation phase breaks down, then firm i will choose
the retaliation quantity qir = 1 to make sure P = 0.
If both firms choose the same trigger strategy, then they will cooperate forever. If
neither one could benefit from changing to a different strategy, then (σ RT , σ RT ) is a
subgame-non-perfect Nash equilbrium. It is not a perfect equilibrium because retali-
ation will hurt oneself and is not a credible threat.

Formally, a retaliation trigger strategy for firm i is

σ RT = (σ0RT , σ1RT , . . . , σtRT , . . .), σ0RT = 0.5Qm ,


0.5Qm if qj (τ ) = 0.5Qm ∀ 0 ≤ τ < t, j = 1, 2
σtRT =
1 otherwise.

1
6.7.4 (σ RT , σ RT ) is a NE for δ >
9
Proof:
At every period t in the cooperative phase, if the opponent does not violate the
cooperation, then firm i’s gain to continue cooperation is (same as the trigger strategy
case)
1
Π∗ = .
8(1 − δ)
3
If firm i chooses to stop the cooperative phase, he will set qt = (same as the trigger
8
strategy case) and then trigger the retaliation phase, making 0 profit per period. Firm
i’s gain will be
9 1 9 9δ − 1
Πv = ⇒ Π∗ − Πv = − = > 0.
64 8(1 − δ) 64 64(1 − δ)

Therefore, during the cooperative phase, the best strategy is to continue cooperation.

The retaliation phase is off-equilibrium subgames and never reached. Since the retal-
iation strategy is not optimal, the Nash equilibrium is not subgame-perfect.
9
In summary, if 1 > δ > , then the duopoly firms will collude in a SPNE; if
17
9 1 1
> δ > , then the duopoly firms will collusion in a non-perfect NE. If δ < , then
17 9 9
collusion is impossible.
 41

6.7.5 Folk Theorem of the infinite repeated game

In the above, we consider only the cooperation to divide the monopoly profit evenly.
The same argument works for other kinds of distributions of monopoly profit or even
aggregate profits less than the monopoly profit.

Folk Theorem: When δ→1, every distributions of profits such that the average
payoff per period πi ≥ πic can be implemented as a SPNE.

For subgame-non-perfect NEs, the individual profits can be even lower then the
Cournot profit.

6.7.6 Finitely repeated game

If the duopoly game is only repeated finite time, t = 1, 2, . . . , T , we can use backward
induction to find Subgame-perfect Nash equilibria. Since the last period T is the
same as a 1-period duopoly game, both firms will play Cournot equilibrium quantity.
Then, since the last period strategy is sure to be the Cournot quantity, it does not
affect the choice at t = T − 1, therefore, at t = T − 1 both firms also play Cournot
strategy. Similar argument is applied to t = T − 2, t + T − 3, etc., etc.
Therefore, the only possible SPNE is that both firms choose Cournot equilibrium
quantity from the beginning to the end.

However, there may exist subgame-non-perfect NE.

6.7.7 Infinite repeated Bertrand price competition game

In the above, we have considered a repeated game of quantity compeition. We can
define an infinitely repeated price competition game and a trigger strategy similarily:
(1) In the cooperative phase, a firm sets monopoly price and gains one half of the
monopoly profit 0.5πm = 1/8.
(2) In the non-cooperative phase, a firm sets Bertrand competition price pb = 0 and
gains 0 profit.
To deviate from the cooperative phase, a firm obtains the whole monopoly profit
πm = 1/4 instantly. If 1/[8(1 − δ)] > 1/4 (δ > 0.5), the trigger strategy is a SPNE.

6.8 Duopoly in International Trade

6.8.1 Reciprocal Dumping in International Trade
2 countries, i = 1, 2 each has a firm (also indexed by i) producing the same product.
Assume that MC = 0, but the unit transportation cost is τ .
qih , qif : quantities produced by country i’s firm and sold in domestic market and for-
eign market, respectively.
Q1 = q1h + q2f , Q2 = q2h + q1f : aggregate quantities sold in countries 1 and 2’s market,
respectively.
 42

Pi = a − bQi : market demand in country i’s market.

The profits of the international duopoly firms are

Π1 = P1 q1h + (P2 − τ )q1f = [a − b(q1h + q2f )]q1h + [a − (q1f + q2h ) − τ ]q1f ,

Π2 = P2 q2h + (P1 − τ )q2f = [a − b(q2h + q1f )]q2h + [a − (q2f + q1h ) − τ ]q2f .

Firm i will choose qih and qif to maximize Πi . The FOC’s are

∂Πi ∂Πi
= a − 2bqih − bqjf = 0, = a − 2bqif − bqjh − τ = 0, i = 1, 2.
∂qih ∂qif

In a symmetric equilibrium q1h = q2h = q h and q1f = q2f = q f . In matrix form, the
FOC’s become:
 a+τ 
  h     h 
2b b q a q 3b ,
= , ⇒ = a− 2τ
b 2b qf a−τ qf
3b
2a − τ a+τ
Q = qh + qf = P = .
3b 3
It seems that there is reciprocal dumping: the FOB price of exports P FOB is lower
than the domestic price P .
a+τ a − 2τ
P CIF = P = , P FOB = P CIF − τ = < P.
3 3
However, since P FOB > MC = 0, there is no dumping in the MC definition of dump-
ing.
The comparative statics with respect to τ is

∂q h ∂q f ∂Q ∂P
> 0, < 0, < 0, > 0.
∂τ ∂τ ∂τ ∂τ
In this model, it seems that international trade is a waste of transportation costs
and is unnecessary. However, if there is no international competition, each country’s
market would become a monopoly.

Extensions: 1. 2-stage game. 2. comparison with monopoly.

6.8.2 Preferential Trade Agreement, Trade Creation, Trade Diversion

Free Trade Agreement FTA: Free trade among participants.
Customs Union CU: FTA plus uniform tariff rates towards non-participants.
Common Market CM: CU plus free factor mobility.

Consider the apple market in Taiwan.

 43

Demand: P = a − Q.
2 export countries: America and Japan, PA < PJ .

At t = 0, Taiwan imposes uniform tariff of $ t per unit.

Q2 (a − PA )2 − t2
P0 = PA + t, Q0 = a − PA − t, W0 = CS0 + T0 = 0 + tQ0 = .
2 2
At t = 1, Taiwan and Japan form FTA, assume PJ < PA + t.
Q2 (a − PJ )2
P1 = PJ , Q1 = a − PJ , W1 = CS1 + T1 = 1 = .
2 2
W1 − W0 = 0.5[(a − PJ )2 + t2 − (a − PA )2 ] = 0.5[t2 − (2a − PA − PJ )(PJ − PA )].
Given a and PA , FTA is more advantageous the higher a and the lower PJ .

P P P
6
@ 6
@ 6
@
@ @ @
@ @ @
CS0 @ @ φ @
P0 @ CS1 @ P0 @
@ @ β @
γ
PJ T0 @ P1 @ P1 @
@ @ δ @
PA @ @ PA @
@- Q @- Q @- Q
Q0 Q1 Q0 Q1
W1 − W0 = (φ + β + γ) − (φ + β + δ) = γ − δ.
γ: trade creation effect.
δ: trade diversion effect.
W1 − W0 > 0 if and only if γ > δ.

6.9 Duopoly under Asymmetric Information

6.9.1 Incomplete information game
Imperfect information game: Some information sets contain more than one nodes,
i.e., at some stages of the game, a player may be uncertain about the consequences
of his choices.

Incomplete information game: Some players do not completely know the rule of
the game. In particular, a player does not know the payoff functions of other players.
There are more than one type of a player, whose payoff function depends on his type.
The type is known to the player himself but not to other players. There is a prior
probability distribution of the type of a player.

Bayesian equilibrium (of a static game): Each type of a player is regarded as an

independent player.
Bayesian perfect equilibrium (of a dynamic game): In a Bayesian equilibrium, players
will use Bayes’ law to estimate the posterior distribution of the types of other players.
 44

6.9.2 Modified chicken game

In the chicken dilemma game, assume that there are 2 types of player 2, 2a and 2b,

2a is as before but 2b is different, who prefers 
NN to SS, SS to being a chicken:
S  S 
1 1
     
Q N Q N
Q Q

@ 2a Q
2b Q
 
S @
N S @@N S @@N S @@N
         
5 2 10 0 5 2 10 0
5 10 2 0 2 10 0 5

Asymmetric information: player 2 knows whether he is 2a or 2b but player 1 does

not. However, player 1 knows that Prob[2a] = Prob[2b] =0.5.

If we regard 2a and 2b as two different players, the game tree becomes:



0
Q
 Q
 Q1/2
 Q 
1/2 Q



 1 Q
@ @
 S @ N  S @N 


A
2b @
2a @

A
A
A
S
AN S
AN S
AN S
AN

A
A
A
A
           
5 2 10 0 5 2 10 0
 5   10  2  0   0  0  0   0 
0 0 0 0 2 10 0 5

In this incomplete information game, player 2b has a dominant strategy, N. Hence,

it can be reduced to a 2-person game. The pure strategy Bayesian equilibria are
SNN and NSN. The mixed strategy equilibrium is different from the ordinary chicken
game.
 45

6.9.3 A duopoly model with unknown MC

2 firms, 1 and 2, with market demand p = 2 − q1 − q2 .
MC1 = 1, π1 = q1 (1 − q1 − q2 ).
2 types of firm 2, a and b.
MC2a = 1.25, π2a = q2a (0.75 − q1 − q2a ).
MC2b = 0.75, π2b = q2b (1.25 − q1 − q2b ).

Asymmetric information: firm 2 knows whether he is 2a or 2b but firm 1 does not.

However, firm 1 knows that Prob[2a] = Prob[2b] =0.5.

To find the Bayesian equilibrium, we regard the duopoly as a 3-person game with
payoff functions:

Π1 (q1 , q2a , q2b ) = 0.5q1 (1 − q1 − q2a ) + 0.5q1 (1 − q1 − q2b )

Π2a (q1 , q2a , q2b ) = 0.5q2a (0.75 − q1 − q2a )
Π2b (q1 , q2a , q2b ) = 0.5q2b (1.25 − q1 − q2b )

FOC are

0.5(1−2q1 −q2a )+0.5(1−2q1 −q2b ) = 0, 0.5(0.75−q1 −2q2a ) = 0, 0.5(1.25−q1 −2q2b ) = 0.

    
4 1 1 q1 2
 1 2 0  q2a  =  0.75 .
1 0 2 q2b 1.25
1 5 11
The Bayesian equilibrium is (q1 , q2a , q2b ) = ( , , ).
3 24 24
 46

7 Differentiated Products Markets

7.1 2-Differentiated Products Duopoly
2 Sellers producing differentiated products.

p1 = α − βq1 − γq2 , p2 = α − βq2 − γq1 , β > 0, β 2 > γ 2 .

In matrix form,
         −1    
p1 α β γ q1 q1 β γ α p1
= − , ⇒ = − ,
p2 α γ β q2 q2 γ β α p2
         
1 α(β − γ) β −γ p1 a b −c p1
= 2 − ≡ − ,
β − γ2 α(β − γ) −γ β p2 a −c b p2
where
α β γ
a≡ , b≡ , c≡ .
β+γ β2 − γ2 β2 − γ2
If γ = 0 (c = 0), the firms are independent monopolists. If 0 < γ < β (0 < c < b)
the products are substitutes. When β = γ ⇒ p1 = p2 , the products are perfect
substitutable (homogenous).
β
6
γ2
@ Define δ ≡ , degree of differentiation.
@Complemnts Substitutes β2
@ If δ (hence γ, c) → 0, products are
@
us

@
highly differentiated.
eo

@
en

If δ → β (hence γ → c), products

og

@
m

@ are highly homogenous.

ho

@ -γ

Assume that TCi (qi ) = ci qi , i = 1, 2. In a Cournot quantity competition duopoly

game, the payoffs are represented as functions of (q1 , q2 ):

π1c (q1 , q2 ) = (p1 −c1 )q1 = (α−βq1 −γq2 −c1 )q1 , π2c (q1 , q2 ) = (p2 −c2 )q2 = (α−βq1 −γq2 −c2 )q2 .

In a Bertrand price competition duopoly game, the payoffs are represented as func-
tions of (p1 , p2 ):

π1b (p1 , p2 ) = (p1 −c1 )q1 = (p1 −c1 )(a−bp1 +cp2 ), π2b (p1 , p2 ) = (p2 −c2 )q2 = (p2 −c2 )(a−bp2 +cp1 ).

It seems that Cournot game and Bertrand game are just a change of variables of each
other. However, the Nash equilibrium is totally different. A change of variables of a
game also changes its Nash equilibrium.
 47

7.1.1 Change of variables of a game

Suppose we have a game in (x1 , x2 ):
π1a (x1 , x2 ), π2a (x1 , x2 ).
FOCs of the x-game:
∂π1a ∂π2a
= 0, = 0, ⇒ (xa1 , xa2 ).
∂x1 ∂x2
Change of variables:
x1 = F (y1 , y2 ), x2 = G(y1 , y2 ).
The payoff functions for the y-game is
π1b (y1 , y2 ) = π1a (F (y1 , y2 ), G(y1 , y2 )), π2b (y1 , y2 ) = π2a (F (y1 , y2 ), G(y1 , y2 )).
FOCs of the y-game:
∂π1b ∂π a ∂F ∂π a ∂G ∂π2b ∂π a ∂F ∂π a ∂G
=0= 1 + 1 , =0= 2 + 2 , ⇒ (y1b , y2b ).
∂y1 ∂x1 ∂y1 ∂x2 ∂y1 ∂y2 ∂x1 ∂y2 ∂x2 ∂y2
Because the FOCs for the x-game is different from that of the y-game, xa1 6= F (y1b , y2b )
∂F ∂G
and xa2 6= G(y1b , y2b ) in general. Only when = = 0 will xa1 = F (y1b , y2b ) and
∂y2 ∂y1
xa2 = G(y1b , y2b ).
In case of the duopoly game, a Bertrand equilibrium is different from a Cournot
equilibrium in general. Only when δ = γ = c = 0 will the two equilibria the same.
That is, if the two products are independent, the firms are independent monopolies
and it does not matter whether we use prices or quantities as the strategical variables.

7.1.2 Quantity Game

Assume that c1 = c2 = 0. The payoff functions are
π1c (q1 , q2 ) = (α − βq1 − γq2 )q1 , π2c (q1 , q2 ) = (α − βq1 − γq2 )q2 .
The FOC of firm i and its reaction function are
√

c α − γqj α γ dqi  γ
α−2βqi −γqj = 0, ⇒ qi = Ri (qj ) = = − qj = − = −0.5 δ.
2β 2β 2β dqj Rc 2β
i

qi
6 
√
dqi  γ
= − = −0.5 δ
dqj Rc 2β
i
The larger δ, the steeper the reaction curve.
HH If products are independent, δ = 0,
HH
HH qi is independent of qj .
HHRic (qj )
HH -
qj
 48

In a symmetric equilibrium, q1 = q2 = q c , p1 = p2 = pc ,
α α αβ α
qc = = √ , pc = α − (β + γ)q c = = √ ,
2β + γ β(2 + δ) 2β + γ 2+ δ
and π1 = π2 = π c ,
α2 β α2
πc = = √ .
(2β + γ)2 β(2 + δ)2
∂q c ∂pc ∂π c
< 0, < 0, < 0.
∂δ ∂δ ∂δ
Therefore, when the degree of differentiation increases, q c , pc , and π c will be increased.
When δ = 1, it reduces to the homogeneous case.

7.1.3 Price Game

Also assume that c1 = c2 = 0. The payoff functions are

π1b (p1 , p2 ) = (a − bp1 + cp2 )p1 , π2b (p1 , p2 ) = (a − bp2 + cp1 )p2 .

The FOC of firm i and its reaction function are

√

a + cpj a c dpi  c
a − 2bpi + cpj = 0, ⇒ pi = Rib (pj ) = = + pj = = 0.5 δ.
2b 2b 2b dpj Rb 2b
i

pi
6 
√
dpi 
 = 0.5 δ
dpj Rb


i
The larger δ, the steeper the reaction curve.
b
 Ri (pj ) If products are independent, δ = 0,

pi is independent of pj .
- pj

In a symmetric equilibrium, p1 = p2 = pb , q1 = q2 = q b ,
a a α(β − γ) ab a
pb = = √ = , q b = a − (b − c)pb = = √ ,
2b + c b(2 + δ) 2β − γ 2b − c 2− δ
and π1 = π2 = π b ,
√
b a2 b a2 α2 (β − γ)β α2 (1 − δ)
π = = √ = = √ √ .
(2b − c)2 b(2 − δ)2 (2β − γ)2 (β + γ) β(2 − δ)2 (1 + δ)
∂pb ∂q b ∂π b
< 0, < 0, < 0.
∂δ ∂δ ∂δ
Therefore, when the degree of differentiation increases, pb , q b , and π b will be increased.
When δ = 1, it reduces to the homogeneous case and pb →0.
 49

7.1.4 Comparison between quantity and price games

α
pc − p b = > 0, ⇒ q c − q b < 0.
4δ −1 − 1
As δ→0, pc →pb and when δ = 0, pc = pb .

Strategic substitutes vs Strategic complements: In a continuous game, if the

slopes of the reaction functions are negative as in the Cournot quantity game, the
strategic variables (e.g., quantities) are said to be strategic substitutes. If the slopes
are positive as in the Bertrand price game, the strategic variables (e.g., prices) are
said to be strategic complements.

7.1.5 Sequential moves game

The Stackelberg quantity leadership model can be generalized to the differentiated
product case. Here we use an example to illustrate the idea. Consider the following
Bertrand game:

q1 = 168 − 2p1 + p2 , q2 = 168 − 2p2 + p1 ; TC = 0.

⇒ pi = Rib (qj ) = 42 + 0.25pj , pb = 56, q b = 112, π b = 6272.

In the sequential game version, assume that firm 1 moves first:

π1 = [168 − 2p1 + (42 + 0.25p1 )]p1 = [210 − 1.75p1 ]p1 .

FOC: 210 − 3.5p1 = 0, ⇒ p1 = 60, p2 = 57, q1 = 105, q2 = 114, π1 = 6300, π2 =

6498.
⇒ p1 > p2 , q1 < q2 , π2 > π1 > π b = 6272.

The Cournot game is:

2q1 + q2 2q2 + q1
p1 = 168 − , p2 = 168 − ;
3 3
⇒ qi = Ric (qj ) = 126 − 0.25qj , q = pc = 100.8, pc = 67.2, π c = 6774.
In the sequential game version, assume that firm 1 moves first:
2 1 7
π1 = (168 − qi − qj )qi = (126 − qj )qi .
3 3 12
7
FOC: 126 − q1 = 0, ⇒ q1 = 108, q2 = 81, p1 = 69, p2 = 78, π1 = 7452, π2 = 6318.
6
⇒ p 1 < p 2 , q 1 > q 2 , π1 > π c > π 2 .

From the above example, we can see that in a quantity game firms prefer to be
the leader whereas in a price game they prefer to be the follower.
 50

7.2 Free entry/exit and LR equilibrium number of firms

So far we have assumed that the number of firms is fixed and entry/exit of new/existing
firms is impossible. Now let us relax this assumption and consider the case when
new/existing firms will enter/exit the industry if they can make a positive/negative
profit. The number of firms is now an endogenous variable.

Assume that every existing and potential producer produces identical product and
has the same cost function.
TCi (qi ) = F . P =A−Q

Assume that there are n firms in the industry. The n-firm oligopoly quantity compe-
tition equilibrium is (see 6.1.2)
 2
A A
qi = = P, ⇒ πi = − F ≡ Π(n).
n+1 n+1
If Π(n + 1) > 0, then at least a new firm will enter the industry. If Π(n) < 0, then
the some of the existing firms will exit.
In LR equilibrium, the number of firms n∗ will be such that Π(n∗ ) ≥ 0 ≥ Π(n∗ + 1).
 2  
∗ A ∗ A
Π(n ) = −F ≥0 ⇒n = √ − 1.
n+1 F
Π
6
Π(n∗ ) > 0 and Π(n∗ + 1) < 0.

-n
n∗ n∗+1

The model above can be modified to consider the case when firms produce differ-
entiated products:
!
X X X
pi = A−qi −δ q j , πi = A − q i − δ qj qi −F, ⇒ FOC: A−2qi −δ qj = 0,
j6=i j6=i j6=i

where 0 < δ < 1. If there are N firms, in Cournot equilibrium, qi = q ∗ for all i, and
 2
∗ ∗ A ∗ A
q =p = , π = − F.
2 + (N − 1)δ 2 + (N − 1)δ
The equilibrium number of firms is
     
∗ A 2 1 A 1
N = √ +1− = √ −1 +1− .
δ F δ δ F δ
 51

When δ→1, it reduces to the homogenous product case. When δ decreases, N ∗

increases.

7.3 Monopolistic Competition in Differentiated Products

The free entry/exit model of last section assumes that firms compete in a oligopoly
market, firms in the market are aware of the co-existent relationship. Now we consider
a monopolistic competition market in which each firm regards itself as a monopoly
firm.

7.3.1 Chamberlin Model

There are many small firms each produces a differentiated product and has a nega-
tively sloped demand curve. The demand curve of a typical firm is affected by the
number of firms in the industry. If existing firms are making positive profits Π > 0,
new firms will enter making the demand curve shifts down. Conversely, if existing
firms are negative profits Π < 0, some of them will exit and the curve shifts up. In
the industry equilibrium, firms are making 0-profits, there is no incentive for entry or
exit and the number of firms does not change.

p Π > 0, new firms are p Π < 0, some firms are p In equilibrium, Π = 0,

6 entering, D is shifting 6 exiting, D is shifting 6
D is tangent to AC.
@ down. up.
@ @
@ AC(Q) @  AC(Q) @ AC(Q)
@ @ @
@ @ p∗ @
@ @ @
@ D @ D @ D
@ -q @ -q @ -q
q∗

7.3.2 Dixit-Stiglitz Model

Dixit-Stiglitz (AER 1977) formulates Chamberlin model.
There is a representative consumer who has I dollars to spend on all the brands
available. The consumer has a CES utility function
"∞ #1/α
X
U (q1 , q2 , . . .) = (qi )α , 0 < α < 1.
i=1

The consumer’s optimization problem is

N
X N
X N
X N
X
α α
max (qi ) , subject to pi qi = I, ⇒ L = (qi ) + λ(I − pi qi ).
i=1 i=1 i=1 i=1

FOC is
 
 1
 α−1
∂L λpi I −1 −1
= α(qi )α−1 − λpi = 0 ⇒ qi = = P −α
 pi1−α = Api1−α .
∂qi α pj1−α
j
 52

∂A 1
As N is very large, ≈ 0 and the demand elasticity is approximately |η| = .
∂pi 1−α
The cost function of producer i is TCi (qi ) = F + cqi .

max pi qi − TC(qi ) = A1−α qiα − F − cqi .

qi

1
The monopoly profit maximization pricing rule MC = P (1 − |η|
) means:

|η| c c(1 − α)
p∗i = c = , πi = (p∗i − c)qi − F = qi − F.
|η| − 1 α α

In equilibrium, πi = 0, we have (using the consumer’s budget constraint)

αF I (1 − α)I
qi∗ = , N∗ = ∗ ∗ = .
(1 − α)c pq F

The conclusions are

p−c
1. p∗ = c/α and Lerner index of each firm is = 1 − α.
p
αF
2. Each firm produces q ∗ = .
(1 − α)c
(1 − α)I
3. N ∗ = .
F
4. Variety effect:
1
U ∗ = (N ∗ q ∗α )1/α = N ∗1/α q ∗ = [αα (1 − α)1−α I]1/α F 1− α c−1 .

The size of the market is measured by I. When I increases, N ∗ (the variety)

increases proportionally. However, U ∗ increases more than proportionally.

Examples: Restaurants, Profesional Base Ball, etc.

If we approximate the integer number N by a continuous variable, the utility function

and the budget constraint are now
Z ∞ 1/α Z N
α
U ({q(t)}0≤t≤N ) = [q(t)] dt , p(t)q(t)dt = I.
0 0

The result is the same as above.

 53

7.3.3 Intra-industry trade

Heckscher-Ohlin trade theory: A country exports products it has comparative ad-
vantage over other countries.
Intra-industry trade: In real world, we see countries export and import the same
products, eg., cars, wines, etc. It seems contradictory to Heckscher-Ohlin’s compar-
ative advantage theory.

Krugman (JIE 1979): If consumers have preferences for varieties, as in Dixit-Stiglitz

monopolistic competition model, intra-industry trade is beneficiary to trading coun-
tries.
(1 − α)I
No trade: p∗ = c/α, q ∗ = (1 − α)F/(αc), N ∗ = , U ∗ ∝ I 1/α .
F
(1 − α)2I
Trade: p∗ = c/α, q ∗ = (1 − α)F/(αc), N ∗ = , U ∗ ∝ (2I)1/α .
F
Gros (1987), with tariff.
Chou/Shy (1991), with non-tradable good sector.

7.4 Location Models

The models discussed so far assume that the product differentiation is exogenously
determined. Location models provide a way to endogenize product differentiation.

7.4.1 Product characteristics ß¹Ô

Different products are characterized by different characters. P0, ¯, æH, ¹”.

Firms choose different product characters to differentiated each other.

Vertical differentiation: Consumers’ preferences are consistent w.r.t. the differen-

tiated character, eg.,. ¹”. (Ch12)

Horizontal differentiation: Different consumers have different tastes w.r.t. the dif-
ferentiated character, eg., P0, ¯.

To determine the characteristics of a product is to locate the product on the space of

product characteristics. ß¹ìP

7.4.2 Hotelling linear city model

Assume that consumers in a market are distributed uniformly along a line of length
L.
d r d
A i B
0 a x L−b L

Firm A is located at point a, PA is the price of its product.

 54

Firm B is located at point L − b, PB is the price of its product.

Each point x ∈ [0, L] represents a consumer x. Each consumer demands a unit

of the product, either purchases from A or B.

−PA − τ |x − a| if x buys from A.
Ux =
−PB − τ |x − (L − b)| if x buys from B.

τ is the transportation cost per unit distance. |x − a| (|x − (L − b)|) is the distance
between x and A (B).
The marginal consumer x̂ is indifferent between buying from A and from B. The
location of x̂ is determined by
L − b + a PA − P B
−PA − τ |x − a| = −PB − τ |x − (L − b)| ⇒ x̂ = − . (1)
2 2τ
The location of x̂ divids the market into two parts: [0, x̂) is firm A’s market share
and (x̂, L] is firm B’s market share.

d r̂ d
 A’s share - B’s share -

0 a x L−b L
Therefore, given (PA , PB ), the demand functions of firms A and B are

L − b + a PA − P B L − a + b PA − P B
DA (PA , PB ) = x̂ = − , DB (PA , PB ) = L−x̂ = + .
2 2τ 2 2τ
Assume that firms A and B engage in price competition and that the marginal costs
are zero. The payoff functions are
   
L − b + a PA − P B L − a + b PA − P B
ΠA (PA , PB ) = PA − , ΠB (PA , PB ) = PB + .
2 2τ 2 2τ

The FOCs are (the SOCs are satisfied)

L − b + a 2PA − PB L − a + b PA − 2PB
− = 0, + = 0. (2)
2 2τ 2 2τ
The equilibrium is given by

τ (3L − b + a) τ (3L − a + b) 3L − b + a 3L − a + b
PA = , PB = , QA = x̂ = , QB = L−x̂ = .
3 3 6 6
τ (3L − b + a)2 τ (2L + d + 2a)2 τ (3L − a + b)2 τ (2L + d + 2b)2
ΠA = = , ΠB = = ,
18 18 18 18
where d ≡ L − b − a is the distance between the locations of A and B. the degree of
product differentiation is measured by dτ . When d or τ increases, the products are
more differentiated, the competition is less intensive, equilibrium prices are higher
and firms are making more profits.
 55

¡«Ç¦UhmúD SOGO ¬y—Ë.

∂ΠA ∂ΠB
> 0, > 0.
∂a ∂b
Moving towards the other firm will increase one’s profits. In approximation, the two
L
firms will end up locating at the mid-point . In this model the differentiation is
2
minimized in equilibrium.

Note: 1. In this model, we can not invert the demand function to define a quan-
tity competition game because the Jocobian is singular.
2. The degree of homogeneity δ = βγ is not definable either. Here we use the distance
between the locations of A and B as a measure of differentiation.
3. The result that firms in the Hotelling model will choose to minimize product
differentiation is so far only an approximation because the location of the marginal
consumer x̂ in (7) is not exactly described. It is actually an upper-semi continuous
correspondence of (PA , PB ). The reaction functions are discontinuous and the price
competition equilibrium does not exist when the two firms are too close to each other.
See Oz Shy’s Appendix 7.5.

Welfare index: aggregate transportation costs

@
@
@
@
@
@ @ @
@ @ @
@ @ @
@dd @d @d
@A B A @ @ B

0 x̂ = L/2 L 0 L/4 x̂ = L/2 3L/4 L

Equilibrium transportation cost curve Social optimum transportation cost curve

7.4.3 Digression: the exact profit function

In deriving the market demand, we regard the location of the marginal consumer
x̂ as the market dividing point. This is correct only when a < x̂ < L − b. When
x̂ ≤ a, all the consumers buy from B, QA = 0 and QB = L. The reason is, if the
consumer located at a (the location of firm A) prefers to buy from firm B, since the
transportation cost is linear, all consumers to the left of a would also prefer to buy
from B. Similarily, when x̂ ≥ L − b, all the consumers buy from A, QA = L and
QB = 0.
 56

x̂
6
Consumers in [0, a] move together. L
L−b H
HH
Consumers in [L − b, L] move together. HH
HH
a H
d d 0 - PA
0 a L−b L PB − (L − a − b)τ PB + (L − a − b)τ

The true profit function of firm A is

  
L−b+a PA − P B
 PA − a < x̂ < L − b


2 2τ
ΠA (PA , PB ) =

 0 x̂ ≤ a
PA L L − b ≥ x̂


 PA L
  PA ≤ PB − dτ
 L − b + a PA − P B
= PA − PB − dτ < PA < PB + dτ

 2 2τ
0 PA ≥ PB + dτ,


where d ≡ L − a − b. Consider the case a = b, d = L − 2a. The reaction function of

firm A is (firm B’s is similar):
( (τ L+PB )2 PB
√ √
0.5(τ L + PB ) 8τ [PB −dτ ]
≤ L or τ
∈
/ (3L − 4 La, 3L + 4 La)
PA = RA (PB ) = (τ L+PB )2 PB
√ √
PB − (L − 2a)τ 8τ [PB −dτ ] ≥ L or τ ∈ [3L − 4 La, 3L + 4 La]

For a ≤ L/4, the reaction functions intersect at PA = PB = τ L. When a > L/4, the
reaction functions do not intersect.

ΠA PB RA PB
6 6 6 RA
RB





 RB

















- PA
- PA
- PA
PB − dτ PB + dτ Case: a ≤ L/4, P ∗ = τ L, Case: a > L/4, no equilibrium.
7.4.4 Quadratic transportation costs
Suppose now that the transportation cost is proportional to the square of the distance.

−PA − τ (x − a)2

if x buys from A.
Ux =
−PB − τ [x − (L − b)]2 if x buys from A.
 57

The marginal consumer x̂ is defined by

L−b+a PA − P B
−PA − τ (x − a)2 = −PB − τ [x − (L − b)]2 ⇒ x̂ = − . (3)
2 2τ (L − a − b)
The demand functions of firms A and B are
L−b+a PA − P B L−a+b PA − P B
DA (PA , PB ) = x̂ = − , DB (PA , PB ) = L−x̂ = + .
2 2τ (L − a − b) 2 2τ (L − a − b)
The payoff functions are
   
L−b+a PA − P B L−a+b PA − P B
ΠA (PA , PB ) = PA − , ΠB (PA , PB ) = PB + .
2 2τ (L − a − b) 2 2τ (L − a − b)
The FOCs are (the SOCs are satisfied)
L−b+a 2PA − PB L−a+b PA − 2PB
− = 0, + = 0. (4)
2 2τ (L − a − b) 2 2τ (L − a − b)
The equilibrium is given by
τ (3L − b + a)(L − a − b) 3L − b + a τ (3L − b + a)2 (L − a − b)
PA = , QA = x̂ = , ΠA = ,
3 6 18
τ (3L − a + b)(L − a − b) 3L − a + b τ (3L − a + b)2 (L − a − b)
PB = , QB = L−x̂ = , ΠB = .
3 6 18
∂ΠA ∂ΠB
< 0, < 0.
∂a ∂b
Therefore, both firms will choose to maximize their distance. The result is opposite
to the linear transportation case.

Two effects of increasing distance:

1. Increase the differentiation and reduce competition. Π ↑.
2. Reduce a firm’s turf. Π ↓.

In the linear transportation case, the 2nd effect dominates. In the quadratic trans-
portation case, the 1st effect dominates.
Also, x̂ is differentiable in the quadratic case and the interior solution to the profit
maximization problem is the global maximum.

Welfare comparison:

d d d d
A B A B
0 x̂ = L/2 L 0 L/4 x̂ = L/2 3L/4 L
Equilibrium transportation cost curve Social optimum transportation cost curve
 58

7.5 Circular Market Model

It is very difficult to generalize the linear city model to more than 2 firms. Alterna-
tively, we can assume that consumers are uniformly distributed on the circumference
of a round lake. We assume further that the length of the circumference is L. First,
we assume that there are N firms their locations are also uniformly distributed along
the circumference. Then we will find the equilibrium number of firms if entry/exit is
allowed.
Firm i faces two competing neighbers, firms i − 1 and i + 1. TCi (Qi ) = F + cQi .

d r d d
i−1 i i+1
−L/N x 0 L/N

As in the linear city model, each consumer needs 1 unit of the product. The utility
of consumer x, if he buys from firm j, is

U (x) = −Pj − τ |x − lj |, j = i − 1, i, i + 1,

where lj is the location of firm j. To simplify, let L = 1. Given the prices Pi and
Pi−1 = Pi+1 = P , there are two marginal consumers x̂ and −x̂ with
P − Pi 1 P − Pi 1
x̂ = + , ⇒ Qi = 2x̂ = + .
2τ 2N τ N

d r id r d
i−1 i+1
−1/N −x̂ 0 x̂ 1/N
 
P − Pi 1 c + P − 2Pi 1
Πi = (Pi − c) + − F, ⇒ FOC: + = 0.
τ N τ N
In equilibrium, Pi = P ,
τ τ
P =c+
, Π = 2 − F.
N N
In a free entry/exit long run equilibrium, N is such that
√
r 
∗ τ 1
Π(N ) ≥ 0, Π(N + 1) ≤ 0 ⇒ N = , P ∗ = c + τ F , Q∗ = .
F N

@
@ Aggregate transportation costs:
@ 1 τ τ
@ T (N ) = N ( )= .
@ 2N 2N 4N
@
@ Aggregate Fixed Costs = N F .
@
@
@
−1/2N 0 1/2N
Equilibrium transportation cost curve
 59

Since each consumer needs 1 unit, TVC = c is a constant. The total cost to the
society is the sum of aggregate transportation cost, TVC, and fixed cost.
r
τ τ s τ
SC(N ) = T (N ) + c + N F = + c + N F, FOC: − + F = 0, ⇒ N = 0.5 .
4N 4N 2 F
The conclusion is that the equilibrium number of firms is twice the social optimum
number.

Remarks: 1. AC is decreasing (DRTS), less firms will save production cost.

2. T (N ) is decreasing with N , more firms will save consumers’ transportation cost.
3. The monopolistic competition equilibrium ends up with too many firms.

7.6 Sequential entry in the linear city model

If firm 1 chooses his location before firm 2, then clearly firm 1 will choose li = L/2 and
firm 2 will choose l2 = L/2+ . If there are more than two firms, the Nash equilibrium
location choices become much more complicated.

Assume that there are 3 firms and they enter the market (select locations) sequen-
tially. That is, firm 1 chooses x1 , then firm 2 chooses x2 , and finally firm 3 chooses
x3 . To simplify, we assume that firms charge the same price p = 1 and that x1 = 1/4.
We want to find the equilibrium locations x2 and x3 .

1. If firm 2 chooses x2 ∈ [0, x1 ) = [0, 1/4), then firm 3 will choose x3 = x1 + .

π2 = (x2 − x1 )/2 < 1/4.
xd2 rxd3
0 1 π2 = (x2 + x1 )/2 < 1/4.
x1 = 1/4

2. If firm 2 chooses x2 ∈ (x1 , 3/4) = (1/4, 3/4), then firm 3 will choose x3 = x2 + .
π2 = x2 − (x2 + x1 )/2 = (x2 − x1 )/2 < 1/4.

xd1 x2dx
d3
0 1 π2 = (x2 − x1 )/2 < 1/4.
1/4 3/4

3. If firm 2 chooses x2 ∈ [3/4, 1], then firm 3 will choose x3 = (x2 + x1 )/2.
π2 = 1 − 0.5(x2 + x3 )/2 = (15 − 12x2 )/16.
xd1 xd3 xd2
0 1 π2 = (15 − 12x2 )/16.
1/4 3/4

The subgame perfect Nash equilibrium is x2 = 3/4, x3 = 1/2 with π1 = π2 = 3/8 and
π3 = 1/4.
 60

xd1 xd3 xd2

0 1 SPE: π1 = π2 = 3/8, π3 = 1/4.
1/4 1/2 3/4

7.7 Discrete location model

Consider now that the consumers are concentrated on two points:
'$ '$
N0 transportation cost is T NL
firm A firm B
&% &%

N0 consumers live in city 0 where firm A is located.

NL consumers live in city L where firm B is located.
The round trip transportation cost from city 0 to city L is T .
Given the prices (PA , PB ), the utility of a consumer is
 
−PA −PA − T
U0 = UL =
−PB − T −PB .

nA (nB ) is the number of firm A’s (firm B’s) consumers.

 
 0 PA > PB + T  0 PB > PA + T
nA = N0 PB − T < P A < P B + T nB = NL PA − T < P B < P A + T
N 0 + N L PA < P B + T N0 + NL PB < PA + T.
 

Nash Equilibrium: (PAn , PBn ) such that PAn maximizes ΠA = PA nA and PBn maximizes
Π B = P B nB .

Proposition: There does not exist a Nash equilibrium.

Proof: 1. If PAn − PBn > T , then ΠA = 0, firm A will reduce PA . Similarily for
PBn − PAn > T .
2. If |PAn − PBn | < T , then firm A will increase PA .
3. If |PAn − PBn | = T , then both firms will reduce their prices.

Undercut proof equilibrium (UE):(PAu , PBu , nuA , nuB ) such that

1. PAn maximizes ΠA subject to ΠB = PBu nuB ≥ (N0 + NL )(PAu − T ).
2. PBn maximizes ΠB subject to ΠA = PAu nuA ≥ (N0 + NL )(PBu − T ).

In choosing PA , firm A believes that if PA is too high, firm B will undercut its
price to grab A’s consumers and vice versa.

There is a undercut proof equilibrium: nuA = N0 , nuB = NL ,and (PAu , PBu ) satisfies

PBu NL = (N0 + NL )(PAu − T ), PAu N0 = (N0 + NL )(PBu − T ).

 61

(N0 + NL )(N0 − NL )T
∆P = PBu − PAu = , ∆P ≷ 0 if N0 ≷ NL .
N02 + NL2 + N0 NL
In the symmetric case, N0 = NL and PAu = PBu = 2T .

PA PA PA
6 6
RB (PA ) 6

firm B will RA (PB ) firm A will


 not undercut


undercut
t





UE






firm B will not




firm A will

undercut - PB
undercut - P
- PB
B
 62

8 Concentration, Mergers, and Entry Barriers

ß“ÞG: h‹p; H¢|; ¼¯9; ¼}j

hEß“, Aß“, A¹ß“

Õ2D‚â5É[;
\úÕ2í„; 1. òQßã: #„}j, öŽ¼¯9
2. Š¢HS¦ªÒ®×G¨

8.1 Concentration Measures

We want to define some measures of the degree of concentration of an industry in
order to compare different industries or a similar industry in different countries.

i = 1, 2, . . . , N , Q = q 1 + q2 + · · · + q N .
qi
Market shares: si ≡ × 100%, s1 ≥ s2 ≥ s3 ≥ · · · ≥ sN .
Q
I8 ≡ 8i=1 si .
P
I4 ≡ s 1 + s 2 + s 3 + s 4 ,
PN
Herfindahl-Hirshman Index: IHH ≡ i=1 s2i .
1 P N Pj 1 PN
Gini coefficient: G ≡ j=1 i=1 si = (N − i)si .
N N 1
PN
Entropy: IE ≡ 1 si ln si .

8.1.1 Relationship between IHH and Lerner index

Consider a quantity competition oligopoly industry. P (Q) = P (q1 + q2 + · · · + qN ) is
the market demand. λi is the conjecture variation of firm i, i.e., when firm i increases
1 unit of output, he expects that all other firms together will response by increasing
λi units of output.
X dQe−i
πi = P (Q)qi − ci (qi ), Q−i ≡ q j , λi ≡ .
j6=i
dqi

∂Πei ∂πi ∂πi dQe−i

FOC: = + or P + qi P 0 (1 + λi ) = c0i .
∂qi dqi dQ−i dqi
Lerner index of firm i, Li :
P − c0i −qi P 0 (1 + λi ) −QP 0 qi 1
Li ≡ = = (1 + λi ) = si (1 + λi ).
P P P Q QP
Industry average lerner index L:
N N
IHH + i s2i λi
P
X 1 X
L≡ s i Li = s2i (1 + λi ) = .
i=1
QP i=1
QP
 63

IHH (1 + λ)
If λi = λ for all i, then L = .
P Q
Q−i 1
The case of collusion λi = : FOC is MR = MCi and Li = L = .
qi P Q

8.2 Mergers
Mergers, takeovers, acquisitions, integration.

3 types:
Horizontal mergers: between the same industry
Vertical mergers: between upstream industry firms and down stream industry firms
Conglomerate mergers: other cases.

In US economic history there were 4 active periods:

1901: mostly horizontal and vertical
1920, 1968, 1980: other types, mostly influenced by changes in Anti-trust Law.

Purpose: (1) reduce competition, (2) IRTS, (3) differences in the prospective of
firms between sellers and buyers,
(4) managers’ intension to enlarge their own careers, (5) the insterests of the
promoters.

8.2.1 Horizontal merger

¯9 → concentration ratio ↑ → competition ↓ → Welfare ↓?
Not necessary. If high cost (inefficient) firms are taken-over, efficiency increases.

Example: In Cournot duopoly model, assume c1 = 1, c2 = 4, and P = 10 − Q.

q1c = 4, q2c = 1, P c = 5, π1c = 16, π2c = 1, CS = 12.5, W c = 29.5.

If firms 1 and 2 merge to become a monopoly, the monopoly would shut down the
production of firm 2 so that the MC of the monopoly is c = 1.

Qm = 4.5, P m = 5.5, π m = 20.25, CS = 10.125, W m = 30.375 > W c .

Comparison of Welfare: W m > W c , it is the trade-off between the production effi-

ciency and monopoly inefficiency.
c m
Comparison of IHH : IHH = 6, 800, IHH = 10, 000.
According to Anti-trust law, such a merger is prohibited, but it is good to the society.
However, if the market is originally in Bertrand price competition, the conclusion is
totally different. In a Bertrand competition market, the industry is efficient.
 64

8.2.2 Vertical merger

A B B

A
¯9 ⇒
A
A1

A
? 

U A ? ?

1 2 2

If both upstream and downstream industries produce homogeneous products and

are Bertrand price competition markets, the merger does not affect anything.

Assumption: The upstream was originally in Bertrand competition and the down-
stream was in Cournot competition.
Downsteam market demand: P = α − q1 − q2 . MC1 = c1 , MC2 = c2 .

α − 2ci + cj (α − 2ci + cj )2 2α − c1 − c2 α + c1 + c2
⇒ qi = , πi = ; Q= , P = α−Q = .
3 9 3 3
Upstream: Assume MCA = MCB = 0. Bertrand equilibrium: pA = pB = c1 =
c2 = 0.

Pre-merge Equilibrium:

α α2 α 2α
q1 = q 2 = , π1 = π 2 = , πA = πB = 0; P = , Q = .
3 9 3 3
Post-merge: Assume that A1 does not sell raw material to 2. B becomes an upstream
monopoly. We ignore the fact that 2 is also a downstream monopsony.
c2 (α − 2c2 + c1 ) pB (α − 2pB ) α
πB = c 2 q 2 = = , max πB ⇒ pB = c2 = .
3 3 pB 4
Post-merge Equilibrium:

5α α 5α 7α 25α2 α2 α2
qA1 = , q2 = , P = , Q= , πA1 = P qA1 = , = π2 = , πB = P B q 2 = .
12 6 12 12 144 36 24
The effects of merge: P ↑, q1 ↑, q2 ↓, π2 ↓, πB ↑, πA + π1 ↑, πB + π2 ↓.
Firm 2 and consumers are the losers.

8.2.3 merger of firms producing complementary goods

Firm X produces PCs and Firm Y produces monitors.
A system is S = X + Y. P s = Px + Py .
 65

Market demand: Qs = α − Ps = α − Px − Py , Qx = Qy = Qs .
Pre-merge:

Πx = Px Qx = Px (α − Px − Py ), Πy = Py Qy = Py (α − Px − Py ).

FOC: α − 2Px − Py = 0, α − Px − 2Py = 0.

α α α2
⇒ Px = Py = , Qs = Q x = Q y = , Πx = Π y = .
3 3 9
Post-merge:

α α2
Πs = Ps Qs = Ps (α − Ps ), ⇒ Ps = Qs = , Πs = .
2 4
The effects of merger:

α 2α α α α2 2α2
Ps = < Px + Py = , Qs = > Q x = Q y = , Πs = > Πx + Πy = .
2 3 2 3 4 9
Therefore, one monopoly is better than two monopolies.

Remarks: 1. It is similar to the joint product monopoly situation. When a monopoly

produces two complementary goods, the profit percentage should be lower than in-
dividual Lerner indices. In this case, a higher Px will reduce the demand for Y and
vice versa. After merger, the new firm internalizes these effects.
2. The model here is isomorphic to the Cournot duopoly model with price variables
and quantity variables interchanged.
3. If there are 3 products, X, Y, and Z, and only X and Y merge, the welfare is not
necessarily improving.

8.3 Entry Barriers hªpí®‰

Û¼ (Incumbent) íi‘ <==> hªpí®‰ ==> ç‚â

1. Economy of scale, large fixed cost

2. Production differentiation advantages (reputation, good will)

3. Consumer loyalty, network externalities

4. Absolute cost advantages (learning experiences)

5. Location advantage (sequential entry)

6. Other advantages

Incumbents may also take entry deterrence (ªÒß×) strategies.

 66

8.3.1 Fixed cost and IHH

(1 − α)I
In Dixit-Stiglitz monopolistic competition model, N = :
F
 2
100 10, 000 F ∂IHH
IHH = N = = 10, 000, > 0.
N N (1 − α) ∂F

(A − c)2
In quantity competition with free entry/exit model, N ≈ √ :
bF
√
10, 000 bF ∂IHH
IHH = = 10, 000, > 0.
N A−c ∂F
r
τ F
In the circular city model, N = , IHH = 10, 000.
F τ

8.3.2 Sunk cost

Sunk costs: Costs that cannot be reversed. ÇŸ‘, µ‘, _|, firm specific equip-
ments, etc.
Sunk costs B×, hB.ߪҬ

Stiglitz (1987): Ébøõ Sunk Cost æÊ, ¹ªUh¼ (Potential Entrants) ú—

.‡, UÛ¼ (Incumberts) ?./:ƒÖ´‚â

A: An Incumbent, B: A Potential entrant.

Πm −  if no entry
 
A B 0 do not enter
Π = Π =
− B enters, − B enters,


where Πm is the monopoly profit and  is the sunk cost.


B
@
Enter Stay
@ out
  @ ∗
Πm − 

−
− 0
Proposition: As long as 0 <  < Πm , there exists only one subgame perfect equilib-
rium, i.e., B stays out.
Conditions: 1. A and B produce homogeneous product with identical marginal cost.
2. Post-entry market is a Bertrand duopoly.
3. A cannot retreat.

If B can invest in product differentiation to avoid homogeneous product Bertrand

competition or the post-entry market is a Cournot duopoly, then the proposition is
not valid any more.
 67


If A can resale some of its investments, say, recover φ > 0. The game becomes


B
@
Enter Stay
@ out
 @m

A
@

Π −
@
Stay in Exit
@ 0
   ∗
− φ−
− Πm − 

The only subgame-perfect equilibrium is that B enters and A exits. However, the
result is just a new monopoly replacing an old one. The sunk cost  can be regarded
as the entry barrier as before.
Notice that there is a non-perfect equilibrium in which A chooses the incredible threat
strategy of Stay in and B chooses Stay out.

8.4 Entry Deterrence

8.4.1 Burning one’s bridge strategy (Tirole CH8)
Two countries wishing to occupy an island located between their countries and con-
nected by a bridge to both. Each army prefers letting its opponent have the island
to fighting. Army 1 occupies the island and burns the bridge behind it. This is the
paradox of commitment.
'$

Army 1 island Army 2

&%

8.4.2 Simultaneous vs. Sequential Games

Consider a 2-person game:

Π1 (x1 , x2 , y1 , y2 ), Π2 (x1 , x2 , y1 , y2 ),

where (x1 , y1 ) is firm 1’s strategy variables and (x2 , y2 ) is firm 2’s strategy variables.

Simultaneous game: Both firms choose (x, y) simultaneously.

∂Π1 ∂Π1 ∂Π2 ∂Π2
FOC: = = 0, = = 0.
∂x1 ∂y1 ∂x2 ∂y2
Sequential game: In t = 1 both firms choose x1 and x2 simultaneously and then
in t = 2 both firms choose y1 and y2 simultaneously.
To find a subgame perfect equilibrium, we solve backward:
 68

In t = 2, x1 and x2 are given, the FOC are

∂Π1 ∂Π2
= 0 and = 0, ⇒ y1 = f (x1 , x2 ) and y2 = g(x1 , x2 ).
∂y1 ∂y2
In t = 1, the reduced game is:

π1 (x1 , x2 ) = Π1 (x1 , x2 , f (x1 , x2 ), g(x1 , x2 )), π2 (x1 , x2 ) = Π2 (x1 , x2 , f (x1 , x2 ), g(x1 , x2 )).

The FOC is
∂π1 ∂Π1 ∂Π1 ∂f ∂Π1 ∂g ∂π2 ∂Π2 ∂Π2 ∂f ∂Π2 ∂g
= + + = 0, and = + + = 0.
∂x1 ∂x1 ∂y1 ∂x1 ∂y2 ∂x1 ∂x2 ∂x2 ∂y1 ∂x2 ∂y2 ∂x2
∂Π1 ∂Π2
Since = = 0, the FOC becomes
∂y1 ∂y2
∂π1 ∂Π1 ∂Π1 ∂g ∂π2 ∂Π2 ∂Π2 ∂f
= + = 0, and = + = 0.
∂x1 ∂x1 ∂y2 ∂x1 ∂x2 ∂x2 ∂y1 ∂x2
Compare the FOC of the sequential game with that of the simultaneous game, we
can see that the equilibria are not the same. In t = 1, both firms try to influence
the t = 2 decisions of the other firms. For the simultaneous game, there is no such
considerations.
∂Π1 ∂g ∂Π2 ∂f
and are the strategic consideration terms.
∂y2 ∂x1 ∂y1 ∂x2
Entry deterrence application: In t = 1, only firm 1 exists:

Π1 (x1 , y1 , y2 ), Π2 (x1 , y1 , y2 ).

Firm 1 is the incumbent and tries to influence the entry decision of firm 2.

8.4.3 Spence (1977) entry deterrence model

In this model the incumbent attempts to use strategic capacity investment to deter
the entry of a potential entrant.

t = 1: Firm 1 decides its capacity-output level k1 .

t = 2: Firm 2 decides its capacity-output level k2 .
If k2 > 0, firm 2 enters. If k2 = 0, firm 2 stays out.

k2 (1 − k1 − k2 ) − E enter
π1 (k1 , k2 ) = k1 (1 − k1 − k2 ), π2 (k1 , k2 ) =
0 stay out,

where E is the fixed cost if firm 2 enters.

Back induction: In t = 2, k1 is given. If firm 2 enters, its profit maximization FOC is

(1 − k1 )2
 
∂π2 1 − k1 1 − k1 1 − k1
= 1−k1 −2k2 = 0, ⇒ k2 = , π2 = 1 − k1 − −E = −E.
∂k2 2 2 2 4
 69

(1 − k1 )2
If − E < 0, firm 2 will choose not to enter. Therefore, firm 2’s true reaction
4
function is
√

 1 − k1
if k1 < 1 − 2 E
k2 = R2 (k1 , E) = 2 √
 0 if k1 > 1 − 2 E.
k2
6
1
H
2 HH k2 = R2 (k1 ; E)
HH

√ - k1
1−2 E
In t = 1, firm 1 takes into consideration firm 2’s discontinuous reaction function.
√ 1 1 1
If 1 − 2 E ≤ (⇒ E ≥ ), then firm 1 will choose monopoly output k1 = and
2 16 2
firm 2 will stay out. This is the case of entry blockaded.
1
Next, we consider the case E < . If firm 1 chooses monopoly output, firm 2 will
16 √
enter. Firm 1 is considering√ whether to choose k 1 ≥ 1 − 2 E to force firm 2 to give
up or to choose k1 < 1 − 2 E and maximizes duopoly profit. √
1. entry deterrence: If firm 2 stays out (k1 ≥ 1 − 2 E and k2 = 0), firm 1 is a
monopoly by deterrence:
√
π1d (E) = max√ k1 (1 − k1 ) = kE (1 − kE ), where kE ≡ 1 − 2 E.
k1 ≥1−2 E
√
2. entry accommodate: If firm 2 enters (k1 < 1 − 2 E and k2 > 0), firm 1 is the
leader of the Stackelberg game:
1 1 1 1
π1s (E) = max√ k1 (1−k1 −k2 ) = k1 (1−k1 ), ⇒ π1s = k s (1−k s ) = where k s ≡ .
k1 <1−2 E 2 2 8 2
π1d π1d 1 π d
1 1
6 E < 0.00536 60.00536 < E < 6 E>
r
16 r 16

1
r
π1s = π1s π1s
8

r r - k1 r r - k1 r r - k1
ks kE k s kE kE k s = k m
Entry accommodate Entry deterred Entry blockaded
√ √
p
1 (1 − 1/2)2
π1d (E) = (1 − 2 E)[1 − (1 − 2 E)] ≷ π1s (E) = if E ≷ ≈ 0.00536.
8 16
In summary, if E < 0.0536, firm 1 will accommodate firm 2’s entry; if 0.00536 < E <
1
, firm 1 will choose k1 = kE to deter firm 2; if E > F 116, firm 1 is a monopoly and
16
firm 2’s entry is blockaded.
 70

Spence model is built on the so called Bain-Sylos style assumptions:

1. Firm 2 (entrant) believes that firm 1 (incumbent) will produce q1 = k1 after firm
2’s entry is deterred. However, it is not optimal for firm 1 to do so. Therefore, the
equilibrium is not subgame perfect.
2. Firm 2 has sunk costs but not firm 1. The model is not symmetrical.

8.4.4 Friedman and Dixit’s criticism

1. Incumbent’s pre-entry investment should have no effects on the post-entry market
competition. The post-entry equilibrium should be determined by post-entry market
structure.
2. Therefore, firm 2 should not be deterred.
3. Firm 1’s commitment of q1 = k1 is not reliable. Also firm 2 can make commitment
to threat firm 1. The first-mover advantage does not necessarily belong to incumbent.

8.4.5 Dixit 1980

If firm 2 is not convinced that k1 = q1 if firm 2 enters, then firm 1 cannot choose kE
to deter firm 2’s entry. Consider a 2-period model:

t = 1: Firm 1 chooses k̄.

t = 2 (Cournot competition): Firms 1 and 2 determine q1 , q2 simultaneously.

Firm 1’s MC curve in t = 2 is

MC1
6


0 if q1 ≤ k̄
MC1 = c
c if q1 > c

- q1
k̄
In t = 2, Firm 1’s FOC is 1 − 2q1 − q2 = MC1 . Its reaction function is

 1 − q2


if q1 ≤ k̄  (1 − q2 )/2 if q2 ≥ 1 − 2k̄
q1 = R1 (q2 ) = 2 = k̄ if 1 − c − 2k̄ < q2 < 1 − 2k̄
1 − c − q2

 if q1 > k̄ 
(1 − c − q2 )/2 if q2 < 1 − c − 2k̄.
2
1 − c − q1
Firm 2’s reaction function R2 (q1 ) = has nothing to do with k̄.
2
Firm 1’s reaction function is affected by k̄.
Therefore, the Cournot equilibrium in t = 2 is affected by k̄.
 71

q2 q2
q1
16
@ R1 (q2 )
16
@ R1 (q2 )
6
1−c @ @
2 @ @ @
@ @ @
k̄ @ è™ú| ⇒ @ 1−c @
XXXr
@ 2 XX
@ XXXR2 (q1 )
@ @ @ XX
@ @ @
@ - q2 @ - q1 @ - q1
1 k̄ 1−c
k̄ 1−c
2 2

In t = 1, firm 1 will choose k̄ to affect the Cournot equilibrium in t = 2. There

are 3 cases:

1−c (1 − c)2
(1) k̄ ≤ q c ≡ : q1 = q2 = q c , π1∗ = π c ≡ , i.e., Cournot equilibrium.
3 9
1+c 1 − c − k̄ 1 + c − k̄
(2) q c < k̄ < q̄ ≡ : q1 = k̄, q2 = , p= , π1 = (p − c)q1 =
3 2 2
(1 − c − k̄)k̄
.
2
1 − c − q̄ 1 + c − q̄ ∗ (1 − c − q̄)q̄
(3) q̄ ≤ k̄: q1 = q̄, q2 = ,p= , π1 = pq̄ − ck̄ = − c(k̄ −
2 2 2
q̄).
q2 q2 q2
16
@ R1 (q2 ) 16
@ R1 (q2 ) 16
@ R1 (q2 )
@ @ @
@ @ @
@ @
1−c @ @
@
@Xr X
1−c
XXrX
2 XXX XXX XXX @
r R2 (q1 )
1−c X 2 XXX@
3 @ XXR (q1 )
X2X
X R2 (q1 )
@ XXX
1−c−q̄ XX
@XX
2
@ @
@@ - q1 @ - q1 - q1
k̄ 1−c 1−c
k̄ 1−c q̄ k̄
3 2 2

The reduced profit function of firm 1 is


(1 − c)2
k̄ ≤ q c




 9
(1 − c − k̄)k̄

π1 (k̄) = q c < k̄ < q̄
 2
(1 − c − q̄)q̄




 − c(k̄ − q̄) q̄ ≤ k̄.
2
1−c
Let q1s be the Stackelberg leadership quantity of firm 1: q1s ≡ .
2
1
1. If c > , then q1s < q̄ and firm 1 will choose k̄ ∗ = q1s .
5
1
2. If c < , then q1s > q̄ and firm 1 will choose k̄ ∗ = q̄.
5
 72

π1 π1
6 6
@
1 1
c> 5
c< 5
@
@
@ @
@ @
@ @

- k̄ - k̄
qc q1s q̄ qc q̄ q1s
In Stackelberg model, firm 1 can choose any point on firm 2’s reaction curve R2 (q1 )
to maximize π1 . In Dixit model, firm 1’s choice is restricted to the section of R2 (q1 )
such that k̄ ∈ [0, q̄]. When q̄ ≥ q1s , the result is the same as Stackelberg model. When
q̄ < q1s , firm 1 can only choose k̄ = q̄.

In both cases q1 = k̄. Therefore, firm 1 does not over-invest (choose k̄ > q1 ) to
threaten firm 2’s entry.

8.4.6 Capital replacement model of Eaton/Lipsey (1980)

It is also possible that an incumbent will replace its capital before the capital is com-
plete depreciated as a commitment to discourage the entry of a potential entrant.

t = −1, 0, 1, 2, 3, . . ..
In each period t, if only one firm has capital, the firm earns monopoly profit H.
If both firms have capital, each earns duopoly profit L.
Each firm can make investment in each period t by paying F .
Denote by Rti (Cti ) the profit (cost) of firm i in period t.

∞  0 no capital 
X
t i i i i 0 no invest (NI)
Πi = ρ (Rt − Ct ), Rt = L duopoly Ct =
F invest (INV).
H monopoly

t=0

Assumption 1: An investment can be used for 2 periods with no residual value left.
Assumption 2: 2L < F < H.p
Assumption 3: F/H < ρ < (F − L)/(H − L).

Firm i’s strategy in period t is ait ∈ {NI, INV}. Assume that a1−1 = INV. We
consider only Markov stationary equilibrium.

If firm 2 does not exist, firm 1 will choose to invest (INV) in t = 1, 3, 5, . . .. Given the
threat of firm 2’s possible entry, in a subgame-perfect equilibrium, firm 1 will invest
in every period and firm 2 will not invest forever.

The symmetrical SPE strategy is such that an incumbent firm invests and a potential
 73

entrant does not:

if ajt−1 = NI

INV
ait =
NI otherwise.
Proof that the above strategy is optimal if the oppoent plays the same strategy:
H−F
Π1 = , Π2 = 0.
1−ρ
If firm 1 deviates and chooses a10 = NI, Π1 becomes H < (H − F )/(1 − ρ) (by
Assumption 3), because firm 2 will invest and become the monopoly.
ρ2 (H − F )
If firm 2 deviates and chooses a20 = INV, then Π2 = (1 + ρ)(L − F ) + <0
1−ρ
(also by Assumption 3).

8.4.7 Judo economics üªÒé™ ·5−

The (inverse) market demand is P = 100 − Q.
t = 1: Firm 2 (entrant) determines whether to enter and if enters, its capacity level
k and price pe .
t = 2: Firm 1 (incumbent) determines its price pI . Assume that firm 1 has unlimited
capacity and, if pI = pe , all consumers will purchase from firm 1.

100 − pI pI ≤ p e k pe < pI
 
I e
q = q =
100 − k − pI pI > pe 0 pe ≥ pI

Backward induction: At t = 2, (k, pe ) is given.

1. If firm 1 decides to deter entry, he chooses pI = pe and πD
I
= pe (100 − pe ).
2. If firm 1 decides to accommodate firm 2, he chooses to maximize πAI = pI (100 −
k − pI ).

100 − k (100 − k)2

max pI (100−k−pI ), ⇒ FOC: 0 = 100−k−2pI , ⇒ pIA = = qAI ⇒ πAI = .
pI >pe 2 4

(100 − k)2
Firm 1 will accommodate firm 2 if πAI > πD I
, or if ≥ pe (100 − pe ), whence
4
π e = pe k > 0.
If firm 2 chooses a small k and a large enough pe , firm 1 will accommodate.
πI
6
πAI

I
πD pe (100 − pe )

-k
 Accommodate -
 74

At t = 1, firm 2 will choose (k, pe ) such that

(100 − k)2
max pe k subject to ≥ pe (100 − pe ).
4

8.4.8 Credible spatial preemption #WªÒ2

Suppose that the incumbent is a monopoly in two markets selling substitute products
j = 1, 2. If an entrant enters into one (say, product 1) of the two markets, the in-
cumbent will give up the market in order to protect the monopoly profit of the other
market (product 2).
Reason: If the incumbent stays in market 1, the Bertrand competition will force p1
down to its marginal cost. As discussed in the monopoly chapter, the demand of
product 2 will be reduced.

Example: Suppose that firm 1 has a Chinese restaurant C and a Japanese restaurant
J in a small town, both are monopoly.
There are two consumers (assumed to be price takers), c and j.

β − PC β − λ − P C dine at C
 
c dine at C j
U = J U = β > λ > 0,
β−λ−P dine at J, β − PJ dine at J,

β is the utility of dinner and λ is the disutility if one goes to a less preferred restau-
rant.
Monopoly equilibrium: P C = P J = β, π1 = 2β.

Suppose now that firm 2 opens a Chinese restaurant in the same town.

1. If firm 1 does not close its Chinese restaurant, the equilibrium will be P C = 0,
P J = λ, π1 = λ, π2 = 0.
2. If firm 1 closes its Chinese restaurant, the equilibrium will be P C = β = P J ,
π1 = β = π 2 .

The conclusion is that firm 1 will close its Chinese restaurant.

8.5 Contestable Market of Baumol/Panzar/Willig (1982)

Deregulation trend in US in later 1970s.
Airline industries: Each line is an individual industry.

Contestable market: In certain industries entry does not require any sunk cost. In-
cumbent firms are constantly faced by threats of hit-and-run entry and hence behave
like competitive firms (making normal profits).

Assumption: Potential entrants and an incumbent produce a homogenous product

and have the same cost function TC(qi ) = F + cqi . The market demand is p = a − Q.
 75

An industry configuration: (pI , q I ).

Feasibility: (1) pI = a − q I . (2) π I = pI q I − (F + cq I ) ≥ 0.
Sustainability: 6 ∃(pe , q e ) such that pe ≤ pI , q e ≤ a − pe , π e = pe q e − (F + cq e ) > 0.

A contestable-market equilibrium: A feasible, sustainable configuration.

p
6
@ p=a−q
@
@
@
@
@
I @r
p @ F
@ ATC = −c
q
@ -q
qI
Extension: 1. More incumbent firms. 2. More than 1 products.

Comments: 1. If there are sunk costs, the conclusions would be reversed. See Stiglitz
(1987) discussed before.
2. If incumbents can respond to hit-and-run entries, they can still make some positive
profits.

8.6 A Taxonomy (}é¶) of Business Strategies

Bulow, Geanakoplos, and Klemperer (1985), “Multimarket Oligopoly: Strategic Sub-
stitutes and Complements,” JPE.
A 2-period model: Firm 1 chooses K1 at t = 1 and firms 1 and 2 choose x1 , x2
simultaneously at t = 2.

Π1 = Π1 (K1 , x1 , x2 ), Π2 = Π2 (K1 , x1 , x2 ).

The reduced payoff functions at t = 1 are

Π1 = Π1 (K1 , x∗1 (K1 ), x∗2 (K1 )), Π2 = Π2 (K1 , x∗1 (K1 ), x∗2 (K1 ),

∂Π1 ∂Π2
where x∗1 (K1 ) and x∗2 (K1 ) are the NE at t = 2, given K1 , = = 0.
dx1 dx2
top dog (»−, _5T‘): be big or strong to look tough or aggressive.
puppy dog ({ü, Qm‘): be small or weak to look soft or inoffensive.
lean and hungry look (_|Ï.ñ÷.§, ü−): be small or weak to look tough or
aggressive.
fat cat (]ý×j): be big or strong to look soft or inoffensive.
 76

8.6.1 Deterence case

If firm 1 decides to deter firm 2, firm 1 will choose K1 to make Π2 = Π2 (K1 , x∗1 (K1 ), x∗2 (K1 )) =
0.
dΠ2 ∂Π2 ∂Π2 dx∗1
= + .
dK1 ∂K1 ∂x1 dK1
Assume that ∂Π2 /∂K1 = 0 so that only strategic effect exists.
∂Π2 dx∗1
dΠ2 /dK1 = < 0: The investment makes firm 1 tough.
∂x1 dK1
∂Π2 dx∗1
dΠ2 /dK1 = > 0: The investment makes firm 1 soft.
∂x1 dK1
To deter entry, firm 1 will overinvest in case of tough investment (top dog strat-
egy _5T‘íõÆ%, entrant øªVÿbç.)

and underinvest in case of soft investment (lean and hungry look _|Ï.ñ÷.§
íšä, U entrant JÑÌ‚ªÇ).

8.6.2 Accommodation case

If firm 1 decides to accommodate firm 2, firm 1 considers maximizing Π1 = Π1 (K1 , x∗1 (K1 ), x∗2 (K1 )).

dΠ1 ∂Π1 ∂Π1 dx∗2 ∂Π1 ∂Π1 dx∗2 dx∗1

= + = + .
dK1 ∂K1 ∂x2 dK1 ∂K1 ∂x2 dx1 dK1
Assume that ∂Π1 /∂x2 and ∂Π2 /∂x1 have the same sign and that ∂Π2 /∂K1 = 0.
 1 ∗ ∗  2 ∗
∂Π dx2 dx1 ∂Π dx1
sign = sign × sign(R20 ).
∂x2 dx1 dK1 ∂x1 dK1
There are 4 cases:
∂Π2 dx∗1
1. Tough investment < 0 with negative R20 : “top dog” strategy, be big
∂x1 dK1
or strong to look tough or aggressive. _5T‘J9„ entrant −‘.

2. Tough investment with positive R20 : “puppy dog” strategy, be small or weak to
look soft or inoffensive. Qm‘, .bK@ entrant Jnù–¬.
∂Π2 dx∗1
3. Soft investment > 0 with negative R20 : “lean and hungry look” strat-
∂x1 dK1
egy, be small or weak to look tough or aggressive. _|Ï.ñ÷.§íšäJH
U entrant =−.

4. Soft investment with positive R20 : “fat cat” strategy, be big or strong to look
soft or inoffensive. ]ý×jJî¸ entrant ı/.
top dog: ı%v9Ê,Þí%, Xº6.
puppy: AŠí*üä, IAnÀí/A.
fat cat: (\ö5) ½bí’ŒA, À3. ‹‘5×A. A)<ícA.
 77

8.7 Limit Pricing as Cost Signaling, Milgrom/Roberts (1982)

Incumbent àQgµIV[ýAÐÑQA…, ò^05¼, ñíu® a potential entrant.

8.7.1 Assumptions of the model

t = 1, 2. Each period’s demand is P = 10 − Q.
Firm 1 is the incumbent, a monopoly in t = 1.
Firm 2 decides whether to enter in t = 2. If firm 2 enters, the market becomes
Cournot competition.
c2 = 1, F2 = 9, i.e., TC2 (q2 ) = 9 + q2 if q2 > 0
c1 = 0 with 50% probability and c1 = 4 with 50% probability.
Firm 1 knows whether c1 = 0 or c1 = 4 but firm 2 does not.
It is an incomplete information game: Firm 1 knows both its and firm 2’s payoff
functions but firm 2 knows only its own payoff functions.

8.7.2 Complete information case

If there is no uncertainty, firm 1 will choose the monopoly quantity at t = 1, i.e.,

q1 (c1 = 0) = 5, p1 (c1 = 0) = 5, π1 (c1 = 0) = 25; q1 (c1 = 4) = 3, p1 (c1 = 4) = 7, π1 (c1 = 0) = 9.

At t = 2, the duopoly equilibrium for firm 1 with c1 = 0 and firm 2 (c2 = 1) is

64
p = 11/3, q1 (c1 = 0) = 11/3, q2 = 8/3, π1 (c1 = 0) = 121/9, π2 = − 9 = −17/9.
9
The duopoly equilibrium for firm 1 with c1 = 4 and firm 2 (c2 = 1) is

p = 5, q1 (c1 = 4) = 1, q2 = 4, π1 (c1 = 4) = 1, π2 = 16 − 9 = 7.

Therefore, firm 2 will enter if c1 = 4 and not enter if c1 = 0.

The high cost firm 1 has incentives to confuse firm 2.

8.7.3 The case when firm 2 has no information

If firm 2 does not know the type of firm 1, firm 2’s expected profit when enters is
0.5(7) + 0.5(−17/9) = 23/9 > 0. Therefore firm 2 will choose to enter.
Firm 1 has incentives to let firm 2 know firm 1’s type.

8.7.4 Separating equilibrium

Firm 2 will use the monopoy price at t = 1 to make inference about the type of firm
1, i.e., firm 2 will not enter if p = 5 will enter if p = 7.
However, this is not an equilibrium. If the low cost firm 1 chooses the monopoly
quantity q1 (c1 = 0) = 5, the high cost firm 1 will have incentives to imitate.
To avoid being imitated, the low cost firm 1 will choose a lower price p = 4.17 (q1 =
5.83). In such case, the high cost firm 1 cannot gain by imitating because the gain at
 78

t = 2 is 9 − 1 = 8 whereas the lose at t = 1 due to imitation is 9 − 5.83(4.17 − 4) ≈ 8.

The separating equilbrium is as follows. At t = 1,

q1 (c1 = 0) = 5.83, p1 (c1 = 0) = 4.17, π1 (c1 = 0) = 24.31;

q1 (c1 = 4) = 3, p1 (c1 = 4) = 7, π1 (c1 = 0) = 9.

At t = 2, firm 2 will enter only if p1 > 4.17.

8.7.5 Pooling equilibrium

If the standard distribution of c1 is smaller, a separating equilibrium will not exist.
Instead, the high cost firm 1 will immitate the low cost firm 1 and firm 2 will not
enter.
For example: Prob[c1 = 0] = 0.8 and Prob[c1 = 4] = 0.2.

8.7.6 A finite version

Assume that p11 ∈ {7, 5, 4}, the game tree is


p 
0
 @ 
@ 1-p

 
1a
XX @1b
 X   XX
XXX
    X X
     X1X  XX 

 p1 =   p1 = 5 XXX p1 =X4XXX
 X
  7 




A
1 1
 2  2 X 2 X

A
A
A
A
A
E
AN E
AN E
AN E
AN E
AN E
AN

A
A
A
A
A
A
    
            
 
0 0 34 46 0 0 38 50 0 0 37 49
 10  18 
 0  0  6  14 
 0  0   1   9 
 0  0 
7 0 −1.9 0 7 0 −1.9 0 7 0 −1.9 0

There is a separating equilbrium p11a = 4, p11b = 7 and firm 2 chooses not to en-
ter (N) if p11 = 4 and enter (E) if p11 ∈ {7, 5}.
For p > 7/8.9, there is also a pooling equilbrium p11a = p11b = 5 and firm 2 chooses
not to enter (N) if p11 ∈ {5, 4} and enter (E) if p11 = 7.
 79

8.8 Chain-Store Game

Selton (1978), “The chain-store paradox,” Theory and Decision.
A single long-run incumbent firm (I) faces potential entry by a series of short-run
firms, each of which plays only once but observes all previous play. Each period, a
potentail entrant (E) decides whether to enter (e) or stay out (s) of a single market.
If s, I enjoys a monopoly in that market; if e, I must choose whether to fight (F ) or
to accommodate (A).


E
@

In the 1-period game, the only SPNE is {A, e}.
e @s
@

I In the T -period game, using backward induction,
@

a
F @A the only SPNE is also {A, e} in each period.
@ 0

−1
 
0
 In the ∞-period game, there is a SPNE such
−1 b that {F, s} is chosen in each period.
Single period game 

E1
 e  HH s
 HH

 
I1
H HH
 H A 
HH
F  H HH
 HH
  

E2F H
E2A H
E2s
 @  @  @s
e @ s e @ s e @

 @  
I2F I2A @ I2s @
F @ A

F @ A
 
F @ A
 
@ a−1 @ a @ 2a
@  −1  @  b  @  0 
           
−2 −1 0 −1 0 0 a−1 a 0
 −1   −1   b  b  0  0
−1 b −1 b −1 b
2-period game
Paradox: When T is large, the incumbent is tempered to fight to try to deter entry.
The SPNE above is counterintuitive.

8.8.1 Incomplete information and reputation

Kreps and Wilson (1982), “Reputation and imperfect information,” JET.
Milgrom and Roberts (1982), “Predation, reputation and entry deterrence,” JET.
With probability p0 the incumbent is “tough”, ie., will fight for sure.
With probability q 0 an entrant is “tough”, ie., will enter for sure. The events that
entrants being “taugh” are independent.
δ: the discount factor.
Since the strategies of both “tough” type I and E are given, we need only analyze
the equilibrium strategies of “normal” type.
 80

T = 1: It is easily shown that the NE, {µ∗ (I), µ∗ (E)}, is


b
e if p0 < ≡ p̄

µ∗ (I) = a, µ∗ (E) = 1+b
 s if p0 > p̄.

T = 2: There are three cases.

aδ − 1
(1) q 0 > ≡ q̄: It is not worthwhile for I to fight to deter entry:
aδ
e if p0 < p̄

∗ ∗ ∗ ∗
µ (I1) = µ (I2) = a, µ (E1) = µ (E2s) =
s if p0 > p̄,

∗ e if I accommodates at t = 1
µ (E2e) =
s if I fights at t = 1.
(2) q 0 < q̄ and p0 > p̄: Fighting will deter entry and µ∗ (I1) = F .
(3) q 0 < q̄ and p0 > p̄: Both fighting and accommodating are not equilibrium. I
will randomize. Let β ≡ P rob[F ]. β is such that E2’s posterior probability that I is
“tough” equals q̄:

p0 p0
P rob[“tough”|F ] = = q̄ ⇒ β = .
p0 + β(1 − p0 ) (1 − p0 )b

The total probability of fighting at t = 1 for E1 is

p0 + (1 − p0 )β = p0 (b + 1)/b.

Therefore, E1 will enter if p0 > p̄2 and stay out otherwise.

T = 3: (a) p0 > p̄2 , I will fight and E1 will stay out at t = 1.

(b) p̄3 < p0 < p̄2 , I will randomize at t = 1.
(c) p0 < p̄3 , I will accommodate and E1 will enter t = 1.

T > 3: Entrants will stay out until t = k such that p0 < p̄k .

When δ = 1: (a) q 0 < a/(1 + a), I will accommodate at first entry and reveal
its type. Hence, limT →∞ π/T = 0.
(b) q 0 > a/(1 + a), there exists an n(p0 ) such that I will fight until there are no more
than n(p0 ) entrants remaining. Hence, limT →∞ π/T = (1 − q 0 )a − q 0 .
 81

9 Research and Development (R&D)

R&D àß“ÞG|—Ë

1. hÞßj¶Z‰¼A…!Z, Uß“ Ò20½h|c

2. h߹Ljhä

3. R&D …™Ñ¼¬µI

4. xXØà½æ

5. R&D, _", D%Èê

6. R&D D Merger activities 5É[

R&D2Гç5ª0 (OECD 1980):

NØ 23%, lœ 18%, Úä 10%, »“ 9%, ë¹, ˘“, ˆ, ]E, ×- < 1%

Production and cost functions are black boxes created by economists. Investigat-
ing R&D processes helps us to open the boxes.

Process innovation: An innovation that reduces the production cost of a prod-

uct.
Product innovation: An innovation that creats a new product.

The distinction is not essential. A process innovation can be treated as the creation
of new intermediate products that reduce the production costs. On the other hand,
a product innovation can be regarded as an innovation that reduces the production
cost of a product from infinity to a finite value.

9.1 Classification of Process Innovations

Consider a Bertrand competition industry.
Inverse demand function: P = a − Q.
In the beginning, all firms have the same technology and P0 = C0 . Suppose that an
inventor innovates a new production procedure so that the marginal production cost
reduces to c < c0 .

P P
6 6
@ @

@r
@ @
@ r Pm (c)
P0 = c 0 @ @
@ @
c @ c @
@ @
@ @
@ -Q @ -Q
Q0 Qm
 82

2 Cases:
a+c
Drastic innovation (large or major innovation): If Pm (c) ≡ < c0 = P0 , then
2
the innovator will become a monopoly.
a+c
Non-drastic innovation (small or minor innovation): If Pm (c) ≡ > c0 = P0 ,
2
then the innovator cannot charge monopoly price and has to set P = c0 − .

P P
6 6
c0 @r
@ Drastic innovation @ Non-drastic innovation
r @r
@
@
@ r
P = Pm (c) @
Pm (c)
π @ P = c0
@
π @@
c c
@ @
@ @
@ -Q @ -Q
Qm Q
A drastic innovation will reduce the market price. A non-drastic innovation will not
change the Bertrand equilibrium price. In both cases the innovator makes positive
profits.

9.2 Innovation Race

êp (innovation): …V.æÊí¼¹, %â R&D z…“¨|V

_" (immitation): y¹Aíêp; ¦%âL² ˙ (backward engineering)

½µAíêp (duplication): .ø85-, %â R&D ½µAíêp

Patent right (ù‚ž): á¤#êpð5Öðù“ž; .<_"C.ø8½µ, ·u<"ù

‚ž

ù‚}é: hß¹, hj¶, hA}, hql

¦)ù‚5‘K: à (Usefulness), µÆ4 (non-triviality), hJ4 (novelty)
Oubçt
.?¦)ù‚

ù‚cf: Ñfn<"Aíù‚, êpð.âS¦híûê˜(

Innovation race (ù‚¬ˇ): ÖPêpð¬óûê¦/ø¼¹5ù‚ž
|lêp6¦)

ù‚, Or(6ªS¦ù‚cfG¨ûêHß¹
.¬|lêp6¦}œÖ5¾‘6
y (Consumer Loyalty)

½æ: u´}¨A¬¬, ¨‘ØÖ5ûê’Ä?

Assumptions:

1. 2 firms compete to innovate a product.

2. The value of the patent right to the product is $ V.

 83

3. To compete, each firm has to spend $ I to establish a research lab.

4. The probability of firm i innovating the product is α. The events of firms being
successful is independent.

5. If only one firm successes, the firm gains $ V. If both success, each gains 0.5 V.
If a firm fails, it gains 0.

6. The entry is sequential. Firm 1 decides first and then firm 2 makes decision.

9.2.1 Market equilibrium

Eπk (n): The expected profit of firm k if there are n firms competing.
ik : The investment expenditure of firm k, ik ∈ {0, I}.
n = 1: 
I if αV ≥ I
Eπ1 (1) = αV − I ⇒ i1 =
0 if αV < I.
n = 2: 
 I α(2 − α)
if V ≥I

α(2 − α) 2
Eπ2 (2) = V − I ⇒ i2 =
2  0 α(2 − α)
if V < I.

2
I/V
6
I
Eπ1 (1) = 0 or =α
V

1 firm I α(2 − α)
Eπ2 (2) = 0 or =
V 2
2 firms
-α

9.2.2 Social Optimal

If there are more firms, the probability of success is higher; On the other hand the
investment expenditure will be also higher. Eπ s (n): The expected social welfare if
there are n firms attempting.

Eπ s (1) = Eπ1 (1) = αV − I Eπ s (2) = 2α(1 − α)V + α2 V − 2I,

Eπ s (2) ≥ Eπ s (1) if and only if α(1 − α) ≥ I.

 84

I/V I/V
6
I 6
Eπ s (1) = 0 or =α
V

(I)
1 firm
(II)
s s I
2 firms Eπ (1) = Eπ (2) or = α(1 − α) (III)
-α V -α

(I): Social optimal is 1 firm, the same as market equilibrium number of firms.
(II): Social optimal is 1 firm, market equilibrium has 2 firms.
(III): Social optimal is 2 firms, the same as market equilibrium number of firms.

Area (II) represents the market inefficient area.

9.2.3 Expected date of discovery

Suppose that the R&D race will continue until the discovery.

ET (n): Expected date of discovery if there are n firms.

∞
X α 1
ET (1) = α + (1 − α)α2 + (1 − α)2 α3 + . . . = α (1 − α)t−1 t = = .
t=0
[1 − (1 − α)]2 α

∞
2
X 1
ET (2) = α(2−α)+(1−α) α(2−α)2+. . . = α(2−α) (1−α)2(t−1) t = < ET (1).
t=0
α(2 − α)

9.3 Cooperation in R&D

Firms’ cooperation in price setting is against anti-trust law. Cooperation in R&D
activities usually is not illegal. Therefore, firms might use cooperation in R&D as a
substitute for cooperation in price setting.
In this subsection we investigate the effects of firms’ cooperation in R&D on social
welfare.

A 2-stage duopoly game with R&D

t = 1: Both firms decide R&D levels, x1 and x2 , simultaneously.
t = 2: Both firms engage in Cournot quantity competition.
Market demend is P = 100 − Q.
Firm i’s R&D cost: TCi (xi ) = x2i /2.
Firm i’s unit production cost: ci (xi , xj ) = 50 − xi − βxj .
β R 0; if β > 0, it represents the spillover effect of R&D; if β < 0, it is the interference
effect.
 85

9.3.1 Noncooperative R&D equilibrium

When firms do not cooperate in R&D, they decide the R&D levels independent of
each other. We solve the model backward.

At t = 2, x1 and x2 are determined. The Cournot equilibrium is such that

(100 − 2ci + cj )2
Πi (ci , cj ) = − TCi (xi ).
9
Substituting the unit cost functions, we obtain the reduced profit function of t = 1:

[100 − 2(50 − xi − βxj ) + (50 − 2xj − βxi )]2 x2i

Πi (xi , xj ) = −
9 2
[50 + (2 − β)xi + (2β − 1)xj ]2 x2i
= − .
9 2
At t = 1, firm i chooses xi to maximize Πi (xi , xj ). FOC is

∂Πi 2(2 − β)[50 + (2 − β)xi + (2β − 1)xj ]

=0= − xi .
∂xi 9
In a symmetric equilibrium, xi = xj = xnc :

50(2 − β) 50[4.5 − 2(2 − β)(1 + β)]

x1 = x2 = xnc = , c1 = c 2 = ,
4.5 − (2 − β)(1 + β) 4.5 − (2 − β)(1 + β)

75 252 [9 − 2(2 − β)]

P nc − cnc = Qnc = , Π1 = Π2 = Πnc = .
4.5 − (2 − β)(1 + β) [4.5 − (2 − β)(1 + β)]2

9.3.2 Cooperative R&D equilibrium

When firms cooperate in R&D, they choose x1 = x2 = x so that

[50 + (1 + β)x]2 x2
Πi = Πj = Π(x) = − .
9 2
Then they decide the level of x to maximize Π(x). FOC is

∂Π 2(1 + β)[50 + (1 + β)x]

=0= − x.
∂x 9
Denote by xc the optimal level of x,

50(1 + β) 50[4.5 − 2(1 + β)2 ]

x1 = x 2 = x c = , c1 = c 2 = ,
4.5 − (1 + β)2 4.5 − (1 + β)2

75 252 [9 − 2(1 + β)2 ]

P c − cc = Qc = , Π1 = Π 2 = Π c = .
4.5 − (1 + β)2 [4.5 − (1 + β)2 ]2
Conclusions:
 86

1. Πc > Πnc .

2. If β > 0.5 then xc > xnc and Qc > Qnc .

3. If β < 0.5 then xc < xnc and Qc < Qnc .

When β > 0.5, consumers will be better off to allow R&D cooperation; social welfare
will definitely increase. When β < 0.5, consumers will be worse off to allow R&D
cooperation; but the social welfare also depends on the change in firms’ profits.

9.4 Patents
êp5òQgM: “¨Þß6‚⣾‘6”ì

êp5ÈQgM: óêhêp, ªœ.q©¾

ù‚ž: þ}#êpð5Ñ{, àJ2¥“h

ù‚¨AÖ´, Ou³ù‚„†³—DÓÄV2¥êp

Êù‚„|Û5‡, êpðÉ?à\òíj¶V\ˆAÐ힂
9õ,, ÛHíêpð?
à\òíj¶V¦)ªù‚yÅíÖ´‚â

Wà, Stradivarius Violin, Coca Cola

da Vinci †Ñ_AE7êp

ù‚žÅ: 1Å 17 , r¹ 20 , «É 20 

žDù‚žÅ.°
bçt.?C~ù‚, Oª\ò
Ú7,ñ˘kž

½æ: ù‚žÅbÖýn?Ê2¥“hDÖ´’Ä…5Ȧ)|_~¬?

9.4.1 Nordhous 1969 partial equilibrium model

P = a − Q: Demand function of a Bertrand competition market.
c: Unit production cost before R&D.
x: R&D magnitude. TC(x) = x2 /2: R&D expenditure.
c − x: Unit production cost of the innovator after R&D.
Assume that the innovation is non-drastic.
P
6
@
@ M (x) = x(a − c)
c @ DL(x) = x2 /2
@
@
M (x) @
DL(x)@
c−x @
@
@
@- Q
a−c a−c+x

T : Patent length (duration).

M (x): Innovator’s expected profit per period during periods T = 1, 2, . . . , T .
DL(x): Deadweight Loss due to monopoly (Bertrand competition).
ρ = 1/(1 + r): discount factor. (r is market interest rate.)
 87

Innovator’s problem:
T
X 1 − ρT 1 − ρT x2
max π(x : T ) = ρt−1 M (x)−TC(x) = M (x)−TC(x) = (a−c)x− .
x
t=1
1−ρ 1−ρ 2

1 − ρT 1 − ρT
FOC: (a − c) − x ⇒ x = (a − c).
1−ρ 1−ρ
∂x ∂x ∂x ∂x
Comparative statics: > 0, > 0, < 0, > 0.
∂T ∂a ∂c ∂ρ
Social optimal duration of patents:
" T # " ∞ #
X X x2 (a − c)x 1 − ρT x2
W (T ) = ρt−1 M (x) + ρt−1 DL(x) − = − .
t=1 t=T +1
2 1 − ρ 1 − ρ 2

(a − c)x 1 − ρ2 x2 1 − ρT
max − subject to x= (a − c).
x,T 1−ρ 1−ρ 2 1−ρ
Eliminating x:
2  2 
(a − c) 1 − ρT 1 − ρ2 1 1 − ρT (1 − ρT )3
 
a−c T
max (a−c)− (a − c) = 1−ρ − .
T 1−ρ 1−ρ 1−ρ 2 1−ρ 1−ρ 2(1 − ρ)
ln(1 − z)
Make change of variable z ≡ 1 − ρT , or T = . The problem becomes
ln ρ
z3 3z 2 p
max z − FOC: 1 − = 0, ⇒ z ∗ = 2(1 − ρ)/3
z 2(1 − ρ) 2(1 − ρ)
p
ln(1 − 2(1 − ρ)/3)
⇒ T∗ = .
ln ρ

9.4.2 General equilibrium models

K. Judd (1985) “On performance of patent,” Econometrica is a general equilibrium
model. His conclusion is that T ∗ = ∞:
1. All products are monopoly priced with the same mark-up ratio and therefore there
is no price distortion.
2. The R&D costs of an innovation should be paid by all consumers benefited from
it to avoid intertemporal allocation distortion. Therefore, infinite duration of patents
is needed.

C. Chou and O. Shy (1991) “Optimal duration of patents,” Southern Economic Jour-
nal: If R&D has DRTS, optimal duration of patents may be finite. There are also
many nonsymmetrical factors, eg., some products are competitively priced, demand
elasticities are different, etc.
 88

9.5 Licencing ù‚¤ž

More that 80% of innovators licence their patents to other firms to collect licencing
fees rather than produce products and make monopoly profits.
Kamien 1992
Consider a Cournot duopoly market with demand P = a − Q.
Firm 1 invents a new procedure to reduce the unit production cost from c to c1 = c−x.
Firm 2’s unit cost is c2 = c if no licencing.

9.5.1 Equilibrium without licencing

a − c + 2x a−c−x a + 2c − x
q1 = , q2 = , P = ,
3 3 3
(a − c + 2x)2 (a − c − x)2
π1 = , π2 = π̄2 = .
9 9

9.5.2 Equilibrium with per-unit fee licencing

Firm 1 can make more profit by licencing the new procedure to firm 2 and changing
per-unit fee for every unit sold by firm 2.
The maximum fee is φ = c2 − c1 = x.
Firm 2’s total cost per unit is still c2 (= c1 + x). Therefore, the equilibrium is the
same as without licencing except that firm 1 now collects x dollars per unit of q2 :
(a − c + 2x)2 (a − c − x)x (a − c − x)2
π1φ = + , π2 = π̄2 = .
9 3 9

9.5.3 Equilibrium with fixed-fee licencing

Firm 1 can also choose to charge firm 2 a fixed amount of money F , independent of
q2 .
Firm 2’s total cost per unit becomes c2 = c1 = c − x. Therefore, the equilibrium is
now
a−c+x a + 2c − 2x (a − c + x)2 (a − c + x)2
q1F = q2F = , PF = , π1F = +F, π2F = −F.
3 3 9 9
The (maximum) F is such that π2F = π̄2 . Therefore
(a − c + x)2 (a − c − x)2 (a − c)4x (a − c + x)2 (a − c)4x
F = − = , ⇒ π1F = + .
9 9 9 9 9

9.5.4 Comparison between π1φ and π1F

9 × (π1φ − π1F ) = (a − c)x > 0.

Therefore, firm 1 will prefer per-unit fee licencing.
The reason is: In the case of fixed-fee licencing, q1 + q2 ↑ and P ↓ and therefore firm
1’s total profit is smaller than that of per-unit fee licencing.
 89

9.6 Governments and International R&D Race

9.6.1 Subsidizing new product development
Sometimes governmental subsidies can have very substantial strategical effects.
Krugman (1986), Strategical Trade Policy and the New International Economics.
Boeing (I, a US firm) and Airbus (II, an EU firm) are considering whether to develop
super-large airliners.

Without intervention, the game is:

I \ II Produce Don’t Produce

Produce (-10, -10) (50, 0)
Don’t Produce (0, 50) (0, 0)
There are two equilibria: (Produce, Don’t) and (Don’t, Produce).

If EU subsidizes 15 to Airbus to produce, the game becomes:

I \ II Produce Don’t Produce

Produce (-10, 5) (50, 0)
Don’t Produce (0, 65) (0, 0)
There is only one equilibrium: (Don’t, Produce).
In this case, by subsidizing product development, a governmental can secure the world
dominance of the domestic firm.

9.6.2 Subsidizing process innovation

If we regard the R&D levels x1 and x2 in the R&D cooperation model as the amount
of R&D sponsored by governments 1 and 2, it becomes a model of government subsidy
competition.

9.7 Dynamic Patent Races

Tirole section 10.2.
Reinganum (1982) “A dynamic game of R&D,” Econometrica.

9.7.1 Basic model

2 firms compete in R&D to win the patent of a new product.
xi : the size of R&D lab established (incurring a continuous cost of xi per unit of
time) by firm i, i = 1, 2.
V : the value of the patent per unit of time. r: interest rate.
Ti : firm i’s discovery time.
Assumption: T1 and T2 are independent exponential random variable:

Ti ∼ 1 − e−h(xi )Ti , density function: h(xi )e−h(xi )Ti ,

 90

where [h(xi )]−1 is expected discovery time of firm i.

E(Ti ) = [h(xi )]−1 , h(xi ) > 0, h0 (xi ) > 0, h00 (xi ) < 0.

Industry discovery time: T̂ ≡ min{T1 , T2 } ∼ 1 − e−[h(x1 )+h(x2 )]T̂ because

Prob{T̂ > T } = Prob{T1 > T, T2 > T } = e−h(x1 )T e−h(x2 )T = e−[h(x1 )+h(x2 )]T .

h(x1 )
Firm 1’s winning probability: Prob[T1 = T, T2 > T |T̂ = T ] = :
h(x1 ) + h(x2 )

Prob[T1 , T̂ ∈ (T, T + dt)] h(x1 )e−h(x1 )T dt[1 − (1 − e−h(x2 )T )] h(x1 )

≈ = .
Prob[T̂ ∈ (T, T + dt)] [h(x1 ) + h(x2 )]e−[h(x1 )+h(x2 )]T dt h(x1 ) + h(x2 )

T2
6
A {T̂ ∈ (T, T + dt)} = A ∪ B ∪ C
T +dt A = {T1 ∈ (T, T + dt), T2 ≥ T + dt}
C B
T B = {T2 ∈ (T, T + dt), T1 ≥ T + dt}
C = {T1 , T2 ∈ (T, T + dt)}, Prob[C] ≈ 0.
- T1
T T +dt

Given (x1 , x2 ), the expected payoff of firm 1, Π1 (x1 , x2 ) is (Π2 is similar)

Z ∞" Z ∞ Z T̂ #
h(x1 )
e−rt V dt − e−rt x1 dt [h(x1 ) + h(x2 )]e−[h(x1 )+h(x2 )]T̂ dT̂
0 h(x 1 ) + h(x 2 ) T̂ 0
Z ∞" #
h(x1 ) e−rT̂ V (1 − e−rT̂ )x1
= − [h(x1 ) + h(x2 )]e−[h(x1 )+h(x2 )]T̂ dT̂
0 h(x 1 ) + h(x 2 ) r r
h(x1 )V /r x1 [h(x1 ) + h(x2 )]x1 /r h(x1 )V − rx1
= − + = .
h(x1 ) + h(x2 ) + r r h(x1 ) + h(x2 ) + r r[h(x1 ) + h(x2 ) + r]

FOC for symmetric NE with x1 = x2 = x:

[2h(x)+r][h0 (x)V −r]−[h(x)V −rx]h0 (x) = h(x)h0 (x)V +rh0 (x)(x+V )−2rh(x)−r 2 = 0.

Social welfare when x1 = x2 = x:

!
Z ∞ Z ∞ Z T̂
W (x) = e−rt V dt − e−rt 2xdt 2h(x)e−2h(x)T̂ dT̂
0 T̂ 0
∞
1
Z h i
= e−rT̂ V − (1 − e−rT̂ )2x 2h(x)e−2h(x)T̂ dT̂
r 0
2h(x)V /r 2x 2h(x)x/r 2[h(x)V − rx]
= − + = .
2h(x) + r r 2h(x) + r r[2h(x) + r]
 91

FOC for social optimal:

[2h(x) + r][h0 (x)V − r] − 2[h(x)V − rx]h0 (x) = rh0 (x)(2x + V ) − 2rh(x) − r 2 = 0.

√ √
Example: h(x) = 2 x, h0 (x) = 1/ x.
FOC for NE:
2
p
r(x + V ) √ √ 2V − r + (2V − r 2 )2 + 12r 2 V
2V + √ − 4r x − r 2 = 0, ⇒ xn = .
x 6r

FOC for social optimal:

√
r(2x + V ) √ 2 √ −r 2 + r 4 + 8r 2 V
√ − 4r x − r = 0, ⇒ xs = .
x 4r
√ √
Derivation: Let zn ≡ xn , zs ≡ xs , fn (z) ≡ 3rz 2 + (r 2 − 2V )z − rV , and fs (z) ≡
2rz 2 +r 2 z−rV . fn (zn ) = 0 and fs (zs ) = 0. Direct computation shows that fn (zs ) < 0,
limz→∞ fn (z) = ∞ > 0. Hence zn > zs and therefore
√ √
xn 2 k − 1 + k 2 + 4k + 1 2V
√ = √ > 1, k = 2 .
xs 3 −1 + 1 + 4k r

Therefore, the equilibrium R&D level is greater than the social optimal level.

Extensions: 1. h(x) = λh̄(x/λ) = λ1−a xa .

2. When there are n > 2 firms.
3. n is endogenouse and optimal x and n.

9.7.2 R&D race between an incumbent and a potential entrant

2 firms compete in R&D to win the patent on a new procedure with unit cost c.
Firm 1: Incumbent with initial unit production cost c̄ > c.
Firm 2: A potential entrant.
Πm (c̄): Firm 1’s monopoly profit before the discovery of the new procedure.
Πm (c): Firm 1’s monopoly profit if firm 1 wins.
Πd1 (c̄, c): Firm 1’s duopoly profit if firm 2 wins.
Πd2 (c̄, c): Firm 2’s duopoly profit if firm 2 wins.

Assumption 1: Πm (c) ≥ Πd2 (c̄, c) + Πd1 (c̄, c).

Assumption 2: The patent length is ∞.

Using the same derivation as basic model,

h(x1 )Πm (c) + h(x2 )Πd1 (c̄, c) + r[Πm (c̄) − x1 ]

V1 (x1 , x2 ) = ,
r[h(x1 ) + h(x2 ) + r]
h(x2 )Πd2 (c̄, c) − rx2
V2 (x1 , x2 ) = .
r[h(x1 ) + h(x2 ) + r]
 92

Comparing the payoff functions reveals that firm 2’s payoff function is essentially the
same as that of the basic model. Further comparison between firm 1 and firm 2’s
payoff functions reveals that firm 1’s incentives are different in two ways:

Efficiency effect: Πm (c) − Πd1 (c̄, c) ≥ Πd2 (c̄, c), firm 1 has more incentives to win
the race. and therefore x1 tends to be greater than x2 in this aspect.

Replacement effect: If firm 1 wins, he replaces himself with a new monopoly.

∂ 2 V1
Therefore, firm 1 tends to delay the discovery date. < 0 tends to make
∂Πm (c̄)∂x1
x1 smaller.

The net effect depends on which one dominates. Following are two extreme cases:

Drastic innovation: Πd1 (c̄, c) = 0 and Πd2 (c̄, c) = Πm (c). No efficiency effect and
x1 < x 2 .

Almost linear h(x) case: h = λh(x/λ), λ → ∞, and h(x) ≈ h0 (0)x.

In this case h(x1 ) and h(x2 ) are very large and firm 1 is more concerned with his win-
ning the race rather than replacing itself. Therefore, replacement effect dominates
and x1 > x2 .
 93

10 Network Effects, Compatibility, and Standards æ˜^‹,

ß¹1ñ4, ß¹d™Ä
Aéuþ}4Ó
.ÞßC¾‘·Õ¶4 (externalities)

ÞßÞ: } ¯T, óß×

¾‘Þ: ˇ, þ}>%, >²m7, ów°

U i = U i (xi ) ⇒ U i = U i (xi , x−i , y), F j (y j ) = 0 ⇒ F j (y j , y −j , x) = 0.

Arrow-Debreu general equilibrium model assumes no externalities. With production

and/or consumption externalities, we have to modify Arrow-Debreu model.

Þß, ¾‘Õ¶4Dæ˜^‹, ß¹1ñ4É
%ÈWÑA.âóº¯n?êµ%È^0,

T¯Þº¹”
Wà

xkdå, œb§ (qwerty,dvorak)

¾© (L„, t„)
>¦d† (Ô¬i, Ô˝i, š5, ˜ )
¶ „
…PŒÞ (>q2, p×ÀP)

—áY$ø2ÅQO6$ø¥<d†, ?¹™Ä“ (standardization), êr1ñ4

½æ: Aéʪ¥, ø….iÊÚ , hß¹.i|Û, ج#|™Ä“}®×ê

0ä%ÈVr7“ (globalization), 1ñ4D™Ä“VAѽb{æ
®Å¼·ı

AÐíß¹d?AÑ0ä™Ä, J¦)1ñ4, êµ|×5æ˜^‹, ¦|×5 Ò2
0

Wà: òj&Úe HDTV, Ú7£w¶iß¹, ®æ˜ß¹

10.0.3 3 concepts
1. Compatibility (1ñ4): .°ÉKªJ²Uà (øu°™Ä)

Standardization (™Ä“): FÉK·ªJ²Uà

2. Downward-compatibility: hß¹ªJHHß¹, OHß¹.øì ªJHhß¹

Wà Pentium III vs 486

3. Network externalities (æ˜Õ¶4): ¾‘6^àÓ°UàAb Ó‹7Ó‹

ª1ñß¹5W: ßà ¯

.ª1ñß¹5W: ÎóœœñDŸå ¯

<8”, ª1ñß¹$Aø__üÕÈ, ÕÈqѵI:_¼

VHS vs β, òœ, CD player, DVD, MO, etc.
ASCII Ñ 7-bit ™Ä“å{

Extended ASCII Ñ 8-bit ³™Ä“å{ (rÖ.°™Ä)

 94

10.0.4 A standardization game

Firm A (USA) and firm B (Canada) are choosing a standard for their product.
α-standard (L„d, CÔ¬iW , etc.) β-standard (t„d, CÔ˝iW , etc.)

A\B α β
α (a, b) (c, d)
β (d, c) (b, a)
1. If a, b > max{c, d} (battle of the sexes), then (α, α) and (β, β) are both NE. They
choose the same standard (™Ä“).

2. If c, d > max{a, b}, then (α, β) and (β, α) are both NE. They choose the dif-
ferent standards (®Wwu).

10.1 Network Externalities

10.1.1 Rohlfs phone company model
Rohlfs 1974, “A Theory of Interdependent Demand for a Communication Service,”
Bell Journal of Economics.

Consumers are distributed uniformly along a line, x ∈ [0, 1].

r
0 x 1
Consumers indexed by a low x are those who have high willingness to pay to subscribe
to a phone system.

x n(1 − x) − p if x subscribes to the phone system
U =
0 if x does not,

n, 0 < n < 1: the total number of consumers who actually subscribe.

p: the price of subscribing.
x̂: the marginal consumer who is indifference between subscribing and not.
t
0 subscribers x̂ 1
In a rational expectation equilibrium, x̂ = n and U x̂ = n(1 − x̂) − p = 0, ⇒
Inverse demand function: p √= x̂(1 − x̂).
1 ± 1 − 4p
Demand function: x̂ = .
2
 95

1
p If 0 < p < , then there are two
6 4
possible marginal consumers,
√
U x̂<0 U x̂<0 1 ± 1 − 4p
2 x̂ ↓ x̂ ↓ x̂ = . The smaller is
2
9 2
U x̂>0 unstable. The diagram uses p =
9
b - b
x̂ ↑
 - x to illustrate.
0 1/3 2/3 1
The phone company maximizes its profits:
2 2
max px̂ = x̂2 (1 − x̂), FOC: 2x̂ − x̂2 = 0, ⇒ x̂∗ = , p∗ = .
x̂ 3 9

Dynamic model and critical mass (@ä!…îEb):

2
Assumption 1: The phone company sets p = p∗ = .
9
Assumption 2: At t, consumers expect that nt = x̂t−1 .
2 2 2
nt (1 − x̂t ) = p = , ⇒ nt (1 − nt+1 ) = p = , ⇒ nt+1 = 1 − ≡ f (nt ).
9 9 9nt
1 2
There are two equilibria: n∗∗ = and n∗ = . n∗∗ is unstable and n∗ is stable:
3 3
2 1
f 0 (n) = f (n∗∗ ) = 2 > 1, f (n∗ ) = < 1.
9n2 2
If the initial subscription is n0 < 1/3, nt → 0.
If the initial subscription is n0 > 1/3, nt → 2/3. 1/3 is the critical mass.
nt+1
6

rstable
f (nt )

r
unstable
◦
45 - nt
n∗∗ n∗
 96

10.1.2 The standardization-variety tradeoff ™Ä“CÖj“?

Consumers are distributed uniformly along a line, x ∈ [0, 1].

0 prefer A a prefer B 1
2 brands/standards, A (à¬G) and B (à˝G).
a > 0 consumers prefer A-standard.
b = 1 − a > 0 consumers prefer B-standard.
xA : number of consumers using A-standard.
xB : number of consumers using B-standard.
δ: the disutility of using a less prefered standard.
 
A xA use A-standard B xA − δ use A-standard
U = U =
xB − δ use B-standard, xB use B-standard.
Consumer distribution: (xA , xB ) such that xA , xB ≥ 0 and xA + xB = 1.
A-standard distribution: A distribution such that (xA , xB ) = (1, 0).
B-standard distribution: A distribution such that (xA , xB ) = (0, 1).
Incompatible AB-standards distribution A distribution with xA , xB > 0.
Equilibrium: (xA , xB ) such that none wants to switch to a different brand/standard.
Proposition 10.3: If δ < 1, then both A-standard and B-standard are equilibrium. If
δ > 1, then both A-standard and B-standard are not equilibrium.

Proof: If δ > 1 and every one chooses the same brand (either A or B), then none
wants to switch to a different brand. If δ < 1, then the cost of switching to a preferred
brand is less than the benefit and therefore a single standard equilibrium cannot exist.
1−δ
Proposition 10.4: If a, b > , then (xA , xB ) = (a, b) is an equilibrium.
2
Proof: Given the distribution (xA , xB ) = (a, b), the utility levels are
 
A a use A br. B a − δ = 1 − b − δ < b use A br.
U = U =
b − δ = 1 − a − δ < a use B br. b use B br.
Therefore, none will switch to a different brand.

b
6
@

1−δ r
@
@ 2-standard equilibrium
@
@
@ range
@
1−δ @
@
@
2 @
r
@
@ -a
1−δ
2
1−δ
 97

Social welfare: W (xA , xB ) = aU a + bU b , where U a (U b ) is the utility level of A-

prefered consumers (B-prefered consumers).

W (A) = W (1, 0) = a + b(1 − δ) = 1 − bδ.

W (B) = W (0, 1) = a(1 − δ) + b = 1 − aδ.
W (AB) = W (a, b) = a2 + b2 = (1 − b)2 + b2 = 1 − b − b(1 − 2b) = 1 − 2ab.

Proposition 10.5: If a > b, then W (A) > W (B).

Proof: If W (A) − W (B) = a + b(1 − δ) − [a(1 − δ) − b] = (a − b)δ > 0.

Proposition 10.6: 1. If δ < 1, then W (AB) < max{W (A), W (B)}.

δ
2. If δ > 1 and > max{a, b}, then W (AB) > max{W (A), W (B)}.
2
Proof: Assume that a > b (or 1 − 2b > 0). (case b > a is similar.)
1. If δ < 1, then max{W (A), W (B)} = W (A) = 1 − bδ > 1 − b > 1 − b − b(1 − 2b) =
W (AB).
2. If δ > 2a > 1, then W (AB) − W (A) = b(δ − 2a) > 0.

Proposition 10.7: If δ < 1, then market failure can happen.

Remark: When a > b and δ < 1, the social optimal is A-standard. However, both
the incompatible standards and B-standard can also be equilibrium.

10.2 Supporting Services and Network Effects

Network effects can occur even there is no network externalities. For example, when
there is a complementary supporting industry exhibiting increasing returns to scale
such as PC industry.

10.2.1 Basic model

Chou/Shy (1990), “Network effects without network externalities,” International Jour-
nal of Industrial Organization.

A PC industry with 2 brands, A and B and prices PA and PB .

Consumers are distributed uniformly along a line, δ ∈ [0, 1].
-δ
0 prefer A δ prefer B 1

Let NA and NB be the numbers of software pieces available to computers A and

B, respectively.
 98

The utility of consumer δ is

 √
δ (1 −√δ) NA if δ buys A-system
U =
δ NB if δ buys B-system.
√
In the above, Ni can easily be generalized to Niα .

Marginal consumer δ̂:

√
p p NA
U (A) = (1 − δ̂) NA = U δ̂ (B) = δ̂ NB ,
δ̂
⇒ δ̂ = √ √ .
NA + NB
Market shares: δA = δ̂ and δB = 1 − δ̂.
If NA increases (or NB decreases), δ̂ will decrease, A’s market share will increase and
B’s market share will decrease.
√ √ √
NA NB δB 1 − δ̂ NB
δA = δ̂ = √ √ , δB = 1 − δ̂ = √ √ , = =√ .
NA + NB NA + NB δA δ̂ NA
In this model, there are two monopolistic competition software industries, A-software
and B-software.
Assume that each consumer has Y dollars to spend on a computer system. If a
consumer chooses i-stytem, he has Ei ≡ Y − Pi to spend on software.
There is a variety effect in each software industry and the number of software pieces
is propotional to the aggregate expenditure spent on them:
NA = kδA EA = k δ̂(Y − PA ), NB = kδB EB = k(1 − δ̂)(Y − PB ).
√
v !
u 
1 − δ̂ NB u 1 − δ̂ Y − PB 1 − δ̂ Y − PB EB
=√ = t , ⇒ = = .
δ̂ NA δ̂ Y − PA δ̂ Y − PA EA
Therefore, the equilibrium market shares are
EA Y − PA EB Y − PB
δA = δ̂ = = , δB = 1 − δ̂ = = .
EA + E B 2Y − PA − PB EA + E B 2Y − PA − PB
Network effects: When δ̂ goes down, δA goes down (δB goes up), which in turn will
reduce NA (increase NB ). Finally, A-users’ utility levels will decrease (B-users’ utility
levels will increase).
The network effect here is the same as the variety effects in Dixit/Stiglitz monopolis-
tic competition model.

Duopoly price competition:

The profit functions of firms A and B are

PA (Y − PA ) PB (Y − PB )
ΠA (PA , PB ) = δA PA = , ΠB (PA , PB ) = δB PB = .
2Y − PA − PB 2Y − PA − PB
The price competition equilibrium is derived in Chou/Shy (1990).
 99

10.2.2 Partial compatibility

Chou/Shy (1993) “Partial compatibility and supporting services”, Economic Letters.

In the basic model, A-computers and B-computers are incompatible in the sence
that A-computers use only A-software and B-computers use only B-software. The
model can be generalized to the case when computer firms design their machines in
such a way that some fraction of B-software can be used in A-machines and vice
versa.
Let ρA (ρB ) be the proportion of B-software (A-software) that can be run on A-
computers (B-computers).
Incompatibility: ρA = ρB = 0.
Mutual compatibility: ρA = ρB = 1.
One-way compatibility: ρA = 1, ρB = 0 or ρA = 0, ρB = 1.

nA : number of software pieces written for A-computers.

nB : number of software pieces written for B-computers.
NA − ρ A NBNB − ρ B NA
N A = n A + ρ A nB , N B = n B + ρ B nA , ⇒ nA = nB = .
1 − ρ A ρB 1 − ρ A ρB
(5)
δi Ei = δi (Y − Pi ): Aggregate expenditure on software from i-computer users.
ni ρ j nj
δi Ei + δj Ej : Aggregate expenditure on i-software.
Ni Nj
As in the basic model, the number of i-software, ni , is proportional to the aggre-
gate expenditure on i-software:
   
nA ρ B nA ρ A nB nB
nA = k δA EA + δ B E B , nB = k δA EA + δB EB ,
NA NB NA NB

(1 − ρA ρB )δA EA (1 − ρA ρB )δB EB
⇒ NA = , NB = . (6)
k(1 − ρB ) k(1 − ρA )
As in the basic model,

√
v !
u
1 − δ̂ NB u 1 − δ̂ (1 − ρB )EB
= √ =t ,
δ̂ NA δ̂ (1 − ρA )EA

1 − δ̂ (1 − ρB )EB (1 − ρB )(Y − PB )
⇒ = = .
δ̂ (1 − ρA )EA (1 − ρA )(Y − PA )
Other things being equal, if firm A increases the degree of compatibility ρA , the
number of software pieces run on A-computers will decrease and hence its market
share δA = δ̂ will also decrease.
 100

10.3 The Components Model

Matutes/Regibeau (1988), “Mix and Match: Product Compatibility Without Net-
work Externalities,” RAND Journal of Economics.
Economides (1989), “Desirability of Compatibility in the Absence of Network Exter-
nalities,” American Economic Review.

AS1 2 firms, A and B, producing XA , YA , XB , YB .

AS2 Marginal costs are 0.

AS3 X and Y are completely complementary.

AS4 3 consumers: AA, AB, BB. You need an X and a Y to form a system S.

2 situations:
1. Incompatibility: A and B’s products are not compatible. You have to buy XA YA
or XB YB .
2. Compatibility: A and B’s products are compatible. There are 4 possible systems:
XA YA , XA YB , XB YA , and XB YB .

Consumer ij’s utility, ij = AA, AB, BB, is

 x y 0 0
 2λ − (Pxi0 + Pyj 0 ) i0 j = ij, i.e.,
 X, Y ·¯<

λ − (Pi0 + Pj 0 ) i = i or j = j but i0 j 0 6= ij, i.e., X, Y øá¯<

0

U ij =

 −(Pix0 + Pjy0 ) i0 6= i and j 0 6= j, i.e., X, Y ·.¯<

0 .¾‘



10.3.1 Incompatibility
There are only 2 systems: A-system (XA YA ) and B-system (XB YB ).
PA = PAx + PAy , PB = PBx + PBy : Price of system A and system B, respectively.

Equilibrium: (PAI , PBI ; qAI , qBI ) such that

1. PiI maximizes Πi (Pi , PjI ).
2. (qAI , qBI ) are the aggregate demand of the consumers at price (PAI , PBI ).

Lemma 10.1. In an equilibrium, consumer AA (consumer BB) purchases A-system

(B-system).

Proof: If in an equilibrium consumer AA purchases B-system, it must be PB = 0. In

that case, firm A can set 0 < PA < 2λ to attract consumer AA.

Proposition 10.13. There are 3 different equilibria:

1. (PAI , PBI ; qAI , qBI ) = (λ, 2λ; 2, 1). AA and AB purchase A-system and BB pur-
chases B-system, ΠA = ΠB = 2λ, CS = λ, social welfare is 5λ.
 101

2. (PAI , PBI ; qAI , qBI ) = (2λ, λ; 1, 2). AA purchases A-system and BB and AB pur-
chase B-system. ΠA = ΠB = 2λ, CS = λ, social welfare is 5λ.

3. (PAI , PBI ; qAI , qBI ) = (2λ, 2λ; 1, 1). AA purchases A-system and BB purchases B-
system. AB chooses to do without. ΠA = ΠB = 2λ, CS = 0, social welfare is
4λ.

10.3.2 Compatibility
4 systems: AA-system (XA YA ), AB-system (XA YB ), BA-system (XB YA ), and BB-
system (XB YB ).

c c c c c c c c
Equilibrium: (PAx , PAy , PBx , PBy ; qAx , qAy , qBx , qBy ) such that
c c c c
1. (Pix , Piy ) maximizes Πi (Pix , Piy ; Pjx , Pjy ).
c c c c c c c c
2. (qAx , qAy , qBx , qBy ) are the aggregate demand of the consumers at price (PAx , PAy , PBx , PBy ).

c c c c
Proposition 10.14. There exists an equilibrium such that PAx = PAy = PBx = PBy =
c c c c c c AA AB BB
λ, qAx = qBy = 2, qAy = qBx = 1, ΠA = ΠB = 3λ, U = U =U = 0, and social
welfare is 6λ.

10.3.3 Comparison
1. Consumers are worse off under compatibility.

2. Firms are better off under compatibility.

3. Social welfare is higher under compatibility.

Extension to a 2-stage game: If at t = 1 firms determine whether to design compatible

components and at t = 2 they engage in price competition, then they will choose
compatibility.
 102

11 Advertising
Advertising is defined as a form of providing information about prices, quality, and
location of goods and services.
2% of GNP in developed countries.
vegetables, etc., < 2% of sales.
cosmetics, detergent, etc., 20-60 % of sales.
In 1990, GM spent $63 per car, Ford $130 per car, Chrysler $113 per car.

What determines advertising in different industries or different firms of the same

industry?
Economy of scale, advertising elasticity of demand, etc.

Kaldor (1950), “The Economic Aspects of Advertising,” Review of Economic Studies.

Advertising is manipulative and reduces competition.
1. Wrong information about product differentiations ⇒ increases cost.
2. An entry-deterring mechanism ⇒ reduces competition.

Telser (1964), “Advertising and Competition,” JPE

Nelson (1970),“Information and Consumer Behavior,” JPE
Nelson (1974),“Advertising as Information,” JPE
Demsetz (1979), “Accounting for Advertising as a Barrier to Entry,” J. of Business.
Positive sides of advertising: It provides produt information.

Nelson:
Search goods: Quality can be identified when purchasing. ⇒ .Ûµ

Experience goods: Quality cannot be identified until consuming. ⇒ µª àÛb

Persuasive advertising: Intends to enhance consumer tastes, eg diamond.

Informative advertising: Provides basic information about the product.

11.1 Persuasive Advertising

Q(P, A) = βAa P p , β > 0, 0 < a < 1, p < −1.

A: expenditure on advertising.
a : Advertising elasticity of demand.
p : Price elasticity of demand.
c: Unit production cost.
max Π = P Q − cQ − A = (P − c)βAa P p − A.
P,A

FOC with respect to P :

∂Π cp Pm − c p
= βAa [(p + 1)P p − cp P p ] = 0, ⇒ Pm = , m
= .
∂P p + 1 P −1
 103

FOC with respect to A:

∂Π Pm − c A 1 a A
= a βAa −1 P p (P − c) − 1 = 0, ⇒ m
= , = .
∂A P P Q a p PQ

Proposition: The propotion of advertising expenditure to total sales is equal to the

ratio of advertising elasticity to price elasticity.

11.1.1 Example: β = 64, a = 0.5, p = −2, c = 1

√ √ √
Q = 64 AP −2 , P = 8A1/4 Q−1/2 , ⇒ P m = 2, Qm = 16 A, Am = 64, Π = 16 A,
√ √
Z 16 A Z 16 A √ √
⇒ CS(A) = P (Q)dQ − P m Qm = 8A1/4 Q−1/2 dQ − 32 A = 32 A.
0 0
√
Social welfare: W (A) ≡ CS(A) + Π(A) − A = 48 A − A.
24
Social optimal: W 0 (A) = 0 = √ − 1, ⇒ A = 242 = 576 > Am = 64.
A
Remark: 1. Does CS(A) represent consumers’ welfare? If it is informative adver-
tising, consumers’ utility may increase when A increases. However, if consumers’ are
just persuased to make unnecessary purchases, the demand curve does not really re-
flect consumers’ marginal utility.
2. Crowding-out effect: Consumption for other goods will decrease.
3. A can be interpreted as other utility enhancing factors.

11.2 Informative Advertising

Benham (1972), “The effects of Advertising on the Price of Eye-glasses,” J. of Law
and Economics.
Ê1Å, Š¢iŸµí˚iŸg¦œò

Consumers often rely on information for their purchases. The problem is whether
there is too little or too much informative advertising.

Butters (1977), “Equilibrium Distributions of Sales and Advertising Prices,” Review

of Economic Studies.
Informative Advertising level under monopolistic competition equilibrium is social
optimal.

Grossman/Shapiro (1984), “Informative Advertising with Differentiated Prdoducts,”

Review of Economic Studies.
In a circular market, informative advertising level is too excessive.

Meurer/Stahl (1994), “Informative Advertising and Product Match,” IJIO.

In the case of 2 differentiated products, the result is uncertain.
 104

11.2.1 A simple model of informative advertising

1 consumer wants to buy 1 unit of a product.
p: the price. m: its value.

m − p 

U=
0 .

If the consumer does not receive any advertisement, he will not purchase.
If he receives an advertisement from a firm, he will purchase from the firm.
If he receives 2 advertisements from 2 firms, he will randomly choose one to buy.

2 firms, unit production cost is 0, informative advertising cost is A.

Each chooses either to advertise its product or not to advertise it.

 p − A if only firm i’s ad is received.
 p


− A if both firms’ ad are received.
πi = 2


 −A if firm i’s ad is not received.
0 if firm i chooses not to advertise.


Let δ be the probability that an advertisement is received by the consumer.

 p
 δ(1 − δ)(p − A) + δ 2 ( − A) − (1 − δ)A ≡ π(2)
 if both choose to advertise.
2
Eπi = δ(p − A) − (1 − δ)A ≡ π(1) if only firm i chooses to advertise.

0 if firm i chooses not to advertise.


If p/A > 1/δ, ⇒ π(1) > 0, ⇒ at least one firm will choose to advertise.
If p/A > 2/[δ(2 − δ)], ⇒ π(2) > 0, ⇒ both firms will choose to advertise.

p/A
6
π(2) > 0,
2 firms in equilibrium.
1
fir
m

2/[δ(2 − δ)]
π(1) < 0, 0 firm q 1/δ
-
δ
1
Welfare comparison:

 δ(2 − δ)m − 2A ≡ W (2) if 2 firms advertise.
EW = δm − A ≡ W (1) if only one firm advertises.
0 if no firm advertises.


m 1
If > , ⇒ W (1) > 0, ⇒ social optimal is at least one firm advertises.
A δ
m 1
If > , ⇒ W (2) > W (1), ⇒ social optimal is both firms advertise.
A δ(1 − δ)
 105

m/A p/A = m/A

6
W (2) > W (1) 6
1/[δ(1 − δ)]

market failure
W (1) > 0

q - q -
W (1) < 0
δ δ
1 1
µxXTòU δ Ó‹
When δ → 1, one firm would be enough. However, if m/A > 1,
both firms will advertise.

11.3 Targeted Advertising ‡ú4µ

(1) Consumers are heterogeneous with different tastes. (2) Large scale advertising is
costly. (3) Intensive advertising will result in price competition.
⇒ ̶nßFí¾‘6, .°¼Sà.°4”5‡ú4µ, ‡ú.°5¾‘6íˇ

11.3.1 The model

2 firms, i = 1, 2, producing differentiated products.
2 groups of consumers: E experienced consumers and N inexperienced consumers.
θE of experienced consumers are brand 1 oriented, 0 < θ < 1.
(1 − θ)E of experienced consumers are brand 2 oriented.

2 advertising methods: P (persuasive) and I (informative).

Each firm can choose only one method.

AS1: Persuasive advertising attracts only inexperienced consumers. If only firm i

chooses P , then all N inexperienced consumers will purchase brand i. If both
firms 1 and 2 choose P , each will have N/2 inexperienced consumers.

AS2: Informative advertising attracts only the experienced consumers who are ori-
ented toward the advertised brand, i.e., if firm 1 (firm 2) chooses I, θE ((1−θ)E)
experienced consumers will purchase brand 1 (brand 2).

AS3: A firm earns $1 from each customer.

From the assumptions we derive the following duopoly advertising game:

firm 1 \ firm 2 P I
P (N/2, N/2) (N, (1 − θ)E)
I (θE, N ) (θE, (1 − θ)E)
 106

11.3.2 Proposition 11.5

N N
1. (P, P ) is a NE if only if N/2 ≥ θE and N/2 ≥ (1 − θ)E or 1 − ≤θ≤ .
2E 2E
(If strict inequality holds, the NE is unique.)
N N
2. (I, I) is a NE if only if N ≤ θE and N ≤ (1 − θ)E or ≤ θ ≤ 1 − . (If
E E
strict inequality holds, the NE is unique.)
N N
3. (P, I) is a NE if only if N/2 ≤ (1 − θ)E and N ≥ θE or θ ≤ min{1 − , }.
2E E
N N
4. (I, P ) is a NE if only if N/2 ≤ θE and N ≥ (1 − θ)E or θ ≥ max{1 − , }.
E 2E

θ
N/E
16@ 
H 
@ H  N/2E
@ H (I, P ) 
 
@ HH
@ H 
(I, I) (I,P
@ HH (P, P )
)&(P,I)
@  H
HH
 @ HH
 (P, I) HH1 − N/2E
 1 − N/E@ HH- N/E
1 2

11.4 Comparison Advertising

Comparison advertising: The advertised brand and its characteristics are compared
with those of the competing brand.
It became popular in the printed media and broadcast media in the early 1970s.
EEC Legal conditions: Material (xñ) and verifiable (ª„õ) details, no misleading
(³Ïû), no unfair (t£).

Advantages of comparison ads:

1. Provide consumers with low-cost means of evaluating available products.
2. Makes consumers more conscious of comparison before buying.
3. Forces the manufacturers to build into the products attributes consumers want.

Negative points:
1. Lack of objectivity.
2. Deception and consumer confusion due to information overload.

Muehling/Stoltman/Grossbart (1990 J of Advertising): 40% of ads are comparison.

Pechmann/Stewart (1990 J of Consumer Research): Majority of ads (60%) are indi-

rect comparison; 20% are direct comparison.
 107

11.4.1 Application of the targeted ad model to comparison ad

Plain ad. => Persuasive ad, aiming at inexperienced consumers.
Comparison ad. => Targeted ad, aiming at experienced consumers.

Applying Proposition 11.5 and the diagram, we the following results:

1. Both firms will use comparison ad only if E > 2N .
2. If 2E < N , both firms will use plain ad.
3. Comparison ad is used by the popular firm and plain ad is used by the less popular
firm in other cases in general.

11.5 Other Issues

11.5.1 Can information be transmitted via advertising?
Search goods: False advertising is unlikely.
Experience goods: Producers will develop persuasive methods to get consumers to
try their products.

Facts: 1. Due to assymmetry of information about quality, consumers can not

simply rely on ads.
2. High-quality experience products buyers are mostly experienced consumers.

Schmalensee (1978), “A Model of Advertising and Product Quality,” JPE.

Low-quality brands are more frequently purchased and firms producing low-quality
brands advertise more intensively. ⇒ There is a negative correlation between adver-
tising and the quality of advertised products.

Kihlstrom/Riordan (1984), “Advertising as a Signal,” JPE. High-quality firms have

an incentive to advertise in order to trap repeated buyers. ⇒ the correlation between
ad and quality is positive.

Milgrom/Roberts (1986), “Price and Advertising Signals of Product Quality,” JPE.

A signalling game model with ad as a signal sent by high-quality firms.

Bagwee (1994), “Advertising and Coordination,” Review of Economic Studies. and

Bagwell/Ramey (1994), “Coordination Economics, Advertising, and Search Behavior
in Retail Markets,” AER. Efficient firms with IRTS tend to spend large amount on
advertising to convince buyers that large sales will end up with lower prices. ⇒ ad is
a signal to reveal low cost.

11.5.2 Advertising and concentration

Is there a positive correlation between advertising and concentration ratio?
Perfect competition industry: Individual firms have no incentives to advertise their
products due to free rider effect. Collectively the industry demand can be increased
 108

by advertising. However, there is the problem of free rider.

Monopoly industry: Due to scale economy, monopoly firms may have more incentives
to advertise.

Kaldor (1950): In an industry, big firms advertise more.

Telser (1964) “Advertising and Competition,” JPE. Very little empirical support for
an inverse relationship between advertising and competition.

Orenstein (1976), “The Advertising - Concentration Controversy,” Southern Eco-

nomic Journal, showed very little evidence that there is increasing returns in adver-
tising.

Sutton (1974), “Advertising Concentration, Competition,” Economic Journal. The

relationship between scale and advertising is not monotonic. Both perfect competi-
tion and monopoly firms do not have to advertise but oligopoly firms have to.

ad
6

- concentration ratio

11.5.3 A simple model of dvertising and prices

if Q ≤ Q∗
 
cH Q a1 − Q if advertising
Cost: TC(Q) = Demand: P =
cL Q if Q > Q∗ , a0 − Q if not advertising.

No Ad equilibrium: Q0 = (a0 − cH )/2, P0 = (a0 + cH )/2.

Ad equilibrium: Q1 = (a1 − cL )/2, P1 = (a1 + cL )/2.
If cH − cL > a1 − a0 , then P1 < P0 , i.e., advertising reduces the monopoly price.
 109

12 Quality
12.1 Vertical Differentiation in Hotelling Model
Quality is a vertical differentiation character.

2-period game:
At t = 1, firms A and B choose the quality levels, 0 ≤ a < b ≤ 1, for their products.
At t = 2, they engage in price competition.

Consumers in a market are distributed uniformly along a line of unit length.

r
i
0 x 1

Each point x ∈ [0, 1] represents a consumer x.


ax − PA if x buys from A.
Ux =
bx − PB if x buys from B.

The marginal consumer x̂ is indifferent between buying from A and from B. The
location of x̂ is determined by
PB − P A
ax̂ − PA = bx̂ − PB ⇒ x̂ = . (7)
b−a
The location of x̂ divids the market into two parts: [0, x̂) is firm A’s market share
and (x̂, 1] is firm B’s market share.

r̂
 A’s share - B’s share -

0 x 1
Assume that the marginal costs are zero. The payoff functions are
   
PB − P A PB − P A
ΠA (PA , PB ; a, b) = PA x̂ = PA , ΠB (PA , PB ; a, b) = PB (1−x̂) = PB 1 − .
b−a b−a

The FOCs are (the SOCs are satisfied)

∂ΠA PB − 2PA ∂ΠB 2PB − PA
= = 0, =1− = 0. (8)
∂PA b−a ∂PB b−a
The equilibrium is given by
b−a 2(b − a) 1
PA = , PB = , ⇒ x̂ = .
3 3 3
The reduced profit functions at t = 1 are
(b − a) 4(b − a)
ΠA = , ΠB = .
9 9
 110

Both profit functions increase with b − a. Moving away from each other will increase
both firm’s profits. The two firms will end up with maximum product differentiation
a = 0 and b = 1 in equilibrium.

Modifications: 1. High quality products are associated with high unit production
cost.
2. Consumer distribution is not uniform.

12.2 Quality-Signalling Games ̾êW

There are one unit of identical consumers each with utility function

 H − P à‹ƒò¹”ß¹

U= L − P à‹ƒQ¹”ß¹

0 .



CH > CL ≥ 0: Unit production costs of producing high- and low-quality product.

AS1 The monopolist is a high-quality producer.
AS2 H > L > CH .
L − CL
Signalling equilibrium (̾êW): P m = H and Qm = .
H − CL
Proof: 1. For a low-quality monopolist, to imitate the high-quality monopolist is
not worthwhile:
L − CL
ΠL (P m , Qm ) = (P m − CL )Qm = (H − CL ) = L − CL = ΠL (L, 1).
H − CL
2. The high-quality monopolist has no incentives to imitate a low-quality monopolist:
L − CL
ΠH (P m , Qm ) = (P m − CH )Qm = (H − CH ) > ΠH (L, 1) = L − CH .
H − CL
The last inequality is obtained by cross multiplying.

The high-quality monopolist has to reduce its quantity to convince consumers that
the quality is high.
If the information is perfect, he does not have to reduce quantity.
The quantity reduction is needed for signalling purpose.

12.3 Warranties ¹”\„z

Spence (1977), “Consumer Misperceptions, Product Failure, and Producer Liability,”
Review of Economic Studies.
Higher-quality firms offer a larger warranty than do low-quality firms.

Grossman (1980), “The Role of Warranties and Private Disclosure about Product
Quality,” Journal of Law and Economics.
A comprehensive analysis of a monopoly that can offer a warranty for its product.
 111

12.3.1 Symmetric information model

ρ: The probability that the product is operative.
V : The value to the consumer if the product is operative.
Symmetric information: ρ is known to both the seller and the buyer.
P : Price. C: unit production cost. Assumption: ρV > C.

 V − P ¹”\„z
U= ρV − P ̹”\„z
0 ..


Monopoy equilibrium without warranty: P nw = ρV and Πnw = ρV − C.

C
Expected cost of a unit with warranty: C w = C + (1 − ρ)C + (1 − ρ)2 C + · · · = .
ρ
(On average, 1/ρ units will end up with one operative unit.)
C
Monopoy equilibrium with warranty: P w = V and Πw = V − .
ρ
It seems that the monopoly makes a higher profit by selling the product with a
warranty. However, if the consumer purchases 1/ρ units to obtain an operative unit,
the result is the same. (Oz Shy’s statement is not accurate unless the consumer is
risk averse.)

12.3.2 Asymmetric information with warranty as a quality signal

ρL : The operative probability of a low-quality product.
ρH > ρL : The operative probability of a high-quality product.
Asymmetric information: The quality of a product is known only to the seller.

Bertrand equilibrium without warranty: P nw = C and Πnw

i = 0, i = H, L.
C
Bertrand equilibrium with warranty: P w = C/ρL , QH = 1, QL = 0, Πw w
H = P − >
ρH
C
0, Πw w
L = P − = 0, U = V − P w > 0.
ρL
The high-quality firm has a lower unit production cost of the warranty product.
In the market for warranty product, the low-quality firm cannot survive.
 112

13 Pricing Tactics
13.1 Two-Part Tariff
Oi (1971), “A Disneyland Dilemma: Two-Part Tariffs for a Mickey Mouse,” QJE.
In addition to the per unit price, a monopoly firm (amusement parks Y—Ò, sports
clubs U™E—¶) can set a second pricing instrument (membership dues ‘) in order
to be able to extract more consumer surplus.

P : price, φ: membership dues, m: consumption of other goods.

√
Budget contraint: m + φ + P Q = I. Utility function: U = m + 2 Q.
p 1 1
max U = I − φ − P Q + 2 Q ⇒ demand function: P = d , Qd = 2 .
Q Q P

13.1.1 No club annual membership dues

Club capacity: K. √
Club profit: Π(Q) = P Q = Q
1 √
maxQ Π(Q) ⇒ Qm = K, Pm = √ , Πm = K.
K

13.1.2 Annual membership dues

√
max Πa (φ) = φ subject to I − φ + 2 K > I = I0 ,
φ

√
⇒ φ∗ = 2 K = Π a > Π m .

Using annual membership dues, the monopoly extracts all the consumer surplus,
like the 1st degree price discrimination.
m m
6 6

I raa Ir
66 aa 6
Πm
? aar
aa φ3
ra
φ ∗ aa
aa U1 ? U1
ara
a am − φ3 + P4 (Q − Q3 ) = I
r
m + Pm Q = I
? aa
a aa U4
U-0 aa U-0
a
Q Q
K Q3 Q4 = K
 113

13.1.3 Two-part tariff

There are two problems with membership dues:
1. It is difficult to estimate consumers’ utility function and the profit maximizing φ∗ .
If the monopoly sets φ too high, the demand would be 0.
2. Consumers are heterogeneous.

Therefore, the monopoly offers a “package” of Q3 < K and annual fee φ3 < φ∗ .
In addition, the monopoly offers an option to purchase additional quantity for a price
P4 .

13.2 Peak-Load Pricing ª¼, ׼ρ¦g

High- and Low-Seasonal Demand Structure: P H = AH − QH , P L = AL − QL ,
AH > AL > 0.

Cost Structure: TC(QH , QL , K) = c(QH + QL ) + rK for 0 ≤ QL , QH ≤ K.

c: unit variable cost, K: capacity, r: unit capacity cost.

max Π = P H QH + P L QL − c(QH + QL ) − rK subject to 0 ≤ QL , QH ≤ K.

QH ,QL ,K

FOC:
MRH (QH ) = c + r, MRL (QL ) = c, QL < QH = K.
AH + c + r AL + c
⇒ PH = > PL = .
2 2
Regulation for efficiency: P H = c + r > P L = c.
r
n-period case: MRH (QH ) = c + , MRL (QL ) = c.
n
Modification: Substitutability between high- and low-seasonal demand.
 114

14 Marketing Tactics: Bundling, Upgrading, and Dealer-

ships
14.1 Bundling (¾ù) and Tying (»»)
Bundling: Firms offer for sale packages containing more than one unit of the product.
It is a form of nonlinear pricing (2nd degree price discrimination).

Tying: Firms offer for sale packages containing at least two different (usually com-
plementary) products.
Examples: Car and car radio, PC and software, Book and T-shirt.

14.1.1 How can bundling be profitable?

Monopoly demand: Q(P ) = 4 − P , MC = 0.

Monopoly profit maximization: P m = 2 = Qm , Πm = 4.

Bundling 4-unit package for $8:

(1) The consumer will have no choice but buying the package.
(2) The monopoly profit becomes Π = 8 > Πm .

The monopoly in this case uses bundling tactics to extract all the consumer surplus.

14.1.2 How can tying be profitable?

A monopoly sells goods X and Y.
2 consumers, i = 1, 2 who have different valuations of X and Y.
Valuations: Vx1 = H, Vy1 = L; Vx2 = L, Vy2 = H, H > L > 0.
Assume that consumers do not trade with each other.

Equilibrium without tying:

 
nt nt H if H > 2L nt 2H if H > 2L
Px = P y = and Π =
L if H < 2L 4L if H < 2L.

Equilibrium with tying, PT = Px&y , QT = Qx&y :

PTt = H + L, and Πt = 2(H + L) > Πnt .

14.1.3 Mixed tying

Adams/Yellen (1976), “Commodity Bundling and the Burden of Monopoly,” QJE.

A monopoly sells goods X and Y.

3 consumers, i = 1, 2, 3 who have different valuations of X and Y.
Valuations: Vx1 = 4, Vy1 = 0; Vx2 = 3, Vy2 = 3, Vx3 = 0, Vy3 = 4.
 115

Assume that consumers do not trade with each other.

Equilibrium without tying:

(1) Px = Py = 3, Qx = Qy = 2, Π(1) = 12.
(2) Px = Py = 4, Qx = Qy = 1, Π(2) = 8 < Π(1).
Therefore, Pxnt = Pynt = 3, Qny nt
x = Qy = 2, Π
nt
= 12.

Equilibrium with pure tying, PT = Px&y , QT = Qx&y :

(1) PT = 4, QT = 3, Π(1) = 12.
(2) PT = 6, QT = 1, Π(2) = 6 < Π(1).
Therefore, PTt = 4, QtT = 3, ΠtT = 12.

Equilibrium with mixed tying:

Pxmt = Pymt = 4, PTmt = 6, Qmt mt mt
x = Qy = 1, QT = 1, Π
mt
= 14 > Πt = 12.

But mixed tying is not always as profitable as pure tying.

14.1.4 Tying and foreclosure (‡½)

US antitrust laws prohibit bundling or tying behavior whenever it leads to a reduced
competition. What is the connection between tying and reduced competition?

2 computer firms, X and Y, and a monitor firm Z (compatible with X and Y).
2 consumers i = 1, 2 with utility functions
 
 3 − Px − Pz buys X and Z  1 − Px − Pz buys X and Z
U1 = 1 − Py − Pz buys Y and Z U2 = 3 − Py − Pz buys Y and Z
0 buys nothing, 0 otherwise,
 

Bertrand equilibrium with 3 independent firms:

(1) Px = Py = 2, Pz = 1, Qx = Qy = 1, Qz = 2, Πx = Πy = Πz = 2.
(2) Other equilibria: (Px , Py , Pz ) = (1, 1, 2) = (0, 0, 3) = (3, 3, 0).

Assume that firm X buys firm Z and sells X and Z tied in a single package.

Total foreclosure equilibrium:

tf
Pxz = 3, Qtf tf tf tf
xz = 1, Qy = 0, Πxz = 3 < Πx + Πz = 4, Πy = 0.
Py does not matter. Consumer 2 is not served. The industry aggregate profit is lower
under total foreclosure.

-foreclosure equilibrium:

Pxz = 3 − , Qxz = 2, Py = , Qy = 1, Πxz = 2(3 − ), Πy = .
Consumer 2 buys one X&Z and one Y and discards X.
 116

14.1.5 Tying and International markets segmentation

Government trade restrictions like tarriffs, quotas, etc., help firms to engage in price
discrimination across international boundaries.

A two countries, k = 1, 2, with one consumer in each country.

A world-monopoly producer sells X.
It can sell directly to the consumer in each country or open a dealership in each
country selling the product tied with service to the consumer.
The utility of the consumer in each county (also denoted by k = 1, 2) is
 
 B1 + σ − P1s if 1 buys X & service  B2 + σ − P2s if 2 buys X & service
1 ns 2
U = B1 − P 1 if 1 buys X only U = B2 − P2ns if 2 buys X only
0 if 1 does not buy, 0 if 2 does not buy,
 

where Pks (Pkns ) are the price with service (without service) in country k, k = 1, 2,
and σ > 0 is the additional value due to service.

AS1 B1 > B2 .

AS2 Marginal production cost is 0.

AS3 Unit cost of service provided by the dealership is w ≥ 0.

No attempts to segment the market:

 
ns B2 if B1 < 2B2 ns 2B2 if B1 < 2B2
P = Π =
B1 if B1 > 2B2 B1 if B1 > 2B2

Segmenting the market:

Pks = Bk + σ, Πs = B1 + B2 + 2(σ − w).

Nonarbitrage condition: B1 − B2 < σ.