Forecasting movie attendance of individual movie showings: a hierarchical Bayes approach

Despite the availability of transaction data, most movie theaters nowadays still rely on managers’ gut feeling to decide how many times and when a certain movie will be screened. Eliashberg et al. (2009) suggest that movie theaters could improve their profits by a more data-driven approach such as a movie attendance forecasting model. However, there are two limitations in the model. First, it does not capture both cannibalization and demand expansion effects. Second, it does not accurately access the uncertainty when making predictions for new movies. To address the limitations in Eliashberg et al. (2009), three hierarchical Bayes models of movie attendance are investigated and compared: linear regression model, standard logit model and nested logit model. Hierarchical linear regression model extends Eliashberg et al’s model by accurately assessing the uncertainty in the predicted admissions. The standard logit model captures both the cannibalization and demand expansion effects in a relatively restrictive manner because of the property called independence from irrelevant alternatives, IIA. The nested logit model relaxes the restrictive IIA property and thus better captures the cannibalization and demand expansion effects.