Combinatorial prediction markets for fusing information from distributed experts and models

Markets are a medium for information exchange between buyers and sellers. Prediction markets exploit the information transmission property of markets to improve forecasts of future events. Participants in a prediction market buy and sell assets that pay off if the underlying event occurs. Prices in a prediction market can be interpreted as consensus probabilities for the underlying events. Prediction markets thus provide a promising method to fuse information from a collection of human forecasters and/or computer forecasting algorithms. This paper describes the use of prediction markets as an information fusion mechanism, introduces combinatorial prediction markets as a mechanism for expressing and exploiting dependencies among base events, presents algorithms for performing basic market computations in combinatorial prediction markets, and introduces the SciCast combinatorial prediction market, a combinatorial prediction market for science and technology forecasting available at http://scicast.org.