This is an edition of Over/Under, a newsletter on the latest in prediction markets, gambling, and the evolution of everyday risk. Sign up here.
Democrats aren’t the only ones celebrating this week. Prediction markets—high-speed betting arenas where users can wager on the future of the republic—have been claiming a win of their own.
The platforms foresaw a Democratic sweep, and that’s what they got. “BREAKING: Polymarket was right,” one post on the company’s X account read yesterday morning. Both Polymarket and Kalshi have said that across the entire slate of elections that were called by Wednesday morning, about 98 percent of the markets’ favored candidates won. In many cases, prediction markets were able to identify the winners early and decisively. In Texas, Kalshi and Polymarket gave James Talarico about a 75 percent chance of victory in the days leading up to the election; he won. In Kansas, Adam Hamilton’s chances were hovering around 64 percent; he won too. Even before the major networks began to make their official calls Tuesday night, and before The New York Times’ needle made its biggest swings, some of the probabilities on these platforms had approached 99 percent, indicating near-complete certainty about who the winners would be.
Part of the original idea behind prediction markets was that they’d be able to draw on the wisdom of the crowd—millions and millions of individual hunches—to produce more accurate forecasts than any single source. The question after Tuesday’s elections is whether they can consistently align on outcomes that the polls and forecasts don’t see coming. Are prediction markets oracles or aggregators? So far, the evidence can’t justify the kind of grand victory they’re claiming.
Not all of the markets’ favorites won their election; Democrats were often underestimated, just as they were in the polls. Many bettors failed to predict Darline Graham’s loss in South Carolina and Greg Abbott’s loss in Texas. And in some cases, the markets were jittery: Probabilities wavered up and down at the last minute in Texas, Arizona, and New York for unknown reasons.
The markets proved themselves most useful this week when they could fill an information gap. The markets can be a rough index of public sentiment on Election Night, when the exit polls are out but the race still hasn’t been called—or in the days after when votes are still being counted, as in Alaska and Mississippi. As Bloomberg’s Matt Levine put it yesterday: “It would be nice if someone else would piece everything together, weight it appropriately by how much it matters, and generate one simple answer to the question ‘who won?’” This is essentially what prediction markets have tried to do. But, Levine notes, the markets are largely piecing together information that already exists.
Bettors can theoretically get their data from anywhere; they might make wagers based on polling, late-night cable-news broadcasts, their own surveys, or other reputable forecasts by experts such as Nate Silver (a paid adviser to Polymarket) and Lakshya Jain of Split Ticket. And at least on Polymarket, a small share of traders are handling the majority of the money—1 percent of accounts control 75 percent of the total volume, per the Anti-Corruption Data Collective, a nonprofit. That’s part of why the demographic makeup amd political biases of the people actually using these websites doesn’t skew the probabilities all that much. The especially active traders tend to bet according to who they think will win, rather than who they personally want to win.
In many cases, the markets favored eventual winners about as much as the forecasting models did. Take New Hampshire’s Senate race: The day before the elections, Nate Silver’s FLIPR Deluxe and Split Ticket both gave him a 95 percent chance. At the same time, Kalshi and Polymarket had him at 94 and 95 percent, respectively. Were markets just putting their trust in Split Ticket and Silver? It’s possible, but the markets and the models are not always aligned. In the Nevada governor’s race, where the Democrat also beat his Republican opponent, VoteHub, another model, gave the Democrat a 58 percent chance of winning, and FLIPR Deluxe gave him a 53 percent chance. Kalshi and Polymarket had him at about 75 percent. This time, maybe the markets were more attuned to early voting, as Silver himself suggested—or maybe they just got lucky. (In many cases, overindexing on early voting is ill-advised.)
The prediction-market industry had a lot to lose in these midterms—not only because lawmakers of both parties are preparing for a regulatory crackdown, but also because Tuesday night was a test of what these platforms can actually do. Kalshi and Polymarket have been trying to convince regulators and the American public that their markets have real uses for society beyond speculation. The industry’s position is that its platforms are public tools that can benefit observers just as much as the hedge funds and market makers who trade on them. But whether the markets can prove their worth as forecasters will depend on whether they can consistently produce the kind of discrepancy we saw in Nevada.
For many commentators, this Tuesday was a testament to the power of these markets to forecast elections. That itself is a boon for the companies. In declaring victory, though, Kalshi and Polymarket are exaggerating the extent of their vindication. Despite the outsize media attention and PR, election markets represent a tiny fraction of today’s betting industry. Bloomberg reports that “last month, politics accounted for just 1% of trading volume across Kalshi and Polymarket, down from two-thirds during the 2024 presidential election.” Sports, parlays, and crypto trades make up the majority of that volume—and the majority of the revenue. It might be more accurate to say that these firms are sports-betting platforms with a politics function tacked on. Outside of elections, the social benefits of these markets are harder to see.
Possible Futures
In every edition, I’ll share my thoughts on a few news stories I’m watching.
- Nebraskans voted against a ballot measure that would have legalized online sportsbooks in the state. In-person betting is already legal, as is sports betting on prediction markets, because they’re not technically considered sportsbooks.
- Democrat Vikki Goodwin, who ousted Texas’s incumbent lieutenant governor, Dan Patrick, on Tuesday night, has said in the past that she’d let the state’s voters decide whether they want to legalize sports betting, both online and in person. Patrick had been far more opposed to the practice.
- Back in May, Abdul El-Sayed wrote about his Senate opponent Mike Rogers, “I’ll smoke him by 7. Bet on it.” His margin of victory in Michigan was indeed seven percentage points. My sense is that El-Sayed isn’t actually a betting man—but he seems to have a knack for it.
PS
Although not everyone is comfortable with the idea of turning elections into a casino, betting on American political contests goes back to the 19th century—Henry David Thoreau once said that “all voting is a sort of gaming” and that “betting naturally accompanies it.”
— Will
Leave a Reply