A new academic paper, recently published on the preprint server arXiv, presents compelling evidence that prediction markets can generate more accurate forecasts for tomorrow's high temperatures than leading public weather forecasting systems. The study, titled "Prediction Markets Beat the Weather Forecast on Tomorrow's High Temperature" by Alexander W. Crosier, leverages five years of historical data from the Kalshi exchange to compare market-implied temperature predictions against official forecasts.

Read our Prediction Market Report 2026 here.

The research, made available on September 22, 2026, specifically highlights that prediction markets on Kalshi, where participants trade contracts based on daily high temperatures in seven American cities, consistently outperformed the National Blend of Models (NBM), one of the most accurate public forecasting products. The paper found that at the end of the market's first hour of trading, the prediction market's forecast exhibited approximately 10 percent lower root-mean-square error (RMSE) compared to the best single public product. This advantage was maintained and even slightly increased, reaching 11 percent lower error by the time of the NBM's final posting on the target day.

Detailed Study Findings

The study meticulously analyzed market data from Kalshi, a CFTC-regulated exchange, spanning five years for seven major U.S. cities. By extracting hour-by-hour market-implied forecasts, the researcher was able to directly compare the accuracy of these decentralized predictions against established meteorological models. The findings indicate that in six out of the seven cities examined, the prediction market's forecast proved superior to the most accurate single public forecast.

A key insight from the paper is that the prediction market appears to aggregate dispersed information more effectively and earlier than traditional forecasting systems. The research observed that while both forecasts improved closer to the target day, the NBM's predictions moved four times further toward the market's forecast between its postings than the market adjusted towards the NBM. This suggests that the market often possesses information that public forecasts later incorporate, rather than reacting to public forecast updates itself.

Implications for Forecasting Accuracy

These findings carry significant implications for the broader utility of prediction markets as forecasting instruments. The ability of a market, driven by individual traders, to consistently outperform sophisticated meteorological models underscores the power of collective intelligence and decentralized information aggregation. This suggests that prediction markets could serve as valuable complementary tools, or even superior alternatives in certain contexts, for forecasting a wide array of real-world events beyond just weather. The study reinforces the idea that market mechanisms can effectively distill diverse knowledge into accurate probabilistic outcomes.

Background on Prediction Market Research

The academic community has long explored the potential of prediction markets for their forecasting prowess. Research in this area often focuses on how these markets aggregate information, their resistance to manipulation, and their accuracy compared to expert panels or traditional statistical models. This latest arXiv paper contributes to a growing body of evidence supporting the efficacy of prediction markets in generating robust forecasts, particularly as platforms like Kalshi provide the scale and liquidity necessary for rigorous empirical analysis. The study highlights the ongoing academic interest in understanding the mechanisms through which these markets process and reveal information.

What to Watch

The publication of this preprint on arXiv marks a notable contribution to the field of prediction market research. Future developments may include peer review and formal publication in a scientific journal, which could further validate these findings. Researchers may also explore similar comparisons in other domains where prediction markets operate, such as economic indicators, political outcomes, or technological adoption, to assess the generalizability of these superior forecasting capabilities. The study's insights could also prompt further investigation into how traditional forecasting methods might integrate or learn from prediction market signals to enhance their own accuracy.

Original announcement: arXiv