As polls closed in Brazil last week, online bettors appeared convinced that Flávio Bolsonaro would become president, placing his odds of winning around 80 percent.
They were wrong.
As ballots were counted, it became clear that Luiz Inácio Lula da Silva would prevail. The odds rapidly shifted as traders scrambled to place new bets on the winner.
This year, the power of prediction markets is being tested like never before. Though the platforms, including Polymarket and Kalshi, have been around for several years, they exploded in popularity more recently by allowing bets on everything from who will win the World Series to how much consumers will spend on Cyber Monday.
The platforms are hoping the U.S. midterm elections will be their moment to shine — and help business boom — with a spate of accurate predictions that surpass polls and election forecasts in their ability to anticipate what will happen.
To attract new users and rake in fees associated with their betting activity, the companies have conducted a marketing blitz. Kalshi rented a giant billboard in Times Square, blaring odds about which party will control the Senate. Polymarket produced a TV spot depicting a failing political campaign with the tagline: “Watch the midterms move.”
Producing accurate predictions this year would be a boon for an industry that is facing increasing scrutiny from lawmakers, including over concerns regarding market manipulations and insider trading.
“I think in a week from now, this is going to be a pretty big story,” said Brett Goldstein, a research professor at Vanderbilt University and co-leader of a team that studies prediction markets. “And we’ll see which way the story goes.”
Studies have shown that prediction markets like Polymarket and Kalshi can be highly accurate. That does not mean that the odds-on favorite always wins, however.
Prediction markets allow users to place wagers on the outcome of an event. Much of the betting activity is focused on the outcome of sports, but elections have emerged as a popular category because campaigns take months to resolve and can accumulate millions of dollars in bets from amateurs and pros alike. As users place bets, the price for each option fluctuates, creating an always-moving set of odds that reflect the likelihood that something will happen.
A review of more than one million markets, published in the Proceedings of the National Academy of Sciences this year, found that those odds were “well calibrated.” Strong favorites usually won, but not always.
But researchers believe they are on the brink of developing an even richer understanding of the markets’ predictive power once all the votes are tallied this year. Past findings do not necessarily reflect the platforms as they are today, with far more users and money coursing through the sites.
Dani Lever, a spokeswoman for Kalshi, compared betting markets to the stock market, saying that investors have always traded around elections. “By shifting the focus from what people want to happen to what they think will happen,” she said in an emailed statement, prediction markets can “filter out social media noise and pundit bias.” Polymarket did not immediately respond to a request for comment.
More than $221 million has flooded into markets related to the midterms, up from $92 million in the 2024 presidential race, according to a report called “Gambling on Democracy” from the Anti-Corruption Data Collective, a research group based in Washington focused on identifying and reducing corruption in private investment.
The rising popularity of prediction markets has been met with plenty of pushback over whether they possess any special predictive power at all.
In the 2024 presidential election, prediction markets showed far sooner than polls that Donald J. Trump would triumph over Kamala Harris, according to a study by the researchers at Vanderbilt University. Polls had narrowly but consistently favored Ms. Harris, the researchers had found.
“We need to watch these very closely,” said Mr. Goldstein, who worked on the study. “They may be telling us something closer to the truth.”
The platforms have also had notable misses this year. During this year’s primary elections, prediction markets gave Spencer Pratt, a conservative candidate for mayor of Los Angeles, a 75 percent chance of being on the Nov. 3 ballot. He came in third place during the primary, trailing by nearly 30,000 votes, and did not advance to the general election.
High quality polls released before the vote were divided, with half showing that Mr. Pratt would come in second, and half showing that Nithya Raman, the eventual second-place finisher, would be second.
Pollsters and election modelers have taken special umbrage with the prediction markets, arguing that they mostly capture the beliefs of online gamblers and are otherwise largely reliant on existing polls.
“I don’t think we should be looking at prediction markets at all,” said G. Elliott Morris, an election forecaster who runs FiftyPlusOne, a polling and election forecasting website. “I think we have great direct measurements of public opinion from the polls.”
Mr. Morris said that some races on prediction markets showed signs that bettors believe polls would underestimate Republican support. Polls have made such mistakes in the past — undercounting support for President Trump in particular. But they were more precise during midterm elections over the past decade.
That may help explain a difference between modelers, polls and prediction markets in the Iowa Senate race. Polls have shown a race that is evenly split. Modelers, like Mr. Morris, have favored Josh Turek, the Democratic candidate. Prediction markets, including Polymarket, Kalshi and PredictIt, have favored Ashley Hinson, the Republican.
In most other tight Senate races, prediction markets have favored Democrats, including in Maine, where Susan Collins, the Republican incumbent who has held the office since 1997, trails behind Troy Jackson, a Democrat. Ms. Collins has trailed in every high quality poll taken in October, except one, where she was tied with Mr. Jackson.
In 2020, Ms. Collins had trailed in every poll released during the final four months of her re-election campaign. But on Election Day, she triumphed by more than 8 percentage points over the Democratic challenger — reflecting an overall miss in polling that underestimated support for Republicans that year.
Campaign managers are also beginning to watch prediction markets for signs of how to pivot their messaging.
For years, campaign strategists have studied how to read polls for such guidance. Eric Wilson, a Republican strategist, said that seeing a lot of undecideds in a poll was an instant sign that campaigns should spend more on name recognition, and avoid mentioning their opponent, for example.
Prediction markets might hold similar insights, he said, but strategists are still learning what to do with them.
“In many ways, it’s as valuable a signal as polling,” Mr. Wilson said. “We just don’t know what it’s saying.”
The post Prediction Markets Face Their Biggest Accuracy Test Yet: The Midterms appeared first on New York Times.
