ECO 310: Empirical Industrial Organization

Lecture 11: Models of Market Entry

Introduction

Victor Aguirregabiria (University of Toronto)

December 3rd, 2018

Victor Aguirregabiria () Competition December 3rd, 2018 1 / 29

 Outline

Models of Market Entry: Outline

1. What is a model of market entry?

2. Why do we estimate models of market entry?
3. Entry models with homogeneous …rms
4. Entry models with heterogeneous …rms

Victor Aguirregabiria () Competition December 3rd, 2018 2 / 29

 Outline

Main References

Bresnahan and Reiss (JPE, 1991)

Bresnahan and Reiss (Journal of Econometrics, 1991)

Tamer (REStud, 2003)

Seim (RAND, 2006)

Victor Aguirregabiria () Competition December 3rd, 2018 3 / 29

 What is a model of market entry?

————————————————————————————

1. What is a Model of Market Entry?

————————————————————————————

Victor Aguirregabiria () Competition December 3rd, 2018 4 / 29

 What is a model of market entry?

Main features of a model of market entry

(1) The dependent variable is a …rm decision to operate or not in

a market.
- Entry in a market can be understood in a broad sense
- e.g., entry in an industry; opening a new store; introducing a new
product; adopting a new technology; release of a new movie;
participate in an auction, etc.

(2) There is a …xed sunk cost associated with being active in the
market;

(3) The payo¤ of being active in the market depends on the number
(and the characteristics) of other …rms active in the market, i.e., the
model is a game.

Victor Aguirregabiria () Competition December 3rd, 2018 5 / 29

 What is a model of market entry?

Main features of a model of market entry [2]

Consider a market where there are N …rms that potentially may to

enter in the market.

ai 2 f0, 1g is a binary variable that represents the decision of …rm i of

being active in the market (ai = 1) or not (ai = 0).

Pro…t of not being in the market is zero.

Pro…t of being active is: Vi (n ) Fi where Vi (.) is the variable pro…t,

n is the number of …rms active, and Fi is the entry cost.

The number of active …rms, n, is endogenous:

N
n= ∑ ai
i =1

Victor Aguirregabiria () Competition December 3rd, 2018 6 / 29

 What is a model of market entry?

Main features of a model of market entry [3]

Under Nash assumption, every …rm takes as given the decision of the
other …rms and makes a decision that maximizes its own pro…t.

The best response of …rm i under Nash equilibrium is:

8
< 1 if Vi 1 + ∑j 6=i aj Fi 0
ai =
:
0 if Vi 1 + ∑j 6=i aj Fi < 0

where 1 + ∑j 6=i aj represents …rm i’s Nash-conjecture about the

number of active …rms.

Victor Aguirregabiria () Competition December 3rd, 2018 7 / 29

 What is a model of market entry?

Example

Two potential entrants: N = 2. With V1 (n ) = V2 (n ) = 100 20 n.

And F1 = F2 = 50.

Vi ( 1 + a j ) Fi = 30 20 aj
Best responses are:

1 if 30 20 a2 0 1 if 30 20 a1 0
a1 = and a2 =
0 if 30 20 a2 < 0 0 if 30 20 a1 < 0

Payo¤ Matrix:
a2 = 0 a2 = 1
a1 = 0 (0 , 0) (0 , 30)
a1 = 1 (30 , 0) (10 , 10)

Victor Aguirregabiria () Competition December 3rd, 2018 8 / 29

 What is a model of market entry?

Example [2]

a2 = 0 a2 = 1
a1 = 0 ( 0 , 0 ) (0 , 30)
a1 = 1 (30 , 0) (10 , 10)
With this payo¤ matrix, the unique Nash equilibrium is
(a1 , a2 ) = (1, 1). Duopoly.
Suppose that the …xed cost were larger, F = 90. Then,
Vi (1 + aj ) Fi = 10 20 aj .
a2 = 0 a2 = 1
a1 = 0 ( 0 , 0 ) (0 , 10)
a1 = 1 ( 10 , 0) ( 30 , 30)
With this payo¤ matrix, the unique Nash equilibrium is
(a1 , a2 ) = (0, 0). No entry.

Victor Aguirregabiria () Competition December 3rd, 2018 9 / 29

 What is a model of market entry?

Example [3]

Suppose that the …xed cost is not as small as 50 and not as large as
90: F = 70. Then, Vi (1 + aj ) Fi = 10 20 aj .
a2 = 0 a2 = 1
a1 = 0 ( 0 , 0 ) (0 , 10)
a1 = 1 (10 , 0) ( 10 , 10)
With this payo¤ matrix, the model has two Nash equilibria: Monopoly
of …rm 1: (a1 , a2 ) = (1, 0); Monopoly of …rm 2: (a1 , a2 ) = (0, 1).

Victor Aguirregabiria () Competition December 3rd, 2018 10 / 29

 What is a model of market entry?

Example [4]

For general value of F :

a2 = 0 a2 = 1
a1 = 0 ( 0 , 0 ) (0 , 80 F )
a1 = 1 (80 F , 0) (60 F , 60 F )
We can see that the model has di¤erent predictions about market
structure depending on the value of the …xed cost:
- If F 60 — > Duopoly is unique Nash equilibrium
- If 60 < F 80 — > Monopoly of 1 or 2 are Nash
equilibria
- If F > 80 — > No …rm in the market is unique Nash
equilibrium
The observe actions of the potential entrants reveal information
about pro…ts, about …xed costs.

Victor Aguirregabiria () Competition December 3rd, 2018 11 / 29

 What is a model of market entry?

Two-stage game

Where does the variable pro…t Vi (n ) comes from?

It is useful to see a model of market entry as part of a two stage

game.

In a First stage, N potential entrants simultaneously choose whether

to enter or not in a market.

In a Second stage, entrants compete (e.g., in prices or quantities)

and the pro…ts Vi (n ) of each …rm are determined.

Example (Exercise): Cournot competition with linear demand

P = A B Q and constant MCs, c, implies:
2
1 A c
Vi (n ) =
B n+1

Victor Aguirregabiria () Competition December 3rd, 2018 12 / 29

 Why do we estimate models of market entry?

————————————————————————————

2. Why do we estimate
Models of Market Entry?
————————————————————————————

Victor Aguirregabiria () Competition December 3rd, 2018 13 / 29

 Why do we estimate models of market entry?

Why do we estimate models of market entry?

[1] Explaining market structure.

- Why di¤erent industries (and di¤erent markets within the same
industry) have di¤erent number of active …rms?

[2] Identi…cation of entry costs parameters.

- These parameters are important in the determination of …rms
pro…ts, market structure, and market power.
- Fixed costs do not appear in demand or in Cournot or Bertrand
equilibrium conditions, so they cannot be estimated in these type of
models.

[3] Data on prices and quantities may not be available.

- Sometimes all the data we have are …rms’entry decisions. These
data can reveal information about pro…ts and about the nature of
competition.

Victor Aguirregabiria () Competition December 3rd, 2018 14 / 29

 Entry models with homogeneous …rms

————————————————————————————

3. Entry Models
with Homogeneous Firms
————————————————————————————

Victor Aguirregabiria () Competition December 3rd, 2018 15 / 29

 Entry models with homogeneous …rms

Market entry with homogeneous …rms

We start with an empirical model of entry in an homogeneous

product industry and where all the …rms have the same costs.

There are several reasons why we start with this case.

1. This is the simpler empirical model of entry, and where this

literature started with the seminal work by Bresnahan & Reiss (JPE,
1990).

2. The model with heterogeneous …rms typically has multiple

equilibria, and this makes the estimation more complicated.

3. Sometimes we have very limited information about …rms’

heterogeneous characteristics.

Victor Aguirregabiria () Competition December 3rd, 2018 16 / 29

 Entry models with homogeneous …rms

Market entry with homogeneous …rms: Data

Suppose the researcher has data from M markets in the same industry.

For instance, the supermarket industry. The M markets are M

neighborhoods from di¤erent Canadian cities.

Markets are indexed by m.

The dataset consists of:

Data = f nm , Sm , Xm : m = 1, 2, ..., M g

nm = number of active …rms;

Sm = market size;
Xm = other exogenous market characteristics a¤ecting demand or
costs.

Victor Aguirregabiria () Competition December 3rd, 2018 17 / 29

 Entry models with homogeneous …rms

Market entry with homogeneous …rms: Model

All the potential entrants in a market have the same pro…t function:
- Same costs, and same demand (homogenous product).

The pro…t function of a …rm in market m is:

Vm ( n ) Fm

where Vm (n ) is the variables pro…t, Fm is the …xed cost, and n is the

number of active …rms in the market.

We describe below the speci…cation of Vm (n ) and Fm in terms of

observable variables and unobservables.

A key feature is that Vm (n ) is a strictly decreasing function of n.

Victor Aguirregabiria () Competition December 3rd, 2018 18 / 29

 Entry models with homogeneous …rms

Market entry with homogeneous …rms: Model [2]

Under Nash-equilibrium, we have the following conditions:

Vm 1 + ∑j 6=i ajm Fm 0 for …rms with aim = 1

Vm 1 + ∑j 6=i ajm Fm < 0 for …rms with aim = 0

Then, nm is an equilibrium i¤:

Vm ( n m ) Fm 0 Active …rms are in their best response

Vm ( 1 + n m ) Fm < 0 Inactive …rms are in their best response

Victor Aguirregabiria () Competition December 3rd, 2018 19 / 29

 Entry models with homogeneous …rms

Market entry with homogeneous …rms: Model [3]

We can write the Nash-equilibrium conditions also as:

Vm (1 + nm ) < Fm Vm ( n m )

The equilibrium conditions imply restrictions on …xed costs and more

generally on the parameters in the pro…t function.

Using these restrictions and the data, we estimate the parameters in

the pro…t function.

Victor Aguirregabiria () Competition December 3rd, 2018 20 / 29

 Entry models with homogeneous …rms

Speci…cation of the variable pro…t function

Bresnahan and Reiss (JPE, 1990) do not model explicitly the form of
price/quantity competition and consider a ‡exible model for the
variable pro…t.
Vm (n ) = Sm [Xmv βv α(n )]
Sm represents market size.

Xmv is a vector is observable market characteristics a¤ecting variable

pro…ts, e.g., income, prices of variable inputs, and βv is a vector of
parameters.

The parameters α(1), α(2), ... capture the competitive e¤ect. We

expect:
α(1) < α(2) < α(3) ... < α(N )

Victor Aguirregabiria () Competition December 3rd, 2018 21 / 29

 Entry models with homogeneous …rms

Speci…cation of the …xed cost

The speci…cation of …xed cost is:

Fm = Xmf βf + δ(n ) + εm

Xmf is a vector is observable market characteristics a¤ecting …xed

costs, e.g., prices of …xed inputs, and βf is a vector of parameters.

εm is unobservable of the researcher; and error term.

The parameters δ(1), δ(2), ... capture possible competition e¤ects in

…xed costs, as well as potential collusive motives.

δ(1) < δ(2) < δ(3) ... < δ(N )

Victor Aguirregabiria () Competition December 3rd, 2018 22 / 29

 Entry models with homogeneous …rms

Equilibrium conditions

The total pro…t function is:

Vm ( n ) Fm = (Sm Xmv ) βv Xmf βf Sm α ( n ) δ (n ) εm

Equilibrium conditions: nm = n is an equilibrium:

Vm (1 + n ) < Fm Vm ( n )

or equivalently:

(Sm Xmv ) βv Xmf βf Sm α ( n + 1 ) δ (n + 1)

< εm
(Sm Xmv ) βv Xmf βf Sm α (n ) δ (n )

Victor Aguirregabiria () Competition December 3rd, 2018 23 / 29

 Entry models with homogeneous …rms

Equilibrium conditions [2]

Suppose that εm is independent of (Sm ,Xm ) and iid N (0, 1).

Let Pm (n ) represent the probability Pr(nm = n j Sm , Xm ):

Pm (n ) = Φ Sm [Xmv βv α(n + 1)] Xmf βf δ (n + 1)

Φ Sm [Xmv βv α(n )] Xmf βf δ (n )

Victor Aguirregabiria () Competition December 3rd, 2018 24 / 29

 Entry models with homogeneous …rms

Estimation of the model parameters

Let θnbe the vector of the parameters of the model.

o
θ = βv , βf , α(1), ..., α(N ), δ(1), ..., δ(N ) .

We estimate these parameters using a Maximum Likelihood estimator

(MLE).
The likelihood function of this model and data is:
M
L(θ ) = ∏ Pr(nm j Sm , Xm ; θ )
m =1
2 3
Φ Sm [Xmv βv α(n + 1)] Xmf βf δ (n + 1)
M 6 7
= ∏ 64 7
5
m =1
Φ Sm [Xmv βv α(n )] Xmf βf δ (n )

The MLE is the value of θ that maximizes L(θ ).

Victor Aguirregabiria () Competition December 3rd, 2018 25 / 29
 Entry models with homogeneous …rms

Answering empirical questions using estimated model

[1] Ratio of Entry costs to Variable pro…ts.

Fm
We can construct the ration: , e.g., in market m, the entry
Vm (1)
cost is 46% of the variable pro…t of a monopolist in this market.

[2] How strong is competition? How quickly pro…ts decline with

n?

De…ne the function ratio:

( n + 1 ) Vm ( n + 1 )
rm ( n ) =
n Vm ( n )

This is the ratio between total variable pro…ts with n + 1 …rms and
with n …rms, e.g., rm (1) = 1.45 means that total variable pro…ts
under duopoly are 45% larger than under monopoly,
Victor Aguirregabiria () Competition December 3rd, 2018 26 / 29
 Entry models with homogeneous …rms

Answering empirical questions using estimated model [2]

Economy theory has several predictions on the ratio

(n +1 ) V m (n +1 )
rm ( n ) = n V m (n )

[1] It is greater or equal than 1, rm (n ) 1;

[2] As n increases, if …rms compete and we converge to the

competitive equilibrium, then rm (n ) converges to 1.

[3] As n increases, if …rms collude, then rm (n ) does NOT decline and

it does not converge to 1.

[4] Contestable markets hypothesis. It is possible to achieve the

competitive outcome even with a small number of …rms in the
market. For instance, if rm (4) = 1, then market m achieves the
competitive outcome with only four active …rms.

Victor Aguirregabiria () Competition December 3rd, 2018 27 / 29

 Entry models with homogeneous …rms

Bresnahan & Reiss (JPE, 1990): Empirical results

M = 202 local markets (small towns)

Five industries: dentists, doctors, drug stores, plumbers and tire

dealers.

Main Findings:
- Entry thresholds converge quite fast after the second entrant.
- After three or four …rms, an additional entrant doesn’t a¤ect
much competition.

Victor Aguirregabiria () Competition December 3rd, 2018 28 / 29

 Entry models with homogeneous …rms

Bresnahan Reiss (JPE 1990)

Victor Aguirregabiria () Competition December 3rd, 2018 29 / 29