A groundbreaking academic study published last month has upended the conventional wisdom about prediction markets, revealing that only a tiny fraction of traders — 3.14% — are responsible for the majority of price accuracy on Polymarket, the world's largest prediction market by volume. The working paper, titled “Prediction Market Accuracy: Crowd Wisdom or Informed Minority?”, was authored by researchers from London Business School and Yale University, and posted on SSRN on April 20, 2026, with a revision on April 25.
Using a comprehensive dataset covering 98,906 events, 210,322 markets, and $13.76 billion in total trading volume across 1.72 million accounts, the team applied a sign-randomization test to classify traders based on whether their profits reflected genuine skill or random chance. The results directly challenge the industry’s favorite narrative — that accuracy emerges from the collective intelligence of a diverse crowd.
The 3.14% Skilled Minority
Only 3.14% of Polymarket accounts qualified as skilled winners. These traders earned persistent profits that held up out-of-sample tests, traded across an average of 79 markets each, and consistently positioned in the direction of final outcomes. The remaining 96% of accounts either broke even by luck or lost money. The study found that skilled traders’ order flow predicted both next-period price changes and final market outcomes at statistically significant levels. A one-percentage-point increase in skilled net buying corresponded to an 8 basis point increase in the probability of the correct final outcome. Lucky winners, despite posting positive account balances, showed no meaningful predictive power.
Polymarket’s growth has been explosive — monthly trading volume surged from $3.3 million in December 2023 to $1.98 billion in December 2025, a nearly 600-fold increase. Active accounts grew from roughly 1,600 to more than 519,000. Yet the concentration of skill remained narrow, suggesting that platform growth did not dilute the influence of the informed minority.
Skill Persistence and Anti-Skill
The researchers split events into training and test sets to measure skill persistence. Among traders classified as skilled in training, 44% retained that classification in the test set. For unskilled losers, 51% remained in that category. By comparison, skilled mutual funds in a parallel test retained their classification only 10% of the time. The authors describe prediction markets as showing unusually high persistence of both skill and anti-skill, meaning that poor traders tend to remain poor traders — a stark contrast to traditional finance.
Skilled traders also reacted first to scheduled news. In tests involving Federal Open Market Committee (FOMC) announcements and corporate earnings releases, only the skilled group shifted its order imbalance in the direction of the news surprise within a narrow window around each release. Other groups showed no consistent response.
Insider Trading Case and CFTC Action
The paper also examined potential insider trading. Researchers identified 1,950 accounts that met timing and conviction criteria, suggesting they traded on non-public information. These accounts averaged roughly $15,000 in profits each and had large price effects when they traded. One documented case involved three accounts that took positions in a contract tied to Venezuelan President Nicolas Maduro hours before a secret U.S. military operation on Jan. 3, 2026, collectively earning more than $630,000.
On April 23, 2026, the U.S. Commodity Futures Trading Commission (CFTC) filed a complaint alleging that an active-duty U.S. Army service member engaged in insider trading using one of those accounts. Despite the size of these profits, the researchers concluded that insider activity was too concentrated in isolated events to account for broad price discovery across the platform.
Who Bears the Losses?
The study also detailed a stark asymmetry: the majority of participants fund accuracy rather than produce it. Unlucky and unskilled losers made up 67% of all accounts and absorbed the entirety of aggregate losses. Market makers and skilled takers together represented fewer than 3.5% of accounts but captured more than 30% of total gains. The authors emphasize that prediction market accuracy is not a product of the crowd, but of a small, identifiable group of informed traders whose participation is the mechanism behind price formation.
Whether these skilled traders will continue to participate as platforms grow in popularity and fees increase remains an open question that the paper leaves for future research. The findings have significant implications for the design of prediction market platforms, regulatory oversight, and the broader debate about the efficiency of decentralized information aggregation.