taken with respect to film, cinema, time, and individual characteristics. This set of
moment conditions thus places discipline on the model’s predicted conditional attendance
probabilities of different demographic groups.
Defining ˆg(θ) = [ ˆg
1
(θ) ˆg
2
(θ)]
0
as a vector of sample equivalents of our population
moment conditions, we can write our GMM estimator as
ˆ
θ = arg min
θ
G(θ) = ˆg(θ)
0
ˆ
Φ
−1
ˆg(θ), (12)
where
ˆ
Φ is a consistent estimate of E[g(θ)g(θ)
0
]. Intuitively, the weighting matrix, Φ
−1
gives less weight to moments with higher variance. Because we include the film fixed
effects, φ
f
, in equation (1), our GMM estimator does not identify the role of time-invariant
film characteristics in consumer choice. Following Nevo (2001), we perform an auxiliary
regression to recover these additional parameters.
An important component of the empirical strategy is the choice of instrumental vari-
ables. A great deal of intertemporal price variation stems from the common cinema
practice of offering discounted ticket prices on Tuesdays. We include a dummy variable
for the cheap ticket day in our instrument set. For most cinemas in our sample the cheap
ticket day is a Tuesday, for a small minority of four it is a Monday, and for a single theatre
it is a Thursday. Average attendance is relatively constant during the week with the ex-
ception of Fridays, weekends and opening days. We include dummy variables for Friday,
Saturday, Sunday, and opening day in our set of explanatory variables. Effectively then,
our maintained assumption is that the choice to offer cheap tickets on Tuesdays instead
of Mondays, Wednesdays, or Thursdays, is unrelated to demand conditions. BLP suggest
that rival product characteristics may provide useful instruments. Davis (2006) considers
the characteristics of rival theatres within five miles of the theatre, such as consumer
service, DTS, SDDS, Dolby Digital, Screens, THX, weeks at theatre, first week of na-
tional release, and local population counts (of different definitions). Accordingly, we also
include a range of other instruments which relate to i) the characteristics of the nearest
rival cinema including number of seats, number of screens and distance from the reference
cinema; and ii) the characteristics of all rival cinemas within a certain distance of theatre
h (e.g. total number of cinema screens, seats, or shopping centre theatres within [0,5],
and [0,10] kms of h).
For our additional moment conditions, we use information about attendance rates
conditional on age and income. In particular, we match attendance rates for the age
brackets {15-24}, {25-34}, {35-49}, and {≥ 50}; and the weekly income brackets {<
400}, {400-600}, {600-800}, {800-1000}, {1000-1300}, {1300-1600}, {1600-2000}, and
{≥ 2000}, where all figures are in Australian dollars.
We close this section by briefly discussing the nature of variation in our data that
identifies our parameter estimates. In principle, we can exploit time-series variation,
cross-section variation within the greater Sydney market, and, because consumers face
transport costs, some variation between local markets within Sydney. In practice, the
variation in price takes a restricted form. The primary source of time-series variation is
the common practice of offering cheap Tuesday tickets. There is very little other time-
series variation in price, with a small number of small theatres offering cheap tickets on
Mondays instead. This time series variation allows identification of the average price
14