Abstract
Prediction markets provide real-time probability estimates that are increasingly used by managers, analysts, journalists, and investors. Using 1.5 billion trades across 779,000 Polymarket markets, we examine when these prices can be trusted as probability estimates. Overall, prices are highly informative and improve with market maturity, with higher trading activity associated with better forecasts, particularly at early stages. However, accuracy varies across categories: weather markets are most precise, whereas mentions and geopolitics markets are least precise. We also find no general market-level longshot bias. Instead, low-probability outcomes are initially underpriced before reversing late in the lifecycle. Default outcomes are persistently overpriced. In sports markets, Polymarket matches professional bookmakers' accuracy. However, live trading shows that rapid information arrival can generate temporary distortions even as overall forecast accuracy improves. Prediction market prices therefore offer valuable probability estimates, but their interpretation should account for when, where, and under what trading conditions prices are observed.

This publication has 32 references indexed in Scilit: