Actionable Signals in Prediction Markets: Evaluating Public and Private Information through Systematic Trading

Abstract

Prediction markets aggregate the private beliefs of many participants into publicly observable probability estimates. Beyond prices alone, the trading behavior they generate may reveal public and private signals that are actionable for automated trading agents. We study whether these observable market signals reliably identify profitable trading opportunities. Specifically, we evaluate two complementary classes of trading signals: Risk Arbitrage, which exploits periods where publicly available information has effectively determined an outcome before prices fully converge, and Information Asymmetry, which exploits large positions accumulated while market prices still reflect substantial uncertainty. Using the complete history of matched trades from Polymarket, spanning 26,964 markets between January 2024 and January 2026, we evaluate the reliability of these two classes of signals. We find that they exhibit markedly different levels of reliability. Risk Arbitrage identifies substantial opportunities around effective resolution but suffers significant losses from false positives, yielding a net loss of $23.4M despite $15.3M in gross profit. Information Asymmetry is considerably more concentrated, generating $139.3M in gross profit but only $7.8M net, with activity overwhelmingly driven by the 2024 U.S. presidential election. Across both regimes, profits concentrate among a small number of accounts, while position size reflects capital commitment more strongly than predictive skill.