Combining Prediction Markets and Forecasting Contests

Abstract

Two popular methods for aggregating individual forecasts are prediction markets, where participants bet on the outcome of future events, and forecasting contests, where participants are ranked according to the accuracy of their forecasts. Can these methods be used in concert to produce more accurate forecasts? We analyse 1.79 million forecasts on oddsportal.com, a social network for sports tipsters. Tipsters are ranked according to the betting return on their tips. We find that an aggregation of these tips predicts sporting outcomes, after controlling for betting/prediction market prices. Rank-order forecasting contests, even without tangible prizes, are useful tools for eliciting crowd forecasts.