A Stress Test for Prediction Markets

Sign up for Over/Under, a newsletter on the latest in prediction markets, gambling, and the evolution of everyday risk.

Brazil’s incumbent president, Luiz Inácio Lula da Silva, was reelected on Sunday night. Not everyone saw it coming.

Visit casino-promo.biz for more information.

For weeks, many U.S. pundits had been suggesting that he would lose, thanks to the growing power of the South American right and the lingering curse of incumbency. This thesis was also plainly supported by recent Brazilian election data: Lula’s main opponent, Flavio Bolsonaro (son of Jair), had won the first round of the election, in early October, by more than 2 million votes, and subsequent polling consistently showed that the right-wing challenger had a slight edge. On both Kalshi and Polymarket, Bolsonaro’s odds of victory were bouncing within the mid-70- and high-80-percent range since the first round—more than a slight edge. Traders were betting on Bolsonaro across markets, even when, in the final days of the race, polling showed Lula gaining ground.

That divergence between the prediction markets and the polls has led to some discussion about whether the markets can be trusted. It’s a critique that keeps coming up: Polymarket traders gave Francesca Hong, a candidate for the Wisconsin governorship, about a 95 percent chance of winning her August primary—right up until the day she lost. But these platforms deal in probabilities, and economists tend to advise against drawing conclusions from a sample size of one. Maybe Lula really did have a one-in-five chance of winning, as the prediction markets suggested, and maybe this was the one-in-five outcome. In industry parlance, probabilistic forecasts are said to be well “calibrated” if their odds correspond with the frequency of actual outcomes, and prediction markets do tend to be well calibrated. But in the case of Brazil, the markets heavily favored the candidate who didn’t win.

That’s the technical explanation, anyway. The problem is that Kalshi and Polymarket have taken great pains to establish themselves as authoritative election forecasters. Both platforms are now running midterms “hubs,” where traders and observers can find data for the upcoming U.S. elections on a single page. Even outside of politics, these executives have consistently pushed the idea that they are running predictive platforms—prediction is in the name, after all. Polymarket CEO Shayne Coplan has said that these markets are “the most accurate thing we have as mankind right now, until someone else creates some sort of a super crystal ball.” This leaves some rhetorical wiggle room—the markets are not crystal balls—but it also sets an expectation of decisiveness and reliability.

When the odds don’t appear to align with the outcomes, these firms can conveniently retreat to the fact that they’re purely dealing in probabilities. This argument is available to them as long as their forecasts don’t read zero percent or 100 percent. Kalshi CEO Tarek Mansour told me yesterday that the “strong drag” toward Bolsonaro may have had to do with Brazil’s electoral history; traders could have been hung up on the fact that a presidential candidate had never lost after winning the first round. Still, the “market priced 20% chance of whatever happened to happen,” he said.

Partly because of how prediction markets advertise themselves, some observers now ascribe them a predictive power that’s similar to polling, or to hybrid forecasts such as Decision Desk HQ, which gathers data from a variety of sources. But there are fundamental differences. Sophisticated polls are a snapshot of a moment in time—a representative sampling of people indicating how they plan to vote. Prediction markets, meanwhile, derive their probabilities from the trading data of users across the world. The kinds of people who typically use these platforms are mostly young and mostly male. Kalshi and Polymarket are banned in Brazil. Is it possible that traders without much knowledge of Brazilian politics, or who had their own political biases, skewed the results?

In this case, it’s hard to know. Recent academic research suggests that prediction markets are still pretty good at producing reliable forecasts even when the demographics are slanted. That’s because, in theory, the incentives of the most active traders are financial rather than political. The idea is that they bet big when they’re sure of an outcome and bet small when they don’t really know what they’re talking about, regardless of what they’d personally rather happen. For example, one Polymarket user who put $55 on Bolsonaro commented, “I LOST MONEY BUT IM SO HAPPY I DID LETS GO LULAAAA.”

Still, the reliability of these markets can depend on the amount of money sloshing around. Mansour denied that Kalshi’s Brazilian-market volatility problem had to do with a lack of trading volume, but volume really only picked up at the very end; as it did, the odds started to change. Something similar played out on Polymarket. As one Bloomberg columnist pointed out, when prediction markets beat the pollsters during the 2024 U.S. presidential election, trading volume on Polymarket was in the billions. This time, the platform saw about $250 million in all.

Forecasting swings like those in Brazil and Wisconsin will almost certainly keep happening. The coming U.S. midterms will be a moment of increased visibility for an already visible industry—these probabilities are now baked into some live election coverage. And once exit polls are in, real-time markets might end up filling the information gap as everyone awaits the final tally. It may not really matter whether people understand the methodological differences between prediction markets and other forecasts, or whether people understand probabilities at all. Most Americans already oppose betting on elections. If Republicans somehow stage a comeback next week despite prediction markets’ forecasts of a blue wave—if these platforms seem to miss at all—these firms may end up losing the little credibility they’ve managed to build thus far.

Possible Futures

In every edition, I’ll share my thoughts on a few news stories I’m watching.

• The NBA announced a formal partnership with Kalshi and Polymarket on Monday, naming the firms its “preferred” prediction markets and giving them access to its data. The league also reached a memorandum of understanding with the Commodity Futures Trading Commission, agreeing to share information and cooperate with its regulations. The NHL already has a league-wide partnership with both firms, and UFC, the MLS, and the MLB have deals with Polymarket; after last year’s betting scandals, the NBA has had a somewhat-fraught relationship with gambling. Monday’s news suggests that the league is leaning in anyway.
• Front Office Sports reports that the NFL told Saquon Barkley, a running back for the Philadelphia Eagles, to divest his equity stake in Polymarket. Different leagues have different rules about whether players can invest in these platforms; the NBA, for example, allows up to 1 percent stakes.
• Kalshi is planning to launch prediction markets for the admission rates of top U.S. colleges. Among them: MIT, Johns Hopkins, Duke, the University of Chicago, and all of the Ivies. Why not? Kalshi invents random markets all the time. Here’s one about a law having to do with Californian goatherds. Both are part of an industry-wide effort to attach markets to much of modern life.

Read full story at source