CBS finds Kalshi correctly favored most midterm winners, but underestimated Democrats

Kalshi’s election markets correctly favored the winner in 95% of the House, Senate and gubernatorial contests examined, but the network’s analysis also found that traders understated Democrats’ overall performance and showed uneven calibration in closer races.

CBS reviewed 481 congressional and state governor races in which it had projected a winner by 10 p.m. Eastern Time on Monday. It used each market’s final price before midnight Eastern on Oct. 28, seven days before Election Day, and defined the favorite as the candidate with the highest yes-price. On Kalshi, a yes-price represents the market’s implied probability that an event will occur.

The favored candidate won 95% of House contests, 91% of Senate races and 97% of governors’ races. Across the sample, eventual winners had received an average 91% probability. Kalshi traders exchanged more than $2.5 billion across individual candidate, ballot-measure and other election markets, almost twice the volume of its 2024 general-election markets, according to the company.

The high headline hit rate partly reflected the number of races that were not competitive. In the safest group of contests, every market favorite won, even though the probabilities implied that a small number should have lost. Candidates in the likely range also won slightly more often than their quoted odds suggested.

The picture changed in less certain races. Where the favorite was less clearly established, the higher-priced candidate won less often than forecast, meaning the markets did not display a general bias toward favorites in those contests. The analysis combined its lean and toss-up categories because there were too few races in each to assess separately.

There were 23 races in which the market favorite lost. Traders had backed Republicans in every one, but Democrats won instead. Kalshi gave Annie Andrews a 23% chance in South Carolina’s Senate race, one of the biggest upsets in the analysis. It gave Gina Hinojosa a 33% chance in Texas’s gubernatorial contest, which she won over incumbent Greg Abbott.

The aggregate results pointed in the same direction. A week before the election, Kalshi forecast that Democrats would win 56.7% of the congressional and gubernatorial races covered by the analysis; they ultimately won 60.4%. Elisabeth Diana said all forecasters had underestimated Democrats, while Kalshi had done so less than any other professional forecasting model.

The analysis treated each race as an individual observation, a simplifying assumption because results are correlated. Tom Rietz of the University of Iowa noted that if many people turn out to vote for Democrats, Democrats become more likely to win multiple races.

Eric Zitzewitz of Dartmouth College called the markets the biggest and most liquid political prediction markets yet, and said their generally strong calibration was unsurprising. Analysts cited broad participation and access to high-quality information, including polling, as reasons prices can improve: traders who believe an odds price is wrong have an incentive to trade against it.

That does not settle whether markets beat conventional forecasting. A New York Times analysis found Kalshi trailed other forecasting models, including Nate Silver’s, for most of the campaign before improving near Election Day. Koleman Strumpf of Wake Forest University similarly told CNBC that prediction markets performed fairly well but did not outperform the best traditional forecast polls.

The results are consistent with a longer-running research finding that political markets often show favorite-longshot bias, in which traders give underdogs more chance than they merit. A National Bureau of Economic Research working paper covering elections from 1880 through 2025 found no systematic overpricing of left- or right-leaning candidates, but said it would be revised to incorporate the 2026 midterms and Polymarket data.