WASHINGTON, Nov 4 (Reuters) – Prediction markets largely got the big picture right in the US midterm elections, although a few races delivered surprises as Democrats seized control of the Senate and House of Representatives.
With billions wagered on Tuesday’s elections, the results offered a test of whether prediction markets could translate their odds into accurate electoral forecasts.
Bets placed on a Democratic midterm sweep largely paid off despite some mispriced races, as tens of millions of voters delivered a stinging rebuke to US President Donald Trump.
“Prediction markets priced the midterm outcome within hours,” said Kelly Ye, head of research at digital asset trading platform Talos.
Several closely contested Senate races had attracted millions of dollars in bets, putting prediction markets under scrutiny. In many cases, the results tracked expectations, with Democrats winning Senate seats in Texas and Ohio.
“This was the first true prediction markets election,” said Harry Crane, professor of statistics at Rutgers University.
“The markets projected Democratic wins across the board throughout the closing weeks of the campaign, and were quick to react in real time to voting information as it rolled in,” he added.
Prediction markets rose to prominence as they diverged from traditional opinion polls by signaling a Trump victory in 2024, and analysts were scrutinizing their performance this week.
According to prediction market Kalshi, traders led the way in predicting that Democrats would take control of the Senate in this year’s elections, forecasting this outcome as early as April. The company’s own assessments of its accuracy showed that on the whole it was more accurate than other leading forecasting models, according to statistics it released Wednesday morning.
However, both prediction markets and polls appeared to face difficulty on Tuesday night in forecasting some of the same races: In the first prediction market upset of the night, Democrat Annie Andrews of South Carolina bested Republican Darline Graham, who had been favored to win with market odds of about 70%.
Likewise, Texas Democrat Gina Hinojosa defied similar odds to unseat the state’s incumbent Governor Greg Abbott. Meanwhile, as of Wednesday morning in Mississippi, Democrat Scott Colom was a hair’s breadth ahead of incumbent Republican Senator Cindy Hyde-Smith, whom markets had heavily favored to win.
The analytics firm BETINT also pointed to prediction market upsets in House races in Indiana, Ohio and Missouri, where Democrats with market odds of 25-40% ultimately won. Election markets for races in the US House of Representatives often have lower volume, which experts say can make their forecasts less reliable.
Those upsets also followed the high-profile market miss last month in Brazil’s presidential election.
Prediction markets allow users to buy and sell contracts tied to real-world events, with prices reflecting traders’ collective expectations. Critics, however, say those odds can be skewed by speculation, thin liquidity and concentrated bets.
Kalshi’s total trading volume on election markets reached over $2.5 billion this cycle, compared with $1.3 billion during the 2024 election cycle, which included the presidential race.
Traders on Kalshi were assigning a 55% probability to a “blue tsunami” as early as April. Meanwhile, Polymarket’s election forecasts were about 98% accurate, a person familiar with the matter told Reuters.
“Prediction markets can be useful because they aggregate the views of people who have an incentive to be right, not just to express an opinion,” said Brian Jacobsen, chief economic strategist at Annex Wealth Management.
“They can give a good indication of where the weight of information is pointing, but a prediction isn’t a promise. Whether it’s politics or sports, you still have to hold the election or play the game to know the result,” he added.
(Reporting by Douglas Gillison and Michelle Price in Washington and Manya Saini in Bengaluru; Editing by Deepa Babington, Lincoln Feast, Alexander Smith and Aurora Ellis)
Comments