Kalshi says 479 of 487 midterm predictions were right, but critics question prediction market accuracy

With the midterm elections finished, Kalshi has pointed to its results as evidence that its markets work: the platform said 479 out of 487 predictions made by traders on its venues turned out to be correct. According to the report, those outcomes were being signalled before any votes were cast, which the source says also underlines the risk of third parties abusing such information for personal gain.

The figure comes with caveats, however. Prediction markets are frequently wrong, and the source cites several examples. Polymarket gave Aleksandar Vučić, described as the ousted prime minister of Serbia, a 70% chance of winning. He lost to the Student List, a group formed after two years of mass protests, taking only 39.2% of the vote, so the race was not close.

A second example comes from the Wisconsin Democratic governor's primary, where both Kalshi and Polymarket gave Francesca Hong a 95% chance of victory. David Crowley won instead, by less than a percentage point. The article argues that the platforms have made boastful claims about their accuracy that events have repeatedly contradicted.

The report also points to structural weaknesses. Prices are driven by traders' own expectations, and those who trade are not necessarily representative of the wider electorate. One study that examined 2,500 political prediction markets concluded that the markets behaved inconsistently with rational information aggregation, suggesting a possible bias in the trades placed. In practice, the source says, platforms attract many participants who hope a particular outcome will occur.

Platform operators often defend their products with the idea of the wisdom of the crowds, arguing that many opinions together outperform individual experts. The article challenges this by noting that only 1% of traders hold three-quarters of all profits. It adds that concerns about insider trading matter too, since some traders follow so-called whales who place large bets on specific outcomes.

The source concludes that prediction markets are probably a useful indicator of likely outcomes, but not the infallible tool that marketing can suggest.