A Bayesian Dymimic Model for Forecasting Movie Viewers

Abstract

In this study, we explore the issue of how to enhance forecast of the box office sales, an all-time question for managers in the motion picture industry. The conceptual core of our approach is the expected sales. The expected sales of agents in the movie market (i.e. screen managers at supply side or potential moviegoers at demand side) play important role in predicting the actual sales. We pay attention to the uniqueness of the expected sales; it is latent and evolving over time. This leads to a quick sense that incorporating these components into the model is a natural choice and thus critical for proper forecast of the movie for future period. Based on this notion, we proposed a simple DYMIMIC model for forecasting the box office sales. The model based on a simplified intuitive story of movie consumption behaviors, spontaneous demand and socially driven demand, was calibrated and tested on the actual movie data in the United States and other countries. The model allows for evolution of the latent expectation in the simplest way, and it captures both the cross-sectional unobserved heterogeneity across countries and the effects of the sequential releases over countries. Compared to previous forecasting models, the suggested approach offers a simple yet informative platform of the model that one can add variables such as movie attributes and marketing activities.