Dynamic linear models for motion pictures box-office forecasting

We use a Bayesian dynamic forecasting model to predict the weekly gross box-office of motion pictures, particularly trying to replicate the "Top 10" charts for results at the box-office. We use multiple regression to estimate means of the prior distributions of the parameters of the exponential decay models used to model the revenue stream. We then use Dynamic Linear Models (DLM) to dynamically update the forecast as data is gained on the stream of revenue of the different movies. We compare the results of the DLM with those of an exponential smoothing model with trend, and the results from a "complete recalibration" method. We also use the "attraction model" to fine tune our Top 10 predictions and account for seasonality and competition. The use of our model need not be restricted to movie box-office forecasting. Indeed, the model can be applied to many instances in the new product introduction framework, and could also be used for inventory control. We formulated the model so that it could be used as a component in an optimisation model, within the framework of MARKOV Decision Processes.Business, Sauder School ofGraduat