Box Office Predictions From Robots Find A-List Actors Hurt Film Profits

Robots can now predict when movies will be profitable. Michael T. Lash and Kang Zhao, a pair of researchers at the University of Iowa, published a paper last week saying that they have built a machine-learning algorithm called the Movie Investor Assurance System that can tell when a film is going to be profitable before it even starts shooting. Just give it the cast, the synopsis, and its projected release date.

MIAS uses a base of knowledge it acquired by examining publicly available data for over 14,000 movies released between 1920 and 2013, then a final data-set of 2,506 films released between 2000 and 2010. Both data sets, which excluded documentaries and films that were either unrated or rated NC-17, allowed MIAS to understand the relationships between the actors and directors, the films’ plots and synopses. The publicly available data was sourced from IMDb and Box Office Mojo.

“Based on historical data that's available, we can do a decent job predicting success based on pre-production,” Zhao told International Business Times.

Value Of A-Listers

While some of what MIAS discovered is obvious, it uncovered a few surprises. For example, while a cast with lots of A-list actors does correlate with higher revenue for a film, it does not correlate as strongly with profit. “If you have a lot of stars, you'll probably have a good chance of producing revenue,” Zhao said. “But on the other hand, it's expensive to get those people together.”

Ultimately, MIAS found that the thing that correlates most strongly with profitability is hiring a director who’s made profitable movies before; films made by those directors are more likely to be profitable than films directed by people who haven’t. That’s a big reason why a director like Steven Spielberg can always get a movie made.

Lash and Zhao are not the first researchers to tackle this problem. Because of how much money it costs to produce and market a feature-length film, film studios, financial analysts and other academics have been trying to build models that can predict the financial success of movies for decades. “Every studio, every investor is looking for a magic solution,” said Paul Dergarabedian, senior media analyst at media measurement firm Rentrak.

Yet as Lash and Zhao said that as they surveyed the field, what they found was a number of products that only assessed how a movie was going to do after it had been made. Those tools, they thought, would be of limited utility to people looking to fund a project.

“You can't make the decision to invest in a movie after it's been produced,” Zhao told International Business Times. “Many of the formulas only use information available very late in the process. When we can get that data, it’s already too late for investors.”

Next Stop, Hollywood?

But relying on historical data, especially when dealing with popular culture, is tricky. American moviegoers have gone through several phases, and they may be going through another one – box office revenues are trending in the wrong direction, as more and more people opt for entertainment on their mobile devices rather than in movie theaters, and tastes can change on a dime. “People's ways of consuming movies may change.”

But Lash and Zhao are both confident that MIAS can adapt; as it’s fed new information, it will come to new conclusions that reflect the new knowledge it acquires. “This is a system that can be fully automated,” Lash told International Business Times. “You could write batch scripts to automatically rerun it based on newly acquired data.”

Lash and Zhao do not have any plans for marketing or selling MIAS. “This was kind of a 'can we do this?'” Lash said, “a proof of concept kind of thing.”

Whether they’d put it into use for a giant studio remains to be seen. “I haven't taken any calls in LA or anything like that,” Lash said.