Your March Madness Soothsayer: Prediction Technologies


“I’ll bet you $50 that an Oregon player will fall on the ground before a Saint Joseph’s player!” said a 40-ish guy wearing a Syracuse basketball jersey to a younger guy with a Utah jacket and beer-stained shorts. They were both waiting in about an hour-long line to make March Madness bets in Las Vegas. “I’m here to bet on Syracuse but this line is taking forever,” said the Syracuse fan. “We might as well make it interesting and place a couple more bets while we’re waiting in line watching the game, right? So I say an Oregon player trips and falls on the court before a Saint Joe’s player. You in?”

People will bet on virtually anything, and not just during March Madness. Will Donald Trump win the Republican primary? Will the Republican Party throw him under the bus if he wins? If Trump were to be elected, would the United States look like Hill Valley in Back to the Future 2 under Mayor Biff Tannen?

If people could predict the future, we could accomplish a lot more than just knowing which basketball player will screw up first or knowing who will win an election. We could solve some really important world problems, like preventing the spread of HIV, stopping drug abuse, and preventing crime. But even though people love making predictions, and betting on anything and everything, they’re actually pretty bad at it. What, then, can be done to help make better predictions about the future?

In one study, doctors and nurses were given the opportunity to bet on whether they thought a flu outbreak would occur. They bought and sold actual contracts over the Internet and would make or lose money based on whether they were correct. This “prediction market,” or a stock market where you bet on events, was built by researchers at the University of Iowa to test whether healthcare workers could forecast an outbreak of seasonal influenza.

The idea behind the study was that, just like people in the stock market bet on the perceived future value of Apple, Google, or Coca-Cola, the market participants in this “influenza market” would bet on the perceived future likelihood of an outbreak of the flu. And just like the current stock price of Apple or Google or the price of oil futures contracts, the current price of “the flu” would give the best estimate of how hard the flu might hit that season.

Here’s how it worked: Each trader started with $100 in their account. They would be able to buy or sell the likelihood of a Centers for Disease Control and Prevention (CDC)–reported flu outbreak hitting within 8 weeks. The prediction market was open 24 hours a day, and prices were updated in real time and visible to all participants. At the end of the 8 weeks, the market would close and report the flu outcome based on data from the CDC. People who had bet correctly would make money.

Because betting is illegal, at the end of the flu season, instead of giving the participants cash, the money in the “winners'" account was converted into an educational grant. This was a way of incentivizing people to take the bets seriously using real money, but preventing the market from being an illegal betting site.

So how did the market perform? Compared to using historical data to predict the flu, the market did a lot better. It reached about 70% accuracy in predicting the CDC flu statistics. The study had some limitations (e.g., a small number of traders did the bulk of the trading; the pool of traders was from a small area), but overall the market successfully predicted flu outbreaks 2 to 4 weeks in advance.

In recent years, specialized trading prediction markets like this influenza market have been used to help predict the outcome of uncertain events. The University of Iowa markets, which started in 1988, has accurately forecasted the sales of computer products and events in popular culture such as the Oscars. For elections, the prediction record of the Iowa model has been significantly better than standard statistical models.

So how can we improve our ability to predict events in order to solve real-world problems? One way is to incorporate the science behind prediction markets. While there are existing markets that can be used for prediction like Iowa’s Electronic Markets and the U.K.-based Betfair, the general principle of crowdsourcing people’s guesses can be incorporated into almost any product or service in order to improve predictions. For example, a product could crowdsource bets on how much exercise people get and use that information to help intervene. Prediction markets and crowdsourcing are just a few tools to help with prediction.

Our Institute is developing a bunch of these “prediction technologies," with the goal to bring together leading experts across disciplines to create digital technologies that solve important real-world problems.

But back to what inspired this post: college basketball. Even with the availability of accurate predictive tools, there are always surprises. For example, take a look at this tweet that showed a near 100% winning chance for a team that was leading by 12 points with 44 seconds left in the game. It turns out, the team lost in overtime. (Watch the video—it’s amazing.) So even with the best tools, we can’t always be accurate, but I’m confident that as time goes by prediction technologies will help us solve more and more difficult problems.

Sean Young PhD

UCLA Center for Digital Behavior, Medical Plaza, Los Angeles, CA, 90024, United States

Sean Young, PhD, MS is the Executive Director of the UCLA Center for Digital Behavior. I'm a scientist, innovator, and UCLA medical school professor. I study the science behind human digital behavior (see for more info about this field of research).I also assemble technology teams and solutions to improve UCLA Family Medicine patient care. For more info or to contact me: