A forecasting system for movie attendance

The main objective of this paper is to develop a system that uses historical data to forecast new movie attendance. In contrast to most models in the literature that consider aggregated prediction or the demand for a cross-section of movies, this paper analyzes the dynamic behavior of attendance at the movie level. The paper considers two alternative models for the weekly adoption or consumption of newly released movies. The Bass (1969) explains adoption through innovation and imitation effects. The Sawhney and Eliashberg (1996) model characterizes the adoption process through time-to-decide and time-to-act effects. The basis of the paper's results is a sample of 117 movies exhibited in Chile between 2001 and 2003. The two models present very similar results. For the Bass model, the innovation effect is greater than the imitation effect; but, in Sawhney and Eliashberg's model, the time-to-act is more significant than the time-to-decide. The sample prediction errors of these models present values between 2.7% and 17.1%, depending on the prediction horizon and the amount of historical data available.