Support Long-term Thinking
Support Long-term Thinking

Prediction Market as weather forecaster

by Kevin Kelly on March 22nd, 02007

This is a very short article on how economists are using prediction markets to predict weather at least as good as meteorologists, which is not very good.

Penn State Researchers Testing Futures Markets For Weather Forecasting

UNIVERSITY PARK, PA (March 21, 2007) – Economists at Penn State’s Smeal College of Business and College of Earth and Mineral Sciences are testing whether futures markets can be used to accurately forecast the weather, and, so far, they’ve found the markets to be just as accurate as major forecasting services.

The 60 participants in this predictions market experiment, which is in the midst of a two-year run at Smeal’s Laboratory for Economics Management and Auctions, are mostly students studying business or meteorology at Penn state. They use allotted funds to bet on what they believe the high and low temperatures will be in different U.S. cities on a given day. As the going rates for various temperatures fluctuate within the market, the researchers can weigh the market’s confidence in what temperatures will be reached.

To date, the weather markets have been as accurate as the major public forecasting services that serve as benchmarks for the research: AccuWeather, the BBC, CNN, and the National Weather Service. On average, the temperatures predicted by the markets have been off by only about 6.6 percent.

In addition, the results so far show the market to be far more profitable for traders who are studying meteorology than those who are studying business. This is consistent with economic and financial models that suggest that better informed traders can reap substantial profits from their inside information.

“Investors, political pundits, and sports fans have all used futures markets to successfully predict everything from presidential elections to World Series champions,” said Anthony Kwasnica, associate professor of business economics at Smeal. “Our experiment is along those same lines. By using markets to forecast something as unpredictable as the weather, we’re testing the limitations of futures markets.”

Each participant in the experiment, which is funded by the National Science Foundation, has the opportunity to buy and sell contracts that represent a particular temperature range. For instance, the markets for March 8 were focused on Tucson, Ariz., and participants could buy or sell contracts betting on a high temperature of 74 degrees or less, 75 to 76 degrees, 77 to 78 degrees, 79 to 80 degrees, 81 to 82 degrees, or 83 degrees or more.

Each contract represents a commitment by the seller to pay the buyer $1 if the temperature falls within the specified range of the contract. Participants can base their decisions to buy or sell contracts on any information available to them, including weather forecasts.

In the Tucson example, if a participant thinks that the high temperature will be 81 degrees on March 8, he offers to buy an 81- to 82-degree contract for a price of up to $1, which he determines based on his level of confidence in his prediction. A seller can then accept or reject the offer.

If the offer is accepted and 81 degrees is indeed the high temperature on record at the end of the day, then the seller of the contract pays the buyer $1. Thus, the seller loses and buyer gains the difference between $1 and what he paid for the contract.

If the high temperature is outside of the 81- to 82-degree window, then the buyer loses what he paid to the seller for the contract and the seller wins.

The going price of the contracts, which are listed on a Web site run by LEMA, determines the market’s confidence in the likelihood of a particular high or low temperature being recorded – a high price on a particular temperature reflects a high degree of confidence that the temperature estimate will be accurate. Thus, the markets can provide a measure of uncertainty associated with the prediction, something not typically provided by weather forecasters.

The cities and dates chosen for the experiment follow those used in the WxChallenge, a weather forecasting competition developed at the University of Oklahoma. The market’s forecasting accuracy will ultimately be judged in comparison to the accuracy of the forecasters participating in the WxChallenge. So far, the predictions markets are just as accurate as the WxChallenge consensus forecast.

In addition to Kwasnica, the Penn State researchers leading this study are Gary Bolton, professor of business economics, Smeal College of Business, and Andrew Kleit, professor of energy and environmental economics, College of Earth and Mineral Sciences.

But as Stephen, one of the Freakonomics guys notes on his blog:

Which makes me wonder if the story here is simply that the typical Penn State meteorology student is as good a forecaster as the typical AccuWeather meteorologist. I wouldn’t be too surprised: the national weather map that’s published every day in the N.Y. Times is supplied by … Penn State.

READ NEXT
The qubits entangle
Alexander Rose - Twitter: @zander
Blogs vs. New York Times
Alexander Rose - Twitter: @zander
Decision: Blogs vs. New York Times
Alexander Rose - Twitter: @zander