The academic record on prediction markets is unusually deep. The Iowa Electronic Markets have run since 1988 as a research project, giving economists more than three decades of real-money forecasting data, and the literature that grew from it is the foundation for everything the modern industry claims about accuracy. The findings below are the ones that matter.
The core evidence
The landmark result comes from the Iowa Electronic Markets. Berg, Nelson and Rietz compared IEM presidential-election prices against 964 major polls across five election cycles and found the market was closer to the final result than the polls roughly three-quarters of the time — and its advantage grew with distance from election day. Wolfers and Zitzewitz’s widely cited 2004 survey in the Journal of Economic Perspectives reached the same conclusion across markets generally: prices aggregate dispersed information quickly and outperform most single-source forecasts. The strength of the evidence is why 22 leading economists — including several Nobel laureates — publicly urged US regulators in Science (2008) to give prediction markets room to operate. Modern data points the same way: industry analysis of 2025 platform data put average Brier scores near 0.09, strong calibration by any forecasting standard, as covered in our statistics page.
Key studies at a glance
| Study | What it tested | Finding |
|---|---|---|
| Berg, Nelson & Rietz (IEM) | Market prices vs 964 election polls, 1988–2004 | Market closer to the outcome ~74% of the time; edge grows with forecast horizon |
| Wolfers & Zitzewitz (2004) | Survey of prediction-market performance | Markets rapidly aggregate dispersed information; prices are well-calibrated probabilities |
| Servan-Schreiber et al. (2004) | Real-money vs play-money markets on NFL games | Both were similarly accurate; incentives matter less than aggregation itself |
| Cowgill & Zitzewitz (corporate) | Internal markets at Google and Ford | Informative but showed an optimism bias that faded as traders gained experience |
| Hanson & Oprea (manipulation) | Can manipulators distort prices? | Manipulation attempts often improve accuracy by adding liquidity for informed traders to trade against |
| Arrow et al., Science (2008) | Policy statement by 22 economists | The evidence justified regulatory room for real-money forecasting markets |
Why they work
The mechanism is incentive-weighted aggregation. A poll weights every response equally and costs nothing to answer carelessly. A market weights each view by the money behind it, pays accuracy, and punishes error — so information held by confident, informed traders moves the price more than noise does. That is also why manipulation is harder than it looks: as Hanson and Oprea showed, a manipulator pushing a price away from the evidence is simply offering profitable trades to everyone who knows better. The 2024 French whale episode is the modern case study — what looked like manipulation was better information, priced in early.
Where they fail
The literature is equally clear about failure modes, and honest traders should know them. Thin markets aggregate little because there is little to aggregate; accuracy findings come from liquid contracts. The favourite-longshot bias means very low-probability contracts tend to trade above their true odds — buying every 3-cent longshot is a losing strategy. Calibration weakens at the extremes, where prices near 0 or 100 move on fees and float rather than information. And corporate-market research shows participant bias travels: traders who all share an optimism will price it in together. Our guide to market accuracy covers how to weigh a price accordingly, and the strategy library covers how to trade the biases rather than pay them.
Go deeper
These guides cover the evidence, the tooling and the timing behind serious trading:
Frequently asked questions
Are prediction markets more accurate than polls?
Usually, yes. The Iowa Electronic Markets studies found market prices were closer to the final election result than major polls roughly 74% of the time across 1988–2004, with the advantage growing further from election day.
Do prediction markets work with play money?
Research by Servan-Schreiber and co-authors comparing real-money and play-money NFL markets found both similarly accurate — the aggregation mechanism matters more than the stakes, though real money attracts more sustained research effort.
Can prediction markets be manipulated?
Attempts happen, but research by Hanson and Oprea found manipulation often improves accuracy: pushing a price away from the evidence hands profitable trades to informed traders, who push it back.