The seemingly infallible reputation of platforms like Polymarket and Kalshi, which had accurately called every state in the 2024 presidential election, has encountered a significant challenge. A recent Democratic primary in Wisconsin saw their forecasts, along with traditional polling firms, dramatically misfire, leading to a reevaluation of their role as definitive prognosticators. While these prediction markets are increasingly integrated into mainstream media and financial analysis, the surprise victory of David Crowley over the heavily favored Francesca Hong has prompted a closer look at their limitations.
Before the votes were tallied, both Polymarket and Kalshi had given progressive candidate Francesca Hong approximately a 95% chance of winning the Wisconsin governor’s primary. This high probability aligned with conventional wisdom and traditional polling. However, when the results came in, Crowley, the underdog, secured the nomination by less than half a percentage point. The immediate aftermath included Polymarket deleting a social media post that had boastfully claimed Hong had a 96% chance of victory, despite her lack of endorsements from prominent Democratic socialists like Bernie Sanders or Alexandria Ocasio-Cortez. This quick retraction highlighted the discomfort the platforms felt in the wake of such a pronounced miss.
In defense of their models, Kalshi’s founders acknowledged the unexpected outcome but maintained that a 95% probability still inherently allowed for a 5% chance of an upset. Kalshi CEO Tarek Mansour articulated this by stating that the underdog had a one-in-twenty chance of winning, a low probability but one explicitly accounted for within the market’s framework. His co-founder, Luana Lopes Lara, echoed this sentiment succinctly: “5% is not 0%.” This argument underscores a core principle of probabilistic forecasting, even as it struggles to assuage critics who viewed the platforms as nearly infallible.
This recent misstep, however, is not an isolated incident for prediction markets, though the probabilities in previous cases offered more room for an upset. Last June, for instance, Kalshi and Polymarket had given reality-TV personality Spencer Pratt about a 75% chance of advancing in Los Angeles’s nonpartisan mayoral primary. He ultimately finished third, failing to make the general election. Similarly, in Kentucky’s 4th Congressional District, Kalshi had assigned Representative Thomas Massie a comparable chance of winning the Republican primary, only for him to lose to a Trump-backed challenger, Ed Gallrein. These instances, while not as stark as the Wisconsin primary’s 95% miss, contribute to a pattern that suggests prediction markets, like any forecasting tool, are susceptible to error.
The credibility of these platforms had soared in the lead-up to and immediate aftermath of the 2024 presidential election. They were widely credited with accurately forecasting Donald Trump’s edge in a race that many traditional polls had depicted as a toss-up. This perceived accuracy led to significant partnerships, including CNN naming Kalshi its official prediction-markets partner and Dow Jones planning to incorporate Polymarket data across its suite of publications, including The Wall Street Journal, Barron’s, and MarketWatch. Such integrations signal a growing trust in prediction markets as a valuable complement to traditional polling and surveys for real-world events.
Yet, the Wisconsin outcome serves as a stark reminder that prediction markets are forecasting tools, not omniscient crystal balls. Statistician and political forecaster Nate Silver, a long-time observer of electoral data, recently weighed in on this perspective. He acknowledged the utility of prediction markets but cautioned against treating them as magical or as a complete substitute for traditional polling in models. His perspective suggests that while they offer a unique lens, particularly in aggregating collective wisdom, they are still subject to the inherent uncertainties of human behavior and political dynamics. The recent events in Wisconsin underscore the ongoing debate about the precise role and limitations of these increasingly influential forecasting platforms.
