Prediction without markets

Abstract

Prediction markets are frequently advocated for corporate and government decision-making due to their theoretical efficiency and ability to aggregate diverse information. However, a comparative analysis of thousands of sporting events and movie box-office outcomes reveals that the performance gains offered by these markets over conventional methods, such as polls and simple statistical models, are surprisingly small. In professional football and baseball, prediction markets provide only marginal improvements in accuracy—measured by squared error, calibration, and discrimination—relative to models utilizing just two or three parameters. For example, the Las Vegas market for professional football is only 3% more accurate than an elementary statistical model and 1% more accurate than a simple poll. These domains exhibit sharply diminishing returns to information, suggesting that fundamental limits to predictability exist regardless of the forecasting mechanism employed. While markets often rank as the top-performing tool, the modest magnitude of their advantage implies that the significant costs of market implementation may outweigh the practical benefits for many policy and business applications. – AI-generated abstract.