Predicting Movie Success
Have you ever noticed how some big-budget movies are widely advertised and seem like they will be a guaranteed success, but surprisingly flop at the box office? Or how some smaller-budget movies that you’ve never heard of somehow become famous?
The movie industry is a very uncertain industry – it’s very hard to predict the revenue of a movie based on just its production data. At first glance, this may seem a little wrong; there should be something that helps predicts the movie’s success, like the actors in the movie. If there are famous actors or stars in a movie, shouldn’t the movie be more successful than others?
It turns out that a study (Vany, Walls, Uncertainty in the Movie Industry: Does Star Power Reduce the Terror of the Box Office?) showed very little correlation between the number and popularity of the stars in a movie and its success. Having a movie star doesn’t even guarantee a positive profit. Some movies that had stars did not return their budget/production cost in revenue. A prediction model based on if a certain actor was in a movie was not an accurate predictor of profit. In fact, the study found that profits in excess of $10 million are Pareto-distributed (inverse power law) with an infinite variance. This variance makes profits almost impossible to predict using only actor data.
However, there are indicators to whether a movie will be successful; they’re just based more on the general public instead of only the movie itself. This is because a movie’s success almost solely depends on the audience. The size of the audience is determined by social popularity of the movie. Information cascades are involved in a significant amount of the spreading of social popularity of a movie and the decision making process of a moviegoer. There are many ways by which information can be spread today, including online methods such as social networks and blogs. Some of these online methods are public and can be used to measure the social popularity of a movie. An information cascade can influence a person’s decision to go to a movie. If a person sees many positive blogs or tweets about a new movie, he/she may decide to go to the movie because of that extra information. Therefore, it may be possible to measure the success of a movie using its social popularity.
Another study (Marton Mestyan, Taha Yasseri, Janos Kertesz, Early Prediction of Movie Box Office Success Based on Wikipedia Activity Big Data) showed how Wikipedia activity could predict the box office success of a movie before it is released. Since a Wikipedia page is entirely user-generated, it can serve as a metric of how socially popular the movie is. A simple model of page activity is having a number out of all people who know about the movie at a certain time view the page of that movie before it comes out. A similar fraction out of all people edits that page. Those numbers with other data empirically produced a very good model to predict revenue, with a R-squared of 0.925 as early as a month before release.
The theory of movie profits has the revenue of a movie being unpredictable without any social data due to its distribution shape. If it were not, then producers would try to maximize the profit of a movie and most movies would be gaining positive profit. However, the spread of information in a network on the scale of this world is very hard, if not impossible, to predict today, and so the success of movies remain unpredictable.
http://marshallinside.usc.edu/Mweinstein/teaching/fbe552/552secure/notes/devany%20new%20one.pdf
http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0071226#pone-0071226-g006