Cornell seminar to examine how accurately prediction markets forecast

Cornell University will host Kalshi researcher Rubens Baiocchi on Thursday for a seminar on whether prediction-market prices are accurate forecasts, and how their performance should be measured against experts and artificial-intelligence systems.

The CFEM and UBS AI & Data Research Seminar, titled “Prediction Markets from the Inside: How Accurate are they?”, runs from 6:30 p.m. to 7:30 p.m. on Oct. 29. It is organized through Cornell’s Department of Operations Research and Information Engineering.

Baiocchi’s work at Kalshi examines the relationship between market prices and forecasting. His presentation will consider predictions of events including Federal Reserve meetings and midterm elections, and will close by applying prediction-market concepts to politics, elections and economics.

A central theme is calibration: whether stated probabilities correspond to eventual outcomes across many forecasts. A market that assigns events a 50% chance should, under that test, see those events occur about half the time. The seminar description says prediction markets generally pass the test, though not invariably.

One exception is the favorite-longshot bias, in which low-priced contracts win less often than their prices imply. Baiocchi researched and traded against that bias before joining Kalshi. He has traded prediction markets personally and at a hedge fund since 2024, and now studies the field from an academic perspective.

The talk will frame the subject through information theory, behavioral finance and political theory. It will also address comparisons with alternative consensus estimates, experts and AI systems, rather than treating a single correct prediction as sufficient evidence of forecasting skill.

The subject has also drawn recent macroeconomic research. A January National Bureau of Economic Research working paper comparing Kalshi-implied forecasts with survey and market-based measures found the markets produced well-calibrated, rapidly updating density forecasts for inflation, payrolls and Federal Reserve decisions. It found Kalshi’s rate forecasts 150 days ahead were broadly similar in mean absolute error to professional forecasters, while its expectations for core CPI and unemployment were statistically similar to the Bloomberg consensus.

That paper described Kalshi as the largest federally regulated prediction market overseen by the Commodity Futures Trading Commission. It found a statistically significant improvement over the Bloomberg consensus for headline CPI forecasts, while also identifying real-time distributional forecasts for measures including GDP, core CPI and unemployment.

Baiocchi holds a master’s degree from Yale and a bachelor’s degree from King’s College. Originally from Italy and Slovakia, he is based in New York.