Progressively Predicting New Product Performance
Lights, camera, action! The movie industry becomes a marketing model for forecasting sales performance.
Research by Pradeep K. Chintagunta
"Sure things" flop. Small-time films soar. Despite the best efforts and big budgets of the studios, marketing remains an inexact science, and movie moguls can't always predict how their products will fare at the box office. New research by a University of Chicago Graduate School of Business professor can help clarify their marketing departments' cloudy crystal balls.
One big challenge of predicting new product performance in global markets - in this case, movies - is the overabundance of competing information sources. Additional challenges arise when products are launched sequentially. Information becomes available at different points and requires new forecasts to be generated at each stage.
Pradeep Chintagunta, a professor of marketing at the University of Chicago Graduate School of Business, and Professor Ramya Neelamegham from the University of Colorado at Boulder, examined these issues in their paper, "A Bayesian Model to Forecast New Product Performance in Domestic and International Markets."
Using the movie industry as a framework, Chintagunta developed a statistical approach that forecasts what movies will do well in which countries, particularly during the first week of release.
"The movie industry is ideally suited to studying whether information in one market can be used to predict success in other markets," says Chintagunta.
Furthermore, Chintagunta's methodology can be applied to almost any industry that uses estimates. In the domestic market, the model could predict launches in different states or regions, perhaps to forecast video sales of a movie using data from the stage performance. Internationally, it can be used in any product category where there is a sequential launch. Examples include Citibank's sequential launch of credit cards in the Asia-Pacific region and the launch of the nicotine patch first in Ireland and then the United States.
The Making of a Successful Model
Chintagunta's model works as follows. Movies are first released in the U.S. and in Canada before they are in other countries. As a result, information on the performance of a new movie in these regions can be used in making forecasts for overseas markets. Information from the U.S. and Canada is combined with historical performance of previously released movies in those international markets to arrive at a forecast for new movies in those markets.
Choosing 35 movies, Chintagunta analyzed their data for the United States, Australia, Brazil, Canada, France, Germany, Italy, Japan, Mexico, the Netherlands, Spain, South Africa, Sweden, and the United Kingdom. The last 13 countries represent more than 80 percent of the overseas box office for American films. In most countries, average first-week viewership almost doubles that of average overall weekly viewership, so he focused on forecasting first-week performance.
Chintagunta gathered much of his data from the domestic and overseas box office receipts reported in Variety Magazine from January 1, 1994 to May 26, 1996. In addition to revenue, the data included the number of screens showing a film, the number of weeks of the run, the rank, the number of first-run engagements, and the name of each film's foreign-rights holder and local distributor. To obtain information on attributes such as genre, the presence or absence of major stars, critical reviews, and Motion Picture Association of America (MPAA) ratings, he turned to the Cinemania Index.
The Making of a Successful Movie
Decision makers in the film industry need forecasts at five stages, so Chintagunta tested his model at each stage. When first evaluating the market, information is linked to a historical database of previous performance releases in international markets. Studios use forecasts at this point to help them pick a movie and the markets for release. After the movie has been produced, characteristics such as genre and stars are known, and managers are looking for pre-launch forecasts to help formulate distribution strategy. The third stage is the pre-domestic launch, at which point Chintagunta incorporated distribution strategy.
This forecast helps executives plan marketing activity over the movie's life cycle and make decisions about the international launch. After the domestic release, domestic performance data can be used to hone the forecast still more. Finally, they do a fifth forecast just before international launch, using all possible information, to fine-tune strategic decisions and assess the competition.
Plugging existing data into his model, Chintagunta determined that the key factor in a movie's success in the United States is, as expected, the number of screens showing it. "This finding seems to apply to international markets as well," says Chintagunta.
Big-name stars have a significant positive impact, especially in
In terms of genre preferences, the United Kingdom, Canada, Australia, and Italy like action movies best; Japan and Korea prefer thrillers; and audiences in the United States, Sweden, Germany, and South Africa are partial to romances.
"What this means," explains Chintagunta, "is that it is possible to create non-geographic groupings of countries for movie marketing based on their relative preferences. The way the groupings came out was somewhat of a surprise."
To test the model's accuracy in making specific predictions, Chintagunta developed first-week viewership forecasts for 10 movies: The Net, Waterworld, Dangerous Minds, 12 Monkeys, Heat, Get Shorty, Seven, Toy Story, Jumanji, and Nine Months. Comparing his predictions to actual viewership, the mean absolute percentage error (MAPE) turned out to be 45.2 percent for the United States, 44.5 percent for Canada, and 43.3 percent across all international markets.
The markets with the largest MAPEs were Brazil at 69 percent and the United Kingdom at 64.5 percent. Chintagunta's estimates were much closer for Japan and Germany, at 21 percent and 22.4 percent, respectively. Admittedly, these estimates could stand improvement.
"One of the reasons for the current performance levels," explains the author, "is that we do not have access to information on other variables influencing movie viewership, such as advertising, production budgets, and release schedules."
Compared with other marketing models, however, Chintagunta's is a great improvement from the guessing game currently played by the studios. Because Chintagunta's approach not only combines different types of information sources, it also works with incomplete information, allowing for predictions early in the movie-making process. As more information becomes available, accuracy improves. Finally, the prediction it gives is an estimated range, rather than a specific point estimate, and more accurately reflects forecasting errors.
Chintagunta's research, then, should prove useful to movie executives who are planning and finalizing their marketing activities and negotiating contracts.
"Our results underscore the theme that each movie is unique, as is each country-and viewership results from an interaction of the product and the market," Chintagunta writes. "Hence, the motion picture industry should use both product-specific and market specific information to make new movie performance forecasts."
From Movies to Videos and Even Camcorders
So as the industry struggles to understand and predict sales of new movies in domestic and overseas markets, Chintagunta's research is much needed. In fact, says Chintagunta, some of the decision makers he interviewed in his research have asked him to implement customized models.
Chintagunta, however, is moving ahead to new areas of study. Applying his methods to technology products such as digital cameras and camcorders, the wide range of products his model can be used for is clear.
Today more than a hundred models are available in the market. So as new products are rapidly introduced, firms need to come up with forecasts for these models.
"Technology products seemed to be a natural next step," he says. "If one looks at a category like digital cameras, for instance, we started off with just a couple of models a few years ago."
The work is complicated, he notes, by the fact that technology product attributes are evolving constantly and the attributes must be determined before forecasting sales performance.
In addition, because technology products such as digital cameras are a relatively new category, consumer preferences are evolving as well. Looks like Chintagunta will be busy for some time to come. - K.S.
Pradeep K. Chintagunta is Robert Law Professor of Marketing at the
University of Chicago Graduate School of Business.