introduction to the course

ECO 2901
EMPIRICAL INDUSTRIAL ORGANIZATION
Lecture 1: Introduction to the Course
Victor Aguirregabiria (University of Toronto)
Toronto. Winter 2015
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
1 / 44
Organization of the Course
Class Meetings: Fridays, 9:00-11:00am; room GE 100
O¢ ce hours: Tuesdays and Thursdays 3:00-4:00pm
Evaluation: Problem Sets (25%); Referee & Presentation (25%);
Final Exam (50%)
I expect that you: (1) attend every class meeting; (2) read the
papers/material before each lecture; (3) participate in class; (4) go
through class notes and understand them; (5) do the problem set on
time; (6) referee & presentation; (7) prepare for the …nal exam.
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
2 / 44
Some General Features of this Course
This courses deals with models, methods, and applications in
Empirical Industrial Organization (EIO).
This course emphasizes with the same intensity modelling,
econometric techniques, and empirical applications.
This year, the course concentrates in dynamic models of
competition in IO.
We will study empirical models of market competition that involve
dynamic considerations: market entry/exit; capacity choice;
advertising; inventories; storable and durable products; adoption of
new technologies; geographic location; network competition; etc.
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
3 / 44
Topics
1.
Some General Ideas on Empirical IO
2.
Dynamic Structural Models of Industrial Organization
3.
Single-Agent Models of Firm Investment and Supply
4.
Structural Models of Dynamic Demand
5.
Empirical Dynamic Games of Oligopoly Competition
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
4 / 44
Lecture 1: Outline
1.
Some Basic Ideas in IO
2.
Data in Empirical IO
3.
Speci…cation of a Structural Model in EIO
4.
Equilibrium and Comparative Statics
5.
Identi…cation and Estimation
6.
Extension and the Rest of the Course
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
5 / 44
Some Basic Ideas in IO
(1)
IO studies the behavior of …rms in markets, their strategic
interactions, and the implications on pro…ts and consumer welfare.
Some examples of type of …rm decisions that we study in IO are:
-
Price and Quantity choice;
Investment in capacity, inventories, physical capital, ...;
R&D, patents;
Advertising;
Geographic location of plants and stores;
Product characteristics;
Entry in new markets;
Adoption of new technologies;
Contracts with suppliers, and customers;
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
6 / 44
Data in EIO:
The "old days"
Aggregate industry level data
The typical study estimated a linear regression model where each
observation n in the sample represents an industry:
Pn
MCn
= β0 + β1 HHIn + εn
Pn
P n MC n
Pn
is the Lerner Index that measures of market power, and HHI
is the Her…ndahl-Hirschman index that measures of market
2
n
concentration, i.e., HHIn = ∑N
i =1 sni , where sni is the share of …rm i’s
output in total industry n output.
Estimations using industry-level cross-sectional data from very diverse
industries typically found a positive and statistically signi…cant β1 .
The interpretation of the estimated regression functions was causal
and not just as an equilibrium relationship.
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
7 / 44
Data in EIO:
New Empirical IO
Motivated by the limitations in the previous empirical literature based
on aggregate industry-level data.
[1]
Market power and concentration are endogenous variables
that are jointly determined by exogenous factors in demand,
technology, and regulation.
[2]
Industries are very heterogeneous in their exogenous
characteristics. There is not a common relationship between market
power and concentration across industries.
[3]
Both endogeneity [1] and unobserved industry heterogeneity
[2] imply a biased in the estimation of the "average elasticity" β1 .
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
8 / 44
Data in EIO:
New EIO
Emphasizes the need to:
[1]
Study competition separately for each industry.
[2]
Use richer data at a more disaggregate level. Data at the level
of individual …rms, products, and markets, on prices, quantities,
number of …rms, and exogenous characteristics a¤ecting demand or
costs.
[3]
Estimate game theoretical models of oligopoly competition.
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
9 / 44
Data in EIO:
Sources of Sample variation
No sample variation from multiple industries, from many …rms
(oligopoly markets), or from many periods (short panels).
Main sources of sample variability in empirical studies in IO
Most of the sample variation in these studies come from observing
multiple products and multiple local markets within the same
industry.
Typical dataset in EIO: cross-section or panel data of many products
and/or local markets from the same industry, with information on
selling prices, produced quantities, product attributes, and local
market characteristics.
Data on …rms’costs are very rare.
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
10 / 44
Speci…cation of a Structural Model in EIO
(1)
To study competition in an industry, EIO researchers propose and
estimate structural models of demand and supply where …rms
behave strategically.
What is an structural model in empirical IO?
Models of consumer and …rm behavior where consumers are utility
maximizers and …rms are pro…t maximizers.
The parameters are structural in the sense that they describe
consumer preferences, production technology, and institutional
constraints.
Under the principle of revealed preference, these parameters are
estimated using micro data on consumers’and …rms’choices and
outcomes.
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
11 / 44
Speci…cation: Typical Structure of IO Models
1. Model of consumer behavior (Demand)
- Product di¤erentiation?
2. Model for …rms’costs
- Economies of scale; Economies of scope? Entry costs? Investment
costs?
3. Equilibrium model of static competition
- Price (Bertrand), Quantity (Cournot).
4. Equilibrium model of market Entry-Exit
5. Equilibrium model of dynamic competition
- Investment, advertising, quality, product characteristics, stores, etc.
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
12 / 44
Speci…cation: Example
Example based on Ryan (Econometrica, 2012).
We start with an empirical question.
US cement industry. Evaluation of the e¤ects in this industry of the
1990 Amendments to the Air Clean Act.
The new law restricts the amount of emissions a cement plant can
make.
It requires the adoption of a "new" technology that implies lower
marginal costs but larger …xed costs than the "old" technology.
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
13 / 44
Speci…cation: Key Characteristics of the Industry
The model here, though simple, incorporates some important features
of the cement industry.
1. Homogeneous product. (We abstract from spatial di¤erentiation).
2. Substantial …xed costs from operating a plant (cement furnace).
3. Variable costs increase in a convex way when output approaches full
capacity.
4. Capacity investment is an important strategic variable.
5. Industry is very local (due to high transportation costs per dollar
value). It can be characterized as a set of many "isolated" local
markets.
6. Oligopolist industry. Small number of …rms at a local market.
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
14 / 44
Speci…cation: Data
The speci…cation of the model depends importantly on the data that
is available for the researcher.
M local markets (e.g., towns) observed over T consecutive quarters.
We index markets by m and quarters by t.
For every market-quarter observation, the dataset contains
information on: the number of plants operating in the market (Nmt ),
aggregate amount of output produced by all the plants (Qmt ), market
price (Pmt ), and some exogenous market characteristics, population,
average income, etc (Xmt ).
Data = f Pmt , Qmt , Nmt , Xmt : m = 1, 2, ..., M; t = 1, 2, ..., T g
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
15 / 44
Speci…cation: Structure of the model
Our model of oligopoly competition has four main components:
(a) demand equation;
(b) cost function;
(c) model of Cournot competition;
(d) model of market entry.
An important aspect in the construction of an econometric model is
the speci…cation of unobservables. In general, the richer the
speci…cation of unobservables in a model, the more robust the
empirical …ndings.
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
16 / 44
Speci…cation:
Demand Equation
Demand is linear in prices and in parameters.
Qmt = Smt
β0 and β1
β0 + βX XD
mt
β1 Pmt + εD
mt
0 are parameters.
Smt represents ‘true‘demand size.
Smt
Smt expfεSmt g
εSmt and εD
mt are random variables with zero mean.
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
17 / 44
Speci…cation:
Demand Equation (2)
For some of our derivations it is convenient to represent demand
using the inverse demand curve:
Pmt = Amt
where
Victor Aguirregabiria ()
Bmt Qmt
Amt
β0 + βX Xmt + εD
mt
β1
Bmt
1
β1 Smt
Empirical IO
Toronto. Winter 2015
18 / 44
Speci…cation:
Cost Function
Every …rm, either an incumbent or a potential entrant, has the same
cost function:
C (qimt ) = VCmt (q ) + FCmt
The Variable Cost is quadratic:
MC
MC
γMC
+ γMC
1
X Xmt + εmt
VCmt (q ) =
MC
MC
with γMC
1 , γX , and γ2
q+
γMC
2
q2
2
0 are parameters.
The marginal cost is:
MCmt (q ) = MC mt + γMC
2 q
where MC mt
MC
MC
γMC
+ γMC
1
X Xmt + εmt is the exogenous MC .
The Fixed Cost is:
FC
FC
FC
FCmt = γFC
1 + γX Xmt + εmt
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
19 / 44
Speci…cation:
Cournot Competition
Suppose that there are Nmt plants active in local market m at
quarter. We assume that …rms active in a local market compete with
each other ala Cournot.
The pro…t function of a …rm is:
e ) = Pmt (q + Q
e) q
Πmt (q, Q
VCmt (q )
FCmt
This best response output is characterized by the following condition
of optimality:
Pmt +
e)
∂Pmt (q + Q
q = MCmt (q )
∂q
An given our speci…cation of demand and costs, this condition implies:
qmt (N ) =
Victor Aguirregabiria ()
Amt MC mt
Bmt (N + 1) + γMC
2
Empirical IO
Toronto. Winter 2015
20 / 44
Speci…cation:
Cournot Competition (2)
The equilibrium price-cost margin is,
Pmt
AVCmt = Bmt + γMC
2 /2
qmt (N )
And the Cournot equilibrium pro…t of an active …rm (with N …rms in
the market) is:
Πmt (N ) =
Bmt + γMC
2 /2
Amt MC mt
Bmt (N + 1) + γMC
2
2
FCmt
This Cournot equilibrium pro…t function is continuous and strictly
decreasing in the number of active …rms, N.
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
21 / 44
Speci…cation:
Model of Market Entry
The equilibrium entry condition establishes that every active …rm and
every potential entrant is maximizing pro…ts.
Active …rms should be making non-negative pro…ts: Πmt (Nmt )
0.
Potential entrants are not leaving positive pro…ts on the table:
Πmt (Nmt + 1) < 0.
There is a unique value of N that satis…es the equilibrium conditions
Πmt (N ) 0 and Πmt (N + 1) < 0.
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
22 / 44
Speci…cation:
Model of Market Entry (2)
Let Nmt be the real number that (uniquely) solves the condition
Πmt (N ) = 0.
s
MC
1 + γMC
γ
2 /2Bmt
+ Amt MC mt
Nmt
1+ 2
Bmt
FCmt Bmt
The equilibrium number of …rms is the largest integer that is smaller
than Nmt : Nmt = int (Nmt )
The entry equilibrium condition, Πmt (Nmt ) = 0, is equivalent to:
(qmt )2 =
Victor Aguirregabiria ()
Nmt
int (Nmt )
Empirical IO
2
FCmt
Bmt + γMC
2 /2
Toronto. Winter 2015
23 / 44
Structural Equations
The model can be described as a system of three equations with three
endogenous variables, N, P, and q Q/N,
Demand equation:
P
= A
Cournot Equilibrium Condition:
q
=
Entry Equilibrium Condition: q 2 =
B N q
A MC
B (N + 1) + γMC
2
FC
B + γMC
2 /2
where A, B, MC , γMC
2 , and FC are exogenously given.
This system of equations is denoted as the structural equations of
the model.
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
24 / 44
Reduced Form Equations
This is a system of simultaneous equations. The solution to this
system of equations determines the value (or values) of the
endogenous variables fN, P, q g for given values of the exogenous
variables X and ε (εD , εS , εMC , εFC ), and the structural
parameters θ f β0 s, γMC 0 s, γFC 0 s g.
N
= fN (X, ε, θ)
q
= fq (X, ε, θ)
P
= fP (X, ε, θ)
This system of equations is denoted as the reduced form equations
of the model.
The model has well-de…ned reduced form functions/equations if for
every value of (X, ε, θ) an equilibrium exists and it is unique.
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
25 / 44
Equilibrium Existence and Uniqueness
Given the assumptions of our model, we have that for every value of
(X, ε, θ) an equilibrium exists and it is unique.
Entry equilibrium condition determines q,
s
FC
q=
B + γMC
2 /2
Plugging q into Cournot eq. condition, we get:
s
MC
γ
1 + γMC
2 /2B
N=
1+ 2
+ A MC
B
FC B
And plugging these expressions for N and q in the demand equation
we obtain the equilibrium price:
s
FC
P = MC + (γMC
+ B)
2
B + γMC
2 /2
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
26 / 44
Identi…cation
The researcher wants to use data and this model to estimate the
vector of structural parameters θ f β0 s, γMC 0 s, γFC 0 s g.
Econometric model (without εSmt ):
Qmt
Smt
Pmt
1 qmt
β1 Smt
2
qmt
+ β1 γMC
qmt
2
Smt
= βX XD
mt
β1 Pmt + εD
mt
MC q
MC
= γMC
XMC
mt + εmt
mt + γ2
X
FC
FC
= γFC
X Xmt + εmt
Assumption 1: Mean independence between exogenous observables
and unobservables:
E (εmt j Xmt ) = 0
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
27 / 44
Identi…cation
(2)
We say the parameters of the model are identify if there is a feasible
estimator of θ that is consistent in a statistical or econometric sense.
A standard approach to prove identi…cation consists in using the
moment restrictions implied by the model to show that we can
uniquely determine the value of θ as a function of moments that
include only observable variables.
For instance, in a classical linear regression model Y = X β + ε under
the assumptions E(X ε) = 0 and E(X X 0 ) is non-singular, we have
that
β = E(XX 0 ) 1 E(XY )
such that this expression shows that the vector of parameters β is
identi…ed using data of Y and X .
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
28 / 44
Endogeneity
In our model, the mean independence assumption E(εmt j Xmt ) = 0
is not su¢ cient to identify the mode.
The three structural equations include endogenous regressors.
From the reduced form (equilibrium conditions of the model), we
know that Pmt , qmt , and Nmt depend on the unobservables εmt such
that:
E(εmt Pmt ) 6= 0 ; E(εmt qmt ) 6= 0 ; E(εmt Nmt ) 6= 0
Therefore, OLS estimation of any of the structural equations, for
instance demand
Qmt
Smt
= βX XD
mt
β1 Pmt + εD
mt
generates inconsistent estimates.
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
29 / 44
Endogeneity
(2)
Under the mean independence assumption E(εmt j Xmt ) = 0 we can
estimate consistently "reduced form parameters".
However, without further restrictions, the parameters that we can
identify from the reduce form equations are not su¢ cient to
separately identify the structural parameters in demand, variable
costs, and …xed costs.
For instance, in the reduce form equation for Nmt we have:
s
MC
1 + γMC
γ
2 /2Bmt
+ Amt MC mt
Nmt
1+ 2
Bmt
FCmt Bmt
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
30 / 44
Some identi…cation approaches in empirical IO
To identify the model we need more information, either in the form of
data or/and additional restrictions in the model.
Some identi…cation approaches used in EIO include:
1.
Randomized experiments
2.
Exclusion restrictions (Instrumental Variables)
3.
"Natural experiments" as exclusion restrictions
4.
Restrictions on covariance-structure of unobservables
4a.
Arellano-Bond instruments
4b.
Hausman-Nevo instruments
4c.
Zero-covariance between unobservables
5.
Partial identi…cation (bounds)
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
31 / 44
Randomized experiments
The implementation of a randomized experiment is an ideal situation
for the identi…cation of an econometric model.
However, the careful design of a useful randomized experiment is not
a trivial problem.
The structural model is a useful tool in the design of the randomized
experiment.
Suppose that we want to estimate …rst the demand equation.
We need to design an experiment that generates sample variation in
price that is not perfectly correlated with Xmt it is independent of the
unobserved demand shock εD
mt .
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
32 / 44
Randomized experiments
(2)
Suppose that experiment consists in a …rm subsidy per unit of output
produced and sold in the market, τ mt
τ mt is determined as random draw from some distribution.
We need also to assume that the implementation of the experiment
does not introduce any change in the behavior of consumers.
Under these conditions, we have that E(τ mt εD
mt ) = 0; no perfect
collinearity between τ mt and Xmt ; and E(τ mt Pmt ) 6= 0.
These conditions imply that we can use τ mt as an instrument for the
Pmt in the demand equation, to identify all the parameters in the
demand.
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
33 / 44
Randomized experiments
(3)
Possible issues.
Agents behavior may change if they know that they are the subjects
of an experiment.
Firms may change the way they compete during the time that
experiment is implemented. For instance, they may decide to agree
not to change their levels of output such that the subsidy will not be
pass-through to the price and they will keep the subsidy as a pure
transfer.
Most importantly, if some consumers are aware of the existence of
this experiment, and given the temporary nature of the experiment,
they may decide to buy cement for inventory. In that case, the
experiment will a¤ect the demand and the estimates of the demand
parameters based on this randomized experiment will be biased.
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
34 / 44
Exclusion restrictions (Instrumental Variables)
In econometrics, the most common approach to deal with endogeneity
problems is using instrumental variables.
In a system of simultaneous equation, we can get instrumental
variables if we impose exclusion restrictions: some exogenous variables
do not enter into some structural equations.
For instance, if XMC or/and XFC includes variables which are not in
XD , then we can use these variables in (XMC ,XFC ) as instruments for
Pmt in the estimation of the demand.
For instance, the price of limestone and coal could be exclusion
restrictions in our application.
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
35 / 44
"Natural experiments" as exclusion restrictions
Consider an unexpected natural shock at period t that a¤ected the
production cost of …rms in a speci…c region.
Let Imt be the indicator of the event “market a¤ected by the natural
experiment”.
Imt = 1ft t g Em
Em is the binary indicator of the event "market m belongs to the
region a¤ected by the natural event".
The key identi…cation assumption to use Imt as an instrument for
price is that the natural event did not a¤ect demand such that E(Imt
εD
mt ) = 0.
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
36 / 44
"Natural experiments"
(2)
Assumption E(Imt εD
mt ) = 0 is typically implausible.
Though the natural event is completely exogenous and unexpected, it
may have occurred in markets that have relatively high (or low) levels
of demand, or during a period of high (or low) demand.
For this reason, most applications using identi…cation from ’natural
experiments’assume a particular structure of unobservables.
D
D
D
εD
mt = ω m + δt + umt ,
The researcher can control for ω D
m using market dummies, and for δt
using time dummies.
D ) = 0.
The identi…cation assumption is that E(Imt umt
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
37 / 44
Restrictions on covariance-structure of unobservables
Suppose that:
εD
mt
D
D
= ωD
m + δt + umt ,
εMC
mt
MC
MC
= ω MC
+ umt
m + δt
This structure together with restrictions on the serial or/and the
D or u MC , can be exploited to
spatial correlation of the shocks umt
mt
obtain exclusion restrictions and instrumental variables estimators.
We distinguish two cases depending on whether the restrictions are on
the serial correlation of the shock (i.e., Arellano-Bond Instruments),
or on the spatial correlation (i.e., Hausman-Nevo Instruments).
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
38 / 44
Arellano-Bond instruments
Consider the demand equation in …rst di¤erences:
∆
Qmt
Smt
= βX ∆XD
mt
D
β1 ∆Pmt + ∆δD
t + ∆umt
D is not serially correlated over time such
Suppose that the shock umt
D
D
D
that ∆umt is correlated with ∆umt
1 but not with ∆umt 2 , ...
Arellano-Bond instruments. Under this condition, the lagged
endogenous variables fPmt 2 , Qmt 2 , Nmt 2 g are not correlated with
D , and they are potential instruments to estimate
the error ∆umt
demand parameters.
In our example, given that the model does not incorporate dynamics
in demand or supply, the key identi…cation assumption is that shocks
MC in the marginal cost are more time persistent that demand
umt
D .
shocks umt
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
39 / 44
Hausman-Nevo instruments
Suppose that we can classify the M local markets in R regions.
Local markets in the same region may share similar supply of inputs in
the production of cement and similar production costs.
D is not spatially correlated, such
Suppose that the demand shock umt
that local markets in the same region have independent demand
shocks.
Under these assumptions, the average price in region R (excluding
market m):
1
P ( m )t =
Pm 0 t
∑
MR 1 m 0 6=m,m 0 2R
can be used as an instrument to estimate demand parameters.
The key identi…cation assumption is that the MC has spatial
D .
correlation that is not present in demand shocks umt
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
40 / 44
Zero-covariance between unobservables
In simultaneous equations models, an assumption of zero covariance
between the unobservables of two structural equations provides a
moment condition that can be used to identi…ed structural
parameters.
For instance, consider the restrictions:
D
FC MC
E(εFC
mt εmt ) = 0 and E( εmt εmt ) = 0.
These conditions imply two additional moment restrictions that
together with the E(Xmt εmt ) = 0 can identify all the parameters of
the model.
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
41 / 44
Zero-covariance restrictions: Example
Suppose that the unobserved demand, εD
mt , and the unobserved …xed
cost, εFC
,
are
not
correlated.
mt
Under this assumption, the model implies that
1
Smt
Qmt
Nmt
2
is not
correlated with εD
mt .
1
Qmt 2
as an instrument of Pmt in the
Smt Nmt
estimation of the demand. That is, we have two moment conditions
two estimate two parameters:
Therefore, we can use
E
E
Victor Aguirregabiria ()
1
Smt
Qmt
Nmt
2
Qmt
Smt
Qmt
Smt
Empirical IO
β0 + β1 Pmt
β0 + β1 Pmt
= 0
!
= 0
Toronto. Winter 2015
42 / 44
Extensions
The previous model is quite restrictive in di¤erent economic and
econometric aspects. Some possible extensions:
1
Heterogeneity in …rm costs.
2
Product di¤erentiation (vertical or/and horizontal).
3
Forward-looking behavior: dynamic entry and exit decisions.
4
Endogenous costs: a …rm’s investment can reduce its MC .
5
Endogenous product quality: a …rm’s investment can improve the
quality of its product.
6
Product portfolio: e.g.,store locations.
7
Mergers.
8
Collusive behavior
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
43 / 44
An overview of the course (Topics)
The course concentrates on the following topics.
1
Dynamic models of …rms’decisions (under perfect of monopoly
competition)
Capital; capacity; inventories; prices; advertising; stores, ...
2
Dynamic models of consumer demand
Storable and durable products; consumer switching costs;
learning; ...
3
Dynamic games of oligopoly competition
Entry-exit games and deterrence; quality/capacity races;
endogenous product characteristics; networks; ...
Victor Aguirregabiria ()
Empirical IO
Toronto. Winter 2015
44 / 44