Are prediction markets accurate? What the research says
Prediction markets are usually accurate, with known blind spots. The Iowa Electronic Markets came closer to the final vote than 964 national polls about 74% of the time across five presidential elections (Berg, Nelson and Rietz, 2008). A 2025 study of Kalshi found prices grow more accurate as markets near their close, but buyers of contracts priced at 10¢ or less lost over 60% of their money on average.
Accuracy depends mostly on liquidity and on how close the market is to closing. The papers below are linked so you can check the numbers.
The studies at a glance
| Study | Data | Main finding |
|---|---|---|
| Berg, Nelson & Rietz (2008), *International Journal of Forecasting* | Iowa Electronic Markets, 1988–2004 presidential elections | Market beat 964 polls about 74% of the time |
| Erikson & Wlezien (2008), *Public Opinion Quarterly* | Same IEM markets vs. adjusted polls | Properly discounted polls beat the markets |
| Page & Clemen (2013), *Economic Journal* | Intrade, 1,787 markets, 512,612 trades | Well calibrated near expiration, biased far out |
| Bürgi, Deng & Whelan (2025), working paper | Kalshi, 2021 to April 2025 | Informative prices, clear favorite-longshot bias |
| Rothschild & Sethi (2016), *Journal of Prediction Markets* | Intrade, 2012 presidential race | Evidence suggestive of manipulation by one large trader |
Prediction markets vs. polls
The Iowa Electronic Markets (IEM), run by the University of Iowa, are the longest-studied case. Berg, Nelson and Rietz compared IEM prices with 964 national polls from 1988 through 2004. The market's forecast of the two-party vote share was closer about 74% of the time, and the authors report that "the market significantly outperforms the polls in every election when forecasting more than 100 days in advance."
Justin Wolfers and Eric Zitzewitz summarized the same record in the *Journal of Economic Perspectives* (2004): IEM vote-share markets missed by about 1.5 percentage points on average, against 2.1 points for the final Gallup poll.
The counterargument is strong. Robert Erikson and Christopher Wlezien (2008) showed that raw poll numbers are the wrong benchmark, because early poll leads tend to shrink by Election Day. Once poll leads are "properly discounted," they wrote, "poll-based forecasts outperform vote-share market prices." Their projected polls beat IEM vote-share prices on about 55% of days, and beat the winner-take-all markets on about 87% of days. They also called the IEM a small market with thin volume and trader limits, so the result may not carry over to larger venues.
Markets beat raw polls. Well-built poll models can match or beat the markets.
Calibration: do 70% events happen 70% of the time?
A forecaster is well calibrated if events it prices at 70% happen about 70% of the time. Accuracy on single races matters less than this long-run pattern.
Lionel Page and Robert Clemen (2013) tested Intrade prices across 1,787 markets. They found the markets "reasonably well calibrated when time to expiration is relatively short," but "prices are significantly biased for events farther in the future," with "high-likelihood events" underpriced and "low-likelihood events" overpriced.
Kalshi data shows a similar pattern. Bürgi, Deng and Whelan studied 313,972 contract prices from Kalshi's launch in 2021 through April 2025. Their findings:
- Prices are informative and improve as markets approach closing, with a steep gain on the final day.
- Contracts priced at 10¢ or less lost buyers over 60% of their money on average.
- Contracts above 70¢ earned small but statistically significant positive returns after fees.
- The average return was −20% before fees and −22% after.
- Makers, who post resting orders, averaged −12% after fees; takers averaged −31%.
- The bias appeared weaker in 2025 than in earlier years.
A February 2026 preprint by Nam Anh Le, not yet peer reviewed, analyzed 292 million trades on Kalshi and Polymarket. It reports that long-horizon prices tend to sit too close to 50%, and that political markets on both exchanges were the most underconfident.
The favorite-longshot bias
The pattern in those studies has a name. In the favorite-longshot bias, long shots are overpriced and favorites underpriced. Erik Snowberg and Justin Wolfers studied it in horse racing (*Journal of Political Economy*, 2010) and concluded that "misperceptions of probability drive the favorite-longshot bias," rather than a taste for risk.
Worked example. Bürgi, Deng and Whelan illustrate the cost with a 5¢ contract that wins 3% of the time.
- You pay 5¢ for a contract worth $1 if it wins.
- Expected payout: 0.03 × $1.00 = 3¢.
- Expected return before fees: 3 ÷ 5 − 1 = −40%.
- Add Kalshi's taker fee, 0.07 × 0.05 × 0.95 ≈ 0.33¢ per contract, and the loss grows.
The same paper's 95¢ example, winning 98% of the time, returns about +3.1% before fees. When you read a price below 10¢, treat it as an upper bound on the real chance. Our guide to reading prices as probabilities shows how fees shift break-even points.
Thin markets and manipulation
Wolfers and Zitzewitz reviewed known manipulation attempts and found that "none of them had much of a discernible effect on prices, except during a short transition phase." They added that how easily a market can be manipulated depends on how thin it is.
The best-documented case came on Intrade in 2012. David Rothschild and Rajiv Sethi analyzed account-level data and found one trader who accounted for more than a third of all bets on Mitt Romney to win, and who lost about $6.9 million. On Election Day, the trader's large resting orders held Romney near 28¢ to 30¢ while the same contract on Betfair moved toward Obama. The authors estimate the trader spent over $375,000 holding prices during a 90-minute window. When the orders were pulled around 9 p.m. ET, Intrade jumped to Betfair's level. The authors found manipulation plausible but did not establish a motive.
If one platform's price sits far from another's, ask why before trusting either. Our live comparison and price-gap statistics make that check quick. A gap is not free money, either; our arbitrage guide explains why.
How did prediction markets do in 2024?
Polymarket, Kalshi and other markets favored Donald Trump for weeks before the 2024 election while polling averages showed a coin-flip race, Semafor reported, citing Reuters. In late October, Polymarket priced Trump at 61%, up from 50% in early October. A French trader who bet through several accounts drew manipulation questions; Fortune reported the trader netted about $40 million.
The record was mixed. On the eve of the election, Kalshi gave Kamala Harris a 76% chance of winning the popular vote and Polymarket 73%, against 71% from FiveThirtyEight's model, Sethi noted. Trump won the popular vote. As Sethi put it, relative accuracy "cannot be answered by looking at one or two events."
How to use market prices sensibly
- Give more weight to liquid markets that are close to resolution.
- Treat prices under 10¢ as generous to the long shot; in the studies above, favorites won slightly more often than priced.
- Check the spread and volume before trusting a number; our liquidity and spreads guide shows what to look for.
- Compare the same question across platforms.
This is general information, not investment advice. Prediction market contracts can lose their full value, so trade only what you can afford to lose.
FAQ
Are prediction markets more accurate than polls?
They beat raw polls: Iowa Electronic Markets prices were closer than 964 polls about 74% of the time from 1988 to 2004. Erikson and Wlezien (2008) found that polls adjusted for how leads shrink before Election Day did better than those markets.
Did prediction markets predict the 2024 election?
Polymarket and Kalshi favored Donald Trump for weeks while polling averages showed a toss-up, and he won. The same markets gave Kamala Harris roughly 73% to 76% odds of winning the popular vote, which she lost.
What is the favorite-longshot bias?
It is the tendency for unlikely outcomes to be overpriced and likely ones underpriced. On Kalshi from 2021 to April 2025, contracts priced at 10¢ or less lost buyers over 60% of their money on average.
Can prediction markets be manipulated?
Briefly, sometimes. In 2012 one Intrade trader lost about $6.9 million and held Romney's price steady on Election Day until withdrawing, after which the price jumped to match Betfair, according to Rothschild and Sethi (2016).