Organizations that track election polls, including The New York Times, currently show Vice President Kamala Harris and former President Donald J. Trump in a dead heat. Election forecasters like FiveThirtyEight, which consider the polls and other data to calculate the candidates’ odds of victory, also see the race as essentially tied.
Another set of websites has a different take. Prediction markets, which allow users to bet real money on election results — more than $100 million on one site alone — are much more bullish on Mr. Trump. These markets are drawing attention, but are they more accurate than the polls, as proponents claim? Do the markets know something that the polls don’t?
Researchers have tried to answer this question over the last few election cycles, though the results have been mixed. Advocates and skeptics agree that there are pros and cons to markets when compared with poll-based forecasts. Markets can sometimes be just as accurate in predicting election outcomes — but there are also good reasons not to take them at face value.
Polls and prediction markets could hardly be more different. Polls attempt to measure what actual voters think by asking a group of them, a sample, how they plan to vote. Polls don’t assign a probability to a candidate’s likelihood of winning — they’re snapshots in time, representing the share of people supporting a candidate at that moment. At their best, polls can be quite accurate, but they are also subject to error and are best understood in aggregate.
Election forecasts — like the ones published by FiveThirtyEight, The Economist, The Hill and election analyst Nate Silver — use the polls and other data such as economic indicators and historical trends to project the likely winner of the election, and the probability that each candidate will win.
Election prediction markets, in contrast, allow their users — who could be anyone, often anywhere in the world — to buy “shares” predicting that a certain outcome will occur. If it does, the price is paid out at $1 per share. So, if the price of a share showing Mr. Trump winning the election is 40 cents, and a user thinks Mr. Trump has higher odds, the user would purchase shares in hopes of making a profit if Mr. Trump were to in fact win.
The volume of users investing or divesting in an outcome drives the cost of the shares up or down accordingly. Those prices are then generally looked upon as a proxy for what the market feels is the likelihood of a given event; an event with shares priced above 50 cents, for example, is seen as more likely than not to occur.
“Right now, I think, we’re at 60 percent Trump versus Kamala — but 60 percent is not 100 percent,” said Tarek Mansour, the chief executive of Kalshi, a prediction market that recently won regulatory approval to allow users to bet on elections in the United States. “I think a lot of people are confusing this. FiveThirtyEight is a probabilistic forecast. Kalshi is a probabilistic forecast. Polls are not.”
So, do these markets perform better than polls? Since polls don’t assign the candidates probabilities of winning, it’s difficult to make a direct comparison. Instead, researchers have looked at past election cycles and compared prediction markets with election forecasts. The results have not provided a conclusive determination as far as any one system’s being more accurate than another.
An analysis of the 2008 election cycle looked at the presidential results in all 50 states as well as 24 Senate races and found that forecasts by Intrade, a prediction market active at the time, performed similarly to FiveThirtyEight that year. A similar analysis of the 2020 cycle found comparable results.
The nature of these markets has made research difficult. Betting on elections has long been prohibited by regulators in the United States; the markets currently operating either have special permission, as Kalshi does, or are offshore. And prediction markets have come and gone, making it difficult to do a long-term analysis.
There are plenty of examples of prediction markets that completely missed the mark, though. In 2016, they were even more confident than some forecasters, such as FiveThirtyEight, that Hillary Clinton would defeat Mr. Trump. Prediction markets can also amplify the opinions of speculators: In 2020, some investors were still betting that Mr. Trump would be the next president even after the race had been called for Mr. Biden.
And in the 2022 midterm elections, prediction markets foretold a “red wave” of Republican victories and priced a win in Pennsylvania by the Republican Senate candidate, Dr. Mehmet Oz, at 63 cents, despite polls showing that the race was a tossup. (Dr. Oz lost to John Fetterman by five percentage points.)
Proponents of prediction markets as election prognosticators believe that having real money on the line, and a large crowd of investors, encourages a more accurate market. The theory is that if an outcome is likely to happen, investors won’t let the corresponding shares sit there at a low price when they could be making a profit. Proponents also note that investors are aware of the polls and forecasts and integrate that knowledge into their bets.
“One thing that the markets have that the polls don’t is individuals in the market interacting with one another,” said Harry Crane, a statistician at Rutgers University who has researched prediction markets. “I buy a share, which means somebody else sold it to me. And we kind of go back and forth in that way that collectively settles on an equilibrium.”
Experts who study prediction markets note that the platforms react much more quickly to news events than do traditional polls, which can take several days to conduct.
After President Biden’s lackluster debate performance in June, for example, rigorous pollsters needed time to find out whether voters thought he should drop out of the race. But on the prediction markets, bettors quickly bought up shares. On Polymarket, one of the major betting platforms, shares predicting that Mr. Biden would drop out were priced at 21 cents the day before the debate. By the next morning, the price had jumped to 43 cents. By July 3, less than a week after the debate, the price was up to 60 cents.
This strength can also be seen as a weakness. Prediction markets can react more quickly to election news but can also be more volatile, noisy and subject to outside influences. In early October, Elon Musk posted on X that Mr. Trump’s shares on Polymarket were priced above Ms. Harris’s. Afterward, the price of Trump shares on Polymarket began to climb, and they are outpacing Trump shares even on other prediction markets.
Polymarket, which is based on cryptocurrency and does not limit how much a user can bet, can also be affected by a small number of large investors. This phenomenon appears to have happened in the presidential election market. PredictIt, another major market, will not allow a user to bet more than $850, while Kalshi limits a user’s total bets to $100 million.
When the polls are essentially tied, looking for any other source of predictive data is tempting, whether it’s markets or macaque monkeys staring at photos of the candidates. But there’s no clear evidence yet to suggest that prediction markets can outperform other election forecasts or are inherently worse.
“In principle, one could accumulate enough data to answer the accuracy question. But I don’t think we’ve answered it yet,” said Rajiv Sethi, an economist at Barnard College. “The jury’s still out.”
The post Betting Markets Favor Trump. But Their Record of Accuracy Is Mixed. appeared first on New York Times.
