Skip to content
Advertisement
Prediction markets and polls mostly agreed in the 2026 midterms, and both missed some races

Prediction markets and polls mostly agreed in the 2026 midterms, and both missed some races

coindesk phone

CoinDesk's review found Polymarket and Kalshi favorites won about 98% of settled races, but Iowa and a few other contests broke the pattern

The 2026 midterms gave prediction markets their biggest public exam in years.

A review in CoinDesk’s State of Crypto newsletter, published November 15, 2026, found that prediction markets and traditional polls largely converged by the end of the election cycle. Both tools pointed to the eventual winner in most of the key races CoinDesk tracked. Both also whiffed on a handful of contests.

The scorecard

According to figures reported by Polymarket and Kalshi, the favorite won approximately 98% of settled races.

Kalshi’s own tally was 479 wins out of 487 races. That leaves eight contracts where the market’s favorite came up short.

On the big-picture question, the platforms delivered. Both Polymarket and Kalshi priced in Democratic control of both chambers of Congress before official results confirmed it.

CoinDesk flagged Wisconsin, Alabama and Iowa as races where markets either split sharply from expectations or underrated the eventual winner.

Advertisement

Where the markets got it wrong

Iowa’s Senate race is the clearest example. Prediction markets favored Republican Ashley Hinson.

Polls, by contrast, described the contest as a toss-up. Democrat Josh Turek won the seat.

In this case the polls were the more honest instrument. A toss-up reading admits uncertainty, while the markets expressed a confidence the result did not back up.

Mississippi offered a different kind of lesson. Polymarket’s contract kept incumbent Republican Cindy Hyde-Smith as the favorite right up until election night.

Kalshi, meanwhile, did not list a contract on that race at all. That detail shows that platforms do not cover the same menu of races, so their accuracy figures are not drawn from identical baskets.

Wisconsin and Alabama rounded out the list of races where the markets stumbled.

A six-year argument that is not over

Prediction markets have been drawing attention for at least the past six years. The question of whether they outperform established polling models has been a live one since the 2020 election cycle.

This year’s results reignited that argument on social media after the votes were counted. Supporters pointed to the near-perfect win rate and the correct call on Congress. Skeptics pointed to Iowa.

The CoinDesk review lands somewhere in the middle. By the end of the cycle, markets and polls were mostly telling the same story, and they shared some of the same blind spots.

What this means

The real test of a forecasting tool is how it performs when the outcome is uncertain. Iowa suggests markets can be overconfident in exactly those moments, while polls that called the race a toss-up arguably described reality more accurately.

Their correct call on Democratic control of both chambers ahead of official results shows they can surface the big picture quickly.

Coverage gaps also deserve scrutiny. Kalshi’s lack of a Mississippi contract means each platform’s accuracy claim rests on its own selection of races, so comparing headline percentages across platforms is not quite apples to apples.

Disclosure: This article was edited by Estefano Gomez. For more information on how we create and review content, see our Editorial Policy.