Why Being Right Doesn't Always Pay: New Research Fixes a Broken Link in Prediction Markets

An 80.33% return over a single month, with a Sharpe ratio of 3.35, is the kind of number that would turn heads on any trading desk. It came not from a hedge fund's proprietary model but from a live deployment on Kalshi, the regulated prediction market, built around a research finding that challenges a basic assumption of how these markets are supposed to work.

The assumption in question is simple: if your forecasts are more accurate than the market's implied odds, you should make money betting on them. A new paper, submitted to arXiv on July 7, 2026, shows that this isn't reliably true in practice, and pinpoints exactly why.

Prediction markets exist to turn scattered, individual beliefs about future events into a single price that functions as a collective forecast. Classical market design theory says that accuracy and profit should move together, but only under specific automated market maker (AMM) structures. The dominant real-world venues don't use those structures. Instead, the largest prediction market exchanges run on central limit order books, the same matching mechanism used in stock and futures trading. The paper's authors find that on these order-book-based platforms, forecasters who are genuinely well-calibrated and informed routinely still lose money. Being right about the future and profiting from being right turn out to be two different skills.

Closing the Gap Between Accuracy and Profit

The core contribution of the research is a formal betting strategy that restores the link between forecasting skill and financial return, grounded in strictly proper scoring rules, the same statistical tools used to evaluate whether a probabilistic forecast is honest and well-calibrated. The strategy's sizing and direction depend only on two inputs: the forecaster's own predicted probability and the current market price. When the forecaster's estimate diverges from the market in the right direction, and given enough market liquidity to execute at reasonable prices, the strategy generates positive expected profit.

To test this, the researchers ran the approach against thousands of forecasts generated by AI models. Across that entire sample, the proper betting strategy was the only method that consistently turned forecasting accuracy into trading profit. Other, more naive betting approaches, including ones that seem intuitively reasonable, failed to reliably capture the edge even when the underlying predictions were good.

Why This Matters Beyond the Lab

The live Kalshi trial is the part of the paper most likely to draw attention outside academic circles. A month-long deployment with an 80.33% ROI and a Sharpe ratio of 3.35 is a striking result for what amounts to a systematic application of scoring-rule theory to real order-book markets. It suggests that AI-driven forecasters, if paired with the right execution strategy rather than ad hoc betting logic, can extract value from prediction markets in a way that current order-book designs otherwise suppress.

For platforms like Kalshi, Polymarket, and similar exchanges, the findings raise a structural question: if skilled forecasters are being systematically underpaid for accuracy under current market microstructures, that misalignment could be shaping who participates and how much informational value these markets actually deliver.