SOURCE: MIT Sloan School of Management

MIT Sloan School of Management

June 28, 2011 07:30 ET

Larger, More Diverse Groups Offer Better Forecasts, Says MIT Sloan Academic

Even if Individually Accurate, Smaller Groups of Forecasters Fare Less Well

CAMBRIDGE, MA--(Marketwire - Jun 28, 2011) - An MIT Sloan School of Management academic has some surprising advice for those who use professional forecasters to help predict gas prices, housing costs, or other trends: A large and diverse group of forecasters -- even including ones who have gotten things wrong in the past -- produces better results than a smaller number of individual forecasters, even when those individuals have records of accuracy.

"By drawing from a large pool, you are able to collect information from forecasters who are different from one another and can present a fuller range of views and outlooks," says MIT Sloan Senior Lecturer P. J. Lamberson, co-author of a forthcoming research paper in Management Science. "Even if some of the individuals in the large group are not that accurate, their errors will cancel each other out. This law of large numbers logic doesn't hold when forecasts are drawn from small groups."

Lamberson says his work is especially relevant as corporations and other players seek more accurate ways to anticipate trends. "Given advances in information technology, the results have greater practical significance than they might have had even a decade ago," he notes. "Many companies now use prediction markets to aggregate the forecasts of large numbers of people." And, Lamberson says, such variation among a larger group matters more than accuracy among a smaller group.

According to Lamberson, whose research focuses on social and economic systems in which individuals' choices are influenced by the behavior of others, both forecasting and its applications have grown exponentially with technology. "Predicting the future is pretty hard, but we are getting better at 'predicting the present' -- that is, being able to measure what is happening right now," he says. "Google, for example, uses a trend-spotting technology based on search requests and is able to look at people entering search terms such as 'sore throat.' When they see big spikes in such searches, they are able to predict that a flu outbreak may be going on, whereas traditional measurements require waiting for doctors to make reports."

Lamberson cautions that there also dangers of online manipulation or too much unfiltered input from the crowd. "You do run the risk of people trying to game the system and create a self-fulfilling prophecy," he says, citing the example of the 2010 election for Senator Edward Kennedy's seat in Massachusetts. The Democratic candidate had a steady 20-point lead, Lamberson says, "but then there was this Twitter bomb, an automated system sending out a lot of messages that said forecasts were showing [Republican challenger] Scott Brown moving ahead. Though the Tweets looked like they came from individuals, they were from a computer using fake addresses. Suddenly, four times more Twitter posts are now mentioning Brown. Once you start postulating something like that, people join in."

Contact Information

  • For further information, contact:
    Paul Denning
    Director of Media Relations
    (617) 253-0576 office
    Email Contact

    Patricia Favreau
    Associate Director of Media Relations
    (617) 253-3492 office
    Email Contact