New academic research challenges the widely held belief that prediction markets derive their accuracy from the collective wisdom of a broad participant base, instead attributing it to a tiny fraction of informed traders. The study, conducted by researchers at London Business School and Yale University, analyzed the entire trading history of Polymarket, the world's largest prediction market by volume.
Massive Dataset Analysis
The working paper, titled "The Accuracy of Prediction Markets: Wisdom of Crowds or Informed Minority?" was posted on SSRN on April 20, 2026, and revised on April 25. Authored by Roberto Gomez-Cram, Yunhan Guo, Howard Kung of London Business School, and Theis Ingerslev Jensen of Yale University, the study covers 98,906 events, 210,322 markets, and total trading volume of $13.76 billion generated through 1.72 million accounts on Polymarket.
Using a statistical method called the sign-randomization test, the researchers classified traders into groups based on whether their profits reflected genuine skill or luck. The findings were striking: only 3.14% of Polymarket accounts were classified as skilled winners. These traders earned persistent profits out-of-sample, traded in an average of 79 markets, and consistently built positions aligned with final outcomes. In contrast, the remaining 96% of accounts either broke even or suffered losses driven solely by chance.
Key Findings: Skill Concentration
The order flow of skilled traders was found to significantly predict both subsequent price movements and final market resolutions. A 1-percentage-point increase in skilled traders' net buying ratio raised the probability of correctly predicting the final outcome by 8 basis points. Meanwhile, lucky traders—despite positive account balances—showed no significant predictive power in either test.
Polymarket's monthly trading volume surged from $3.3 million in December 2023 to $1.98 billion in December 2025—a roughly 600-fold increase in two years. Active accounts expanded from about 1,600 to over 519,000. Despite this explosive growth, the concentration of skill remained remarkably narrow.
The study also tested skill persistence. Among traders identified as skilled in the training set, 44% retained that classification in the test set. In comparison, only 10% of skilled mutual funds retained their classification in parallel tests. The authors note that prediction markets exhibit unusually high persistence in both skill and non-skill categories.
Insider Trading and Regulatory Action
Skilled traders were the fastest to react to scheduled news events, including FOMC announcements and corporate earnings releases. The research also examined insider trading patterns. It identified 1,950 accounts meeting timing and conviction criteria consistent with trading on non-public information. These accounts earned an average of approximately $15,000 each and caused significant price moves. In one documented case, three accounts took positions on contracts related to Venezuelan President Nicolás Maduro hours before a secret U.S. military operation began on January 3, 2026, collectively profiting over $630,000.
On April 23, 2026, the U.S. Commodity Futures Trading Commission (CFTC) filed a complaint alleging that an active U.S. military service member used one of these accounts to engage in insider trading. Despite such isolated incidents, the researchers concluded that insider activity was too concentrated to explain broad price discovery across the platform.
Conclusion
The study concludes that the accuracy of prediction markets reflects the behavior of a small but identifiable group of informed traders, whose participation forms the mechanism of price formation. Whether these traders will continue to participate as the platform grows and fees increase remains an open question for future research.