The highly anticipated U.S. presidential election is just 22 days away, and polls show an increasingly tighter race. Two national polls released Sunday show an ultra-close race with Vice President Kamala Harris just barely in the lead, while a third poll shows her tied with former President Donald Trump.

But prediction markets tell a different story. PredictIt shows Trump’s campaign in the lead as the 2024 presidential election winner. It also has the GOP winning the presidency and the electoral college margin. As of Monday afternoon, Polymarket, a cryptocurrency-based prediction marketplace, also shows Trump has a better chance of winning the election on Nov. 5.

While prediction markets have been around for a while, they’ve grown in popularity—especially for political events. These betting sites allow traders to buy and sell shares on the outcome of a given event, like whether Trump will get impeached or Biden will leave the presidential race.

Traders then buy shares depending on which outcome they think is more likely. Share prices or “odds” rise and fall depending on demand. If the event happens the way the trader predicted, the contract or “bet” will rise to $1 and pay out—or fall to $0 if it does not.

Prediction markets can be a more sophisticated and accurate way to foresee political events, Thomas Miller, a data scientist at Northwestern University, told Fortune’s Shawn Tully. Miller is renowned for his accurate predictions during the 2020 election, and much of his methodology comes from examining prediction markets.

“Political betting sites are the best at predicting the wisdom of the crowd,” he told Fortune.

While polls tell you about how people responded in the recent past, the odds on sites like PredictIt and Polymarket show real-time sentiment and are better for mapping the future, according to Miller. Plus, polls don’t show who will win the electoral college, while traders on prediction markets can wager on its outcome.

How polls and prediction markets compare

Three national polls released Sunday show Harris’s lead is slimming.

Last month, an ABC/Ipsos poll of likely voters showed Harris with a six-point lead over Trump. But that gap has closed to 50% support for Harris to 48% for Trump. A CBS/YouGov poll shows Harris up 51% to 48%, down from a four-point lead last month. And NBC’s poll of registered voters shows a complete tie between Harris and Trump, both at 48%.

“As summer has turned to fall, any signs of momentum for Kamala Harris have stopped,” Democratic pollster Jeff Horwitt, who conducted this survey with Republican pollster Bill McInturff, told NBC. “The race is a dead heat.”

Meanwhile, Polymarket shows Trump has a 54% chance of winning the presidential election with Harris trailing behind at just a 45.3% chance.

PredictIt’s markets are a little closer. For the market called “Who will win the 2024 US presidential election?” the “yes” outcome for Trump is currently at $0.54 and the “yes” outcome for Harris is $0.50.

“If you wanted a single best predictor, one source of information to know what’s going on in the elections, it’s the prediction market,” Harry Crane, a statistics professor at Rutgers University, told CNN. “Any given piece of information that you might have, such as a poll or an aggregator, or news, or a pundit’s opinion, all of that is being combined in the participant’s markets who have an incentive to get it right and to make money off of it.”

But if prediction markets are as accurate as some experts say, why would there be such a discrepancy with what polls are showing? It comes down to what each is measuring, Ryan Waite, vice president of public affairs at public affairs consultancy Think Big, told Fortune.

“Polls tell us how voters feel right now, while prediction markets attempt to predict what will happen in the future,” he said. “Prediction markets are quicker to respond to new information, such as economic shifts or debate performances, while polls take time to update.”

But prediction markets aren’t perfect, Waite said, explaining that they can be manipulated by someone with a lot of money who wants to sway perception and that they aren’t always accessible to everyone.

Still, polls have issues of their own, including sampling bias, leading questioning, and nonresponses that can also produce inaccuracies. That’s why—at least for now—using both polls and prediction markets can help us understand future political events.

“While prediction markets are gaining traction, they’re not going to replace traditional polling anytime soon,” Waite said. “They provide a complementary perspective. Polls help us understand how different voter segments feel about candidates or issues, while prediction markets can provide a dynamic view of how the broader public thinks things will play out. Together, they give a fuller picture of the political landscape.”

This story was originally featured on Fortune.com