A study finds Polymarket’s predictive accuracy comes from a small group of better-informed traders rather than the collective intelligence of many users.

The Block Crypto reported on April 26 that a paper analysing all Polymarket trading from 2023 to 2025 concluded the platform’s accuracy reflects the “wisdom of an informed few,” not the “wisdom of crowds.”

The paper, Prediction Market Accuracy, was written by Roberto Gomez-Cram (로베르토 고메스-크람), Yunhan Guo (윤한 궈) and Howard Kung (하워드 쿵) of London Business School and Teis Ingerslev Jessen (테이스 잉어슬레브 옌센) of Yale University. It analysed 1.72 million accounts, 210,322 markets and about $13.76 billion in trading volume.

The researchers classified only 3.14 percent of all accounts as “skilled winners.” These accounts consistently predicted short-term price moves and final outcomes through their order flow. This group, along with market makers, took more than 30 percent of total profits, while accounting for less than 3.5 percent of accounts.

By contrast, 67 percent of accounts were classified as unlucky or unskilled loss-making accounts. They were calculated to have absorbed all losses across the platform. The researchers said it was hard to gauge skill based on profit and loss alone. When they randomly reversed each trader’s buy and sell direction and recalculated 10,000 times, the overlap between top earners and the skilled group was only 12 percent. About 60 percent of “lucky winners” also turned into losses in a separate sample of events.

Skill also showed strong persistence. Of the accounts classified as skilled in a training sample, 44 percent remained skilled in a separate sample. In a parallel test using the same method on active mutual funds, that ratio was about 10 percent.

The researchers also identified 1,950 accounts suspected of insider trading. These accounts were opened just before a single event, then became dormant after the outcome was confirmed, and their price impact per dollar was 7 to 12 times larger than that of skilled traders. The researchers judged it was hard to view them as a factor determining overall platform accuracy because they were concentrated in specific events.

The paper also covered three accounts opened between Dec. 27 and Jan. 3. The accounts earned more than $630,000 by betting on Maduro’s ouster before a U.S. military operation was made public. The case coincides with the first insider-trading case in which the U.S. Commodity Futures Trading Commission sued U.S. Army Sergeant Major Gannon Ken Van Dyke on Thursday on allegations of trading event contracts using confidential information.

The researchers also questioned industry explanations that prediction markets harness “collective intelligence.” The paper also cited related remarks by Kalshi CEO Tarek Mansour and Polymarket CEO Shayne Coplan.