prediction. Note that the two pricing first-order conditions implicitly define equilibrium
prices as a function of quality, y
t
. Again, following Fan, the derivative of prices with
respect to quality, say dp
t
/dy
t
and dr
t
/dy
t
, can be found via the implicit function theo-
rem applied to the pricing first-order conditions (
3.26 and 3.27). This in turn allows Fan
to make use of first-order conditions for newspaper quality that look forward to the later
equilibrium in prices, without computing the “second-stage” pricing equilibrium.
To complete an empirical specification,
Rosse (1970) assumes that incremental costs
@
C=@q
t
,@
C=@a
t
,@
C=@y
t
ðÞ
are all linear in data and parameters. The three first-order
conditions (
3.26–3.28), plus linear subscriber and advertiser demand equations, then
make up the five linear equations that
Rosse (1970) takes to the newspaper data.
7
Iden-
tification then follows from classic linear simultaneous equations arguments.
Fan (2013)
uses the oligopoly versions of these same equations, modified for a first-stage quality deci-
sion together with carefully chosen functional forms and a more realistic demand side.
BLP suggest estimating the cost parameters via classic Generalized Method of
Moments (GMM) techniques that are closely related to nonlinear (in the parameters)
IV methods. As usual, we need excluded instruments that are assumed to have zero
covariance with the unobservable ωs. The moment conditions used in estimation are
then the interaction of the implied unobservable ωs with the instruments.
To implement the GMM method, given any guess for the demand and cost param-
eters, it must be possible to solve the first-order conditions (
3.26–3.28) for the cost shocks
ω. This is very close to the supply-side estimation technique in BLP, where the first-order
pricing conditions for multi-product firms have multiple cost errors in each equation and
one must show how to “invert” the multi-product oligopoly first-order conditions to
solve for the cost shocks. Conditional on the solution for the ωs, the GMM econometric
techniques of BLP carry directly over to the case of two-side markets.
Note that the first-order conditions are linear in the cost unobservables (with coef-
ficients that are total demand derivatives). A sufficient condition to solve
(3.26)–(3.28) for
the costs shocks is that the first two equations can be solved for the quantity and ad
demand errors, as the third equation can then be solved for the quality error. Since
the equations are linear in the ωs, this solution is unique as long as the appropriate deter-
minant condition is satisfied; here that condition is
@q
t
@p
t
@a
t
@r
t
6¼
@q
t
@r
t
@a
t
@p
t
:
7
Rosse (1970), in the manner of that day, simplifies the problem by simply “tacking on error terms”
(disturbances) at the end of each first-order condition, but this obscures a discussion of exclusion restric-
tions, which ought to be based on the idea that there are various incremental costs on the right-hand side of
the first-order conditions.
113
Empirical Modeling for Economics of the Media