MARKET STRUCTURE AND ENTRY: WHERE'S THE BEEF? - CEPR

+θiS3OWNNBjt+θiS4RIVALNBjt +s(mijt,θSi)+ρi ηj.
In (6), OWN and RIVAL are the numbers of existing outlets in the district ρi determines
the variance share of the random effect, ηj is the market-specific time-invariant error
term, and βi, θSi and ρi are firm specific (vectors of) parameters to be estimated. In
standard fashion, we assume that ηj is i.i.d. normally distributed with zero mean, that
2
cov(ηj,εijt)=0, and that σ ≡ 1.
ε
Population is likely the leading determinant of market size, but because different
age groups may display different tastes for eating hamburgers, we add age group controls
into S(.). We include own and rival outlets in neighboring markets in S(.) since if tastes
are correlated among neighboring markets, firms may use information they learn in
adjacent markets to update their predictions of market size in a particular market. The
coefficient of POPulation in S(.) is normalized to one following BR. The coefficients of
the S(.) variables can then be interpreted as increases or decreases in expected market
size, measured in population equivalents.
We specify variable profits per customer as
(7) Vijt(.) = γi1AREAjt+γi2WAGEjt+θiV1OWNjt+θiV2RIVALjt+θiV3OWNjt*RIVALjt
+θiV4OWNNBjt +θiV5RIVALNBjt.
In (7), γi and θVi are firm specific parameter vectors to be estimated. Incorporating
the set of neighborhood variables into V(.) has two motivations. First, the number of own
outlets in adjacent markets allows a control for possible economies of scale in
distribution. The firms have some regional facilities that may serve several of our
markets. If these operations are characterized by economies of scale, we would expect
θiV4 to obtain a positive value. Insofar as the rival’s operations have this characteristic,
21