Democrats weren’t the only big winners on election night. Prediction markets celebrated their own victories, accurately predicting most of the races, while raking in billions of dollars in bets.
Polymarket, the largest prediction market in the world, claimed to be 98.4 percent accurate on pre-election eve, with 97 percent accuracy in the Senate and governorship races and 98.6 percent accuracy in the House races.
Kalshi, which sold the most contracts during the midterms, scored 96.6 accuracy for the House and 97 percent for the Senate and governors’ races.
Polymarket and Kalshi incorrectly predicted Darline Graham would win the Senate seat in South Carolina and Greg Abbott would win the governor’s race in Texas. The Mississippi Senate race hasn’t been called yet, but if the Democrat wins it will be another incorrect pick.
Prediction markets work like this: bettors can buy and sell contracts on almost any event, from midterm elections, to sporting events to how many inches of snow New York City will receive this winter. Like the stock market, it’s the buyers and sellers who determine the price. For example, when the polls closed Nov. 3, Byron Donalds’ probability of winning was 66 percent. The buyers who purchased contracts at 66 cents each received $1 per contract if they picked Donalds to win, or a 34-cent profit. A bettor who bought contracts on the 34 percent probability that David Jolly would win received no money.
Betting can continue until four of eight designated media sources have called an election, said Nicole Kagan, head of research at Kalshi.
This was the first midterm election since Kalshi and Polymarket received certification from the Commodity Futures Trading Commission. U.S. citizens did bet on the 2024 presidential election on Polymarket’s international website.
Kalshi received about $2.5 billion in contracts during the midterm cycle; Polymarket received about $600 million. Florida’s governor’s race was heavily traded with $44 million worth of contracts sold on Kalshi and $141,000 on Polymarket, according to their sites. Bettors placed $11.6 million in contracts for the Senate race on Kalshi and a little over $138,000 on Polymarket.
Local congressional races even received bets. Almost $32,000 was bet on the District 19 race between Jim Schwartzel and Victor Arias and approximately $332,000 on the more competitive District 22 race between Pia Dandiya and Casey Askar.
Who was more accurate in calling the races — the prediction markets or the polls — was as much a post-election day debate as why a blue wave swept the country. The comparisons were especially important after studies showed that Polymarket was more accurate than polls during the 2024 presidential election.
Vanderbilt University political scientists found that 95 percent of the contracts were favoring Donald Trump by midnight, several hours before the Associated Press called the election. The study also showed the prediction markets were showing Trump winning, as early as October, while polls were giving Kamela Harris the edge or calling it a toss-up.
A New York Times review of the numbers for last week’s election said the prediction markets did better closer to Election Day but were no more accurate than polls for most of the midterm season.
Kalshi, in its post-election report, said as of 9 a.m. Election Day it was 75 percent more accurate in the close races than poll models.
Three months out Kalshi was incorrect on four Senate races, three gubernatorial races and 25 House races.
Polls and prediction markets differ in several ways. The Vanderbilt study looked at the wisdom of the crowd theory where a large number of people can be more accurate than a single poll. The larger number can see something change faster than one poll.
Prediction markets also differ in that polls ask a person who they will vote for; people betting on the prediction market are putting down money on they think who will win. Also, polls are a snapshot of a moment in time; prediction markets are continuous and can change minute by minute and take into account new information, such as a news story or looking at a poll results.
FGCU political science professor Sean Ewing believes polls and prediction markets often align because bettors look at polls to help them decide how to bet.
This year, however, he said the polls, even though they predicted the outcomes correctly they were massively wrong on the margins.
Nationally they were an average of 5 percent off. Polls were off by 10 percent in Texas.
“It was actually a pretty big polling miss,” Ewing said.
Florida was another big miss, where Democratic candidates Angie Nixon lost by 2.9 percent and gubernatorial candidate David Jolly by 1.3 percent.
“I think Angie Nixon is a great example, where we didn't get as many polls in Florida statewide as we did in some other swings states that looked closer, but I mean, the polling averages had Moody ahead by 6 or 7 points, and certainly Nixon got way closer than that, and I think Jolly too, so, to me, I think the polls are probably more wrong in this election than they've been in any midterm in the past couple of cycles,” Ewing said.
He thought the pollsters were off because they overcompensated for past misses when they underestimated Republican votes. This year they overestimated the Republican vote and underestimated the Democrats.
KALSHI PREDICTIONS 3 MONTHS, 1 MONTH, 1 WEEK AND 1 DAY BEFORE ELECTION
Senate Races
3 MONTHS OUT
30 of 34 races (Wrong Iowa, Kansas, Texas, South Carolina)
1 MONTH AND 1 WEEK OUT
31 of 34 races (Wrong Iowa, Kansas, South Carolina)
1 DAY BEFORE
32 of 34 (Wrong Iowa, South Carolina)
ELECTION NIGHT POLLS CLOSED (Wrong South Carolina)
House Races
3 MONTHS OUT
Correct 384 of 409
1 MONTH OUT
Correct 389 of 409
1 WEEK OUT
Correct 391-409
1 DAY OUT
Correct 395 – 409
POLLS CLOSED
Correct 400-409
Governors Races
3 MONTHS OUT
Correct 31 of 34 (Wrong Ohio, Texas, Wisconsin)MONTH OUT
1 MONTH OUT
Correct 31 of 34 (Wrong Nevada, Texas, Georgia)
1 WEEK OUT/1 DAY OUT/POLLS CLOSED
Correct 33 of 34 (Wrong Texas)
WGCU is your trusted source for news and information in Southwest Florida. We are a nonprofit public service, and your support is more critical than ever. Keep public media strong and donate now. Thank you.
