Guesswork can pay off for box office predictions

The Hollywood studios have refined the art of predicting the opening weekend box office tallies of their ever-more expensive movies although occasionally their estimates are astray, given such variables as the weather, release date, the competition and whether the marketing campaign cuts through.

Given the high stakes, the studios might want to take a look at a predictive tool created by two academics, David Court and American Charles Plott, later developed with Dr. Jordi McKenzie of the University of Sydney – School of Economics.

They say their model has proved to be remarkably accurate and after further testing they hope it will be adopted as a risk analysis tool by distributors and to help guide investment decisions.

“Ultimately we hope the industry will use it,” said McKenzie, who presented the latest findings at the Not Fluffy conference at the Australian Film Television & Radio School (AFTRS) on Saturday. “It would be relatively easy and costless to implement. We just need to prove there’s useful information.”

The tool, codenamed BOP for Box Office Prophecy, involves pari-mutuel information aggregation mechanisms. In layman’s terms, that means combining betting systems used by the TAB with aggregate information collected from a panel of people who provide their ideas, intuition and opinions on B.O. results.

McKenzie says similar ‘prediction markets’ are used internally by giants such as Google, Yahoo, HP and Intel. Likewise, the Hollywood Stock Exchange, in which players use fake money to buy and sell ‘shares’ of actors, directors and upcoming films, has proven to be a reliable predictor of US ticket sales.

Court, Head of Screen Business at AFTRS, and Plott, Professor of Economics and Political Science at the California Institute of Technology, launched BOP in 2006. Each Friday 10-15 players made simulated bets on films that were due to open the following week, the odds fluctuating according to the size of the wagers.

Plott said that trial showed “amazingly accurate predictions” can be made, demonstrating that 10 or more participants are better informed about B.O. outcomes than any individual.

In 2010, McKenzie joined the project for BOP2, extending the forecasts to two-four weeks before release and up to six months in advance. Members of a panel drawn from the AFTRS, distributors and exhibitors were given a pot of make-believe money to buy tickets predicting B.O. outcomes.

Last month the team completed a third iteration, BOP3. “We are extremely pleased with our results,” reported McKenzie. “Although the mechanism sometimes misses the mark, statistically speaking, there is strong evidence that the panel possess solid information which can be aggregated to provide meaningful predictions of B.O. outcomes.  Moreover, we are able to quantify the uncertainty surrounding B.O. outcomes because our predictions are in the form of a probability distribution and not a single point prediction.” 

This initiative differs from a study undertaken by researchers at Tottori University in Japan, who plotted the number of times a movie was mentioned in blogs and on Facebook pages from several weeks before the release of movies such as Spider-Man 3 and The Da Vinci Code, until the end of the first run weeks or months later. Combining the social media data with figures on the daily advertising spending, they said they “can predict the [daily] revenue of the corresponding movie very well."

One of America’s most successful producers, Ryan Kavanaugh of Relativity Media, uses a form of statistical analysis known as ‘Monte Carlo’, which measures the probability of a film’s success based on past scenarios with similar variables including talent rank, release date, genre, budget, B.O. results and home-video performance. Kavanaugh says the formula is only one tool he uses but “if the results come back showing that a certain percentage of the time we’d lose money, then it is too risky.”