IO @ UTD: Third session

Meisam Hejazinia

1/29/2013

Couse conjecture: price of monopolist will reach
to marginal cost, as price adjustment become more
frequent (in small period), when the product would
be durable good.
pt+1 would be expectation of the next period price.
In this period consumers would purchase that
pt −δp
have the relation of v > vt+1t+1

Monopolist make sense to lease the good. The pt−1 −δpt

State of economy would be: vt = 1−δ
monopolist has option to either lease, and then get
everything back, and then lease again. the producer knows that v ∈ [0, vt ]

Gul paper: According to linear rule, pt (vt ) = µ.vt

We have an infinite number of periods. It is Quantity sold in period t: qt = vt − vt+1

monopoly of durable goods. Consumers have unit
demand for one item. last consumer who would buy would be:
vt+1 = λpt (vt ) if my valuation is more than
Resell market in the previous examle exist, but λ.p I am going to purchase.
here that does not exist, and anyone purchased
would be out of market. qt = vt − λ.pt (vt ) = vt − λ.µ.vt

Consumer has valuation v, and consumers with qt = (1 − λ.µ).vt

v > λ.p; λ > 1 will purchase.
Producer have to decide what price to charge.
Producers with V̄ have already purchase, and all
consumers lower than this will pruchase.
producer rule therefore would be p = µ.v̄, µ < 1
We must have had the assumption that λ.µ < 1
π = .... + δ t .pt .qt + δ t+1 .pt+1 + δ t+2 .pt+1 .qt+1 =
The ... + δ t .pt [vt − λ.pt ] + δ t+1 pt+1 [vt+1 − λ.pt+1 ] +
δ t+2 .pt+2 [vt+2 − λ.pt+2 ] + ...

so > 1, and µ < 1 vt was assumed to be uniformly distributed

between zero and one. We want to miximize profit
We want to find µ∗ , λ∗ with respect to p.

In some period we must have v − pt ≥ δ(v − pt+1 ) We want to find µ, and we want to maximize over
all p’s.
δ would be the discount factor.
FOC with respect to pt : = δ t [vt − λ.pt − λpt ] +

1
δ t+1 .λ.pt+1 = δ t vt − 2λpt + δλpt+1 zero profit.

FOC: ⇔= δ t+1 λ.pt − 2λpt+1 + δλpt+2 δ → 1 ⇒ u∗ = 0

FOC: ⇔= δ t+2 λ.pt+1 − 2λpt+2 + δλpt+3 v1 = 1 means p1 = µ∗ .v1 = 0.1 = 0.M C

We start to see patterns here.
λ.pt − 2λ.pt+1 + δ.λ.pt+2
λ.µ.vt − 2.λ.µ.vt+1 + δ.λ.vt+2
λ.µ.vt − 2.µ.λ.µ.vt + δ.λ.µ.λ.µ.λ.µ.vt
λ.µ.vt [1 − 2λ.µ + δ.λ2 .µ2 ] = 0
Let’s assume that λ.µ < 1
Intuitively story makes sense, and we showed
theoritically.
We tried to keep things simple here.
Rubenstein has infinite horizon game, and players
play until infinity, and he introduces discount factor.
Stahl has different approach, and assumes barganin-
ing lasts one pariod, but offers stand form limited
time. The number of offers with this approach goes
to infinity.

1
µ= √
λ[1+ 1−δ]
Shapiro 1983 paper:

Equilibrium at period t: pt+1 = µ.vt+1 = Experience goods, consumers ex ante is not much,
µ.λ.pt (vt ) = µ.λ.µ.vt and only by using it they will learn about it. Search
good you don’t know what the quality is and by
pt+1 = µ2 .λ.vt inspection without using it you wil learn about it.
Experience goods you have to purchase and use
Marginal consumer will be: vt+1 − pt = them to find out about them.
δ(vt+1 − pt+1 ) = δ(vt+1 − µ.vt+1 )
Experience good introduced by Nelson, and
vt+1 − pt = δ(1 − µ).vt+1 consumer have imperfect info. until they purchase
and use it. The sales is the dynamic one. If the
pt = [1 − δ(1 − µ)]vt+1 buyer learns, the demand function shifts. The seller
is bundling information with actual product. With
pt
vt+1 = 1−δ(1−µ) the product the information about product and
attribute is selled. With purchasing favorability of
pt
λ.pt = 1−δ(1−µ) information will be understood.

1
δ= 1−δ(1−µ) Consumers are pessimistic q > R R would be
expectation. Redbull coke for example. Prior is
µ∗ = pessimistic. Second case is when consumers are
λ∗ = √1−δ1
optimistic. The true quality would be smaller than
expected quality q < R. For the first case result is
Linear non stationary equlibrium so would be in that there would be introductory offer to inform con-
this form. sumers. It would be two stage introductory scheme.
Monopoly charges high prices at the beggining, and
Quose conjecture says that the price will go down, then they realize that the good is good one. Once
and monopoly then will charge marginal cost, and the introductory expires, the monopolist will charge

2
 p
higher price. Pricing scheme is milking reputation. if f θ ≥ 1

p
When consumers are optimistic, everybody think uninformed: purchase if f θ ≥ R
that it is better than it is, so you charge a lot,
and once they bought they will more or less out The cumulative distribution is left hand usually
of market. Then you skimm. You slowly skim the and here it is not so, and we have right hand.
market, and you start with high price and slowly the F (θ) = 1 f (t)dt. This is for simplicity.
R
θ
price will go on, and you jump back up.
F (bar
Here assumption is that everybody is optimistic theta) is mass of consumer with valuation of θ̄ or
or pessimistic, and it means everybody are homoge- higher θ ≥ θ̄
neous.
Demand function for each price will gie us quan-
All expectations are homogneous, and consumers tity demanded. Each consumer will need one unit.
have point expectation, mean there would be no S(p) = F ( p ) gives us consumers whose valuation is
R
uncertainty. higher than p. Mean consumers who would purchase.

The product is non durable. Jump back would be This is for no information case, and for full
to monopoly price. information we will have Z(p) = F (p/q)

Consumers are myopic. If price would be 2 in one period, and consumers

will find out, and if price change the demand will
Monopolist who sells the product of quality q not change, and everybody has bought, and know
and produces quantity x. There is a cost function the true quality. If the price increases the structure
c(x, q). There are standard assumption; it is either u of the economy will not change.
shape, average cost curve, or u shape constant aver-
age cost curve, means decreasing marginal cost curve. Lemma 1. if pT > pT −1 then, pt = PT t >> T

The choice variable is the p, and it is discrete time
model, so it would be discrete time r > 0 would be
the interest rate.
It is discrete time model, infinite horizon. Con-
sumers have unit demand. In each t, and they are
indexed by their taste of quality. θ ∈ [0, 1].
There is density function f (θ).
Consumer surplus is θ.q − p.
The economy will not change, since there would
be no additional consumer that learned. If it was
optimal in PT , then it will still be optimal for all
periods.
The implication is that we only need to consider
the declining path of price. If it was optimal for
the consumer to charge the lower price in the next
period, and I did not change the market today, and
helped consumer learning, for the next one it will
not change anymore.

R > 0, if R = q then it would be no learning, and Learning is trhough experiment, and is complete.
we are back into static monopoly.
The pessimistic consumers are not necessarily low
In updating, it is instantaneously and perfect. type. They are distributed between zero and one.
Consumers with information purchase if and only if Pessimistic consumers expect quality R, and the real

3
quality would be q, they expect R < q. price.

If monopolist informs a lot of consumers, then PH = Z −1 (x) = Z −1 (S(PL ))

they will have higher willingness to pay, and they
can charge higher price in the later periods.
We have to consider only declining price path. The
Lemma applies for everything that happens before,
and it applies to both optimistic, and pessimistic.
Z is the informed demand function. X consumer
will know what is the true quality, and the second
price is that all this consumers who are informed are
willing to the purchase. All this consumers will have
second demand function.

R < q, it is not informing all of consumers, There is a trade off for the producer. Producer
but part of consumers. It is two stage pricing pol- sees profit in two stages, or periods. By producing
icy. Introductory price is low, and high in the second. the amount between the uninformed, and informed,
you would sacrifice for later greater profit.
Price path is kind of clear. When there are fully
informed demand. Suppose that all consumers are The profit for this two stage policy would be
−1
fully informed. Then π(x, q) = z .x − c(x, q). It is v(x) = π(x, R) + π(x, q)/r
for fully informed case.
It is infinite horizon game, so the future profit is
−1
For incomplete informed π(x, R) = S (x)x − discounted.
c(x, q) dV (x)
The first derivitive would be dx = 0 ⇔
X (R) < X < X ∗ (q)
∗
φ(x, x̂) = π(x, q) when x ≤ x̂, and π(x, R) when
x > x̂
If all consumers are pessimistic, you want to make
p sure that they become informed quickly.
S(p) = F(R )
The curve is because if you increase the price the
S −1 (x) = R.F −1 (x)
people who are uninformed will not purchase.
Z −1 (x) = q.F −1 (x) The strategy is dynamic pricing without letting
information out too quickly when consumers are
Marginal revenue of the informed lies above the optimistic.
uninformed one.
R>q
If people are uninformed the quantity would be
much less than the informed. There are two types: The first one is declining
price path: P1 ≥ P2 ≥ P3 ≥ ...
For uninformed the profit would be lower. Both in
the form of reverse U. The second one is P1 > P2 > ... > PT −1 < PT =
PT +1
The right leg of the informed, and left leg of the
uninformed would be monopolist profit curve. We would have two uninformed would be higher
profit revers U, and informed would be lower. You
PL : X = S(PL ) Monopolist chooses quantity and skimm uninformed. The price and quantity will go
not the price, and this quantitiy will determine the to perfect information case. The price is lower when

4
perfect information price.
At some point you jump up. Optimal strategy is
to jump back up for perfect information cost, and it
would be discrete jump.
valuation. It is unlikely, and not that likely. Usually
have incomplete information. The monopolist should
have perfect information of valuations. Arbitrage
would be easy. If you can not prevent first degree
will not work.

You skimm uninformed, you slowly lower the price 2. Second degree would be the signal that we
until it is not profitable anymore, and you have can not observe it. How we put incentive so that
enough consumer. The heterogeneity is on marginal each group choose the packages that belongs to their
value of quality, and how much they value quality. group. There should be compatible designs for this.

Now will go over price discremination, and con- 3. Third degree: Direct and observable signals
tinue next session. could be used for discrimination. Getting the hair
cut. There is some direct signals that are observable.
Problem set with questions will include one price Signal is directly observable, so it is controlable.
discrimination would be in two weeks. You don’t give your student ID, you would not be
chared. This was the third type.
When we say coke today is different from coke
tommorow, mean same price to different consumers, First Degree
or different price to same consumer.
Determine willingness to pay. If the monopolist
Different prices to different consumers due to has all information, he can charge prices for different
different costs. units. There is different case that we have identical
downward slopping demand curves. Different will-
The possibility of arbitrage hinders the price
ingness to pay for different units. q = D(p)
n , where n
discrimination. If the good is transferable, then price
is number of consumers.
discrimination does not work. Other consumers
will not purchase from monopolist, since he could
purchase from another. Aggregate demand is Q = D(p). Monopolist
comes with pricing schedule, but some transfer T (q)
Services is hard to transfer, e.g. counselling, is the total amount that consumer pays.
movie, financial assistance, medical treatment. If
consumers are offered two different packages, or It could be linear T (q) = p.q
contents, then we will have transferability, and price
discrimination will not work. Certain price quanity, It could be affine linear: linear with intercept:
and quality menues to different consumers. If all fixed fee for each consumer T (q) = A + pq. Now
consumers could purchase the preferable ones, then what we want to show is that with appropriate
discrimination work. Consumers will self select. i pricing scheme, monopolist can increase profit with
will target to two different, so that there would be discrimination.
no transferability.
If monopoly can charge two prices, then consumer
Incentive compatibility constraint would be an would be better off.
important issue here. There are trhee degree of price
descrimination: We are thinking about stealing consumer sur-
plus. Monopolist wants to steal it, and needs to
1. perfect price discrimination: monopolist knows find how big it is and charge it in the form of fixed fee.

5
 The best way is to charge the marginal cost. If he
charges price that equals to marginal cost then he
charges fixed fee. Fixed fee of consumer surplus, and
variable cost that maximizes the consumer surplus.
This would be Aggregate demand, and for each it
should be A/n. Due to dead weightloss the in one
part tariff can not take suprlus, but in two part
monopolist can do this.

Paper of shmalinze, and the related papers in

sylabus.

6
 IO @ UTD: Fourth session
Meisam Hejazinia

02/05/2013

Linear pricing scheme: T (q) = p.q price would be

monopoly price. Therefore, S SM < SS F D

Two part tarriff: fix fee plus the variable fee: excercise 3.1 is streight forward. Competitive
T (q) = A + p.q fringe: means there are samll number of firms
that are price taking, and are competing with the
The firm has too choose price, and once price monopolist. pc < p0 < pm . If the conopolist charges
is selected. With appropriate pricing scheme the lower than their price which is p0 then they would
consumer will extract the consumer surplus. go out of market. If monopolist charges more than
this amount, monopolist will not get any maret share.
How to maximize social suprlus:
p̃(q) monopolist decides to charge price less than
1. Price equal to competitive price pc . If we maxi- this price then the monopolist price would be lower,
mize the social suprlus, there would be no consumer and if monopolist decides to produce more, then the
surplus. T (q) = pc .q + A. The consumer surplus price would be lower. For any quantitiy, so less than
R qc 0 0
would be S c = 0 [p(q) − pc ]dq. License fee in the q the price of the monopolist will be p̃(q) = p . if
0
form of fixed premium. A would be individual fee and q ≥ D(p ) the price of monopolist would be p(q).
c
it should be lower than each individual share A ≤ sn .
If MC is equal to zero we will have p = 0 = pc ,
c c
R qc
Total two part tarriff would be T (q) = s /n + p .q and A = S c (pc ). pc = 0 then A = 0 [p̃(q)]dq/n.
if q >c 0, and zero if q = 0. Each consumer consumes
q = qn . Each consumer consumes the share. It means that from the triangle created from the
two axis and the demand function, now the top
The monopolist profit is higher than uniform triangle created with (p0 , q 0 ) can be removed, since
monopoly pricing: consumer will go to the cometitor if the monopolist
charges the price higher than them.
c
π F D = sn .n + q c .pc − C(q c ) profit with first degree
price descrimination. Here we assumed that consumers are identical.
What if the consumers are not identical. In this case
= sc + [q c .pc − c(q c )] consumer must charge Ai for each of the consumers.

π m = pm .D(pm ) − C(D(pm )) You have to find a way to remove the arbitrage

possibility. Most of the time it would be intermarket
Does π F D > π m . social surplus would be arbitrage, or personal arbitrage. Intra personal
ss = π m + sm + (DW L)
m
personal arbitrage is that I claim to be low type and

1
 pi −c0 (q)
then I pay the low price, but I am really high type. pi = − DDi (p
i (pi )
i ).pi
= 1
i

Di0 (pi ).pi )

Third degree price discrimination. Suppose we are i = − Di (pi )
single product, and the cost of production is c(q).
We divide the market by m groups. This is based Group with smaller elasticity have higher markup,
on some exogeneous information: gender age, status. and pay more. Rich people pay more. They don’t
This is exogeneous information or signal. care about price. Addiction, and less sensitivity is
This correlated with price elasticity or dmeand intuition. Means these consumers have less alterna-
elasticity. tive and they can not avoid the prices. Consumers
in third degree price elasticity excercise with less
Your signal is correlated with their characteristics. elasticity play more.
m groups, would be m distinct demand.
Higher elasticity means you have more alternative,
We assume there is no arbitrage between group. and they pay lower price.
Mean they can not sell to other group.
With linear demand the total quantity does not
Also there is no discrimination within the group. change. ∆qi = qi − q m then some group consume
You know their characteristics. You don’t have any more and some groups consume less. As a result for
detail about consumer in each group. If you know high elasticity group it will have one direction and
you can split it into other groups. for low elasticity it would go another
Pdirection. With
linear demand curve we will have δqi = 0 means
Linear tarriff for each group, ti for each group, there would be no welfare change.
where Ti (q) = pi .q. Old and young people for
example will pay different prices pi . We did not made an ssumption about what actual
demand will look like. An increase in total quantity
We have to find price vector p̄ = (p1 , , pm ), would be necessary, but not sufficient. Even if we
q̄ = (D1 (p1 ), D2 (p2 ), , Dm (pm )). have increase it is no gaurantee that we will see an
Pm improvement.
Total quantity would be q = i=q Di (p i ). Mo-
nopolist will sell one product. Therefore there is What there is intermediate good and final goods
only one cost function c(q). If there is different type the structure would change, and here all discussion
of product for different group, we also would have were about final prices.
product differentiation problem.
From first degree , you had i consumers, and in
WeP have to set P up monopolist profit third group you had groups. In the first you recieve
n m
max[ i=1 pi .Di (pi ) − c( i=1 Di (pi )] signal about each consumer, but in the third you do
not know anything about individual within group,
This is like multiple product with independent you recieve signals about each of the groups only.
demand, but the cost is not independent. We as- Second degree you know there are group, and the
sume that regularity condition holds, and everything structure of demand for each group, but you don’t
behave nicely, and second order condition holds. know who is who. Also first degree can move to
second if you know the demand distribution, yet
First order condition would be Di (pi ) + pi .Di0 (pi ) − don’t know who is who you will move to the second
c0 (q).Di0 (pi ) = 0 degree.
second degree there is private information. First

2
degree are direct price discrimination, and second We try to say something about the shape of the
R D (p)
and third are indirect. function: Si (p) = 0 i [pi (q) − p]dq

We will compare third degree with uniform pric- The distance between cost curve and the demand
ing. First degree in terms of welfare improves, unless curve would be consumer surplus. The cost for con-
all are identical. sumer is price.
R b(x)
We apply the Libnitz rule: V = a(x) f (x, t)dt
What we know is that the smallest inverse dmeand b(x)
elasticity mini 1i =≤ p̄−c 1 then ∂V = inta(x) ∂f∂x (x,t)
dt − ∂a(x)
∂x f (a(x), t) +
p̄ ≤ max i . This means all ∂x
∂b(x)
the consumers will buy. For third degree linear de- ∂x f (b(x), t)
mand will not change total quantity, assuming that
∂si (p) R D (p)
everybody is willing to buy. If one group is not buy-
∂p = ∂D∂p
i (p)
[pi (Di (p))−p]− 0 i dp = −Di (p)
ing then total quantity changes, since you increase
the quantity by introducing one group to the market. S 0 (p) = −D (p) < 0
i i

This does not mean that everybody buys, but we Si0 (p) = −Di0 (p) > 0
assume that everybody buys under uniform pricing.
Figures for this are in the book.
John robinson (193?) denoted that strong markets
are the ones that pay more under the third degree Si (p̄)−Si (pi )
We know how the slopes are. p̄−pi ≥ Si0 (p̄)
price discrimination, than under uniform pricing,
and there is at least one group that pays less, called
Si (pi ) − Si (p̄) ≥ Si0 (p̄)[pi − p̄]
weak markets, than under uniform pricing.
Si (pi ) − Si (p̄) ≥ −Di (p̄)[pi − p̄]
We give one group more social surplus, and the
other group pays less. Some changes in monopolist
pricing, and some change in consumer suprplus. It Si (pi ) − Si (p̄) ≥ −q̄i (pi − p̄)
depends on the value and distribution.
Si (pi ) − Si (p̄) − q̄i .p̄ ≥ −q̄i .pi
P
Constant marginal cost: M C = C( qi ) =
Si (pi ) − Si (p̄) + qi .pi − q̄i .p̄ ≥ (qi − q̄i )pi
P
c.( qi )

To get whole consumer suprlus Si (pi ) − Si (p̄)+

Pm Pmwe add up to have
CS = i=1 Si (pi ), and π = i=1 (pi − c)Di (pi ) =
Si (pi ) − Si (p̄) + qi (pi − c) − q̄i (p̄ − c) ≥ ∆qi (pi − c)
P
(pi − c)qi . Constant marginal cost we had here.
If this inequality holds for all i, it will also hold
P
Under unifor price we have c̄s = Si (p̄) for sum of them.
P P P P
P Si (pi ) − si (p̄) + qi (pi − c) − q̄i (p̄ − c) ≥
P
Profit would be π̄ = (p̄ − c)q̄i
∆qi (pi − c)
∆qi = qi − q̄i P
∆w ≥ ∆qi (pi − c) Lower bound.
The P change in welfare
P would
P be W̄ = P
Si (pi ) + (pi − c)qi − [ Si (p̄) + (p̄ − c)q̄i ] ∆w ≤ (p̄ − c). ∆qi Upper bound.

3
 Two part tarriffs
Instead of offering the fixed quanity or the price,
the manufacturer will define price and quanity
slabs. There is personal arbitrage, since you can
not identify who the person is. Different menue valuation.
to different people will be offered. As a result
there is self selection problem. The manufacturer is
interested to design the incentive. We have to design
incentive compatible constraints.
Potential to continium of contracts quanity and
quality will be in the form of {T, q}, T (q) = A + p.q.
A > 0.
A is not consumer suplus since we don’t know who
you are, it is fixed fee, e.g. just a licence fee.
The total tarriff is decreasing. For consumers
utility would be in the form of :
U (Di (p)) = Si (p) = θi .v(Di (p)) − pDi (p)
2
−p)
Si (p) = (θi2θ i
v would be valuation, and θ would be marginal
s1 (p) < s2 (p)
Always know that there is some type two that will
have higher surplus if sees this price.
For monopolist to choose optimal price we have to
determine aggregate demand.
p
There is D(p) = λD1 (p) + (1 − λ)D2 (p) = 1 − θ
If perfect discrimination (First order Discrimi-
nation) would be possible. pF D = c, Ai = Si (c) =
Si (P F D )

u = θv(q) − T if the consumer accepts it. Problem of personal arbitrage or full arbitrage
is that you can not target one price to one specific
and zero if the consumer does not buy. group. What is the optimal pricing to the monop-
olist. mxp (p − c)D(p) then the price of monopolist
We assume that v(0) = 0, and v 0 (.) > 0, v 00 (.) < 0. should be pm = c+θ 2
θ represents consumer valuation of quality. θ1 is 2
probability λ, and θ2 with probability 1 − λ. θ2 > θ1 . πRm
= (θ−c)
4θ
1 λ 1−λ
θ = θ1 + θ2
S1 (pm ) ≥ 0 and θ1 ≥ pm = c+θ
2
2

1−(1−q)2
v(q) = 2
m
πR ≥ π2m
v(q) = 1 − q
Two part tarrif:
We want to derive the mand function for the
consumer: I can not identify consumers with fixed fee, and I
can only offer one fixed fee. There is some price fee,
Consumer will max[θi .v(q) − [A + pq]] so that and monopolist should find it. What is the highest
u(q) ≥ 0 level of fixed fee?

θi v 0 (q) − p = 0 A ≤ S1 (p)

θi (1 − q) = p = p(q) inverse demand function. A = S1 (p)

p
Di (p) = qi − 1 − θi The monopolist profit would be maxp [(λS1 (p) +
(1 − λ)S1 (p) = S1 (p)) + (p − c)D(p)]

4
 θ1 , λ , θ2 , (1 − λ)
To maximize you must get first order condition:
π m = λ[T1 − c.q1 ] + (1 − λ)[T2 − c.q2 ]
tp
p =
The first comparison is profit comparison: You can select from the menues.

πF D ≥ πT P ≥ πm We have two rationality constraints. We design

the contract so that we make sure that group 1 will
We allo the monopolist to choose the price and not go for group 2 contract.
fixed fee in two part tarrifs, so we are back to
uniform pricing. θ1 v(q1 ) − T1 ≥ θ1 v(q2 ) − T2 (IC1)

pF D = c < pT P < pm If this holds true the contracts are incentive

age 146 you can read the intuition behanid this. compatible. The same should hold for group 2.
Compatibility constraint should also hold for group
FOC m: D(p) + (p − c)D0 (p) = 0 2as following:

FOC p: D(p) + (p − c)D0 (p) = D1 (p) > 0 θ2 v(q2 ) − T2 ≥ θ − 2v(q) − T1 (IC2)

We know that S10 (p) = −D1 (p) We assume that individual rationality also holds:

S 0 1(p) = −D1 (p) θ1 v(q1 ) − T1 ≥ 0 (IR1)

Price in two part tarriff is strictly less than θ2 v(12 ) − T2 ≥ 0 (IR2)

monopoly pricing.
if IC2, IR1 → IR2 not binding
D(p) − D1 (p) + (p − c)D0 (p) = 0
if IC1, IR1 → IC2 not binding
λD1 (p) + (1−)D2 (p) − D1 (P ) = (1 − λ)[D2 (p) −
D1 (p)] θ2 v(q2 ) − T2 ≥ θ2 v(q1 ) − T1 > θ1 v(q1 ) − T1 ≥ 0

We have something positive plus (p − c)D0 (p) = 0, θ2 v(q2 ) − T2 > 0 IR2 not binding
and this means (p − c) should be greater than zero
since D0 (p) < 0. IR1

Two part tarriff is not optimal for monopolist, and θ1 v(q1 ) = T1

non-linear is much better.
θ1 v(q1 ) − T1 = 0 ≥ θ1 v(q2 ) − T2
For non-linear prices. I charge you much much
more for larger quantity than smaller quantity. T2 > θ1 v(q2 )

We have two values, we offer one menue of contract Conclusions:

to first type, and one menue to the second.
1. The low demand consumers, low types, will go
{T1 , q1 }, {T2 , q2 } home with zero surplus.

5
 high types will go with positive surplus (informa-
tion effect).

They know they are good, and this helps them to

go home with good contract.

2. Relevant IC prevents high demand from buying

low demand.

You have to offer the high demand guys sufficient

contract, so that they do not go for the low demand.

3. There is efficiency at the top: High demand by

socially efficient quantity.

Low demand guy pay cost more than marginal cost.

Read chapter on vertical integration: chapter 4

that we will start next week. Problem set 1 is due
next week, and next one will be posted next week,
and you have 2 weeks for.

6
 IO @ UTD: Fifth session
Meisam Hejazinia

02/12/2013

New problem set is posted for price discrimination High value consumers pick the one targetted for
that we will finish today, and the vertical integration them, and Low type also select one targetted to
that we will discuss. them. High valuation buy one that targetted to
them.
There is guide to say how to solve. Non linear
optimal pricing or contract, and assumption of This was part of incentive compatibility constraint.
downward sloping demand curve is also there.
θ = θ1 , θ2 are valuation of consumers either high
We looked at two part pricing, and said that two value or low value. θ2 > θ1 . share are per following:
part pricing is good since it allows monopolist to
extract more welfare. It is still not optimal. θ2 is 1 − λ

A = S1 (p) < S2 (p) θ1 is λ

First type of consumersurplus. Second type (T1 , q1 )

consumer go home with something positive.
(T2 , q2 )
Tarriff is function of quantity T (q) = A + pq for
linear tarrif. For non linear tarrif we do not have The monopolist has interest and they make sure
this anymore. You have n group, but you can not everybody buys.
identify them.
There is share that consumer one purchases.
In second degree prevention of arbitrage not
possible. You know that this consumer type are Maximize over bundels that stated
there, and you have to offer a contract to them so
that they choose based on fit. π m = λ[T1 − cq1 ] + (1 − λ)[T2 − c.q2 ]

You extract everything from low type, then high Maximization problem comes with constraing,
type are happy since they go home with positive incentive compatibility, and rationality constraint.
suprlus.
θ1 V (q1 ) − T1 > 0 Individual Rationality 1
(T1 , q1 ), (T2 , q2 ), (T3 , q3 ), (T4 , q4 ), you offer differ- θ2 V (q2 ) − T2 ≥ 0 Individual Rationality 2 (IR2)
ent set of contracts and let them self select for non
linear pricing. θ1 V (q1 ) − T1 ≥ θ1 V (q2 ) − T2 Incentive compatibil-
ity 1

1
 FOC 2.: (−λ)[θ2 V (q2 ) − c] = 0
θ2 V (q2 ) − T2 ≥ θ2 V (q1 ) − T1 Incentive compatibil- θ2 V (q2 ) = c
ity 2 (IC2)
There is optimal time condition by consumer.
1) IR1 & IC2 → IR2 is not binding
→ IC1 is not binding. It is socially efficient or optimal by type two
consumer.
θ2 V (q2 ) − T2 = θ2 V (q1 ) − T1 > θv(q1 ) − T1 = 0
→ IC2 > IR1 Socially optimal q2

→ θ2 V (q2 ) − T2 > 0 What do we know about denominator of q1 is

lower than one, so θ1 V 0 (q1 ) > c
2) IR1 → θv(q1 ) = T 1
V 0 > 0 yet, V 00 < 0, since q1 is small, this
θ1 v(q1 ) − T1 ) = 0 > θ1 V (q2 ) − T2 ends up to be quality. → q1 < q1 ∗, means distor-
tion for the low types, and efficiency on the high type.
T2 > θ1 V (q2 )
We were not able to extract all the benefit from
θ2 V (q2 ) − T2 = θ2 V (q1 ) = T1 high type.

θ2 V (q2 ) − T2 = θ2 V (Q1 ) − θ1 V (q1 ) q2∗ > qq∗

θ2 V (q2 ) − T2 = (θ2 − θ1 )V (q1 ) q2∗ such that θ2 V (q2∗ ) = c

θ2 V (q2 ) − [θ2 − θ1 ]v(q1 ) = T2 > θ1 v(q2 ) q1∗ such that θ1 V (1∗2 ) = c

θ2 v(q2 ) − θ1 v(q2 ) − (θ2 − θ1 )v(q1 ) > 0 θ identified consumer types.

(θ2 − θ1 )(v(q2 ) − v(q1 )) > 0 Two incentive compatibiltiy constraint, and two
individual rationality constraints. The high demand
θ2 > θ1 > 0 if q2 > q1 consumers go home with strictly positive surplus,
due to information rent.
IR1 binding: T1 = θ1 v(q1 )
Incentive compatibility binding does not allow
IC2 is binding: T2 = θ2 V (q2 ) − (θ2 −1 )V (q1 ) incentive compatibility to be binding.
Monopolist will choose πm (T1 , T2 )
θ ∈ [θl , θ̄]
maxq1 ,q2 π m = λ[θ1 .v(q1 ) − cq1 ] + (1 − λ)[θ2 v(q2 ) −
(θ2 − θ1 )v(q1 ) − cq2 ] f (θ) density

We assumed IC1 is binding by not adding it to F (θ) cdf

this problem.
nonlinear tarriff T (q)
FOC 1 : λθ1 V (q1 ) − λc − (1 − λ)(θ2 − θ1 )V (q1 ) = 0
→ θ1 V 0 (q1 ) = c
1−λ θ2 −θ1 consumers θ buys q(θ)
1− λ θ1
T (q(θ))

2
 Monopolist wants each to purchase based on it’s
We assume that q(θ) % θ type mean if it is θi we want him to select qi , Ti .
T (.) % θ By mimicing I would choose the quantity that is
R θ̄ not meant to be to me. By incentive compatibility
max θl
[T(q(θ)) − c.q(θ)]f (θ)dθ we will try to make self selection to be truth revealing.
dU (θ)
so that (IR), (IC) dθ = U 0 (θ) = V (q(θ)) + [θ dv(.)
dq −
dT (.) dq(θ)
dq ] dθ

(IR) θV (q(θ)) − T (q(θ)) ≥ 0θ [θ dv(.)

dq −
dT (.)
dq ] =0

θl : θl v(q(θl )) − T (q(θl )) = 0 → U 0 (θ) = u(q(θ))

(IC): θ = θ̃ U (θ) = θV (q(θ)) − T (q(θ))

θv(q(t)) − T (q(θ)) ≥ θv(q(θ̃)) − T (q(θ̃)) for θ1 ,˜ T (q(θ)) = θV (q(t)) − U (θ)

R
Lowest type: θ̃ = θ − dθ π= θlθ̄ [T (q(θ))−c(q(theta))]dθ
Rθ
θV (q(θ))−T (q(θ))
− θ v(q(θ−dθ))−T (q(θ−dθ)))
≥0 U (θ) = θl
V (q(u))du + U (θl )
d d

limd→0 θ[ V (q(θ))−Vd(q(θ−dθ)) − By individuality rationality constraint U (θl ) = 0

limd→0 T (q(θ))−Td(q(θ−dθ)) ≥ 0 So: =
R θ̄
V (q(u))du
θl

θ.V 0 (q(θ)) − T 0 (q(θ)) ≥ 0 FOC θ θv 0 (q 0 (θ)) = c + 1−F (θ) 0

f (θ) .V (q(θ))

θ̃ = θ + dθ For all theta, other than the highest type,

we have marginal cost on the other side. Quan-
0 ≥ θv 0 (q(θ)) − T 0 (q(θ)) tity on the other part of tarriff is not socially optimal.

θv 0 (q(θ)) − T 0 (q(t)) =) for θ IC. I don’t know who you are, but I have to offer
you a contract. Best quantity or quality would be
Means that imitation would not be possible. derieved by something like this.

Merlis 1971 Chapter 4

If monopolist were to maximize their util- Go through book, and make sure you understand
ity by choosing the type that they mimic things in the book.
maxθ̃ θV (q(θ̃)) − T (q(θ̃)) would be suprlus.
Plug function in and go through them by yourself.
It is optimal for me that I don’t deviate and the θ
that I select would be my own. Go through those, and try to change things to find
out how things would change.
U (θ) = θV (q(θ)) − T (q(θ)) = the one above means
the maximized θ̃ Upstream monopolist

3
 Intermediary product : Retailer
Retailer sells to consumer.
We have upstream market and then we have
downstream market.
Upstream is between manufacturing and retailer
Downstream is between retailer and consumers.
Verticaly integration means, the upstream firm or
downstream firm integrates.
If upstream firm has control directly or indirectly possible.
through the whole chane.
This could be by deligation or by contract.
The aggregate profit of the upstream and down-
stream is the profit.
Whatever retailer does will be internalized by the
manufacturer, and vice versa if they are integrated.
Production M C = c.
Whole sales price pw
Retailer sells good to downstream, and the quan- cient examples.
tity sold is whatever demand function is q = D(pr)
The demand from manufacturer is D(pr)
The quantity demanded in the upstream market
would be whatever it is needed in the downstream.
There are number of different contract and tech-
nologies. One is linear pricing.
T (pw ) = pw .q
Two part tarriff would be T (.) = A + pw .q
There would be contractually retail price controls/
retail price maintenance.
Firms could fix quality/quantity
There would be some contractual solutions.
Whether or not feasible depend on the environ-
ment. You might not know enough about retailer to
fix quantity or quality.
The vertical constraint also depends on what is
feasible.
You might select territory to allow them to have
profit in downstream, and extract as much as
Another way is tie in supplier. If they would need
many input factors and other firms provide those, it
would be constraint.
You get good from me, and then I would be your
exclusive contractor.
Decision variables: whole sales price, then francise
fee A, another is quantity purchased by retailer
(another decision variable), consumer price, retail
location are all decision variables.
Instruments, targets, control problem, and suffi-
Instrument: something the retailer can use, try to
implement vertically integrated market.
Target directly affect aggregate problem.
Retail quantity, amount of service is a target.
Whatever retailer does affects the profit.
Any of these instruments is taken from retailer
and given to manufac.
How to use instrument to reach desired value of
target.
4
 Sufficient instrument is the one that maximizes
integrated profit.
Retailer monopolist, manufacturer is monopolist.
The benchmark is vertically integrated profit,
profit of entire structure.
D(pr ) = D(pr (pw )) = D(pw )
The monopolist will pw = argmaxpw =
(pw − c)D(pw ) → pw > c → first marginalisa-
tion
pr > pw → second marginalisation

Profit for upstream is: c.q + T − T + q.p Problem of marginalisation. The downstream
T what recieves from the retailer. starts with markup.

Total revenue of vertically integrated would be : The first marinalization enforces second marginal-
isation.
π = p.q − c.q = (p − c)D(p)
D(p) = 1 − p
If we want to maximize the entire profit we have
to mazimize this. pr = argmaxp (p − pw )(w − p)
pr = 1+p
2
2

pm = argmaxp (p − c)D(p)
1−pr
qr = 2
q m = D(pm )
π = ( 1−pw 2
2 )
The upstream monopolist does not have full
control on retailer. pw = argmaxp (p − c)( 1−p
2 )

1+c
What if it is decentralized, and each make their pw = 2
own decision?
1−c 2
πm = 2
First manufacturer and then retailer, using back-
(1−c)2
ward induction: πr + πm < πI = 4

Retailer has to set price pr = argmax(p − pm )D(p) non integrated profit:

3
FOC: D0 (p)(p − pw ) + D(p) = 0 π ni = 16 (1 − c)
2
which is lower than the integrated
one
Price is not decreasing in marginal cost. The
higher marginal cost the higher marginal cost. If upstream is monopolist and downstream firm is
competitor then pr = pw
If c < pw , the whole sales price larger than the
actual cost of producing. pr > pm and manufacturer pw is marginal cost in downstream market.
does not want it.
In this case monopolist maxp (pw − c).D(pw )
The target is not sufficient, the target is not equal
to vertical integration price. pw = pm means equal to benchamrk price

5
 There is no markup in the downstream, and This was very basic, now lets talk about vertical
we are still not in deadweight loss. We maximize instruments.
decentralize profit.
Retailer sets price greater than marginal cost
Vertical integration does not increase profit. pw . Manufacturer also does this, so would be
double marginalization. Target is price and
Upstream is perfect competitor, and downstream we want to implement, and charge this price
competitor. pm = argmaxp (p − c)D(p)

Then we will have pw = c First one is franchise fee.

pr = argmax(p − c)D(p) The two part tarrif would be T (q) = A + pw .q

First order condition for the monopolist: pw = c

For manufacturer: D0 (p)(p − c) + D(pw ) = 0 → pr = pm

For retailer FOC: πr = πm − A

D0 (p)(p − pw ) + D(p) = 0
A = π m Franchise fee, which would be equal to
Marginalization has an effect on monopolist. The
consumer surplus.
monopolist will make lower profit.
Manufacturer profits are πw = A + (pw − c)D(p2 )
Horizontal externality external 4.2 in the book.

Briefly we will talk about this set up. A = π m , so every one would be happy.

Two firms each produce one good. Goods are per- Retailer gets everything, and is claimer of what is
fect complement. One firms sells to the downstream left.
market.
Another draw back is if there is private informa-
One decision maker, what happens if both firms tion, we may not know what A should be. Two part
sell the complementary product when there is only tarriff could be used as screening price.
one customer in the market. Combine profits from
two, and you will see decision variable. pr = pw , if you have many retailer, the will charge
whole sale price.
Horizontal structure, but firm 1 makes decision
first, and firm 2 makes his decision based on the price Manufacturer could require retailer to charge the
firm 1 decides. This could be solved by backward target price.
induction.
pT = pm
Two firms choose prices simultaneously. Show
that the price is equal to some of these experessions. How can we use retailer price maintenance, and
In this excercise you will have sequential decision whole sales pricing that include all vertically inte-
making. grated profit.

6
 We could have pw = pm for the input. Our
manufacturer charges target price. Then manufac-
turer require retailer to charge pr = M Cr = pw = pm
That implements target price, and it insures that
manufacturer has all profit, and retailer walks home
with nothing.
This is vertical integration by contractual control,
and not by ownership.
Retailer exerts some promotional effort.
FOC: D(pw ) + (p − c)D0 (p, s) = 0
Retailer’s: maxp,s (p − pw − Φ(s))D(p, w)
FOC, p D(p, s) + (p − pw − Φ(s)) ∂D(p,s)
∂p =0
FOC, s (p − pw − Φ(s)) ∂(p,s) 0
∂s − Φ (s).D(p, s)
The retailer of service would be lower than bench-
mark case. There are so two distortion. Second one
would be double marginalization.

Manufacturer has incentive not just price but

service. Service is not contractable. Either because
it is not observable, or costly to control. (customer
service, promotion, and advertising are sample of
these services).

Retailer don’t want to, since they don’t acquire

the service and should send to manufacturer.

This is model of vertical integration. Higher

service results in higher quality of product.

q = D(p, s)

q & p, q % s

Cost of service is Φ(s) % s

P hi(s) per unit of the good

maxp (p − c − Φ(s))D(ps )s)

π m = (pm − c − Φ(sm ))D(pm , sm )

FOC with respect to s would be (p − c −

Φ(s)). 2D(p,s)
s − Φ(s) − D(q, s) = 0

with p: (p − c − Φ(s)) 2D(p,s)

2p + D(p, s) = c

Manufacturer charges the price:

Manufacturer:

7
 IO @ UTD: Sixth session
Meisam Hejazinia

02/19/2013

The instruments that monopolist can use. Monopolist wants the retailer to charge the price
that maximizes the vertically integrated means
Sufficient vertial restraints. pr = pm .

Downstream moral hazard. s = sm

Franchise fee as an instrument. The retailer profit π m = pin − A

T (q) = A + pw .q For manufacturer we have π in = A

For franchise fee to work, the proposed instrument
is as stated. Monopolist will charge pw = c
Double marginalization is problem since does not
allow the vertically integrated profit.
Retailer can not set pr , but we wanted to still
provide efficient level of service. We knew that the
price that retailer will charge is pin
The input price would be pw in this case.

The retailer first order condition are now RPM

First order condition for the price. Retailer price maintenance is that retailer sticks
to that price.
D(p, s) + (p − c − Φ(s)) D(p,s)
∂p =0
p̄r = pin
c is marginal cost for the retailer.
The monopolist makes profit by charging whole
First order condition for the service: sales price pm > c

−Φ0 (s)D(p, s) + (p − c − Φ(s) ∂D(p,s)

∂s =0 Monopolist wants to engage in markup pricing.

The retailer has efficient incentive to provide the Retailer will pay something more mean pw > 0
service. and retail price maintenance would not be proper,
since retailer would not provide enough level of
pm is the price retailer pays for the input. service.

pr would be the price that retailer charges. Quantitative forcing

1
 sufficient vertical restraint ex. 4.3. The joint profit of manufacture one and the
retailer. For vertically integrated is manufacture one
The moral hazard problem is that the retailer and whatever retailer produces. We are interested in
provides service, which is not observable by the the optimal input factor. If the optimal input fac-
manufacturer, and manufacturer want to extract tor would be there the incentive would be to charge p.
maximum profit.
(x, x̃) = argmaxx,x̃ [p(f (x, q))f (x, x̃) − cx − c̃x̃]
There is double side mean there is state when
manufacturer also wants to provide the service. In First order condition on x would be
this case in excercise 4.4 D(p, S, σ) fx0 [p0 (x, x̃) + p] = c marginal benefit factor would be
equal to the the marginal cost, and for the second
σ could be advertising expenditure of mcdonalds. one we will have fx̃0 [p0 .f (x, x̃) + p] = c̃

Demand function is increasing in both σ and s Marginal rate of technical substitution would be
M RT S = − c̃c
It is called, bilateral Moral hazard.
M Px fx0 c
|M RT S| = M Px̃ = fx̃0 = x̃
We have upstream manufacturer who produced
two input factors. Upstream manufactuerer produces pw price for input factor x
two goods x and x̃, and there is second manufacturer
produced substitute good in the production process pw > c
x̃.
actual price ration = pp̃w = pc̃w > cc̃
The retailer in downstream needs x or x̃ for output w

process. The output function would be f (x, x̃). You

Since there would be competition for the second
can used more of x, and less of x̃, and this is called
factor, the price would be equal to marginal cost,
substitute in the production process.
but for the first input factor we have monopolist, so
pw would be greater than its cost.
f (0, x̃) = 0 and f (x, 0) = 0

Also the marginal cost for the first manufacturer We want the retailer to use xxin , and x̃ → x̃in
is c and for the second one is c̃.
pw = c
The first intermediate good has marginal cost of c
and if it is producing x̃ then the marginal cost would p̃w = c̃
be c̃.
A franchise fee that extracts profit.
The retailer set price p, and the produce output
level of f (x, x̃) Frachise fee is sufficient vertical restraints.

Inverse demand would be p(.) = D−1 (.) resulting The second one is tie in with resale price
from the demand of q = D(p). Then the price in the maintencance
donstream market would be p(.) = D−1 (f (x, x̃))
The downstream unit, purchases both of the
What would be the benchmark? factors. Monopolist of first factor, says if you are
not going to purchase second factor, I will not give

2
you first input factor. problem. The first is what is the production capacity,
and second they decide about the price. This is on
The upstream manufacturer should decide pw and the day to day, firms decide to how set their prices.
pw . On the daily basis they set prices. This will back to
cournot equlibrium. Firms compete in capacity and
The upstream manufacturer will select pp̃w w
= c
c̃ then on prices. Then we look at quantity and price
to force the retailer to manufacture the amount leadership, and finanlly on conjecure of prices.
required x = xin , and x̃ = x̃in
Cournot
The retailer has to charge pin , and then upstream
monopolist will go home with zero profit. Firms compete in quantities of non differentiated
products (identical). Goods are perfect substitutes.
We use retail price maintenance, to make sure re-
tailer charges this price, so that monopolist extracts π i (qi , qj )
profit.
P (Q) would be demand function
Tie in would be in the form of exclusive contract.
Q = qi0 + qj
The next section is interbrand compeition that we
will not consider here. π i (qi , qj ) = p(qi + qj )qi − ci (qi )

If you don’t use instrument, you will have double π is concave, and we want it to be twice differen-
marginalization problem. tiable in qi

Through franchise fee you solve this problem. We look at the best response or reaction function.

The tarriff would be in the form of three part We denote by Ri (qj ).

tarriff T (x, x̃) = A + pw .x + p̃w .x̃, which will give
you the same condition as price maintenance. The quantity that firm i will produce would be,
qi = Ri (qj ) so that πi0 (R(qj ), qj ) = 0
Set the retail price may not be problem due to the
anti trust law. π i (qi ,qj )
∂qj = 0 Ri (qj )

We leave the real of monopolist, and now we look Ri (qj ) is unique, and single valued.
at the strategic interaction between two firms. First
wee look at the static form, one shot. In three weeks Slope of the reaction function Ri0 (qj ) =
∂Ri (qj )
=
∂qj
after spring break, we will look at the dynamic form. i
πij (Rj (qj ),qj )
i (R (q ),q )
−πij i j j
We start with the cournot model, then bertrand
competition of prices, and then we look at the By cancavity assumption the denominator is
capacity constraints. The problem with bertrand is positive.
that we have imperfect competition. Both firm will
earn zero profit in Bertrand. Bertrand paradox in sgn(Ri0 (qj )) = sgn(πij
i
)
Nash equlibrium will tell us that this is not case.
Once product are differentiated, both firms will earn If the cross derivative is negative, this means the
positive benefits. The third solution is two stage marginal profit for firm i is decreasing the more j

3
 αi
produces, and this is defintion of strategic substitute Li = 
πi0 j < 0
Lerner index would be the following. The more
We can draw reaction functions of each quantity firm you have αi would be smaller.
based on other quantity, and when two reaction
functions intersect, we would have equlibrium. Lcourn
i < Lmonop
i

Quantity competition is not always strategic Lcourn

i > Lcomp
i =0
substitute, but most of the time.
In most IO papers, people have assumed linear
Nash equlibrium qi = Ri (qj ) and qj = Rj (qi ), demand.
and we want these two to be mutual best responses.
qi∗ = Ri (Rj (qi∗ )) and qj∗ = Rj (Ri (qj∗ )). P (Q) = 1 − Q = 1 − qi − qj
Ci (qi ) = ci .qi
π i (qi , qj ) = p(qi + qj )qi − ci (qi )
For nation equlibrium we must find (qi∗ , qj∗ ).
dP ∂Q 0
πoi (qi , qj ) = q i dQ ∂qi + p(qi + qj ) − Ci (qj ) = 0
Then we must find π i (qi∗ , qj∗ )
p(qi + qj ) − c0 (qj ) + qi .p0 (qi + qj )
Where p(qi + qj ) − c0 (qj ) is profitability of extra unit. π i (qi , qj ) = q i (1 − qi − qj ) − ci .qi

qi .p0 (qi +qj ): losess of all individual marginal units. FOC: 1 − qi − qj − qi − ci = 0

1−qj −ci
Qi = = Ri (qj )
p(qi + qj ) − c0i (qj ) + (qi + qj )p0 (qi + qj ) 2

The same symmetric things happen for j, by

’social costs > ’individual costs’
replacement. As a result:
The total quantity in cournot model would be qi∗ =
1−2ci +cj
3
greater than monopoly.
1−2cj +ci
qj∗ = 3
We can compare three different structure, compet-
itive, cournot, and monopoly. p=
1+ci +cj
3

comp q > Cournot Q > monopoly Q The profits are π i = (

1−2ci +cj 2
)
3

comp p < cournot P < monpoly p. For linear demand we usually have this quadratic
form of quantity as the profit.
First order condition p(Q) − c0i (qi ) = −qi .p0 (Q)
0 0
What if we have Cournot with n firms? identical
− p(Q)−ci (qi )
p(Q) = − qip(Q)
.p (Q)
= − qQi p P(Q).Q
(Q) firms
Pn
= −p(Q)p0 (Q).Q Q= i qi
qi P
Q = αi Q−i = j6=i qj

4
 π i = p(Q).qi − ci .qi
1 1
a a
1. D(Q) = Q
 = qi +qj


Assuming different marginal costs. Assuming

identical we can get rid of i here. We take the first derivative, and we get
1
1 
(1 − 1 q1q+q
1
) = 1c
π i = p(Q) + qi .p0 (Q) = c qi +qj 2

We take the cross derivative:

We assume linear demand so:
∂ 2 πi α 1 q1 −q2
π i = 1 − Q − qi = c q1 ∂q2 = ( q1 +q2
) 2 (q1 +q2 )2

= 1 − Qi − qi − qi = c There would be two sectors cutting each of the

axis, and their center on each axis, and their inter-
Therefore, best response function, given what section will give us the equlibrium.
everybody else produces would be:
 is constant elasticity.
1−Q−i −c
qi = 2 = Ri (Qi )
Stackleberg equlibrium (1934)
Symmetric equlibrium, so would be: √
P (Q) = 100e1/10) Q
qi = qi
√
R1 (q2 ) = 200 + 20 100 + q2
Q−i = (n − 1)q
R1 (q2 ) % q2
⇒= 1−(n−1)q−c
2 = q
U shape Average cost curve.
q∗ = 1−c
n+1
Convergence example.
The equlibrium price would be p∗ = 1 − nq ∗ =
1−c
c + n+1 >c
− c(q)
q

π ∗ = ( n+1
1−c 2
)
Minimum efficient scale at 1.
We have lot of firms, then limn→∞ q ∗ = 0 and
limn→∞ p∗ = c, and limn→∞ π ∗ = 0 All firm produce at minimum efficient scale which
we assume 1.
In long run, each firm will produce at minimum
efficient scale. Marginal cost at q = 1 is c.

We skip the existance of equilibria, and uniqueness. The minimum efficient scale is 1 and M C = c.

The reaction function was downward sloping, and p = D(Q)

here we had strategic substitute.
We allow them to enter at minimum efficient scale
Inverse demand function give us strategic comple- α.
ments.

5
 Then we reduce α. All of them who enter should to take contradiction.
produce α, and as we decrease it, the number of
firms will go to infinity, and this is convergence. Q < Q∗ − α. If this was an equilibrium, an extra
firm will get the negative profit. If one firm enters
cα (q) = α.c( αq ). for this minimum efficient scale Q + α < Q∗ → Q < Q∗ − α. Means, one firm can
would be α. enter the market, so that everyone makes positive
profit.
We minimize minq cαq(q) = min c(q/α)
q/α
The total amount that is produced has to be
q somewhere in [Q∗ − α, Q∗ ].
α =1

q=α As we take limα→0 [Q∗ − α, Q∗ ] = Q∗

Mc = c This would be competitive equlibrium.

The smaller α, more firms would be in this market. Raffen (1971) in review of economics and statistics:

The number of firms would be Q c(q) = c − d.q 2 + e.q 3

α

a − c > 0, d − b > 0
It depends on how you can squeez them, for
example 92 would be 4.
p(Q) = a − bQ
We fix α, and then we change the number of firms.
The long run equlibrium would be where price=
marginal cost = Average cost. p = M C = AC.
We are going to show that Q ∈ [Q∗ − α, Q∗ ].
d
q = c, and q = 2.e .
Q∗ = perfectly competitive quantity.
d
qcLR = q = 2e
Q∗ = D(c).
d2
pLR
c = AC(qcLR = c − 4e
∗
Q>Q :
Q−c d2
cα (qi
QLR
c = b + 4bc
P (Q) < c ≤ qi , mean not greater than average
cost of each of the firm. What’s the total quantity of all the other firms
given what I produced.
Negative profit for operating firm.
Q−c+2q(d−b)−3eq 2
Q−i (qi ) = b
This will not work.
We have many firms here. π i (qi , Q−i ) →
∗ i
Q<Q −α π (qi , Q−i (qi ))

d−b
The firm that is not yet in the industry makes π i (qi ) = 2eqi2 (qi − 2e )
zero profit. If Q < Q∗ − α, then no firms wants to
enter, since any additional firm will not get profit, we want qi so that πi (qi ) =? 0
otherwise there would be no equilibrium. We want

6
 π i (0) = π i ( d−e
2e ) = 0 c1 < c2

∂π i
∂qi qi =0 >0 p1 = c2 − 

LR
qcor = d−e
2e <
p2 = c2 ; pm m
1 if p1 < c2 − 
2
−b2
pcor = c − d 4e
LR
> pLR
c
pm
1 < c2 − 
2 2
a−c d −b
QLR
cor = b + 4be < QLR
c
π1 = (c2 − c1 )D(c2 ) = (pm m
1 − c1 )D(p1 ) Monopoly
LR QLR
Ncomp = c

LR QLR Look at 5.3, 5.4, and 5.5. The solutions are at the
Ncor = cor
LR
qcor back of the book.
LR LR
Ncor > Ncomp Three solutions for Bertrand Paradox.

In this equlibrium of cournot, we have more firms, 1. repeated interaction

and less is produced in long term than competitive
market. 2. product differentiation (not identical products,
gives market power to them, leading positive profit).
In second case, we had zero profit, in long run, and
everybody only produced minimum efficient scale. 3. Capacity constraints.

The number of firms in all are indogeneous. The
cost structure would be different.
Bertrand
Firms make simultaneous pricing decision.
Cournot and bertrand are simultaneous, with
price or quantity leadership, we will have leader
and follower model, and we will look at subgame
equilibrium. Variation in supply function, and
conjectural variation will be considered last.

Identical products. Everybody produces the same.

π i (pi , pj ) = (πi − c)Di (pi , pj )

Question is what is demand function:

Di (pi , pj ) = D(pi ) pi < pj
D(pi ) D(pj )
2 = 2 pi = pj

0 pi > pj

Nash equlibrium would be π i (p∗1 , p∗j ) ≥ π i (pi , p∗j )pi

if c1 = c2 = c → p1 = p2 = c

Bertrand Paradox.

7
 IO @ UTD: Sixth session
Meisam Hejazinia

02/26/2013

Familiarize yoursef with mathematica or maple Equlibrium in two stage game is cournot equlib-
to simplify the problem, if calculation becomes rium.
cumbersome.
Suppose we have a situation of decreasing return
D[f [x], x] = f 0 [x] to scale, increasing average cost. The firm will not
want to produce more.
You can put assumptions also there.
Suppose q̄i = Si (p) = D(p) is firm supply that
Example solution is online. It is with the same the firm is willing to supply at price p. Firm will
name. face residual demand in this case. D(p) − Si (p).
Sometimes we denote this by Djr (p) = D(p) − Si (p)
We went over static oligopoly, and we went residual demand of firm j.
through cournot analysis.
The result is that firms have capacity constraing,
Law for repeated interaction, in the market so we have firm Rationing
situation when we have two firms, oligopoly.
Π = p(q1 + q2 ).q1
Two firm that have market power yet the result is
price equal to marginal cost. What if p = 1 − Q, and p = 1 − q1 − q2 . The
price would be to the demand I face which is not full
Two solution to bertrand paradox: demand but residual demand.

1. Product differentiation (imperfect substitute): Firm 2 supplies certain side of the market. Who
a week after spring break. are my customers? What is the rationing rule?
Efficient rationing and proportional rationing are
2. dynamic game: in multiple periods. two types.

Final solution is capacity constraing, and then Efficient rationing You have consumers with
decide on prices. The result is Bertrand, but the ca- lower marginal valuation and so on, and the customer
pacity constraints equals the equlibrium to cournot. who is the most eager will buy, and whoever is left
firm j will face. Mean most eager consumer will buy
Kreps-Shenkman in 1983 worked over this model. from i, and that shifts qj .

In Tirole 5.7.2 explains this. Proportional Rationing Rule: Probability of

not having bought earlier is (D(p) − Si (p))/D(p).

1
 We can read the price of the demand function.
The probability of being part of the residual Lemma 1 is that both firms select from this price.
demand is (D(p) − Si (p))/D(p).
Lemma 2: pi ≥ p(q̄j + Ri (q̄i ))
The large part of discussion is on rationing rule.
From these two lemma q̄i ≤ Ri (q̄j ) impure strate-
At stage one firm choose capacity, and that gies.
capacity is fixed.
mixed strategies: Tirole
On stage 1 set the capacity, and on stage 2 set
prices, and all firms do that simultaneously. We will have interest of little capacity. We have
a join interest. If we hit the price, it would be
It is not first cournot then bertrand, but it is lower than capacity. The competition for lower price
about production capacity. I choose my storage or would not be there. Jointly have interest to keep the
production capacity, and once the pre stage is over, capacity lower.
I then compete with someone else on quantity. You
can do that as inventory. In cournot that was not We look at the capacity game stages. There are
the case, since we through in the market and let the cost of producing capacity. The cost for selling later
market decide the price. on will be equal to zero c = 0.

We try to solve this by backward induction. Nash equlibrium is such that q̄i = qj∗∗

First stage: find Nash equlibrium in pricing game. q̄j = qj∗∗

Second stage: Nash equilibrium in capacity, qi∗∗ = argmaxqi qi [p(qi + qj∗∗ ) − co − c]

giving, or anticipating price equlibrium.
This happens to be cournot equlibrium of quan-
First we think ahead, and then think about the tity.
capacity. I maximize my cost margin, given your best
quanity response, and vise versa.
Pricing game:
At first stage firms will select cournot quantities
Both firms have some capacity constraints. There as their capacities. qi∗∗ , qj∗∗ and on the second stage
are capacity constraints q̄i , q̄j of pricing they will have pi = pj = p(qi∗∗ + qj∗∗ )

We assume c = 0, mean marginal cost equal to The basic intuition is that both firms want to
zero, p” < 0 constraint themselves at lower capacity. In equlib-
rium, capacity is simultaneously non-cooperatively
Efficient rationing, the highest willingness to selected.
pay buys first, and the low willingness to pay will
purchase then. This is efficient, with those with The outcome of two stage game boils down to one
higher valuation to purchase, if it was with limited stage cournot game.
supply.
The quantity decision is long run, and price is day
p1 = p2 = p(q̄1 + q̄2 ) to day decision. Capacity could not be configured on
daily basis, but price could. Cournot is simple way

2
of representing this two stage game. D2 (p) = a + c.p1 − b.p2

We will look at the product differentiation in more Imact on both are the same.
detail in one week after spring break.
After spring break we will think about the decision
For cournot model, the demand function is to differentiate the product.
p(q) = 1 − q1 − q2 for homogeneious goods.
Make your life simple and use this structure
We can add one parameter for differentiation. rather than decision to differentiate, only with one
p(q) = 1 − q1 − γq2 , if competitor increases his price parameter, to make your life simple.
by one unit, the price will go down by less than
one unit γ < 1. This is for non homogeneous case, Suply function competition
or product differentiation. When goods are not
perfectly substitutable. Firm compete in supply function. They do not
produce certain price or quantity, but the choice
γ < 0 complement variable is supply function. It was introduce first by
Hert (1982), and Grossman (1981). Whey should
γ = 1 homogeneous competition be in price or quanity. How firm decide
to choose in price or quanity. Given the market price
γ = 0 monopoly I will decide how much I will produce.

This would be differentiated cournot model. Firms produce homogeneous goods. Firm do
legally binding contracts, saying that if market price
You can write for differentiated Bertrand model in this, I will produce this much, if another I will
the form of D(p) = 1 − p1 + c.p2 The only difference produce another quantity.
is that we have positive sign for the bertrand model,
yet we had negative sign in cournot model. The assumption is that firms are able to commit.

If c > 0 we will have substitute, and if c < 0 we We have n-firm oligopoly, and we have standard
will have complements. downward slopping demand curve.

Leviton (1920s), and Boley (1924) We have a collection of supply functions

(Si (.), i = 1, n)
You can have the cournot one in the form of
p(q) = α − βq1 − γq2 . For the Bertrand we will have There is a demand function and the market
Pn clears
D(p) = a − b.p1 + c.p2 at the price that demand will be D(p) = i=1 Si (p)

There is inverse demand function p2 (Q) = This is the market clearing condition, and there
α2 − β2 .q2 − p2 .q1 would be such a price.

Q would be function of q1 , q2 , but not summ of The firm’s profit is i = p.Si (p) − C(Si (p))
them. p2 (q1 , q2 ) = p2 (Q)
These may not necessarily be identical firms, the
D1 (p) = a − b.p1 + c.p2 supply are different, but the costs are identical.

3
 The firms are identical in term of marginal cost, demand.
but they do not supply the same quantity.
In equlibrium θ will drop out. The profit will be
You want to play your best response to whatever πi = p(θ)Si (p(θ)) − c(Si (p(θ)))
everybody else does.
The choice variable, and action set is, the set of all
If marginal cost would be different, the equlibrium linear supply functions. They make choices simulta-
would not be symmetric anymore, and the model neously, we look for equlibrium supply function. We
would not be tractable anymore. know this is going to be unique supply function.

Firm choose supply function simultaneously. b would be as parameter in supply function Si (p),
since p(θ) you supply function will have θ inside.
Given supply function we will find out the market Those prices depend on θ, the quantity will depend
price, and given market price we will find out on it, but indirectly through price.
equlibrium quantities.
D(p) = θ − b.p, and S(p) = l.p linear supply
There is multiplicity of equlibrium. I see your function in the form of aggregate. p = θ
l−b
supply function, and given the anticipation of your
supply function, I will decide my supply function. With different realization of θ we will have differ-
ent price.
Klemperes, eyer (1989): added uncertain demand.
How do we choose the supply function?
Strategy is supply function, and you can have
mixture of supply function.
Sj (.)
D(p, θ) = D(p) + θ b > 0 θ[0, ∞) r
PResidual demand is Di (p, θ) = D(p, θ) −
S
j6=i j (p)
If we do this we will get unique equlibrium in
linear supply function, if demand is linear.
maxp [p.Di (p, θ) − C(Dir (p, θ))]
If D(p) = −b.p, D(p, θ) = θ − b.p there would be
unique supply function in the form of linear. So you Now we assume that for given θ, there would be a
don’t look for any other form. price. We have to consider all possible θ.

The paper is on the sylabus and you can check it We assume that a unique invertible p(θ) exists.
out there.
The supply function is the collection of all these
The firm commits to twice differentiable function. prices (pi (θ), qi (θ)) where qi (θ) = Djr (p, θ)
They select quantity price competition.
Instead of writing residual demand you can write
The firm commite to supply function. Commit- your supply Si (p)
ment before θ is observed. This helps us to boil
down to unique equlibrium. First order condition would be p − c = − ∂DrS(p,θ)
i
i
∂p
∂Dir (p,θ)
= D0 − Sj0
P
Once θ is P
realized there would be p(θ) such that ∂P j6=i
n
D(p, θ) = i=1 Si (p), since theta will shift the

4
 (n − 1)S 0 (p) = s
p−c(s) + D0 (p) The total revenue is T R(q1 , q2 ) = p(q1 + q2 ).q1 ,
and marginal revenue is M R1 = p(q1 + q2 ) +
∂Q
For two firms: ∂q1
dP dq1
+ ∂dq2 dq1 = 1 + dq2
q1 . dQ [ dq1 dq1 ]
maxp [p.[D(p) + θ) − q2 (p)] − C(D(p) + θ − q2 (p))] I form belief about how my quantity will affect
my rivals quantity. It is still static. This is the place
First order condition is D(p) + θ − q2 (p) + {p − that classic IO theory crashes with game theory.
c0 (D(p) + θ − q2 (p))}[D0 (p) − q20 (p)] = 0 Hicks (1936), and Bowley (1924).

dq2
Where D(p) + θ − q2 (p) = q1 (p) which is residual. dq1 = λ1

Rewriting this will result in qi (p) + {p − If λ1 = 0 we will have cournot.

c0 (q1 (p))}[D0 (p) − q20 (p)] = 0
In the original paper θ was not there so we If λ1 < 0 → q1 &⇒ q2 %
removed.
If λ1 > 0 → q1 &⇒ q2 &
q(p)
q 0 (p) = p−c0 (q(p)) + D0 (p)
FOC:
Show (Problem set 3):
q1 = q2 = q
D(p) = −bp
λ1 = λ2 = λ
D(p, θ) = θ − bp; θ[0, ∞)
p−c0 (q) 1+λ
p = 2
c.q 2
c(q) = 2 ;c >0
This is Lerner index.
S(2) = d.p
q λ = −1 → competitive
d = 12 [ 1c − b + ( 1c − b)2 + 4b
c )
λ = 0 → cournot
Compare with cournot and Bertrand.
= 1 → monopoly . collosive
Conjectual variation
Conjectural elasticity.
Our assumption in cournot was that I choose my
q1 dq2 dlogq2
quantity given my rivals quantity. My best response α1 = q+2 dq1 = dlogq1
is q1 . In bertrand, yet we did not have any quantity
dq2 q2
effect. dq1 = q1 .α1

p−c0 (q) α+(1−α) 12

The cournot was static game, simultaneous select. p = 
Static game does not allow us how does my quantity
affect the quantity my rival selects. My choice will α = 1 monopoly case
not affect my rival’s choice. My choice should not fit
back. α = 0 cournet

5
 α = −1 competition

Elasticity here is quantity elasticity

Frish (1933)

R2 (q1 )
dq2
R20 (q1 ) = dq1

Numerical example:

Leontiff p(q) = b − q1 − q2
q12
c(q1 ) = 4

∂πi
∂q1 = b − (5/2 + λ1 )q1 − q2 = 0
∂πi
∂q12
= −(5/2 + λ1 ) < 0
6−q2
q1 = 5/2+λ1

6−q1
R2 (q1 ) = q2 = 5/2+λ2

−1
R20 (q1 ) = 5/2+λ2

−1
R10 (q + 2) = 5/2+λ1

λ1 = R20 (q1 )

λ2 = R10 (q2 )

λ = −2

λ = − 12

This gaves the same result without going through

full fledged dynamic game.

6
 IO @ UTD: Eighth session
Meisam Hejazinia

03/05/2013

player 2 coop player 2 defect theory point of view, since we need enforcable curtail
player 1 coop 3,3 −1, 4 agreement, which is usually illegal. It is not really
player 1 defect 4, −1 1, 1 interesting to look at the institution economist point
of view. We look at the overt colusion.

Asset colussion/ static colusion It is an open agreement. They have quantity

setting, and they decide what is the optimal quantity
How many firm expected that they should produce.

Selthon q is quantity per firm

Dynamic colusion n is the number of firms

repeated game Q is the total quantity

super games Different size firms may end up with different

quantities. They sit together and try to maximize
Payoff matrix of normal form game profit. They maximize total profit minus cost of
production:
cooperation q m /2, each will produce. When Pn
defection we will have couple of production quantity maxq̄,Q p(Q).Q − i=1 ci (qi )
(q m /2, q m /2).
All the individual
Pn quantities should be less than
In one shot game equilibrium tells us there is no total quantity: i=1 qi − Q ≥ 0. It is an agreement,
cooperation. Cooperation means firms will collude, and this is to have it make sense.
since they would be better off jointly.
You also assume that everybody produces non
Today we are interested in condition in which negative quantity qi ≥ 0.
collusive equiliubrium is sustainable.
First order condition with respect to individual
Then we will look at the super game, repeated in- quantity qi : p0 (Q).Q + p(q) − c0i (qi ) = 0 ∀i
teraction game, what is the parameter δ that we must
have to make sure that we have collusive equilibrium. We run this true. There is proposition. Going
through this analysis we find out that they produce
Static colusion is not that interesting from game less than cournot, so Cortel: higher profit, of non

1
cooperative equlibrium. Also Coutel produces lower Cortel would be leader, and the rest are followers.
quanity. Stackleberg view that first Qk amount of Cortel will
be set, and then the rest will decide how much they
The probalem is that without external enforce- want to produce.
ment the Cortel is not sustainable. If you do not get
punished, you will deviate. STACKELBERG

Selten, (1973), IJGT (international Journal of The price would be function of the amount the Cor-
Game Theory). Something was previously in ger- tel would produce and the rest of members
Pn produce.
man. ’Four or two few, six or too many’ was the The quantity would be fixed. P (Qk , i=k+1 qi )
name.
When the firm want to enter, and when the firm
n firms in the market wants to exit, sit the thing that we are searching for.

k firms in the Cortel Stage 3 we get the Qk given, and then each
members profit N : non member, of firm i:
k≤n P
πiN = qi (1 − Qk − j 6= iqj − qi )
It does not say anything about size of Cortel. It is
about size of the market. We got the first order condition :

n ≥ 2 at least two firms. maxqi πiN : (1 − Qk − (n − k − 1)qN ) − 2qi = 0

MC = 0 qj = qN
P 1−Qk
Linear demand: p(Q) = 1 − Q Q ≡ qi qN = n−k+1

(n−k)(1−Qk )
three stage QN = n−k+1

1. joint or not Stage 2:

2. The curtail writes an agreement: Qk Mk = (1 − QN (Qk ) − Qk ).Qk

1
3. Firms play a Cournot game FOC: Qk = 2

1
→ Cortel firms: qk stick to what is agreed. This is general result of stackelberg. πk = 4(n−k+1)

1
Al other firms qi = Ri (qj , Qk ) where Qk is the πk (k, n) = 4k(n−k+1)
amount that Cortel produces.
1
πN (k, n) = 4(n−k+1)2
There are k Cortel members, they range from
1 . . . k {1, . . . , k} Stage 1: All inclusive: k = n

non members would be {k + 1, . . . , n} pik (k, n) ≥ πN (k − 1, n)

2
 n>4:k<n PIn each period the firm makes the profit
T
t=0 i (ait , ajt
π
Internally stable. When there is no incentive to
quit. pik (k, n) ≥ πN (k − 1, n) Do not quit δ<1

The external stability is do not joint the Cortel: Low delta means high impatience, or low patience
πN (k, n) ≥ πk (k + 1, n)
Cournot game ait = qit
n = 5/n = 6 ⇒ k = 4
In bertrand would be price: ait = pit
n = 9/n = 10 ⇒ k = 6
The profit function does not depend on t. We
Cortail all inclusive agreement works only when allow the actions to be function of the history
market is small.
Ht = (ai0 , aj0 , ai1 , aj1 , , ait−1 , ajt−1 )
Super Games
For bertrand we will have:
Repeated Games
T <∞
We have a stage game: In each stage firm have
price or quantity decision
ait = pit
Stage game is played T times
pit = c
T <∞
ci = cj = c
T =∞
We do backward induction.
Super game is stage game that is played T times.
Payoff in each stage game is time invariant. Read 6.5.(.2) section Kleps, Milgram, Roberts,
Wilson (1982), Gang of 4. They considered the
We need to specify what to do in each of the simple stage game. Super game, and they played T
priod, on each of the contingency. periods. There is a share α of crazy people. Not
mean completely rational, but they have strong
Let (ait , ajt ) are firms actions in the stage game. preference for cooperation. There is considerable
πi is the firm’s profit in the stage game. amount of people who will cooperate if you cooperate.

Action is going to be quantity. With small fraction of players who cooperate,

will be sufficient for the collusive equilibrium to be
The profit is the profit from stage cournot game sustainable, even in the finite game.
or Bertrand game.
If there is this type of uncertainty, then even in the
The profit function in itself, is real life and is finite game there would be sustainable equilibrium.
dependent upon action which is dependent upon t.
πit = πi (ait , ajt ). When we say collusion it means tacit collusion.
They do not talk with each other. This is ’coopera-

3
 aN cc aN ash Defect:
aN cc 3,3 -1,4
aN ash 4,-1 1,1 πi,Def + δ.πi,N ash + δ 2 πi,N ash + = πiDef +
δ
1−δ πi,N ash

tive’ equilibrium in the non-cooperative game. 1

≥ πi,Def + δ
1−δ πiN cc 1−δ πi,N ash

Tricky thing is that they do not talk to eachother. δ≥

πi,N CC = πm /2
If T = ∞ playing the Nash equilibrium is subgame
Nash equilibrium, but it is not the only equilibrium. πi,Def = π m
There are many equilibriums.
πi,N ash = 0
There are many subgame Nash equilibrium. Play
that in all equilibrium is Nash equlibrium. Then 1
δ≥ 2
we say Tacit collusion is sustainable equilibrium.
(Friedman, 1971), Fold theorem. Couple of applications:

Always cooperating is jointly Nash equilibria. In 1. with n firms

any stage game, there is strategy aN CC mean non
cooperative collusion, since we did not talked to each 2. information lag (not immediate punishment):
other, but it is tacit collusion. you do not find out quickly

Trigger strategy: 3. Price wars during doom: low demand state.

Lower competition results in collusion, but in boom
1. Start with place aN CC , which is good strategy. the collusion collapsed.
Continue as long as all other player play the same.
1. Tacit collusion is easier, smaller number of firms
2. If somebody deviates from the strategy that
does not play aN CC revert back to aN ash . πm
We assume Bertrand with n firms. πi,N cc = n

I observer whether someone defects at the end of πi,DEF = π m

period, and I will shift to aN ash as soon as I observe
that. pii,N ASH = 0
πi,DEF −πi,N CC
The profits for Nash, or profits if we both play Nash δ≥ πi,DEF −πi,N ASH
equilibrium strategy: πi,N ash = πi (anash , anash )
Tacit collusion would be easier smaller the number
pii,Def = πi (aDef , aN cc ) of firms.

πi,Def > πi,N cc > πi,N ash 2. Long information Lag.

In the Cooperation case: Presented discounted value:

1 1
P DV = piiN cc +δπiN cc +δ 2 πiN cc +. . . = 1−δ πi,N cc P DVN CC = 1−δ πi,N CC

4
 πsm
P DVDEF = pii,DEF + δ.πi,DEF + δ 2 .πi,N ash + πi,N CC = 2
δ2
δ 3 . . . = (1 + δ)πi,DEF + 1−δ πi,N ash
Present discounted value if I always collude would
The longer the lag, the harder the sustainability be:
of equilibrium. P∞ t 1 π1m m
1 π2
t=0 δ [ 2 2 + 2 2 ]
P DVN CC ≥ P DVDEF π1m +π2m
r P DVN CC = 4(1−δ)
q
πi,DEF −πi,N CC 1 1
δ≥ πi,DEF −πi,N ASH = 2 > 2 The cost of defection δ
π1m +π2m
4(1−δ)

Demand can be high or low in period t. Gains of defection undercutting.

1
P r(highDemand) = 2 If I defect I make the full monopoly profit, so my
πm πm
gain from defection are πsm − 2s = 2s
Demand is iid.
Equlibrium would be sustainable:
D2 (p) > D1 (p)
pim π1m +π2m
2
s
≤δ 4(1−δ)
s = 1, 2
π1m pim
2 < 2
2

There would be three phases in each stage game.

At the beggining of each stage demand is observed. π2m π1m +π2m
⇒ 2 ≤δ 4(1−δ)
Then given that observation price is set. In the third
period profits are realized. At the end of the period 2π2m
δ>δ= 3.π2m +π1m
profit are realized and actions are observed.
1 2
2 <δ< 3
What do we want to find? We are interested in
(p1 , p2 ) price in low demand state and high demand What happens if δ < δ ?
state.
Then π1m , π2m would not be sustainable
1. Both firms set price ps , s = 1, 2, when demand
is s. This collusive agreement would not be sustainable
since the condition does not hold, but this does
2. Sustainable in equilibrium. not mean that there would not ever be collusive
equlibrium.
3. PDV may not be pareto-dominated by the
other equilibrium payoffs. We have to find p1 , p2 such that the firms profit
are maximized:
What is the price that we want?
We have to find collusive prices that maximize the
Each of this two ps will maximize the profit. firm’s profit.

1 π2 (p2 )
ps ≡ argmaxp p.Ds (p) max[ 12 .π1 (p1 )/2 + 2 2 ]/(1 − δ)

5
 What are the prices that maximize each firms qi = qj (impose symmetry)
profit, subject to not deviation?
1−c
qc = 2+θ(n−1)
π1 (p1 )
S.t 2 ≤ δ[ 12 π1 (p
2
1)
+ 1 π2 (p2 )
2 2 ]/(1 − δ) (Not
1−c
Binding) πc = ( 2+θ(n−1) )2 ⇒ πi,N ASH

π2 (p2 )
2 ≤ δ[ 21 π1 (p
2
1)
+ 1 π2 (p2 )
2 2 ]/(1 − δ) (Binding) πiN CC :
P
max[π1 (p1 ) + π2 (p2 )] (p(.) − c)qi + i6=j (pj (.) − c)qj

So that: π1 (p1 ) ≤ k.π2 (p2 ) FOC ⇒ Ri (qj )

π2 (p2 ) ≤ k.π1 (p1 ) qi = qj = qN CC

1−c
k= δ
≥1 qN CC = 2[1+θ(n−1)]
2−3δ
1−c
πN CC = ( 2[1+θ(n−1)] )2
p1 = p∗1
maxqi [(1 − qi − θ(n − 1).qN CC − c).qi ]
Set p02 < pm 0 m
2 so that π2 (p2 ) < π1 (p1 )

In Bertrand we said best response is  below the

The last thing we will do today is look at the price
collusive price. In cournot I need to look at the best
versus quantity:
response of others.
So far we have looked at the Bertrand. Now we What is my defection strategy? qDEF
will look at the Bertrand when the product is differ-
entiated. The products are not perfect substitute. qDEF = 1−c−θ(n−1)qN CC
2

Now our profit from Nash equilibrium would be πDEF = [ 2+(n−1)θ 1−c 2
1+(n−1)θ . 4 ]
different.
1
δ= 1+r
We will compare the delta we found from Bertrand
2
and Cournot, and show that collusion will be easier 1
≥ πiDEF −πiN CC
= 14 [ (2+(n−1)θ)
r πiN CC −πiN ASH 1+(n−1)θ ]
to sustain when it is quantity competition.
P For price competition, we must have:
pi = 1 − qi − θ j+i qj 2
1 1 1 (2+(n−3)θ)
4 ≥ 4 1−θ 1+(n−2)θ
Symetric game. Assume n firms. Take first order
condition. 1+(n−2)θ 0 1+(n−1)θ
4(1 − θ) [2+(n−3)θ] 2 < r < 4 (2+(n−1)θ)2

You can assume different level of substitutability It satisfied for quantity case, but not satisfied for
between different products, but once you go more price.
than two firm it will become intractable.

F OCqi = Ri (qj )

6
 IO @ UTD: Ninth session
Meisam Hejazinia

03/19/2013

Today we will talk about product differentiation.
Chapter 3.6 − 4.2. It is two stage game, first price
contract or quantity, and then they compete over the
type of contract they have selected. Linear demand
function with differentiated product.
Inverse dmeand function would be p1 =
a − b(q1 + θq2 ) a is shift parameter, and b is
slope. If θ is equal to zero then independent demand,
if θ ≡ 1 then we will have perfect susbstitute, and if
θ < 0 shows complement.

We refer to ancient paper that discussed demand p2 = a − b(q2 + θ.q1 )

function by different product.
Inverse linear demand function.
We start with the representative model that is
(1−θ)a−p1 +θ.p2
aggregate model, although it has some product q1 = b(1−θ 2 )
differentiation.
q2 = (1−θ)a−p 2 +θ.p1
b(1−θ 2 )
Some degree of off differentiation, and how the Demand function with respect to other quantity
market works. The product differentiation is part of the slope is positive, with respect to mine it is
the firm choices, and the other half is it is not part negative.
of choice.
Profits are:
We have some degree of market power, but it
does not seem to be that firms use market power π1 = q1 (a − b(q1 + θ.q2 ) − c)
excessively. What type of decision the firm is used
is important. If instead of looking at the perfectly π2 = q2 (a − b(q2 + θ.q1 ) − c)
differentiated product to imperfect substitute the
conclusion of bertrand paradox will collapse, and First order condition with respect to good one:
we will see similar pattern. The price would be
above the marginal cost in cournot. With imperfect 2q1 + θ.q2 − a−cb
substitute we will see that two models are close.
First order with respect to second good is:
Representative model of product differentiation:
a−c
θ.q1 + 2.q2 = b
Bowley 1924, singh-Viven (1984)
This is good with strategic substitute.
Consumer utility would be: U (q1 , q2 ) =
b[q 2 +2θq q +q 2 ]
a(q1 , q2 ) − 1 21 2 2 The game is symmetric:

1
 1 a−c
q1 = q2 = q = 2+c . b As n increases profit for firm will also go down.
We same the same qualitative result as a result.
1
p1 = p2 = p = c + 2+θ (a − c)
Qualitatively we get the same prediction from this
If we have n firms we can do the same thing, we model.
get the FOC for each, and then we impose symme-
try, saying that qi = qj = q, then the result would be: Ri0 = ππiji ””

1 a−c
qc = 2+(n−1)θ . b n=2:p%θ

1
pc = c + 2+(n−1)θ .(a − c) n>5:p&θ

The less product differentiation the lower the In cournot model higher theta gets lower price, and
price, means less product differentiation we will have lower theta givs higher price, but here in bertrand
more competition. Higher product differentiation more differentiation gives lower price at the n = 2.
means I will have more markup pricing, and I will
have more monopoly in the market. Here the only decision variable were quantity or
price, in the next stage we want to give decision
Qualitatively we will have the same outcome here. variable of product differentiation.
We talked about cournot to this point, which was
quantity competition. Now we will discuss Bertrand We will look at the horizontal product differentia-
about the price competition. tion and vertical product differentiation.

π1 = (p1 − c).q1 (p1 , p2 ) The goods are differentiated, since consumers have
different reservation value, and you add another
π2 = (p2 − c).q2 (p1 , p2 ) level of complexity, and then the argument could be
if some consumers have higher value for some, and
We take First order condition with respect to price: other lower for others, probably it is vertical dif-
ferentation. As a result horizontal means consumers
FOC1: (1−θ)a+c+θ.p
2
2
= p1 have same reservation value for all the goods. Some
consumers have higher value for some and some
FOC2: (1−θ)a+c+θ.p
2
1
= p2 lower, it means it would be same quality. In vertical
form it would be two dimensional, due to consumer
This is reaction function since form is p1 = R1 (p2 ) income we will have different form.

The goods are substitute but prices are strategic horizontal differentiation
complements for this.
Consumers are different in their preferences.
1+θ
p1 = p2 = c + 2−θ (a − c)
Spacial approach: taste adjustment cost comes
1+θ
pb = c + 2+(n−3)θ (a − c) from spacial adjustment cost. The good that one
consumer supplies is within the same category, but
Once the firm choose prices, if goods are differ- is different from what I am looking form. If it is not
entiated, firms can walk home with strictly positive equal to my taste, I have to adjust a little bit. WE
prices. can use hotteling model for both of these forms.

2
 You have two streets, and one end of town is 0,
and the other on 1. All consumers live in the main
street. One firm is at one end of main street, and
hte other at the other end. I am consumer X, and it
is my location. I can not decide whether I want to
purchase from A or B. Both firms produce different
brand characteristic. The only differentiation is the
distance, and products are the same.
Supplying good at two different location differen- From B: PB + c.y
tiates goods. Not only you pay PA or PB , but you
also incure transportation cost PA + t(x).
The consumer will incure transportation cost t(x)
to go to firm A or t(y) to go to firm B. We assume
linear transportation costs. If consumer from A
wants to by from the transportation would be c.x
and from B → c.y.
Total costs:
From A: PA + c.x
a + b + x + y = l, also from (∗).

transportation cost could be linear or quandratic. y = l − a − b − x.

The distance would be for good t included so
PB + t(1 − x). By locating at different points The indifferent consumer x̃(PA , PB ) :
firms differentiate themselvs. This was special PA + c.x̃ − pB + c.(l − a − b − x̃)
differentiation. Firm A and Firm B is not located
geographically, but they locate themslves in term The indifferent consumer is: x̃ = 21 [l − a − b +
of preference. You like Vanila ice-cream, and then PB −PA ]
c
there is no adjustment. If you like choclate then you
have to adjust your taste. This was the distance
As a result demand for firm A would be
interpretation. Why it is maximum product differ-
qA = a + x̃ = 21 (l + a − b + PB −Pc
A
)
entiated when K-mart and wallmart are at different
side of town but mcdonald and berger king are close
qA (pA , pB ) = a + X̃(pA , pB )
to each other.
qB = l − qA (pA , pB )
Cosumers: uniformly distributed on the line with
length l.
Two firms: A, B. Firm A locates at a > 0, unit Full market coverage
from the left end of the line, and then there is firm
B locates at b > 0 which is at the right end of the line. piA = pA .qA (pA , pB )

Eventually these two would be decision variable piB = pB .qB (pA , pB )

for the firm.
mill price means the firm offers the price and you
a + b ≤ l (∗). get the good, and alternative is delivered price which
means the firm encounters the transportation cost.
Firm A will charge mill price PA . The alternative
would be delivered price. Firm A will set mill price We basically get the demand function, and we have
PA , and firm B will set the mill price PB . We have profit function, and then we have price competition
a consumer which is x unit away from firm A, and for bertrand.
y unit away from firm B. If line has unit length,
then consumer is defined as x, and then y would be PA∗ = (b + a−b
3 ).c
∗ b−a
1 − x = y. pB = (b + 3 ).c

3
 ∗ 1 ∗2 c(l−a−b)
πA = 2c .PA PB∗ = 3 (3l + b − a)

pi∗B = 1 ∗2
2c .pB Profits at equilibrium:

∗
pi∗A (a, b) = 1
2c [(b + a−b
3 ).c]
2
πA = c
18 (l − a − b)(3l + a − b)2
∗
∂πA
pi∗B (a, b) = 1
2c [(b + b−a
3 ).c]
2
∂Q
c
= − 18 (l + 3a + b)(3l + a − b) < 0

We have restriction that a + b ≤ l, since the role if a ≥ 0 then a∗ = 0

will flip.
if a ∈ R then a∗ < 0 (to the left of zero)
We will have minimal differentiation.
if b > 0 then b∗ = 0
a+b=l
if b ∈ R then b∗ < 0 (to the the right of l)
A and B will sit over eachother, means minimal
differentiation. Use hotelling mode only when you are really
interested in location, else use the normal aggregate
Gabtzweicz, thisse (1979) showed that for large form.
a and b the pure nash equilibrium in second stage
pricing game may not exist. This means we will not With quadratic transportation cost, we will have
have reduced form for profit. We need to have that maximum differentiation.
so that the firm know where to locate, but if it does
not exist then the whole will fall apart. This would Take the book and go through the examples of
be discontinuity in profit function. Pure strategy hotelling of the problem. Slightly different different,
does not exist, although mixed strategy may exist. and make sure you understand them.

As a result linear transportation cost does not
work, so they showed that if we assume quandratic
transportation cost everything will work. The costs
are quandratic in the distance. As a result we away from the competitor. We sell the same good
assume c.x2 and c.y 2 for the firms.
From A: PA + c.x2
From B: PB + c.y 2
x̃ : PA + c.x̃2 − PB + c(l − a − b − x̃)2
pB −pA
x̃ = 21 [l − a − b + c(b−a−b) ]
We take a and b is given and we derieve equilib- equilibrium.
rium prices PA and PB .
PA∗ = c(l−a−b)
3 (3b + a − b)
There are two effect:
1. strategic effect: Firm want to locate furthure
to the same market, we will have fierece price
competition. Firm want to locate as far away from
the other firm. This alleviates market competition.
2. Market share effect: firm want to locate near
the center of the market since you will have more
consumers, since there would be more demand.
interaction of these two effect will define the
These were equilibrium results. Socially optimal
location will be a∗ = 4l b∗ = 3.l
4 that minimizes the
transportation cost. This is under quandratic cost.

4
 The maximal differentiation is not derived. As
Location that not only reduces the transportation long as entry has positive profit a firm will want to
cost, but also minimizes the prices, will give us the decide to enter.
socially optimal location.
In the free entry equilibrium we will have zero
On linear case, we have multiple equilibria, but profit.
the dominant one is when a = b, mean they sit on
the top of each other, but on that case the price is We derive Nash equlibrium for pricing game, and
not defined. then at the second stage we drive nash equilibrium
in the entry stage. This is backward induction.
If full coverage they will always have same profit.
When they sit on top of each other they equally So first: Pricing:
compete over all the market.
Everybody price at p. We assumed symmetry
Salop 1979 is about circular city. The problem pj = p. I want to know my optimal price, given
with the hotelling is when you sit at one side, there everybody else selects equilibrium price.
would be no consumer and no competitor. Salope
solved the problem with circular city. The consumers The share of the market between two firms would
are all located on the circular line. Each firm will be n1 .
have competitor on either side, the circumfrence is
equal to one. The question is what is the price pi . The indiffer-
ent consumer will sit in x̃ ∈ (0, n1 )
The question of interest is not location of the firm,
but how many firm we will have in this market. indifferent consumer will pay either pay
hotelling allows for any given number of firms. pi + tx̃ = p + t( n1 − x̃)
t
p−pi + n
It is symetric game, and consumers buy one unit x̃ = 2t
of the firm. There is linear transportation cost. t
p−pi + n
Firms will have fixed cost of entry. Fixed cost would Di (pi , p) = 2.x̃ = t
be f , and firm will have marginal cost of product
which would be c. FOC:

πi = (pi − c).Di − f pi = Di (p)

The Salope game is two stage game. Firms pi = p

simultaneously decide to enter the market, and once
t
they decided then they locate equally distanced on p=c+ n
the circle.
The price so should be above marginal cost.
Stage 1: firm decide to enter simultaneously What is number of firms? Now we look at the
Entry stage:
Stage 2: Pricing stage, mean competition
1
Di (pi , p) = n
Maximaly differentiated is when they are euqlly
distanced on the circle. πi = (p − c). n1 − f

5
 t
π = (c − n − c) n1 − f be utility, and if not buy:

π=0 U = 0 otherwise (q = 0)
q
n∗ = ft θ ∼ U [ θ, θ + 1]

In the entry stage all firms will make zero profit, θ ≤ 1, θ̄ = θ + 1

and we determine the number of firms that enter the
market. 2 firms

If transportation cost is zero nobody will enter. If quality Si , Ss > S1

cost of entery is low many firms enter.
√ unit cost of production = c
p∗ = c + t.f
Assumption 1: θ̄ ≥ 2 θ
This is price of free entry equilibrium. Assumption 2: Market is covered: means each
consumers buys at least one of the goods.
Firm still have market power, with strictly positive
transportation cost, firm will have at least some ∆S = S2 − S1
market power.
There is unit demand.
The more firm we will have higher total cost,
although the less would be transportation cost. The high θ will buy high quality
solution is at the end of the book for this socially
optimal and calculate for yourself. low θ will buy the low quality

Vertical differentiation This is like the price differentiation.

Firm do not choose their horizontal location, and There is indifferent consumer.
adjustment, but firm decide the quality of their offer.
θ.si − p1 = θ.s2 − p2 .
Tirol 75
p2 −p1
θ= ∆s
Shaked sutton (1982)
All consumers that sit in θ ∈ [ θ, θ̃] → s1
We have oligopolistic competition and quality
differentiation. We have consumers and consumers θ ∈ [θ̃, θ̄] → buy s2
have following utilities.
p2 −p1
D1 (p1 , p2 ) = ∆S − θ
U = θ.s − p for one unit of the good. For quality
p2 −p1
s and price p. D2 (p1 , p2 ) = θ̄ − ∆S

With vertical differentiation, consumer are differ- πi (pi , pj ) = (pi − c).Di (pi , pj )
ent in marginal valuation of the good. They attach
higher value to quality, and lower value to additional 1. maxpi πi
quality. If the consumer buys something that would

6
 2. pi = Ri (pj )

3. p1 = c + θ̄−2 θ
3 .∆S
p2 = C + 2θ̄− θ
3 ∆S > p

(θ̄−2. θ).∆S
pi1 (s1 , s2 ) = q

(2.θ̄− θ).∆S
pi2 (s1 , s2 ) = q

s ∈ [ s, s̄]

For firm 1:

θ̄−2 θ
pi1 (s1 , s2 ) = q (s2 − s1 )

1. maxs1 π1 (s1 , s2 )

2. s1 = R1 (s2 )

Suppose s1 < s2

s∗1 = s, s2 = s̄

s1 > s2

s∗1 = s̄, s∗2 = s

high quality firm makes higher profit.

Monopolistic competition is not location decision,

but is one where firm supply differentiated product.
They are able to engage monopolistic pricing, but
they engage in markup pricing. elements:
1. markup pricing

2. zero profit

7
 IO @ UTD: Ninth session
Meisam Hejazinia

03/19/2013

1 continue on product differen- their utility, given price.

tiation 3. Free entry. Means competition aspect of the
We talked about horizontal differentiation. We model.
could have companies competed in three dimension
in cube. Three parameters θ and α, and you can In long run we will have zero profit condition.
extend them. You can have multiple number of firms. Each firm makes zero profit. Price would be above
Horizontal and vertical differentiation was discussed, marginal cost. At the end each firm makes zero
you can have combination. Consumer differentiation profit. We have to assume entry cost, otherwised
from horizontal, and vertical differentiation in term infinite firm. Questions: What are entry costs? How
of quality input of firms. many firms in equilibrium?

Fixit, spense, stiglitz have couple of papers. They Unlike monopoly that entry was impossible, here
come with idea of monopolistic competition. Firms is possible, but with cost.
compete with all firms, and not just their neighbors.
The one we usually solve comes with two firms. Assume inverse demand function: pi = A(n)qiβ−1
Two products are substitute, but they are imperfect A(n) < 0, and 0 < β < 1. So there is some constant
substitutes. In hotelling you just compete with the A(n) times quantity.
firm that is next to you. Circle model also does not
let us to have competition in cloud, and competition Production cost c.qi + f .
with neighbors would only be possible.
1. interested in price and quantity?
Tirol has different approach. We have representa-
tive consumer that has utility over different brands. 2. At the end we want to say something about the
We have utility function over n brands. Each firm number of firms in equilibrium?
i = 2, . . . , n produces his hown brand of product.
Monopolistic competitive equilibrium comes in 3 We go from inverse demand function to direct
ways: demand function:
1
1. Goods are imperfect substitutes. Try to
pi
qi = ( A(n) )− 1−β
combine monopoly into competitive market. Firm
1
behave as their are monopolistic as their own brand. Price elasticity in this example would be − 1−β <0
So first is to set monopolistic price.
We know what the markups will be.
2. Consumer choose whichever product maximizes

1
 pm −c 1 1−β
Learner index is going to pm = |c| = 1 A(n) % c → n & c
1
pi
max[(pi − c)qi (pi )] = maxpi [(pi − c)( A(n) ) 1−β ] A(n) % f → n & f
c
pm
i = β Two effect that counter each other for c since
increases the price, at the same time the number of
Means there is some market power, since the price firms decreases, so these would be competing factors.
is not equal to marginal cost.
The fix cost will not affect the price.
Zero profit condition for entry:

Each firm has to make zero profit:

pm m m
i .qi − c.qi − f = 0 2 entry, accomodation and exit
As long as strictly positive profit, firms will enter. condition
As the number of firm increases n will increase, and
One firm that has entered due to the technology
this will affect inverse demand function.
advantage or chance, and this incumbant faces entry.
m m m What happens to some of the previous structures
pi .qi = c.qi + f
that we considered.
We want to solve for quantity qim that each firm
will sell in zero profit condition equilibrium. We look at three types of interaction:

qim = β f 1. Accomodation: accepts the fact that entrant

1−β . c
arrives, and tries to make more. Adjust behavior
If fix costs increase the quantity of each firm in after entry.
the market will increase. Price is markup pricing,
and fixed cost does not depend on it. The higher the 2. Blockade: Entry will not happen. The entrant
fixed cost, the lower your profit after your revenue. has the production cost that incumbant does not
Small number of firms will survive. Then we will have to change his behavior.
have higher value of firms in the industry.
3. Deter the entry: Change the behavior so that
m
q %f it deters the entry.
i
qim & f
Bormel(1982), contestable market
What can we say about n?
Contestable market
We have inverse demand function we have:
pi = A(n).qiβ−1 n firms

c
β = A(n).qiβ−1 Perfect substitutes

β
c
β = A(n)[ 1−β . fc ]β−1 Production cost c(q), c(0) = 0

1 β 1−β β 1−β
A(n) = β ( 1−β ) .c .f fixed cost if q > 0

2
 m incumbands
n − m potential entrants
What is the industry configuration which is
feasable?
qi , . . . , qm that are produced, each with price p
Pn
Feasible if market clears i=1 qi = D(p)
If the firm makes no profit, it will exit the market.
Firms make non negative provit
Each firms revenue not smaller than cost of
production p.qi ≥ c(qi )∀i = 1, . . . , m
Second sustainability: If they are strictly positive
then firm will enter, then the industry is not sus-
tainable. It is sustainable mean nobody out would b
e better off.
q e ≤ p, q e ≤ D(p), such that pe .q e > c(q e ).
If cross subsidization will jeopardize the sustain-
ablity condition.
A contestable market is the one that entry comes
with zero cost.
As soon as the revenue is slightly higher than cost industry {q c , pc }
of production, an entrant could enter the market.
They will exit the market as the soon as profit
becomes negative.
Under single product case, where production cost
is c(q) = f + c.q π̃m ≡ maxq [(p(q) − c)q] − f
unique industry configuration:
{pc , q c } such that AC = p
identical firms
Contestable market is one that the intersection of Natural monopoly characterized with high fixed
demand curve and the average cost curve is where cost.
the average cost would be decreasing.
If p < AC everybody will make negative profit,
violation of feasibility
If p > AC then we will see entrants, violation of
sustainability
The only way that the firm would want to enter,
is to charge the price less than market, means less
than average cost, so get negative profit.
There is only one firm that is able to survive in
this market. It is technological efficiency.
We have seen lot of things but not average cost
pricing. It is socially efficient given that there would
be no subsidy. Nobody wants to produce could be
alternative, since we will still have consumer surplus.
Allocation here is constraint efficient, and not overall
efficiency.
Marginal cost price will not work, since the
products are not substitute.
Excercise 8.1
No allocation will exist if D(pc ) would be larger
than minimum efficient scale
Demand is lower than minimum efficient scale:
mean the quantity that minimizes firms average cost.
M ES ≡ argminq AC(q)
Price is equal to average cost. P = AC
Demand given this price would be higher than
minimum efficient scale. Such industry configuration
would not be sustainable.
Entrant could charge pe = pc −  q e = M ES
3
 t + dt is the last instance in time
WAR OF ATTRITION
Once the second firm leaves the remain firm will be If firm i stays then firm j will exit with probability
monopoly. We will have mixed strategy that each
firm has strictly positive probability of winning. Both firms have to leave with strictly positive
probability.
Tirol p.311
j drops out with probability pr
It is natural monopoly, since there is no enough
room for the firms to survive. It will also drop out from now t to t + dt.

We assume time is continuous. Once it dropped out, I will make monopoly profit.

interest rate r. π̃ m −f
r

Two firms R∞
(π − f )ert dt = π̃ m −f
p r

Cost of productions are c(f ) = f + c.q if q > 0,

If I am staying out (condition for firm i)
and c(0) = 0 m
0 = f.dt + prj . π r−f
If Bertrand p = c profit −f f.r
pij = π̃ m −f .dt
If one firm leaves the market monpoly condition m
−f
will be built pm FOC (j) 0 = −f.dt + prj p̃i r

π̃ m − f We have random length of time which is tech-

nologically inefficient. Means we have encountered
Both firms at the market at the t = 0 fixed cost once.

If the probability was strictly positive they will
not start again.
At any time t a firm can stay or leave. The firm
should be indifferent between staying and leaving.
At each time you need to be indifferent between
dropping out, or staying in.
If t were drop out the profit would be equal to zero.
If I stay in then I would make duopoly losses of f.dt
dt is the time that elapses from now to the next.
f.r
prj = π̃ m −f .dt
First price is competitive, and once one firm leaves
the price will go to the monopoly price.
Excercise 8.2 in the book, look at it. Look at the
contestable market, and then compare the result
with result of contestable market. The welfare of the
contestable market would be higher.
First mover, and Leader follower model
Stackelberg. See some capacity or some costs to
commit certain actions.

t is now You can affect entry by deterring entry, or acco-

modate entry and make best of it.

4
 cost that can deter entry.
Interpretation of capital. How much capital is
produced to make best out of entry. Entry Deterrence: f > 0 π 2 = k2S (1 − k1S − k2S ) − f

1
How the idea is different from quantity interpreta- 16 −f
tion. First I decide how much quantity to produce
1
then why don’t I adjust quantity. The leader can Suppose f is close to 16 , is k1S optimal for firm 1?
again adjust capacity. The capacity can be adjusted,
and is not fixed. If firm one increases its capacity so much, then
that could be optimal. If I increase capacity then
Two firms: the other firm might decide to not enter.

Firm1: level of capital K1 , fixed The quantity is not always optimal. We have to
find K1b that discourages entry by firm 2.
Firm 2: level of capital K2 .
Given this k1b firm two wants to maximzie
0
π (k1 , k2 ) = k1 (1 − k1 − k2 ) maxk2 K2 (1 − k1b − k2 ) − f
1−k1b
π 2 (k1 , k2 ) = k2 (1 − k1 − k2 ) R2 = 2

1−k1b 1−k1b
The higher quantity firm 2 produces smaller would π2 = 2 (1 − k1b − 2 ) −f
be firm 1:
1−k1b 2
1. πji < 0 pi2 = ( 2 ) −f =0−

i
√ 1
2. πij <0 k1b = 1 − 2 f > 2 = K1S

K1s = 12 , k2s = 14 , π 1 = 1
8 π2 = 1
16 We start the game is from stage 1 that incumband
produces an amount, and then on the second stage
k1N = 31 , k2N = 13 , π 1 = π 2 = 1
9 the entrant decides to enter.
√ p
Reaction function: π 0 = (1 − 2 f )(1 − 1 + 2. (f )) ≥ 1
8
√
1−k2 1−k2s 3 3−2 2 1
R1 = 2 = 2 = 8 6= k1S True as long as f > 32 ∝ 0.005 = 16

Entry cost or set up cost is strictly positive, since In section 8.2. stigletz approach to this, which
then we can look at the entry blockade. is quantity rather than capacity would be revised.
Entry deterance as public good and what are other
If firm one knows he can not deter the entry, it re- forms of capital.
duces the scale of firm 2 production, and maximizes
the production. Stackeleberg: Some times we interpret it as
over investment. I produce too much to deter
Entry accomodition, then firm one will produce entry. It is always this type of over investment.
k1S = 12 We have to model it in the form of two firms a
leader and the follower. We have the technology
Entry happens no matter what the fixed cost leader K1 , and then both firms compete in quantities.
would be. We want to find the critical point of fix

5
 1 2
1. investment K1 Sign( ∂pi
∂X2
dX2
dk1
∂π dX1
= Sign( ∂X 1 dK1
).SignR2

2. on second stage quantities is selected simulta- Business strategies, and top dog, puppy dog, lean
neously. and hungry cat, and fat cad explanation in Tirol
book.
m m
Profit for firm one would be π (k1 , X (K1 )) on
case of No Entry Bundeling and entery deterant

Entry: Reduce residual demand, and limite competitor

business opportunity, and with fix cost means you
π 1 (K1 , X1 , X2 ) can deter entry by over investing.

π 2 (K1 , X1 , X2 ) Two papers bundling as an entry barrier, and at

the end we look at bundling as entry deterrant.
X1∗ (K1 ), X2∗ (K1 )
Bundling in chapter 8 example 8 in Tirol
We have Deterence if firm one chooses K1 such
that π 2 (K1 , X1∗ (K1 ), X2∗ (K1 )) < 0 Rulebuff (2004)
Accomodition if π 2 (K1 , X1∗ (K1 ), X2∗ (K2 )) > 0
Two goods A, B
K1 such that π 2
(K1 , X1∗ (K1 ), X2∗ (K1 )) =0
Consumers with unit demand for good A and
∂π 2 ∂π 2 ∂π 2 dX1∗ ∂π 2 dX2∗ good B (for both)
dK1 = + + .
∂K ∂X ∂K ∂X dK
| {z 1} | 1{z 1} | 2{z 1}
The valuation of the consumers are αA , αB ∈ [0, 1],
The first term is demand effect, the second one independently uniformly disributed
would be strategic effect, and the third one would be
equal to zero based on envelop theory. Suppose consumers always want to purchase full
coverage.
dπ 2
Inverted firm 1: ”Taugh” if dK 1
<0
dπ 2 Incumbent produces A and B. Could have been
”Soft” if dK1 > 0
market leader. M C = 0. Entrant enters the market.
Accomodation: The challanger enters either market ’A’ or market ’B’.

π 1 (K1 , X1∗ (K1 ), X2k (K1 )) Entrant will produce either A or B.

dπ 1 ∂π 1 ∂pi1 ∂X1 ∂π 1 ∂X1 Entry cost is common knowledge.

dk1 = + +
∂K ∂X dK . K
| {z 1} | 1{z 1} | {z . 1 }
Timing:
The first one is direct effect, the second one would
be equal to zero, and the third one would be equal First incumbant sets prices (pure bundle, or two
to strategic effect. different cost for them)

dX2
dK1 = ( dX dX1 1 dX1
dX1 )( dK1 ) = R2 . dK1
2
The prices would be fixed for the rest of the game.

6
 Second, the entrant decides to enter or not. Z(PA , PB ) = 2f − , so that you limit entry.
If your profit would be higher you have given the
Three, the entrant sets the prices. entrant the chance to enter.

We look at different situations.

(i) First we look at independent pricing.

(ii) pure bundling

When incumbent alone. Independent prices:

In no entry:

1
PA = PB = 2

1 1
πA = 4 πB = 4

1
πA + πB = 2

Entry

PAe = PA − 

PBe = PB − 

1
πe = 4

1
π inc = 4

Limit pricing

What are optimal prices?

1
PA = 4

13 3
πA = 44 = 16

The reduced decision variable for the incumbent

are the profit that are made in this market.

Set PA and PB such that profit would be equal to

Z(PA , PB ) then the entrant can make half of that
Z(PA ,PB )
2 =f

7
 IO @ UTD: Eleventh session
Meisam Hejazinia

04/02/2013

1 continue on entry, and exit In the uniform pricing case then the entrant needed
to slightly undercut the entrant, say p∗e = pI − .
In pure bundling case we have consumers who are Suppose we have p̄ = 1. Who will buy from entrant?
uniformly distributed b/w zero and one, and two
dimensions independent (in hoteling). Whose surplus is αA + αB − 1 ≥ αB − 21 consumer
will buy from entrant. This holds true for αA ≥ 12 .
p̄ = 1. Consumers are willing to buy bundle if
their valuation is at least one. For β > 12 will buy from incumbent. Buy bundling
incumbent could keep the first quadratic, and this
The says down triangle will not buy, and upper deters some entry. Now the entrant profit is only
triangle will buy in hotelling square. half of the profit we had in the previous case.
Incumbent has to bundle two goods, and this profits
In order to figure out what is optimal price, we on two quadratic. The axis of this system of corrdi-
need to figure out what is the demand. nate that make the quadratic is center of line y = −x.

If price is larger than one then we will have the Suppose we have price p̄ as a line that inter-
intersection of the new line which is higher than sects both αA and αB axis. When entrants entrs
y = −x will intersect the square of hottelling, and with the price pe for αB , all the area under this line
would be in the form of x = 2 − p̄. which is greater than pe for αB will buy from entrant.

Demand if p̄ > 1 would be in the form of αA + αB − p̄ ≥ αB − pe , so αA ≥ p̄ − pe

2 2
D(p̄) = (2−2p̄) , and if p̄ ≤ 1 then D(p̄) = 1 − p̄2 . This means the demand for entrant would be
e
In this√ case we calculate maxp̄ [p̄.D(p̄)], so D (pe ) = (1 − pe )(p̄ − pe ), since we have pure
∗
p̄ = 23 < 1. bundling for the second, so incumbent will not
satisfy the need for p̄ − pe .
Uniform pricing will result in pA = pB = 12 . √
1−p̄+p̄2
Proposition 1 is p∗e = 1+3 p̄ − 3
We will have pure bundling effect. Bundling
discount will be seen here. This will deter incumbent Mixed bundling case as an extension would be in
of entry, but incumbent still allows entry. the form of pA , pB , p̄ mean different price for each of
the goods, and p̄ for the bundle. Until now we just
Incumbent benefits from bundling here, and if discussed the pure bundling.
there was no such benefit then incumbent would
charge p̄ = 1. Now the common theme is what we can use to
deter the entry.

1
 Etnrant will accept price if ce ≤ T . This is how
p. Aghion- p. Bolton (1987) entrant decides.

downstream buyers v = 1. In order for integrated structure to make the offer

is p(ce ≤ T ) = T
Downstream sellers who produce/ incumbent
cost = 12 . T.P r(ce ≤ T ) + 12 (1 − pr(ce ≤ T )) =
minT [T.T + 12 (1 − T )] ⇒ T ∗ = 14 .
We have supplier who produces with ce ∼ U [0, 1]
(private information). What does this mean for efficiency?

We have three stage game. The entrant will supply product if its cost are less
than T .
at first stage t = 1 the buyer will write the
contract. At second stage an entrant arrives t = 2 Incumbent produces at the cost of 12 .
it observses cost ce , and at third stage t = 3 the
entrant makes price offer to the buyer. The incumbent supplying is inefficient. The
incumbent could produce with the lower cost.
The initial contract will specify {p, p0 }. p is
We compare no contract situation with contract
the main price of purchase, and the price 0 is the
situation.
payment penalty for bridging the contract.
When we have no contract:
The good comes with some quality and some
quantity, and it is called WIDGET.
Incumbent makes take it or leave it offers.
The benchmark is that we have integrated struc- It is like monopoly pricing.
ture: Incumbent/ buyer.
Incumbent could extract full surplus from the
Incumbent wants to minimize the production cost. buyer, because the incumbent will set the price equal
to one.
Trade is better than not, but want to minimize
price. When will we see entry?

Internally cost would be 12 . We assume there would be bertrand condition.

If they outsource the cost would be T , mean the 1

cE > 2 the entrant will not enter.
price to give the other supplier, mean buying from
entrant. As a result the only case is cE ≤ 21 .

The benchmark is hypothetical situation. φ = pr(cE ≤ 12 ) = 12 .

If they were integrated we would have no con- Price of bertrand would be p = 12 .

tract. We bechmark with the first institution, and
it maximizes everything, since it has no purturbation. The profit for the buyer would be:

2
 He knows with probability (1 − φ) there would be Incumbant price problem would be:
entry, surplus would be zero.
maxp0 ,p φ.p0 + (1 − φ)(p − 12 ) so that 1 − p ≥ 14 .
With probability φ we will have entry. As a result:

(1 − φ).0 + φ.(1 − 21 ) = φ. 12 = 14 . This is as a result pricing problem for the incum-

bent. p = 34 and p0 = 21 .
Profit for incumbent would be (1 − φ) 21 = 14 .
What are the expected profit for this contract for
Incumbent has two options: either not offer incumbent?
contract, or offer.
11 31 1 3 5
The contract will look as follows: 24 + 44 = 8 + 16 = 16 > 14 .

p is payment if there is trade. This contract allows for entry. The penalty is less
than price, so there is entry with probability of one
p0 if there is no trade. penalty. quarter. In term of efficiency, entrant should enter.
This contract gives us inefficient entry.
If there is no entry the buyer surplus would be 1−p.
3
How would entry blockade look like? p = 4, and
If there is entry, then the buyer will switch if the
p0 = 43 .
entrant offers surplus of at least 1 − p.
3
At stage 3 the entrant makes price offer. The price p could not be higher than 4, since it
violates the condition of 1 − p ≥ 14 .
if entry: buywer switches if p̃ is at least the surplus
is ≥ 1 − p. In this case buyer and incumbent will form a
coalition that entrant compensates not only the
STages would be the following: buyer, but also the incumbent as well. They find
a way to extract additional rent from the contract.
t = 1 accept p, p0 if surplus ≥ 14 . Contracts here is used as entery deterance. You
basically create a contract as barrier to entry.
t = 3 buyer will accept the price offer if p̃ + p0 ≤ p,
since p is the price of staying with the incumbent.
This is all about entry, exit and incumbant.
p̃ ≤ p − p0 .
Because there is entry with no cost, the incumbent
The entrant so first should make this offer, and has not extracted all he could from the buyer.
the entrant should make strictly non negative profit
mean p ≥ CE → p − p0 ≥ Ce .
If they could sit together and negotiate for collu-
5
sion surplus of 16 would also be shared.
{p, p0 }.

Probability of entry would be φ = p(ce ≤ p − p0 ) Spier, whinston Rano 1995, extended this paper,
φ = p − p0 . and discussed investment part of it.

3
2 pricing under asymetric in- π i (pi , pj ) = (pi − ci )(a − bpi + dpj )
formation Firm i knows its type and depend on type maxi-
Static game with asymetric information. mizes profit.

Firm 2 does not know firm 1 cost, and has to

One player knows something that the other player
maximize profit given what it expects the cost of
does not.
firm 1 would be.
t = 1 there is information.
Bayesian Nash equlibrium should be calculated.
t = 2 there is pricing.
Action set is strategy set, and you either go left or
right.
Differentiated duopoly.
If you think there is type then strategy would be
Two firm setting prices. different.
One firm has incomplete information about the The action set contains two actions, but with
cost of rival. different type combination you will have four actions.
It is symmetric in term of demand D(pi , pj ) = What are the action sets?
a − b.pi + d.pj . Goods are substitute but strategy
complement. We have upward sloping reaction Ai ⊂ R+ means any amount could be the action.
function.
S2 ⊂ R+
Firms are risk neutral, since we are dealing with
expectation we need to form some assumption. Con- S1 ⊂ R+ ∗ R+
stant return to scale production. (CSR). Marginal
cost is going to be constant. Suppose 2:

d ≤ b , and d > 0. p∗2 firm suppose k = L, H.

Firm 2 we have c2 is common knowledge. πi∗ (pk1 , p∗2 ) = (pk1 − ck1 )(a − bpk1 + dp∗2 )

Firm 1 would have c1 ∈ {cL H

1 , c1 }. j = 2, i = 1

CqL < C1H +dp∗ k

2 +b.c1
pk1 = 2b for k ∈ {L, H}

pr(c1 = cL
1) = x cL H L H
1 < c1 , we know that p1 < p1

pr(c1 = cH
1 )=1−x pe1 = x.pL H
1 + (1 − x)p1 .

ce1 := x.cL H
1 +(1−x)c1 , would be expected marginal Reaction function is linear in the cost.
cost.
We can take expectation of the price, or take the
Ex-post: would be profit function of the firm: reaction function, since it is linear in the cost we can

4
 a+d.p∗ e
2 +b.c1
use pe1 = 2b
periods. On the first period they will signal type,
and then set the actual value.
Mean the expectation would take this direct form.
There are two cases:
ce1 is expected marginal cost.
1. If you can not deter entry. You would like to
Firm 2 needs to know the reaction function of signal high cost. If you signal high cost, the other
firm 1. We will work with expected reaction function. firm will set the high cost optimally. If you happen
to be someone of low type you would be better off.
We will have to look for the reaction function of
each, and how the firms will react to that. 2. If entry can be detered you will signal low type,
since you want the firm to behave less agressively.
We are looking for the price the firm is going to
charge. The other firm does not know your actual cost, by
setting the price you will signal the the cost.
The full expected payoff would be Ec1 π2 = x[(px −
c2 )(a−bpx +dpL H
1 )]+(1−x)[(p2 −c2 )(a−bp2 +d.p1 )] =
Two firms both in the market with two periods.
e
(p2 − c2 )(a − bp2 + d.p1 )
What would be the expected price of firm 1 and Marginal cost would be zero.
then react to it would be done by firm 2.
Demand would be qi = a − pi + pj . a is unknown,
c a+dpc1 +bc2
p2 p1 = 2b
but distributed on real line. Does not really matter
what it is.
a p +bce
pc1 (p2 ) = d 22b 1
ae is expected value (mean).
∗∗ 3+ce1 +2c2
p2 = 3
Firm one will maxpi Ea [(a − pi + pj ).pi ] −
k
pk1 =
3+2c1 +c2
would be for k ∈ {L, H} maxpi (ae − pi + pj ).pj
3

ae +pj
There is no signaling power here. This is static pi = 2
game. You have interest to signal, but here we do
not allow that. Equilibrium would be p+ +
i = pj = a
e

Let say the type could be verified. The higher the Means what we expect the demand to be.
price firm two charges the higher the profit for firm
one. Suppose we have two period model, where a does
not change and is same for both period.
Low type will not send the message that I am low
type since this will reduce its profit. Highest type They only see the realization of their demand qi ,
wants to reveal the type since the second firm will they know their own price, but they do not know
then set the price higher and it will capture more of they rival price.
the market.
We can not allow firms to observe price, since we
We restrict to what does pricing say about firms want to keep that uncertainty.
price structure. We have price competition over two

5
 a is common for both firms. Expectation since firm A does not have any
knowledge.
Suppose firms price symmetrically.
dπi3 ∂πiB ∂pB ∂πiB
dA
= ∂pB
. 2.piA + ∂πiA
i i i
pA
i = α and it is symmetric. ∂πiB ∂ã
∂πiA
= πiB ∂p B
a = pi
i
If that is the case DiA = a − α + α = a, means on
the equilibrium we can infer what a is. Envelop theorem, or from maximization the first
term would be equal to zero
What if the firm deviates?
Ea [a − 2pA B A
i + α + δ.pi (pi )] = 0

i deviates. Then pA
i 6= α. pA
i −α
Ea [a − 2.pA
i + α + δ[a + 2 ] =0
What is that firm j observes? DjA = a − α + pA
i = pA
i −α
ae − 2.pA e
i + α + δ[a + 2 ] =0
ã = a + (pA
1 − α)
Symmetry , both firm set price α pA
i =α
In the symmetric equilibrium it would be correct.
Means deviation would be the same. Now we assume α = ae (1 + δ)
that one firm deviates and the other things what he
sees is correct value.

Firm j assumes complete information, and is on

the equilibrium path.

DiA = a−pA
i +α means firm i knows true value of a.

What would be the second stage?

in this case B on the second period while A first

period: pB A
j = ã(pi )

maxpi (a − pB A B
i + ã(pi )).pi

pA
1 −α
As a result pB
i =a+ 2

In equilibrium j sets pA B A
j = α and pj = ã(pi )

pA
i −α
pB
i =a+ 2

What is the deviation?

πi = Ea [(a − pA i + α).pi
A
+ δ(a − (a +
pa
i −α A B
2 ) + ã(pi )).pi ] = maxpA
i
Ea [(a − pA A
i + α).pi +
B B A A
δ.πi (pi (pi ), pi )]

6
 IO @ UTD: Twelfth session
Meisam Hejazinia

04/09/2013

1 Limit pricing
For simplicity π1k = π1k (pkm )
Last example we went through was one with stochas-
tic demand. After entery: D1k &D2k is firm 1’s and firm 2’s
duopoly profit given type k.
Two firms (incumbent, and entrant).
Assumption:
Incumbent sets price p. Incumbent knows its cost.
D2H > 0 > D2L The entrants dupoly is smaller
c1 ∈ {cL H L H
1 , c1 }, c1 < c1 than zero if it faces low type firm 1. If firm 1 is high
cost type only firm 2, entrant, will have positive
Firm 2 does not know firm ones marginal cost. profit.

Firm 2 knows ck1 after entry. When firm two wants The problem is that at the time of entery firm 2
to enter does not know cost of entering, but as it does not know cost of firm 1, mean low type or high
enters it knows it. type.

We have assymetric of information in duopoly. Firm 1 wants to signal firm 2 that it is low cost
type.
Firm two will learn firm one’s cost.
π1k > D1k
k ∈ {L, H}
Firm 1 wants to signal low type by setting its
Firm 1: price pL
1.

π1k (p1 ) = (p1 − ck1 ).Qm

1 (p1 ) Setting this price pl1 will result in the negative
profit for the first period for firm 1.
Same demand. Firm 1’s monopoly profit will be
in this form. If the incumbent sets the price pL k
1 that is 6= pm
then losses in stage 1 are offset by game in stage 2
Concave with regularity conditioning. with no entery.

pkm monopoly price for type k There is some price that should convince the
entrant that you are in fact low type.
Because cL H L H
1 < c1 then pm < pm

1
 Perfect bayesian equilibrium: price that low type could charge that is not equal to
pL
1
1. Separating equilibria, where we have two
different types with different prices. π1L (pH
m)

2. Pooling equilibria. Two types set the same π1L (pL

m)
price.
pL L
m 6= p1
Separating equilibrium the high type price induces
entery and low type price does not induce the entery. pL
m is maximized monopoly price for low type

Incumbent knows that there is potential enterant. π1L (pL L H H L

m ) > π1 (pm ) pm 6= pm

In the separating equilibrium we have some type M1L |δD1L

k, high type price will induce entery.
Condition 2 is : π1L (pL L L L
1 ) + δπ1 ≥ π1 + δD1
For high type we will have:
π1L − π1L (pL L L
1 ) ≤ δ(π1 − D1 ) Incentive compatibility
pH
m will induce entery. for Low (ICL)

The low type does not want to signal that it is The paper shows that under reasonable condition
high type. we will have :

The payoff for high type is π1H + δD1H [p̃˜1 p̃1 ] where p̃1 < pL
m

The low types price pH

1 . There is some price that pL ˜ L
1 ∈ [p̃1 , p̃1 ] < pm
if the high type wants to mimic the low type it will
set this price. Equilibrium prices pH L ˜
m , pm ∈ [p˜1 , p̃1 ]

π1H (pL H
1 ) + δ.π1 → pL
1 = p̃1

High type charges high price, and low type will / {pL
Suppose there is p ∈ H
1 , pm } unexpected event
set low price pL .
Arbitrary belief x = 1 pr(H|p) = 1 We will not
We do not want high type to set low price so, have bayesian here.
π1h + δD1H ≥ π1H (pL H
1 ) + δπ1 Incentive compatibility
Condition High (ICH) We needed to provide some believes other than
incentive compatibility that supports separating
π1L (pL L L
1 ) + δπ1 p1 is equilibrium price, and any equilibrium. Here we founded it.
deviation from this means it is not low type. These
are equlibrium payoffs. Social welfare is higher than under symetric
information. The second stage is always identical,
Now the low type deviates → will have entery. for social we look at the first stage. We have lower
monopoly price for the low type. The lower price
Now we want to find off equilibrium payoffs. I here is good.
need to deviate to have entery. What is the best

2
 The entrant will not enter if the condition
xD2H + (1 − x)DxL ≥ 0.
Results are sensitive to assumption. We have
abundance of equilibria. We have abundance of
multiplicity of equilibria.
Once you take all these equilibria in standard
signaling game there is separating equilibria that finance by itself.
survives the standard refinement.
Arbitrary believes help the refinements. The typ-
ical signaling game has one equilibria that survives
that the high selects high education.
You need to motivate something, boil it down to
simple signaling game, and then find equlibria, and
go to the refinement and try to convince that it
makes sense.
Don’t stop at perfect nash equilibria; you need to
go through some refinements.
We skip 9.5 and 9.6 section. We will look at 9.7.
Lemon pricing, you choose the action in early
stage to convince other party that you are of
certain type. Now we affect the actual pay off of
the entery. We will affect the actual pay off of entery.
We assume now that there is entery, and we
engage in predatory pricing to push the rival out
of market. We try to get of entrant once entrant F (D(1 + r))]
entered the market.
You better able to sustain the losses. You have
deeper market. Bearing losses hurts both of you, but
you have deeper pocket and can endure it more.
Rival could go to the bank and tries to convince
that he will run out and then I will be able to repay.
Pay me now and I will recover. There should be
some constraint. There should be some imperfection
in the capital market.
Tirol direction of deep pocket.
Firm 2: financing project to debth.
The cost of the project is k which is total invest-
ment.
E is the firm’s equity. The amount that firm can
D = K − E is the amount of finance by the bank.
There is a random profit. Profits are π̃ ∈ [π, π̄] F
would be CDF and f would be the pdf.
r would be interest rate
The repayment would be D(1 + r)
if π̃ > D(1 + r) then π̃ − D(1 + r) as profit
if π̃ < D(1 + r) then the bank will keep π̃
Bankrupcy will cost the bank π̃ −¯
Firm expected profit would be U (D, r) =
R π̃
[π̃ − D(1 + r)]f (π̃)dp̃ there was second term
D(1+r)
which was equal to zero.
Bank’s profit:
R D(1+r)
V (D, r) = π [−̃B]f (π̃)dπ + D(1 + r)[1 −
Realization less that what the firm will repay
Additional assumptions:
Banks are competitive: so (1 + r0 ).D
r0 would be the banks interest rate
Zero profit condition:
V (D, r) = (1 + r0 ).D
3
 r(D) % D exists The firm one strategy could be pricing strategy or
quantity strategy all leading to reducing the firms
Higher lowen will come with higher cost of equity.
bankrupcy.
Now we want to talk about assymetric information
The firm will invest in the project only if in advertising.
Opportunity cost for the firm is that:
Next week we will tak about search.
The firm holds the equity. The firm could become
a lender, and lend money to other bank. Alkerof (1970). Lemon. We have sellers that know
the quality of the good, buyers that do not.
Firm 2 will hold on the project so if E . . . (1 + r0 )E
quality q: the characteristic of the good.
if profit would be greater than opportunity cost the
firm only will go for the project. U (D, r) ≥ (1 + r0 )E Seller valuation is θs .q if it keeps it, and p if it is
sold.
w = U (D, r) − (1 + r0 )E ≥ 0
Buyers payoff is θb .q − p if bought, and 0 if not
Firm one has effect on E the cash holding of firm bought.
two is lower. Firm one makes sure that it competes
very aggressively drives the profit down so that the θB > θs mean the buying is efficient.
firm two exits profit.
q ∼ U [0, q max ]
W = E π̃ − (1 + r0 ).k − B.F ((1 + r)(K − E)) ≥ 0
Buyer θb .q e − p ≥ 0
The expected value of cash flow minus total
investment cost. The joint opportunity cost the net q e ≡ E[q|sold]
value of project would be positive.
The buyer will buy if the price is p ∈ [0, θB .q e ]
The expected bankrupcy cost will affect the
interest rate. Seller will sell if q ∈ [0, θps ] and p ≥ θs .q

Higher E will give you higher expected payoff for The expected quality of the good would be
the project. E[q|sold] E[q|q < θps ]
p
If in the first stage very aggressive firm’s eq- q e (p) = 2θs
uity will be driven down, and we will have lower
w, and this increases the chance that firm 2 would p ≤ θB . 2θps
not be able to finance the project in the second stage.
θB ≥ 2θs
The higher E the smaller expected bankrupcy
cost, so smaller rate. The assumption here is that advertising reveals
information of the good in verifiable manner.
This is pricing strategy without assymetry infor-
mation. Advertising for search goods: goods that quality
is verifiable before purchase. Hard facts. ascertain

4
the quality of the good before purchase. (p − c).QA − pA = 0

Experience good advertising would be about soft What is advertising elasticity of demand?
facts.
QA .A
A = Q
The quality can be observed only after consump-
p−c pA
tion. p = QA .p

pA QA .A P A .A
The firm could only fool you once, and if fool you A p−c
p = −QA .p Q = p.Q
twice means you are fool.
A pA .A
p = p.Q
Dorfiman Steiner condition (1954).
A, p are choice
Firm sells bundle of two thing the good itself and
information about it. If learner index would be equal to zero then
advertising would be equal to zero. Consequently,
If the information and physical data could be when we have no markup, we will have no advertising.
complementary.
We had bundling of service and product in pre-
vious sessions. Q(s, p) some very simple form of
Higher advertising implies higher demand. Mean
investment cost.
hard fact reduce uncertainty, and come iwth higher
demand.
No markup pricing leads to no advertising. It is
necessary condition. If there is no perfect condition
Provide degree of product information by releasing
there would be no advertising. If advertising elastic-
more information. Release some information about
ity would be zero, we will have no advertising.
your good. You differentiate your good from others.
You update your quality belief, and by releasing
The higher profit associated with new product
information you can control how different your
the higher your advertising. The more you benefit
product is from the othe product.
from attracting people to buy your good you want
to advertise more.
maxp,A π = maxp,A [(p − c)Q(p, A) − F − pA .A]
A is advertising. In static game you will advertising in every period.

P A unit price of advertising. Advertising would be stock of good will. Could

be brand. That consists of all advertising in the past.
Qp < 0, QH > 0
At is advertising in t.
We maximize over price and advertising.
at is the amount of advertising in t
FOC of p:
at = At + (1 − γ)at−1 This is deprication of
p−c 1
Learner index would be p = p advertising.

FOC of A: at = At = (1 − γ)[At−a + (1 − γ)at−2 ] . . .

5
 Pt
at = t=0 (1 − γ)t−τ At + (1 − γ)t+1 a−1 L.

The monopolist cash flow in t is (pt − c)Q(pt , at ) − High quality with probability H, where
pA
t t which depends on stock of good will. L < H < 1.

PDV (presented discounted value) of cost follows: π(p, H, H), price, type, consumer perception.
P∞
maxpt ,at t=0 δ t [(pt − c).Q(pt , at ) − pA
t [at − (1 − Consumer are willing to pay for high quality good.
γ).at−1 ] The higher their perception the more they are willing
to pay.
Here pt , at are choice variables. We select this to
maximize above. The low quality guys want to signal high quality,
and high quality firm want to offer in separate
Here the assumption is that the product is in- equilibrium high advertising that is difficult for the
durable good. low quality to mimic.

FOC w.r.t pt learner index, and we get also w.r.t. Again here we have incentive compatibility con-
at straints.

If I invest today it will effect on stock today, Incentive compatibility constriant for high quality
tommorow, the day after tommorow and . . .. firm. In equlibrium suppose that we have (p, A)
combination. This is in separating equilibrium.
Dorfman stiener condition tells us that:
P∞ π(p, H, H) − PA .A in equilibrium
1−γ t ∂Qt+τ
pA
t = t=0 ( 1−r ) (pt+τ ∂at+τ
In the off equilibrium high quality firm knows that
Consumers have different tastes. Some uncertainty he could not convince that it is of high quality, then
about the characteristics. why should I advertise.

Nelson 1970 − 1974 π(pHL , H, L) In this case off equilibrium the

consumer will believe that I am a low type
Non informative advertising campaign. Just say
your product is around. Invest in non-informative, ICH: π(p, H, H) − pA .A ≥ π(pHL , H, L)
since you want consumer that your product exists.
incentive compatibility in low would be (ICL)
Every informative would be cheap talk, unless you
as the high quality firm find it cheaper to advertise. off equilibrium would be π(p, L, H) − pA .A

If it were to be informative, it has to be truthful. in separating equilibrium: π(pLL , L, L) − 0

The argument here is about repeat purchase. pLL 6= p

Milgram robert (1986) pHL 6= p

Signaling model of advertising. Signal the quality. Incentive compatibility would be now : (ICL)
Low quality means it is satisfactory with probability π(pLL , L, L) ≥ π(p, L, H) − pA .A.

6
 Any deviation from this would not make sense for
the low type either.

7
 IO @ UTD: Thirtheen session
Meisam Hejazinia

04/16/2013

1 Economics of search Theoritical foundation for price dispersion. In

Emanuele Tarantino
Empirical evidance of price dispersion.
Theoritical foundation of search
the standard framework, like bertrand, introducing
search cost, and assymetry of inormation results in
different result.
Consumers have imperfect information about the
prices. The incure cost to get this information.

1. Sequential search and the Diamond paradox Recall gives you the option to go back over the
previous options and search again, or not.
2. Sequential search with heterogeneous agents
Reinganum, Wolinskey These models could be applied to labor economics.

3. Alternative search. Non sequential search. Object could be good or job offer. Recieve
Fixed sample size. Varian 1980. prize yn , if stop at period n. Search with recall
yn = max{v1 , . . . , vn }, search without recall yn = vn .
4. Empirical test of search models The utility you attach to job will be job offer that
you recieved.
Consistency with behavioral consumers in the
market. Stopping rule, describes after which sequence you
should stop.
Motivation. Started back in the 60’s stigler.
Standard models of competition with homogeneous Stopping time N is integer N after which you
products, like cournot and bertrand. There is one should stop.
price in the market. ”‘law of one price”’, yet you
observe dispersion in the market. The rule is to stop after n if an only if yn ≥ y,
where level of y is the reservation price.
Empirical litrature on search and price dispersion,
between 5% to 30%. For books like software, Stopping rule is recursive. Environmet is station-
book, gasoline and so on. There is in the online ary, and at any point the decision is whether to stop
marketplace as well. or continue the search.

Goods are different, but you do not observe these Expected benefit, using marginal conditional.
differences. The
Rb additional benefit of one more search.
y
[v − y]f (v)dv. truncated at y. b is maxi-

1
mum value. y is reservation price. v is the value that 1 − λ of consumers remain uninformed (pick
touched to the given price. Uncertainty is captured at a firm at random), and other λ consumers are
by density function f (v). informed (purchase from the lowest price firm).

Additional cost to visit next store is s. Optimality ((1 − λ)/n + λ) will of uninformed purchase from
Rb
requires that y [v − y]f (v)dv = s. lower price firm, and (1 − λ)/n of uninformed will
buy from greater price firm. No pure strategy.
This will tell us the optimal time to stop search.
πi (pi , F (p)) = pi [ 1−λ
n + λ(1 − F (pi ))
n−1
] = v̄(1−λ)
n
Stopping time N (y) is geometrically distributed: is the expected profit. where the second term is the
1 with F (y)[1 − F (y)] and for n with F (y)n [1 − F (y)]. probability that you would be the lowest firm.

Diamond showed that under certain condition, all In online market, do we have precommitment to
firms will set monopoly prices. fixed sample size, or we have sequential search?

Large number of identical firms. Many identical

consumers. Inelastic demand. Each consumers
before starting the process recieves a price quota.
If you want to know more you undertake search
process. Finally the consumer decides whether
and where to buy. Homogeneity of the firms and
consumers are key assumptions here.

If increase the price slightly below the search cost,

noone will search.

Under diamond equilibrium despite no search price

bertrand that we had pi = 0 we have monopoly price.

How to solve this?

1. Introduce heterogeneity on consumers’ and/ or

firm’s side.

2. Consider alternative search rules: Fixed size

search

Until now we had hetrogeneity on the firm side.

Now we have Wolinskey 1986 which has heterogene-
ity on the consumer side.

The decision to search again depends on the

outcome of the previous search.

Varian 1980. Show price dispersion by assuming

2
 IO @ UTD: Thirtheen session
Meisam Hejazinia

04/16/2013

2 σ2
1 Principle Agent problem For normally distributed variable E(eγ. ) = eγ 2

I missed first part γ = −η.s

2
q contineous .s2 .σ 2 /2
E(e−η.s. ) = eη
2 2
q = a +   ∼ N (0, σ 2 ) → −e−η[t+sa−
c.a2
2 − η.s 2.σ ]
= −e−η.2̂(a)

principle risk rental The first term of exponent of left side is ..., second
term is cost, and third term is risk preimum, or
constant absolute risk aversion volatility..

u(w, a) = −e−y[w−ψ(a)] ˆ = t + sa − 1 c.a2 −

w(a) η.s2 .σ 2
2 2
00
η = − uu0 max − e−η(ŵ(a)) = maxŵ(a)
c.a2
ψ(a) = 2 The optimal would be a0 sc . Agent will put this
much effort.
Linear contracts w − t + s.q
Principle maximizes this: E(q − w)
principle maximizes expected payoffs:
E[a +  − ω] = a − ω + E[]
maxa,t,s E(q − t − s.q)
where E() = 0
Subject to (participation constratints)
E(−e−y[t+s.q−ψ(a)] ) ≥ 0 E[a +  − t − s(a + )] = (1 − s)a − t
c.a2
a ∈ argmaxE(−e−η[t+sq− 2 ]
) Principle tries to extract everything from the
2
agent. Suppose there is outside option which comes
−η[t+s(a+)− c.a
E(−e 2 ) with w̄, means outside option wage, so ω̂(s/c) = w̄.
ca2
E(−e−η[t+sa− 2 ]
).E(e−η.s ) maxs,t (1 − s) sc − t
ca2
(−e−η[t+sa− 2 ]
).E(e−η.s ) Subject to ŵ(s/c) = w̄

1
 a0 = s/c
The principle interested to maximize payoff:
1
s= 1+η.cσ 2 R q̂
max q̄
V (q − w(q)).f (q|a).dq
Fixed fee t will depend on outside auction.
ψ(q) is cost of exerting effort
The principle could only offer outside contingent
contract. Participation agent (aget not negative Given the
R materialization of q, agents problem
profit, or not less than outside option), and incentive would be u(w(q))f (q|a)dq − ψ(a) ≥ 0(IR) Individ-
compatibility(agent will produce optimal level of ual rationality
effort) to agent.
R
a ∈ argmaxâ { u(w(q))f (q|â)dq − ψ(â)}(IC)
Principle maximizes profit conditioned on partici- Incentive compatibility constraint
pation and incentive compatibility constraints. More
effort will increase q. I will earn more on higher q. IC → FOC: (ICa)
If the costs are higher I want to exert less effort. SOC (Second order constraint): (ICb)
Principle can use fixed fee t to exert entier surplus.
Set up lagrangian.
Linear contracts are nice and simple, and you just
need two number. However, they are not optimal. V 0 (q−w(q))
u0 (w(q)) = λ + µ. ffa(q|a)
(q|a)
This is for optimal risk
Optimal are non linear contracts.
sharing (co insurance)
Inefficiency of contract now will be discussed.
If µ is larger than zero we have deviation from
What is non linear contract? co-insurance. Effort of agent will not be verifiable.
Principle is interested in maximizing its expected
There is some output q which is function of effort payoff.
level and state of nature. q = u(θ, a)
The principle could ignore incentive compatibility
θ ∈ ω Natural disaster that affect outcome. The constraint. Principle offers w(q) that leads to
good state of nature bad state of nature, or anything co-insurance.
that will codetermine tha output.
µ > 0 typically
Principle payoff: −v(q − w)
We want to see higher wage for higher outcome.
Agent payoff: u(w) − ψ(a) t was fixed fee. s was positive. We want to see
something similar for non-linear contract, but the
q ∈ [q̂, q̄] problem is that it is not automatically given. We
need the assumption of cdf satisfying first order
CDF F (q|a) The higher effort level the higher stochastic dominance. Intuitively the probability of
outcome of the process (first order dominance) higher outcome is increasing in ’a’.

PDF: f (q|a) If µ > 0 the agent would be risk averse.

Contraction contract w(q) non linear tariff some We also need something that puts some structure
function of outcome. on ffa(q|a)
(q|a)
.

2
 and, second stage, optimization.
Suppose principle is risk neutral. We have
1 The effort levels you can put it could be finite.
u0 (w(q)) = λ + µ, and suppose we have two levels of
effort: aH , aL . How much do I have to pay? For each of effort level
he knows cost, and had cost function and he tries to
fa (q|a) = f (q|aH ) − f (a|qL ) maximize this.

f (q|a) > f (a|aH )

Optimal risk sharing gives us

1 f (q|aL )
u0 (w(q)) = λ + µ.[1 − f (q|aH ) ]

If f (q|aL ) < f (q|aH ) then I will get more if I put

more effort.

This is monotone likelihood ratio. Any contract

is possible, but we need first order stochastic domi-
nance as a structure we put to make sure that aget
gets higher as it puts higher effort.

First order approach is not always valide. We

only consider first order, we can not make sure it
is optimal level of effort, we can pin point local
maximum or minimum and not maximum.

Mirrless(1975) investigates it.

First order is necessary but not sufficient.

Rogoser (1985) added one condition to this

which was convexity of this condition. Maximum
Likelihood Ration Property (MLRP). It comes with
convex CDFC (Distribution Function Condition). In
this case first order stochastic condition works.

Grossman-Hart(1983). Why we started with first

order? Since we had continium of types, and we
had incentive compatibility, and we do not wanted
types to mimic eachother. We had huge amount of
incentive compatibility constraints. They tried to
use discrete type space. Discrete performance space.
q1 < q2 < . . . < qn . It might not be too tractable,
but it is still better than infinite number.

It looked at two stages: first stage, implemenation

3
 IO @ UTD: Fiftheen session
Meisam Hejazinia

04/30/2013

Two stage game. At the second stage you solve Hotelling model
for signaling game. In the signaling game you will Little bit of product differentiation
have information sets. The incumbant in the second Vertical control
stage does not know whether it is new entrant or it Strategic interaciton (Static dynamic duopoly)
is the old one. Prey and accomodate strategies on Limit pricing
each of stage. Prior is given in the form of common Predetary pricing
knowledge. What changes is posterior that the Entry barriers
entrant may be able to update or not. α is common Search
knowledge, and both player know it. The incumbant Strategic decision making
knows his type when he makes the move. α is
entrant beleif about the type. The entrant starts the Bertrand paradox: homogeneous goods. Price
game, and decides whether to enter or not. There is equal to marginal cost
also a second round. Nature draws a type. Enterant Solution: (1) differentiated product, (2) repeated
knows the payoff structure for each one. The only interaction, (3) search cost
uncertainty is α, whether it is old or new. α is for
entrant and non incumbant. One sided assymetry of Linear demand functions
information.

Here we look at the perfect bayesian equilibrium.

What are they player, what they can do. What

do they know. Strategic aspect. Nash equilibrium?
Perfect Bayesian Equlibrium, since players move
sequentially under assymetric information. What is
the objective function.

Perfect equlibrium, add probability of epsilon of

fault, trembeling hand, will equilibrium be equilib-
rium again.

Topics covered:
Monopoly
Price discrimination
Second degree price discrimination
Third degree price discrimination