How Accurate Were Prediction Markets on the 2026 Midterms?

By 6:00 a.m. Eastern on November 4, 2026, the morning after the midterms, Kalshi's pre-election favorites had won 479 of the 487 races that had been resolved, about 98%. RotoWire published the tally from prices taken just before the first polls closed at 6:00 p.m. ET on November 3. Democrats won the House and the Senate, which both Kalshi and Polymarket had favored. So the markets had a good night. The question this scorecard tries to answer is how good, and the 98% figure is the weakest way to measure it.

The short answer: markets picked the right party in both chambers and in almost every race, but most of those races were never in doubt. In the competitive ones, favorites won at least as often as their prices implied. The real flaw was direction. Every known surprise went to Democrats, which means the markets shared one error across the whole map.

By the numbers

• 479 of 487: races won by Kalshi's favorites as of 6:00 a.m. ET on November 4, per figures Kalshi reported and RotoWire published. DeFi Rate flagged these as company-reported figures that it had not audited.
• 93.3% and 75.0%: Kalshi's prices for Democratic control of the House and Senate at 6:00 p.m. ET on November 3. Polymarket had 94.5% and 75.5% (RotoWire).
• 14: Democratic wins over opponents who had been odds-on favorites the week before, among more than 400 races with known results, according to analytics firm BETINT. All known upsets went to Democrats (Reuters).
• 31 minutes: Polymarket's claimed median lead over CNN, Fox News and the Associated Press in calling 10 key races (CNBC).
• $2.5 billion: Kalshi's midterm trading volume this cycle, up from $1.3 billion in the 2024 cycle (Reuters).

Why is "98% accurate" a weak test?

Of the 487 races in the Kalshi tally, 360 were House seats. Most House seats are safe for one party, and a market that prices a safe seat at 99% and watches it come in has learned nothing. Joshua Clinton, a Vanderbilt political scientist whose 2025 study of nearly 2,500 markets from the 2024 election found them well calibrated overall, made the same point in a Reuters preview published on Election Day: broad performance measures do not show how markets do in close contests.

A quick hypothetical shows the problem. If 30 of the 487 races had been pure coin flips and a market had missed half of them, it would still score about 97%. Headline accuracy mostly measures how many easy races were on the ballot.

The better test is calibration: whether outcomes priced at a given level happen at that rate. As Covers put it, outcomes priced at 70% "should occur approximately 70% of the time across a sufficiently large sample." A 70% favorite that loses is not necessarily a bad forecast.

Kalshi published its own midterm study covering races called by 6:26 a.m. ET on November 4. It reports directional accuracy rising from 93.8% three months out to 98.2% at poll close, and its Brier score falling from 0.040 to 0.016. (A Brier score is the squared gap between the forecast and what happened; 0 is perfect and a 50/50 guess on everything scores 0.25.) Those are Kalshi's numbers about Kalshi. As of November 10, we found no independent table that sorts 2026 races into probability buckets and checks each one.

One more reason to read the tallies with care: RotoWire's own breakdown lists the House as 356 of 360 favorites winning and labels it 95%. The fraction works out to 98.9%.

How did markets do in the close races?

To get past the safe seats, we took the 10 competitive Senate races with a published price and a final result. Nine come from DeFi Rate's Election Day snapshot of Kalshi: Georgia (Democrat at 97%), North Carolina (97%), Minnesota (91%), Michigan (80%), Texas (74%), Maine (73%), Ohio (72%), Kansas (63%) and Iowa (53%). The tenth is South Carolina, where Republican Darline Graham was favored at about 70% in market odds, per Reuters. Alaska was still unresolved and is left out.

Our arithmetic: if those prices were calibrated, the favorites should have won about 7.7 of the 10 races. They won 9. On this small set the markets were, if anything, too cautious about their favorites.

But look at which way everything broke. The nine favorites who won were all Democrats, and the one favorite who lost was a Republican. Iowa and Kansas had flipped from Republican to Democratic favorites during the final week, and Democrats won both. Straight Arrow News reported that pre-election polls had Democrats winning the national House vote by about 9 points and that they were on track to win it by about 13 (SAN). BETINT found markets and polls did about equally well at picking winners, and that Democrats repeatedly outperformed their market probabilities.

That pattern matters for calibration. Rutgers statistician Harry Crane told SAN that Abbott's loss was not independent of the other races. When one national swing moves every contest at once, a night of elections behaves more like one large bet than 487 small ones. A forecaster comparing platforms after the 2022 midterms reached the same conclusion, treating each cycle as roughly one data point. Our read: 2026 shows markets got the direction right and underpriced its size, and one cycle cannot tell you much more than that.

Which upsets did the markets miss?

South Carolina Senate. Graham was at 72% a week out, according to DeFi Rate, and around 70% before the result, according to Reuters. Andrews won roughly 51.1% to 46.7%. CNBC reported that Andrews became the Kalshi favorite shortly after 7:30 p.m. ET, about 30 minutes after polls closed.

Texas governor. Kalshi gave Gov. Greg Abbott about a 60% chance, SAN reported, and Covers put him near 65% on the morning of Election Day. Gina Hinojosa won by 7.1 points. On Polymarket, her odds went from 39% to 73% within five minutes of the first poll closings (CNBC). One local TV recap printed a Kalshi price favoring Hinojosa. Figures like these depend on the moment the market was read, and DeFi Rate argued that a proper calibration test needs fixed timestamps.

Texas lieutenant governor. Lt. Gov. Dan Patrick lost to Vikki Goodwin, a result DeFi Rate said traders had not forecast.

Mississippi Senate. Election-day markets had heavily favored Republican Sen. Cindy Hyde-Smith, Reuters reported. By Wednesday, Democrat Scott Colom was narrowly ahead in the count, and CNBC reported that markets gave him 90% odds with the result still outstanding.

The misses also predate November. In Wisconsin's Democratic primary for governor, state Rep. Francesca Hong was favored at about 95% and lost to David Crowley (The National News Desk). CNBC wrote that markets gave progressive candidates very high odds in summer primaries in Wisconsin and Michigan, and that in Brazil, Luiz Inácio Lula da Silva had about 20% odds on Kalshi before polls closed and won.

Were markets faster than the TV networks?

On speed, the markets had the clearest win, though the best numbers come from the platforms themselves. Polymarket said its median lead over CNN, Fox News and the AP in 10 key races was 31 minutes. Polymarket CEO Shayne Coplan posted on X that "It's now routinely hours between when prediction markets call it versus the TV networks."

RotoWire reported that PredictIt had Democrats winning the House by 10 p.m. ET, more than three hours before the AP call, and that Kalshi and Polymarket were both above 99% for both chambers by about the same time. Kalshi spokesperson Jack Such told CNBC that the exchange called a race once a candidate's odds passed 99% and trading on the other side had "dried out."

A market call and a vote count are different things, though. The Mississippi race was still being counted while markets priced Colom at 90%. A price reacting to early returns is a fast forecast of the count, and it can still be wrong.

So what does the scorecard say?

Our read, as of November 10, 2026: on direction, the markets passed. Both chambers went the way they were priced at poll close, and in the competitive Senate races favorites won more often than their odds implied. On size, they failed in the same way the polls did, underestimating how far a Democratic wave would reach. On speed, they were faster than television by the platforms' own accounts.

The 98% number is real, but it mostly counts the easy races. Some caution about the sources is warranted too: CNBC discloses a commercial relationship with Kalshi that includes a minority investment, and RotoWire notes Kalshi's media deals with CNN and CNBC. Several of the cleanest figures in this scorecard come from the companies being scored.