The day Israel voted in its consequential election this week, Kalshi Chief Executive Tarek Mansour was in Midtown Manhattan defending the accuracy of his platform.

Prediction markets, Mansour said, gave a “better forecast” than public polls and would be more trustworthy than media members at shaping public opinion. After exploding in popularity this year, prediction markets are heading into Tuesday’s midterm elections in the U.S. looking to stake their claim as the antidote to years of failures by pollsters to capture accurate estimates on electorates around the globe.

But results this year are showing big surprises. In Brazil, Serbia and Wisconsin, elections which all polled as high as 90% for one candidate, were upset by their supposed long shots. At both Kalshi and its rival Polymarket, traders appear to have gotten it wrong on the next Israeli prime minister—and three days after the polls closed, they still can’t agree.

Election analysts and pollsters point to the same nagging problem: polling data. Coupled with low trading volume, a close race can appear to be anything but in betting markets. The problem is likely to only get worse for prediction markets, the observers say.

“There’s fewer quality polls, so a lot of what you see out there is garbage,” said Caleb Davies, a prominent trader who’s betting about $250,000 on this year’s midterm elections, but avoiding elections that look too concentrated. “When it’s that lopsided, you’re like, maybe the market knows something that I don’t.”Prediction markets are everywhere this year, largely as a venue for sports betting. But both Kalshi and Polymarket owe some degree of their prominence to the 2024 election when the odds markets’ insistence that Donald Trump would win—in the face of polls saying the opposite—vindicated the upstart companies.

Two years later, election betting is happening at scale. On Polymarket, some 4,300 political markets are available to wager money on. They include the geopolitically seismic, like who will control the U.S. Senate, to a handful of curiosities, like the mayor’s race in Waratah-Wynyard, a rural municipality in Tasmania.

The industry is hanging a lot of hopes on the U.S. midterm elections, where polls have oscillated for months about what party will control Congress and how Trump’s final two years in office will go. Internal polling from congressional campaigns show that control of the Senate will likely be closer than prediction market polls suggest, according to one senatorial aide who has seen the data.

In Washington, D.C., metro stations are abundant with Polymarket ads claiming its electoral markets are 99.8% accurate.

Polymarket has a data partnership with Dow Jones, the publisher of The Wall Street Journal.

In a statement, a spokesperson for Kalshi said that prediction markets aren’t “a replacement, but instead a complement, to polling data.”Bedeviling these markets is some confusion about how they work. Odds on a prediction market are set by traders buying and selling contracts on a candidate. The more users spend buying contracts on one side, the better that politician’s chances appear to be. One research paper from Dartmouth economics professor Eric Zitzewitz found that traders tend to bet more on longshots than they should, lifting prices above their real-world probabilities of winning, but filter out political or demographic biases.

“If you want a prophecy about an election, consult your local astrologist,” said Sam Henry Lazarus, a geoeconomics analyst at the Council on Foreign Relations. “Prediction markets provide probabilities, and people are continuing to struggle to think probabilistically when it comes to markets and certainly elections.”Those odds tend to react to old-fashioned polling, which are a snapshot of potential voters—some random, others less so.

“Polling has gotten harder,” said Charlotte Swasey, the director of analytics at the Searchlight Institute, a Democratic think tank. Gone are the days when calling landlines could reliably find people by home at dinnertime to ask what they thought about so-and-so candidate, and expect an honest answer.

In the U.S., those polling issues were apparent this summer when Francesca Hong, a Wisconsin State Assembly member associated with the Democratic Socialists of America, was the clear favorite to win the Democratic nomination against David Crowley, the Milwaukee County executive. In the final days before the election, traders gave her about a 96% chance of winning. Crowley won.

In Israel, the data issues are complex. Some of the country’s most prominent pollsters are more favorable to Prime Minister Benjamin Netanyahu or opposition parties, potentially skewing their results.

Kalshi traders reacted to these polls by favoring Gadi Eisenkot, a former chief of staff for the Israel Defense Forces, who ran just ahead of “no one” as the next prime minister. Polymarket traders were more bullish on Eisenkot, giving him roughly 2-to-1 odds he would win. (That market had no similar nil option). After voting closed, both those leading positions plummeted. Kalshi’s odds have since favored no winner, while Netanyahu became the most likely on Polymarket. While their rules differ slightly, it is unlikely that both could be right.

Several well-timed trades for Netanyahu from newly created wallets have since come under investigation by Polymarket, according to a person familiar. Bloomberg earlier reported on the investigation.

“The Israeli election is a great lab for all kinds of intersections of technology and democracy,” said Tehilla Shwartz Altshuler, at the Israel Democracy Institute. “We need to start asking ourselves: How come the science of polling has become so, I would say, detached from reality?”For next week, trading in the 2026 election on prediction markets has been substantially smaller than in 2024, which included the presidential election.

Polymarket and Kalshi saw about $400 million in trading volume across all U.S. election markets this month through Oct. 28, according to data provider The Block. That is overshadowed by the same month in 2024, when both platforms collectively passed $2.5 billion in volume, which still stands as the largest month for election trading volume, the data show.

Both Polymarket and Kalshi assign more than 60% odds that Democrats take a majority in the Senate after the midterm elections and more than 90% odds the party takes the House. “There will be some surprises,” Davies, the trader, said.