Inside Harry Crane’s Prediction Market Research

One of the most prominent academic voices in prediction markets has sat down with Prediction News.

Harry Crane is a statistics professor at Rutgers University. He has a career as a profitable sports bettor and has conducted research into prediction markets for years. His research covers the mechanics of prediction markets and their forecasting abilities.

Crane is a strong proponent of prediction markets, viewing them as superior to polls. Some of his work in prediction market analysis during the election concerned price differences across platforms. His work measuring different margins of inefficiencies was one clarifying piece of research in this noisy field.

Rational market pricing

During the 2024 election, there were certain markets that seemed mispriced. For example, a PredictIt market only gave Joe Biden a 92% chance of winning California, a low probability given California’s solid blue leaning.
“92% for Biden, feels like you could make eight cents for free,” Crane said. “But you can’t, because the fees are so high.”

In 2024, PredictIt had a 10% fee on gross profits and a 5% withdrawal fee. Those high fees came from the no-action letter terms that allowed PredictIt to launch in 2014. High fees meant that traders had to sell their contracts at higher prices to profit as much as they would on commercial platforms with lower fees. In contrast, Polymarket’s U.S. app only charges a 0.1% taker fee.

Some commentators assumed the difference in prices across platforms meant that the markets were irrational or even just not liquid enough. Crane found the opposite to be true.

“Actually, [it’s] a very rational market if you think about it,” Crane said. “Traders are betting based on the economics that are presented to them.” Factoring in fees can lead to different outcomes and reveal consensus across markets that may not be obvious from pricing alone.

Crane has continued busting myths about prediction markets, including a common misconception about prediction market accuracy.

Accuracy vs Calibration

Prediction markets do a good job of aggregating publicly available information. However, there’s a difference between arriving at an accurate price and a contract resolving at the rate its price implies.

A well-calibrated market is one whose contracts resolve at the same rate as their implied probability. If a contract costs 50%, then it should resolve 50% of the time. Research into both Polymarket and Kalshi has found the platforms to be calibrated well.

However, that does not mean the prices are accurate reflections of a likely future. An accurate price is one that gives the true odds of an event occurring. The market may price a coin toss at 50%, but an accurate forecaster should be able to beat chance.

“What actually matters is, when you said 10% in the past and the market was 5%, how often did it happen?” Crane said.

This difference is how traders can tell whether they have an edge. If a trader is right more often than the market, as Crane explained:

“What you will find generally is when you said 10% and the market was 5%, if you’re good, if you’re really good, it probably happened about 6% of the time. So you actually do have an edge, but it’s relative to the market.”

Fans of prediction markets often mistake calibration for accuracy. Some traders are public about the edges they find in different markets, but these high performers are often secretive to avoid losing them.

Crane’s commentary is a must-watch for anyone who wants a closer look at what even fans of prediction markets get wrong.