Key points to remember:
- Researchers from London Business School and Yale found that only 3.14% of Polymarket accounts are considered qualified, yet generate the majority of price discovery.
- Polymarket’s qualified traders maintained their classification 44% of the time out of sample, compared to just 10% for qualified mutual funds.
- On April 23, 2026, the CFTC filed a complaint for insider trading related to a Polymarket contract relating to the removal of Nicolas Maduro from power.
A study published on SSRN covers 98,906 events on Polymarket
THE working documenttitled ” Prediction Market Clarification: wisdom of the crowd or informed minority? » was published on April 20, 2026 on SSRN and revised on April 25, 2026. It was authored by Roberto Gomez-Cram, Yunhan Guo and Howard Kung of London Business School, and Theis Ingerslev Jensen of Yale University.
Researchers analyzed the entire transaction history on Polymarketthe largest in the world prediction market by trade volume. The study covered 98,906 events, 210,322 markets and $13.76 billion in total. trade volume on 1.72 million accounts.
Using a statistical method called the sign randomization test, the authors classified traders into distinct groups based on whether their profits reflected true skill or chance.
The results run counter to a widely held hypothesis. Prediction Market Platforms, including Kalshi and Polymarket itself, routinely describe their accuracy as the product of the collective intelligence of a diverse group of participants. The study directly challenges this framework.
Only 3.14% of Polymarket accounts qualify as qualified winners. These traders made persistent profits that held up out of sample, traded an average of 79 markets each, and consistently positioned themselves in the direction of the final results. The remaining 96% of accounts broke even by luck or lost money.
The authors found that skilled traders’ order flow predicted both subsequent period price changes and final market outcomes at statistically significant levels. A one percentage point increase in net purchases of skilled workers corresponded to an 8 basis point increase in the probability of a correct end result. The lucky winners, despite showing positive account balances, showed no significant predictive power in either test.
Polymarketit’s monthly trade volume increased from $3.3 million in December 2023 to $1.98 billion in December 2025, an increase of almost 600 times over two years. During the same period, the number of active accounts grew from around 1,600 to more than 519,000. Despite this growth, the concentration of skills remains narrow.
The study also tested skill persistence. Researchers divide events randomly into training and testing sets. Among traders classified as qualified in training, 44% retained that classification in the test set. For non-qualified losers, 51% remained in this category. For comparison, qualified mutual funds under parallel testing only retained their classification 10% of the time. The authors describe prediction markets as demonstrating unusually high persistence of both skills and anti-skills.
Qualified traders also reacted first when scheduled news arrived. In tests covering the Federal Open Market Committee (FOMC) announcements and publications of company results, only the specialized group shifted its order imbalance towards the new surprise in a narrow window around each publication. The other groups showed no consistent response. The paper examined insider trading separately.
The researchers identified 1,950 accounts that met the criteria for timing and conviction, suggesting they traded on non-public information. These accounts averaged about $15,000 in profits each and had significant price effects when traded. One documented case involved three accounts that took a position in a contract linked to Venezuelan President Nicolas Maduro hours before a covert U.S. military operation on January 3, 2026, collectively earning more than $630,000.
On April 23, 2026, the Commodity Futures Trading Commission (CFTC) filed a complaint alleging that an active duty member of the United States Army engaged in insider trading using one of these accounts. Despite these price effects, the researchers concluded that insider activity was too concentrated on isolated events to account for broad price discovery across the entire platform.
According to the study, the majority of participants finance accuracy rather than producing it. Unlucky and unqualified losers accounted for 67% of all accounts and absorbed all of the overall losses. Market makers and qualified takers together made up less than 3.5% of accounts but captured more than 30% of total gains.
The authors conclude that prediction market accuracy reflects the behavior of a small, identifiable group of informed traders whose participation is the mechanism behind price formation. Whether these traders will continue to participate as platforms grow and fees increase remains an open question that the paper leaves for future research.
