A persistent question about prediction markets is why their prices tend to be accurate — is it because many independent people betting their own money aggregate into a collectively smart price (classical "wisdom of crowds"), or because a small number of well-informed traders do the real work of price discovery while everyone else just provides liquidity?
A 2026 paper by Roberto Gómez-Cram, Yunhan Guo, Theis Ingerslev Jensen, and Howard Kung — "Prediction Market Accuracy: Crowd Wisdom or Informed Minority?" — answers this directly, using close to the complete trading history of Polymarket. The paper is currently invited for revise-and-resubmit at the Review of Financial Studies, posted to SSRN as working paper No. 6617059 (DOI: 10.2139/ssrn.6617059). We go through the methodology and findings below; the SSRN listing is here if you want to read the full paper yourself.
The data
The study draws on roughly 1.72 million Polymarket accounts, close to 98,900 events, and about $13.7 billion in trading volume across 2023–2025 — essentially the platform's full transaction history over that window, not a sample.
The method: a coin-flip benchmark, not a vibe check
The central methodological move is building a rigorous counterfactual for luck. For each trader, the authors take their actual sequence of trades and randomize the direction (buy vs. sell) of each one, then resimulate that randomized version 10,000 times to build a distribution of what pure chance alone would have produced for that trader. Lead co-author Theis Jensen described the approach plainly in an SSRN blog feature on the paper: "we simulate how they would have done if they were essentially just flipping a coin." A trader's actual profit is only called statistically significant skill if it sits meaningfully outside that chance-generated distribution — a much higher bar than just "this account made money."
Two further checks sharpen this:
- Persistence. Each trader's history is randomly split in half, and the authors test whether being classified as "skilled" in one half predicts the same classification in the other half. They report roughly 44% persistence — skill that shows up once tends to show up again, which is the difference between a real, repeatable edge and noise that happened to look like skill in one sample.
- News-event behavior. The authors specifically examine trading around scheduled news events — FOMC announcements, earnings releases — and check for patterns consistent with suspected insider-type activity, separately from the broader skill measure.
The finding: about 3% of traders, and they're not insiders
The headline number: roughly 3% of accounts show statistically significant skill under this methodology. In Gómez-Cram's own words, from an op-ed he wrote for CityAM covering the paper: "market accuracy is not driven by the 'wisdom of crowds,' but by a small, highly informed minority." Per the SSRN blog feature, Gómez-Cram put the number directly: "only about 3% consistently earn profits."
That skilled minority, combined with market makers, is under 3.5% of all accounts — and together they capture over 30% of total trading profits on the platform. The paper's description of this group: they react quickly to public news, they systematically trade against known pricing biases (like the favorite-longshot effect that shows up repeatedly in prediction-market research), and they exploit arbitrage where it appears.
What about the other 96–97%? They generate the bulk of trading volume, but contribute comparatively little price-relevant information. Coverage of the paper has summarized the relationship bluntly: the majority doesn't produce the market's accuracy — it funds it, in the sense that their net losses are effectively what the skilled minority's profits come from.
Is it just insiders, then? No — and this is the part that makes the paper's answer more interesting than either simple story. The authors specifically looked for suspected insider-type trades and found they do move prices much more per dollar than an ordinary skilled trade. But those trades are too rare and too concentrated in isolated events to explain the market's overall accuracy on their own. The paper's actual conclusion sits between the two popular narratives: not naive crowd aggregation, and not "it's all insiders" — a small, persistently skilled trading population doing sustained, repeatable price discovery.
Why this matters if you're using prediction-market prices for anything
- A Polymarket price is informative because of a minority, not a majority. If you're treating a market price as an aggregated "wisdom of crowds" signal, this paper is direct evidence that framing is the wrong mental model for Polymarket specifically — the information content is concentrated in a small, identifiable, persistent group of traders, not spread evenly across all participants.
- Persistence (44%) means skill here isn't mostly luck. A one-off "this trader beat the market once" is a much weaker signal than this paper's own benchmark — real, repeatable skill is a minority phenomenon, confirmed by splitting each trader's own history in half.
- It's not an insider story. If your working assumption about prediction-market accuracy was "someone with material non-public information is moving these prices," this paper's own search for that pattern came back limited — informed, legal, fast reaction to public information is the dominant mechanism it finds, not leaked information.
FAQ
Are prediction market prices driven by the wisdom of crowds? Per this paper's findings on Polymarket, no — not in the classical sense. Accuracy is concentrated in a small, statistically identifiable skilled minority (roughly 3% of accounts), not produced by aggregating many independent opinions evenly.
What percentage of Polymarket traders are actually skilled, according to this research? About 3% show statistically significant skill under the paper's coin-flip randomization benchmark. Combined with market makers, under 3.5% of accounts capture over 30% of total trading profits.
Is Polymarket's price accuracy driven by insider trading? The paper looked specifically for this and found suspected insider-type trades do move prices more per dollar than ordinary skilled trades, but concluded they're too rare and concentrated to explain the market's overall accuracy — the dominant mechanism is a persistent skilled minority reacting to public information, not insider access.
Where can I read the full paper? Gómez-Cram, R., Guo, Y., Jensen, T. I., & Kung, H. (2026). "Prediction Market Accuracy: Crowd Wisdom or Informed Minority?" SSRN Working Paper No. 6617059. Available on SSRN. It's currently invited for revise-and-resubmit at the Review of Financial Studies.
Is this the only recent research on prediction-market accuracy? No — it's one of twelve papers we cover in our broader prediction-market research survey, spanning liquidity provision, arbitrage, calibration, and manipulation research from late 2024 through mid-2026.

