Learning about movies: the impact of movie release types on the nationwide box office

Abstract

Major Hollywood studios typically release new movies in North America in one of the two ways, wide release or platform release. In this paper, we investigate how release form affects the demand of a new movie after it is nationally released. In particular, we focus on movies for which the platform release is pre-planned to make the problem tractable. We estimate our model using a sample of Hollywood movies that eventually received nationwide release from 1999 to 2003. Our results show that platform release shifts consumers’ perception of unobservable movie appeal through its first stage performance, which turns out to be a stronger effect than that of advertising. Meanwhile, we find that the demand for platform movies decays faster than for wide release ones after their national release. Using counterfactual analysis, we find that more than half of the platform movies which later went to national release would have earned higher profits if they had been given a wide release.

Appendix: switching platform movies to wide release

In the counterfactual analysis, we compare the profitability of wide release versus platform release for the platform movies in our sample. The decision process of the movie studio is assumed to be as follows. In stage 1, the national release date is set. It is exogenous in our counterfactual analysis, and it is set as the takeoff week of the platform movie. In stage 2, the movie studio decides to adopt a wide release or platform release strategy, given the release date (T _0j ), movie characteristics, X _j , and the unobservable (to the studio) appeal of the movie, $\widetilde{A}_{j}^{u}$. In stage 3, the movie studio decides on the advertising expenditure (Ad _j ) and number of theatres engaged in the opening week. Once the movie is released, the number of theatres in the following weeks is largely determined by box office performance (Krider et al. 2005). In addition, we make the following assumptions:

• A1 On average, $\widetilde{A}_{j}^{u}$ equals the realized actual unobservable appeal of a movie, $A_{j}^{u}$. Specifically, we write $\widetilde{A}_{j}^{u} \sim N\left( {A_{j}^{u} ,\sigma_{{\tilde{A}}}^{2} } \right)$.

• A2 Observed advertising expenditure, number of theatres, and movie life period are assumed to be the equilibrium outcome given X _j and $\widetilde{A}_{j}^{u}$. However, they cannot be determined from the estimated demand model directly because we have no information on such missing factors as the financing ability and risk preference of the studio and the bargaining power of studio versus exhibitors. Instead, we use reduced form regressions to establish the relationships between those equilibrium outcomes and$\left( {X_{j} ,\widetilde{A}_{j}^{u} } \right)$, that is, $\left( {X_{j} ,\widetilde{A}_{j}^{u} } \right)$are the independent variables in those regressions.

Given the above assumptions, we can compare the profitability of wide release versus platform release for movies given their characteristics (X _j , $\widetilde{A}_{j}^{u}$, T _0j . We can assume $\widetilde{A}_{j}^{u}$ = $A_{j}^{u}$ in our analysis to see whether the observed platform releases are optimal.

1.1 Advertising, Number of Theatres, and Movie Life under Wide Release

Before conducting the counterfactual analysis, we have to predict the advertising budget, the length of the theatrical run of the movie (“life” period), and the number of theatres engaged in each week for a platform movie should it switch to wide release. As discussed above, we use a reduced form approach to establish the link between these variables and $\left( {X_{j} ,\widetilde{A}_{j}^{u} } \right)$, which are movie characteristics and perceived unobservable appeal by the studio.

1.1.1 Advertising Expenditure

We assume for a wide release movie, the advertising expenditure is decided by

$$ Ad_{j} = \alpha + \beta_{1} X_{j} + \beta_{2} \tilde{A}_{j}^{u} + \varepsilon_{j} $$

where Ad _j is the advertising expenditure for wide release movie j. Note that we have no observation on studio’s perceived unobservable appeal $\tilde{A}_{j}^{u}$. Instead, we know $A_{j}^{u}$ through our demand estimation. Therefore, we can write the equation as

$$ Ad_{j} = \alpha + \beta_{1} X_{j} + \beta_{2} A_{j}^{u} + \chi_{j} $$

where $\chi_{j} = (\beta_{2} \varepsilon_{{\tilde{A}j}} + \varepsilon_{j} )$, where $\varepsilon_{{\tilde{A}j}} \sim N\left( {0,\sigma_{{\tilde{A}}}^{2} } \right)$. Therefore, χ _j is still i.i.d. normal distribution with mean zero.

1.1.2 Number of Theatres Engaged

We define two regressions for the opening week and the following weeks separately. This is because the number of theatres in the opening week is often determined based on expected demand, while in the following weeks, the number of theatres is driven largely by the last week’s performance (Elberse and Eliashberg 2003; Krider et al. 2005). Therefore, we have

$$ \log (Theatre_{j1} ) = \alpha_{1} + \beta_{10} X_{j} + \beta_{11} \tilde{A}_{j}^{u} + \beta_{12} Ad_{j} + \beta_{13} Season_{j1} + \varepsilon_{j1} $$

for the opening week, and

$$ \log (Theatre_{jt} ) = \alpha_{2} + \beta_{21} X_{j} + \beta_{22} \tilde{A}_{j}^{u} + \beta_{23} Ad_{j} + \beta_{24} \log (BO_{jt - 1} ) + \beta_{25} Age_{jt} + \varepsilon_{jt} $$

for the following weeks, where Season _j1 is a vector of month and holiday variables for opening week and BO _jt−1 is the box office revenue for movie j in the last week. By a similar argument to that for advertising, we rewrite the equations as

$$ \log (Theatre_{j1} ) = \alpha_{1} + \beta_{10} X_{j} + \beta_{11} A_{j}^{u} + \beta_{12} Ad_{j} + \nu_{j1} $$

$$ \log (Theatre_{jt} ) = \alpha_{2} + \beta_{21} X_{j} + \beta_{22} \tilde{A}_{j}^{u} + \beta_{23} Ad_{j} + \beta_{24} \log (BO_{jt - 1} ) + \beta_{25} Age_{jt} + \nu_{jt} $$

where $\nu_{j1} = (\beta_{11} \varepsilon_{{\tilde{A}j}} + \varepsilon_{j1} )$ and $\nu_{jt} = (\beta_{22} \varepsilon_{{\tilde{A}j}} + \varepsilon_{jt} )$, which are i.i.d. normal distribution with mean zero.

1.1.3 Movie life period

The length of a movie’s theatre run under wide release is defined according to the number of theatres engaged. We assume that a movie is out of the market when the predicted number of theatres is less than 95, which is the mean of the observed number of theatres in the last week across all movies in our initial sample.

We estimate the above regression equations using observations of the wide release movies in the sample and therefore predict those variables for the platform movies. In addition, we restrict the number of theatres engaged in the opening week as the minimum between the regression prediction and the actual observation when the platform movies reach national release.