Views On BART Movie Prediction Model

I spent the last weekend reading carefully the paper by three professors discussed in this blog post a week ago. They use a complicated statistical technique known as the Bayesian Additive Regression Tree (BART) model to work out a prediction of the receipts from the box office for movies based entirely on the script. The main factors that the model assesses are the Genre and the content variables which includes the storyline, kind of conflict all in the 23 content questions form. 

One need not be a studio manager or movie director to immediately understand the implications of this model on informing investment decisions across given scripts that are being considered for funding. I also think that this reiterates that a well-considered model can still beats so-called experts at making this kind of decision. To my surprise the model provides guidelines on possible Box Office receipts when the dollar value of making films varies over time. So the question I would ask the authors is how this time function was accounted for and whether the model can be tweaked to predict the number of tickets sold instead. As I have mentioned in a separate blog post, comparing movies by the number of tickets sold is more accurate when they are made in different time periods.