The whims of the box office can be hard to predict. Will that new J-Lo movie be a Maid in Manhattan (which made $94 million) or a Gigli ($6.1 million)? Will a flick with an initially sad box office run be redeemed by its eventual cult status? (Think Blade Runner, which made only $500,000 more than its budgeted cost at the box office, but entered the sci-fi canon after audiences rediscovered it on VHS.)

A new paper by computer scientists at the University of Iowa crunches the data to pinpoint what, if anything, can predict financial success in the sometimes unpredictable movie business. According to authors Michael T. Lash and Kang Zhao, the key to riches doesn’t lie with casting a popular star. Even the biggest stars don’t have enough pull to bring people to truly bad movies (or even good movies that are badly marketed). What's more, popular actors are usually more expensive to hire. The only reliable predictor of a movie’s success, according to Lash and Zhao? Its director. 

The researchers used data mining and machine learning to analyze 14,000 films and 4,000 actors and directors listed in IMDB and on Box Office Mojo. They focused on whether any significant predictions about a film’s financial success (i.e. whether a movie will take in more money than its budgeted cost) could be gleaned just from knowledge available in the pre-release stages—who's involved, what time of year it hits theaters, what it’s about, etc. 

They found that a director’s gross revenue for past films predicted a movie’s success. A director with a high average revenue was likely to keep making money. By contrast, popular stars were correlated with more revenue for a film, but not greater profits, as films with A-list actors cost more to make in the first place. According to this prediction algorithm, while having George Clooney in your film isn’t a recipe for instant success, tapping Steven Spielberg to direct may be.

[h/t: MIT Technology Review]