### Empirical estimation of market power and firm

```Empirical estimation of market power and firm conduct
How can conjectural variation models help?
Alan Crawford and Benoît Durand
OFT Seminar
The questions

Can we use conjectural variation models to
a) measure empirically market power?
b) identify the source of market power? Can we use conjectural variation models
to tell whether market power is the result of product differentiation or collusion?

And if not, what else can we do?
2
Empirical conjectural variations models and the conduct
parameter


Goal: apply empirical conjectural variation models to measure market power
and to draw inferences about firm conduct
Method: estimate a parameter, θ, (a.k.a the conduct parameter)
–
Value of θ measures market power, i.e. the wedge between price and marginal cost
–
The conjectural variations theory links some specific value of θ to firm behaviour
3
Empirical conjectural variations models and the conduct
parameter

BUT not all economists agree that empirical CV models allow us to draw
–
–
If 5 firms compete in a market and θ = 0.5, what can we say about firm conduct?
–
If firms are not cooperating (Nash behaviour) θ = 1/5 = 0.2 < 0.5
–
If firms are perfectly colluding θ = 1 > 0.5
–
Does θ = 0.5 mean that firms are colluding? And if so, how bad is it? And how
do firms sustain a collusive price level?
CV models do not provide an economic interpretation for what θ = 0.5 actually
means. But how important is it to rationalise firm’s conduct beyond perfect
competition, Nash and monopoly?
–
Objective: for ruling out perfect collusion or non-cooperative behaviour,
empirical CV models might have a role to play (subject to practical issues that
have to be overcome)
4
Empirical evaluation of the conduct parameter: Calibration (EALI)

The conduct parameter is simply an Elasticity-Adjusted Lerner Index (EALI):
–
–
Pros:
– Simple and quick to calculate
... BUT appearances can be deceiving...
Cons:
– Measurement issues
– Marginal cost is difficult to measure
– Price-elasticity of aggregate demand must be estimated (or inferred)
– Market shares depend on market definition
–
Injudicious implementation can lead to erroneous calibration
–
–
–
E.g. mixing wholesale margins with consumer demand price elasticities may be
inappropriate
No statistical hypothesis testing
– E.g. cannot test whether θ is different from 1
No corresponding formula in differentiated product industries
5
Empirical evaluation of the conduct parameter: CPM

The Conduct Parameter Method (CPM): use econometric techniques to estimate
a structural model (supply relation and demand function) to recover the conduct
parameter
–
–
Pros:
– Modelling a supply relation involves making (explicit) assumptions about consumer
demand, cost function, vertical relationships etc.
–
Estimate (economic) marginal cost
–
Perform statistical test to determine whether level of market power is different from
that of important benchmarks (i.e. perfect competition, monopoly, etc)
Cons:
– Can only be applied in homogenous product industries
– For differentiated product industries, the approach is too demanding in terms of
data (insurmountable obstacle in practice)
– Model mis-specification may lead to biased measures
– The shape of the demand and marginal cost functions may poorly approximate
reality
– The Corts’ Critique: the reality of the industry is poorly approximated by a CV
model
6
Are empirical CV models the only tool?

Empirical application of conjectural variation models considers that firm conduct
can be measured or estimated
–

Critical issue: economic theory does not provide an interpretation for most values of θ
Alternative: why not compare a “menu” of economic models in which firms
are assumed to behave in a certain way?
–
Question of interest: does market power result from collusion?
–
Compare one model in which firms are assumed to perfectly collude (monopoly) with
a model in which firms are assumed not to cooperate (Nash)
–
Too simple? Firms may be behaving in ways so that they set price above the
Nash level but below the monopoly level (but empirical CV models do not allow
us to do much more)
–
This approach should allow the analyst to determine that part of market power
that is due to product differentiation and/or multiple brand ownership (see Slade
(2004) study of the UK brewing industry)
7

Two-step approach:
1.
Estimate a ‘menu’ of models that differ in the way firms are assumed to behave.
2.
Select the model that best explains the data (different methods can be implemented; nonnesting hypothesis testing is one of them)
–
Pros
–
–
Modelling assumptions are clear
–
Flexible approach that can be applied to differentiated product industries
–
Hypothesis testing to determine which model (i.e. which firm conduct) is best supported by
the data
Cons
–
Require specification of different plausible models. In differentiated product industries, this
could mean “many models” → may be cumbersome to implement
–
Estimation of a demand model in differentiated product industries requires a large dataset
and (often) making some simplifying assumptions about consumer pattern of substitution
–
Nagging feeling that the menu of models does not include an adequate approximation to
the industry...
.... BUT is it reasonable to suppose that investigations require the ‘true’ model?
8
Concluding remarks



None of the methods is perfect (though some are less perfect than others)
–
When there is no clear cut evidence about collusion, Competition Authorities should consider
relying on this type of empirical analysis (if they have no in-house economists just hire us!)
–
Obviously, implementation should follow best practice:
–
Model’s assumptions should fit the industry features
–
Perform sensitivity analysis
–
Contrast results with other type of evidence
Two methods to estimate the conduct parameter in homogenous product industries:
calibrating EALI and CPM
–
Calibrating EALI maybe useful as a ‘quick check’ during the initial phase of an investigation but
care must be taken about the quality and the relevance of the inputs!
–
CPM’s ability to formally test hypothesis against industry data may lead to conclude that some
form of behaviour is not consistent with the industry
Menu approach is a viable alternative in both differentiated and homogenous product
industries
–
Select the model (and thus firm conduct) that is most consistent with observed market
outcomes but this method can be relied upon only during in-depth investigations
9
Empirical illustrations


The conduct parameter method (CPM)
–
Genesove & Mullin (1989) study of the American Sugar Trust in the late XIXth century
shows that the conduct parameter θ is close to zero.
–
The threat of European imports might explain the absence of market power?
–
Nevo (2000) study of the US ready-to-eat cereal industry reveals that firm margins are
consistent with Nash-Bertrand
–
Comparison of predicted margin from different models (Nash-Bertrand, coordinated
behaviour) with observed price-cost margin
10
Locations and contact
London
Brussels
The Connection
198 High Holborn
London WC1V 7BD
Telephone +44 20 7421 2410
Email: [email protected]
Bastion Tower
Place du Champ de Mars 5
B–1050 Brussels
Telephone: +32 2 792 0000
Email: [email protected]
The Hague
Melbourne
Lange Houtstraat 37-39
2511 CV Den Haag
The Netherlands
Telephone: +31 70 302 3060
Email: [email protected]
Rialto South Tower, Level 27
525 Collins Street
Melbourne VIC 3000
Telephone: +61 3 9935 2800
Email: [email protected]
Johannesburg
Augusta House, Inanda Greens