Prediction Markets Aren’t Likely to Replace Polling
When polling forecasters assessed public opinion in the lead-up to the 2024 U.S. presidential election, they were pretty off point: Most suggested a very close race between Donald Trump and Kamala Harris. Online betting platforms known as prediction markets appeared to fare better. Unlike polls, which gather snapshots of potential voters’ preferences, prediction markets allow people to wager money on future events, essentially using the hive mind to make predictions. And Polymarket, one of the largest prediction markets, predicted a Trump win, by 58 to 42 percent. (Trump ended up winning 312 Electoral College votes, to Harris’ 226.)
That election alone pulled in more than $3.5 billion in bets on Polymarket. Polymarket and Kalshi, another sizeable market, have facilitated a wide-ranging gambling economy in which users bet on everything from geopolitical events to interest rates to the price of eggs.
Excluding a few bets deemed beyond the pale — such as an attempt to predict a nuclear weapon detonation — Polymarket representatives have defended their platform weighing geopolitics; on a bet predicting when the U.S. might strike Iran, the website noted that prediction markets could give people affected by attacks “the answers they needed in ways TV news and X could not.” Kalshi, meanwhile, calls for viewing prediction markets as a “complement to polling.” Since December 2025, predictive markets have agreed to partnerships with news organizations including CNBC, CNN, Yahoo Finance, and Dow Jones & Company, which publishes The Wall Street Journal.
Some analysts now expect prediction markets to compete against pollsters themselves. In some instances, prediction market bets, often referred to as “event contracts,” also successfully forecast economic developments better than trained economists or political scientists. Proponents view the markets as valuable sources of information, as useful as professional polls and financial market trends. Critics, however, have observed prediction markets’ inconsistencies, and they worry about the markets skirting gambling and trading regulations. (Kalshi and Polymarket did not respond to Undark’s requests for comment.)
In terms of using the markets in lieu of polls, experts warn that the two are rather different — they don’t measure the same things, nor are they designed for the same purposes. Polls gauge people’s sentiments at a particular time, and they are generally not meant for predictions, even if they’re frequently interpreted that way. Using prediction markets like polls, or vice versa, could lead to misunderstandings. “I never use prediction markets to try to understand public opinion,” said Courtney Kennedy, vice president of methods and innovation at the Pew Research Center. “Where’s the data coming from? I think, scientifically, I’d much rather be looking at rigorous data that’s actually from the public as opposed to people’s guesses as to what that is.”
The most popular prediction markets today arrived following the exploding growth of sports betting apps, but they encompass just about everything, such as whether Ali Khamenei would be out as Iran’s Supreme Leader (before he was killed in an Israel-U.S. airstrike), whether the U.S. would confirm that aliens exist before 2027, or whether Trump would say particular nicknames for political opponents during his 2026 State of the Union address. The markets effectively operate as exchanges, connecting traders wagering “yes” or “no” on specific questions, and they target individual events and outcomes. The approach makes it particularly easy to bet on sports and elections.
In theory, the markets could be viewed as channeling the wisdom of crowds, who eventually pick one side of a question or another through the money they spend on their bets. Such pricing can be used to construct a forecast, said Andrew Gelman, a political scientist at Columbia University.
“Where’s the data coming from? I think, scientifically, I’d much rather be looking at rigorous data that’s actually from the public as opposed to people’s guesses as to what that is.”
This sort of wisdom from a crowd, however, is markedly different from polling, which relies on a person’s opinion on a topic, rather than how they think a particular situation will play out. As Christopher Wlezien, a University of Texas political scientist, puts it, the two are different types of tools. “I’m a big fan of prediction markets, I use them, but these comparisons are kind of silly,” he said. But it’s “kind of funny that we would be comparing a forecasting device with a preference-registering device.” Yet people do make such comparisons, and in previous research he attempted to study the strengths and weaknesses of each approach. In a 2008 paper, he documented cases of long-shot bias, where markets overvalued underdogs, in winner-take-all U.S. elections. And in a historical analysis from 2012, he concluded that, following the advent of opinion polling based on statistical methods in the 1930s, election markets performed no better than polls as election predictors.
Some people misinterpret both polls and prediction markets. Polls aren’t predictive, Kennedy said, because they include errors and unknowns, such as whether the same people who are polled will vote in an election and who will or won’t change their minds. But people inevitably try to use them to predict elections — and some poll forecasters make a living on it. Prediction markets, on the other hand, don’t probe attitudes the way polls do, and they sometimes have biases, because they’re not representative of the broader population, for example. It’s not possible for prediction markets to replace polls or pollsters, Kennedy argued, and in election-related wagers, polls can inform prediction markets.
Kennedy emphasized another distinction: “Polls are not profiting from human suffering or political instability or anything like that. Polls are designed to increase understanding; it’s not about anybody making a profit.”
It didn’t take long for the modern form of prediction markets to become a force in U.S. politics and economics. In May 2018, though sports betting had been banned for decades in most of the country, the Supreme Court struck down a previous ruling that had restricted legal betting primarily to Nevada. Two markets, DraftKings and FanDuel, then quickly expanded into online and retail sportsbooks. Not long after, Kalshi and Polymarket were founded, in 2018 and 2020 respectively, and they moved to include sports betting as well as financial and political events, including some betting on the 2022 midterm elections.
Kalshi and Polymarket look similar and offer similar types of wagers, but they have a couple differences. For one, Polymarket is crypto-based, so its users trade with currencies like Bitcoin and Ethereum, and its main platform operates offshore, outside U.S. jurisdiction, where it moved after settling charges with the federal government for operating without registration, while the New York-based Kalshi is the first major prediction market firm regulated by the Commodity Futures Trading Commission. Both companies are supported by investments from a range of venture capital firms. As fundraising rounds continue, the two companies might each seek roughly $20 billion valuations.
Sometimes prediction markets are lauded in the news media as “economic oracles” that are “beating Wall Street Ph.D.s.” Some economists do believe they’re useful sources of information in particular circumstances. Through wagers and leaderboards, a prediction market gives people incentives to gather and act on information, said David Rothschild, an economist at Microsoft Research. “It gets people to think about a unique question and to pull in dispersed or idiosyncratic information in a meaningful way, so you really get insight into how the probability of that outcome is shifting over time.”
Rajiv Sethi, an economist at Barnard College, Columbia University, believes such markets can be particularly useful during historically unprecedented times. “My view of prediction markets is that they have the potential to generate valuable forecasts for certain events, especially when we’re in uncharted waters and history is not a good guide to the future,” he said. As examples, Sethi cited bets on climate-related risks, impacts of the Covid pandemic, and forecasts that former President Biden would drop out of the presidential race in 2024, long before he actually did and while he was insisting he would not.
In theory, in a prediction market, the well-informed marginal traders can drive the market to where they think the forecast should be, regardless of how many low-information traders there are, Rothschild said. Nevertheless, biased outcomes can emerge for many reasons, according to research he published with Sethi in 2016. The researchers studied nearly 300,000 transactions from 6,300 unique trader accounts on the former Ireland-based prediction market Intrade for the Barack Obama-Mitt Romney 2012 election. The authors noted that traders are predominantly young and male, and prices tended to skew towards the best-funded traders, who need not be the best-informed. The researchers also noted a phenomenon known as the Romney Whale, where a single, unknown trader was responsible for a third of bets on Romney, and wagered and lost nearly $4 million over a two-week period. In that case, the marginal trader wouldn’t have been able to overcome that large whale’s influence or manipulation, as they deliberately or incidentally pushed prediction market prices toward Romney.
In more recent research in 2025, available on the open access research platform SSRN, Sethi and his colleagues analyzed Polymarket forecasts and the statistical models used by three poll aggregators — The Economist, Silver Bulletin, and FiveThirtyEight — during the 2024 Trump-Harris election. Polymarket did successfully forecast a Trump victory, but it did not fare much better than statistical models on who would win the popular vote or the Electoral College, and it fared particularly poorly down the ballot compared to the models, especially for Congressional races. The researchers attribute this partly to the lower trading volumes in those markets.
Sethi stresses that the apparent success of prediction markets on the Trump-Harris election, while notable, is merely a single data point, and they did not correctly forecast other election outcomes. “One needs to look at a larger range of events,” he said. “And when you do that, the jury is still out on the accuracy of markets versus models.”
The future of prediction markets may be uncertain, as criticisms mount. And regulators are also targeting the markets, which they tend to view less as financial exchanges than as gambling platforms that allow market manipulation like insider trading, such as bets on Nicolás Maduro’s removal just before the U.S. military invaded Venezuela and kidnapped him. Sethi also noted the concerning practice of “wash trading,” where people both buy and sell, which he describes as essentially “trading with yourself” to give the impression of a higher-volume market. There’s limited risk of detection of this on crypto-based markets like Polymarket, he said, and it appears to be common, according to a recent study of his and researchers at the Columbia Business School, available on SSRN. To help limit insider trading and wash trading, he calls for imposing the KYC requirement (“Know Your Customer”), which is already in use in financial markets, requiring identity verification for traders.
Gelman, the political scientist, and others have also criticized prediction markets for profiting from war and death, which became apparent in the run up to the war on Iran and afterward. Following controversy about bets over Iran’s Khamenei, Gelman pointed out that, in contrast, traders in financial markets cannot bet directly on wars. But traditional financial markets do allow for investment in oil futures, which are closely linked to conflicts with countries that have significant oil reserves, including the ongoing war with Iran and previous U.S. wars in Iraq.
“My view of prediction markets is that they have the potential to generate valuable forecasts for certain events, especially when we’re in uncharted waters and history is not a good guide to the future.”
Unlike polls or even stock markets, prediction markets do not come with standardized or rigorous estimates of uncertainties. That makes interpreting them challenging, at least for now. Assuming prediction markets’ ethical and legal issues can be resolved, Wlezien thinks that prediction markets could improve, especially once the “Nate Silvers and Elliott Morrises” join in, referring to popular polling and election data analysts known for parsing errors in polls. (Silver joined Polymarket as an adviser in 2024.)
Wlezien also speculates that prediction markets could not just profit from politics, but influence them as well. Whether prediction markets are accurate or inaccurate, they could have impacts on “candidate choice, candidate entry and exit, funder behavior, voter behavior,” Wlezien said. For example, in April, a former campaign manager for Obama joined Kalshi as a policy advisor, with the company announcing on their website that the market was “creating data that can effectively complement polling and expert opinion,” and before that, the president’s son Donald Trump Jr. became an adviser for both Polymarket and Kalshi. In addition, a former California gubernatorial candidate attempted to place two Kalshi bets on the upcoming governor’s race, spurring calls to ban such insider betting.
Prediction markets could affect how investment decisions are made as well, by providing a new source of market data. But it’s hard to predict what kind of impact prediction markets could eventually have on politics and financial markets. Wlezien was skeptical of markets having a substantial influence, but he did offer one carefully hedged prediction himself: “I’m guessing it’s not going to be massive, but it might be greater than zero.”