On midterm election night, both Kalshi and Polymarket were quick to announce their successes, claiming they provided precise predictions.
With billions wagered on these prediction markets that enabled traders to place bets on election outcomes and control of the Senate and House, their reliability gained significant attention. Media outlets like CNN and MSNBC highlighted prediction-market odds during their midterm reporting, while artists such as YG and celebrities like Heidi Montag shared their midterm bets on election night.
Kalshi’s recap of the midterm elections revealed that $2.5 billion US was traded on its markets, nearly twice the $1.3 billion US seen during the 2024 election cycle. Polymarket, as reported by Reuters, noted nearly $700 million US in bets placed throughout this election cycle on both its U.S. and international platforms.
Polymarket announced on X that their prediction-market platform was “correct” and asserted that “98.4 percent of confirmed midterm races were won by the Polymarket favorite the day prior to Election Day.”
In a similar vein, Kalshi claimed that candidates favored on its platform led or won 98 percent of races before the polls closed on election day.
Valerie Cross, who leads emerging verticals at NEXTPredict.io, which observed Kalshi and Polymarket during the election night, remarked that while those statistics may not be entirely inaccurate, “it’s easy to misinterpret.”
She further noted that these figures illustrate how prediction markets effectively incorporated “early returns, exit polls, and county counts swiftly, and mostly moved in the correct direction.”
“They were also early to indicate surprises, as raw vote totals can sometimes make a race appear closer than the odds suggest,” she explained.
Predicting a blue tsunami
Although the platforms managed to accurately predict many winners, they fell short in forecasting some of the evening’s most unexpected outcomes.
Betting Intelligence’s analysis of over 400 races revealed 14 upsets, all attributed to Democratic candidates.
One notable upset occurred in the South Carolina Senate race, where prediction markets had favored Republican contender Darline Graham, giving her a 70 percent chance, before Democrat Annie Andrews unexpectedly emerged victorious.
Other surprising results included Democrat Gina Hinojosa defeating incumbent Republican Governor Greg Abbott in Texas, with markets assigning Hinojosa a mere 40.1 percent chance the day before the election.
The firm concluded that both polling and market predictions “identified the eventual winners with equal frequency, while both underestimated Democratic candidates.”
David Szakyoni, co-founder of the Anti-Corruption Data Collective, which evaluates prediction markets using publicly accessible data, noted that the markets “seem to react to vote counts and polling information, rather than providing unique insights.”
Rajiv Sethi, an economist at Columbia University, mentioned that neither polls nor prediction markets successfully “predicted the magnitude” of Democratic victories on election night.
He cited instances where both polls and markets failed to forecast outcomes, such as the very tight Mississippi Senate race.

(CBC)
As of Thursday afternoon, Democrat Scott Colom was leading Republican Senator Cindy Hyde-Smith by 1,803 votes, with The Associated Press yet to call the race.
Neither the polls nor the markets anticipated the razor-thin margin between the candidates; however, Sethi indicated that prediction markets offered Colom a slightly higher chance than polling suggested, citing their capacity to respond to candidate missteps or successes during the campaign.
“[Polls] fail to account for truly unusual occurrences that markets can identify. That’s an advantage of the market,” he stated.
Yet, despite their quicker reactions, markets still depend on polling data for their predictions.
“The markets react to the models. Without those models, the markets would be pricing things very differently.”
Before the networks
Cross concurred that markets should not always be compared to polls.
“They assess different elements at various times and are more interdependent than headlines might suggest.”
However, she asserted, “For an initial gauge on who is likely to win on Election Day, prediction markets remain unparalleled.”
“Traders integrate all available information into their predictions,” Cross noted. “At 8:12 p.m., with approximately 31 percent of votes counted in South Carolina, Andrews was ahead with 52 percent to 46 percent, yet NBC declared it too early to make a call.”
Polymarket CEO Shayne Copland stated on X, “There’s now often a significant gap of hours between when prediction markets make a call and when television networks do.”
Kalshi CEO Terek Mansour echoed a similar sentiment, stating, “For decades, election night meant tuning in to the TV and waiting for a select few to inform you of the results. Prediction markets revolutionize that.”
According to Betting Intelligence, in most closely contested races, prediction markets reached a 97 percent probability for the eventual winner about an hour before The Associated Press announced the outcome. None of the candidates that achieved that threshold were defeated.
Despite these early predictions, Szakonyki expressed skepticism regarding the added value prediction markets contributed to the electoral discourse during this cycle.
Cross, however, emphasized that the trading volume clearly indicates a high level of interest.
“Midterm elections typically garner a fraction of the attention that presidential races do. If a midterm can generate these figures, the 2028 elections could be significantly larger.”



