The informational content of prediction markets: evidence from Polymarket Bitcoin contracts

Abstract

Prediction markets have grown from niche research focused platforms into a multi-billion-dollar industry, yet the academic literature on the properties of these new platforms has not caught up. Prediction markets on the touch probability of traded assets are a new form of contract, in which traders bet on whether an asset will reach a specified price level within a given window. They represent a structurally distinct and largely unexamined category. Because these contracts reference a continuously traded underlying with a deep options market, the probability they quote may already be embedded in existing market data. This thesis asks if Polymarket Bitcoin price-target contracts contribute to price discovery, or whether they merely repackage the touch probability already implied by the Bitcoin spot and options markets. The analysis uses 258 monthly barrier contracts traded between October 2024 and March 2026 and proceeds in three stages: a Mincer-Zarnowitz calibration regression, per-contract Granger causality tests at the hourly frequency, and a cross-sectional comparison of Polymarket prices against a closed-form one-touch barrier probability evaluated under both implied and realised volatility. The calibration test cannot reject Polymarket prices are unbiased probability forecasts of their resolution outcomes. The Granger causality tests reveal that Polymarket no detectable predictive content for next-hour Bitcoin returns. A closed-form barrier probability computed from the spot price, strike, time to maturity, and a volatility estimate explains over 95% of the cross-sectional variation in Polymarket prices under both implied and realised volatility, though the joint null of exact equality is rejected. The price tracks the realised-volatility benchmark and sits below the implied-volatility one, so the level departure is consistent the variance risk premium embedded in options rather than independent information about Bitcoin. The three results replicate out of sample on 252 Ethereum contracts. Taken together, the findings suggest that financial-asset prediction markets contribute little information beyond what is already priced in the underlying spot and options markets.