Title: Prediction market efficiency: evidence from Kalshi on pricing accuracy, forecasting, and risk

Authors: Tanay Subramanian

Addresses: Department of Applied Mathematics, Brown University, Providence, RI, USA

Abstract: Prediction markets are moving rapidly from academic curiosities to mainstream trading venues. This paper studies Kalshi, the first federally regulated US prediction exchange, using 103,003 resolved event markets across 16 categories as of November 2025. Prices are informative but uneven, since calibration is strong in some domains such as economics, and biased in others, such as entertainment. A simple nonlinear calibration model improves testing accuracy (R² ≈ 0.29; Brier ≈ 0.14), while volume-weighted payoff regressions reveal category- and horizon-specific pricing distortions linked to market frictions. Machine-learning models that treat market prices as inputs, rather than final forecasts, outperform the market baseline using only public features, reducing Brier error by roughly 30% and increasing ROC-AUC from about 0.62 to 0.78. Payoff distributions show negative average returns and positive right-tail skew, driven by outsized returns from a small fraction of contracts, demonstrating lottery-like exposure. Overall, Kalshi resembles a hybrid market combining information aggregation with preference-driven demand for skewed payoffs.

Keywords: prediction markets; Kalshi; market efficiency; calibration; machine learning; risk premia; tail risk.

DOI: 10.1504/IJFMD.2026.157002

International Journal of Financial Markets and Derivatives, 2026 Vol.10 No.4, pp.333 - 359

Received: 02 Mar 2026
Accepted: 16 May 2026
Published online: 14 Oct 2026 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article