Prediction Markets from the Inside: How Accurate are they?
At Kalshi, Rubens Baiocchi works on studying how prediction market prices relate to forecasting. He tries to understand how accurate these markets are at predicting future events such as the next Fed meeting or the upcoming Midterm elections. For instance, one way is to compare prediction markets with alternative consensus estimates. Even more interestingly, prices can be compared with experts (and increasingly AI) who dominate public discourse to determine skill. Some of the worst forecasts in history came from brilliant people, while others built lasting reputations on one lucky call. A fairer test of skill looks across many forecasts and asks whether a forecaster’s confidence matches how often they are right. This question is called calibration. When markets say an event has a 50% chance of happening, does it happen about half the time?
Markets mostly pass that test, but not all the time. For example, one exception is the favorite-longshot bias, in which cheap contracts win less often than their prices imply. Before joining Kalshi, Baiocchi was researching and trading against that bias. The talk closes with several applications of these concepts to various cases in prediction markets applied to politics, elections, and economics. Taken together, these cases show how prediction markets can be studied through many lenses and from diverse fields like information theory, behavioral finance, and political theory at once.
Bio: Rubens Baiocchi is a researcher at Kalshi, the regulated prediction market exchange in New York. He has traded prediction markets since 2024, both personally and professionally at a hedge fund, and his current work at Kalshi studies these markets from an academic angle, with a particular focus on forecasting. Baiocchi holds a master’s from Yale and a bachelor’s from King’s College. Originally from Italy and Slovakia, Baiocchi is now based in New York.