The Green Beret arrested for betting on a classified US raid seemed like a one-off scandal for prediction markets. A new study suggests it may be a more worrying fact: an extreme example of the small group of informed traders who, as the soldier is accused, actually move prices at Polymarket, while the crowd around them loses money.
The study, part of a working paper published this week by Roberto Gómez-Cram, Yunhan Guo, Theis Ingerslev Jensen and Howard Kung of London Business School and Yale, directly tests the industry’s central claim that markets work because of the mass knowledge of their participants.
Using all Polymarket trades between 2023 and 2025, the authors conclude that it is actually a small group of informed traders that moves prices. Researchers analyzed 1.72 million accounts and $13.76 billion in trading volume, and found that only 3% of traders account for the majority of price discovery, meaning they are the ones moving prices toward the correct outcome.
These traders constantly predict outcomes and move prices in the right direction. The remaining 97% mostly do not. They provide liquidity and generate volume, but as a whole, they are on the losing side of trades compared to the informed minority, whose profits come directly from those positions.
The hard part is distinguishing skill from luck. With over a million traders on Polymarket, many will rack up big profits just by chance.
To filter that out, the authors repeated each trader’s bets 10,000 times, keeping everything the same except the direction.
The same markets, the same moments, the same amounts of dollars, but by flipping a coin it was decided whether it was bought or sold. That gave them a benchmark of what each trader’s profits would be like without an actual edge. If the actual results consistently outperform the coin toss, that’s skill. If not, it’s luck.
The findings show that among the biggest winners by gross profit, only 12% surpassed the benchmark, and many apparent winners did not stay that way: about 60% of the “lucky winners” become losers when their performance is compared to a separate sample of events.
Their activity improves market precision. When trained participants account for a larger proportion of trades, prices move closer to the correct outcome, especially in the final stretch before resolution. They are also the first to react when new information arrives, changing positions in response to events such as Federal Reserve announcements or corporate earnings, while other traders show an inconsistent reaction.
The same advantage that makes expert traders valuable for price discovery raises a more difficult question when that information is not public, or not supposed to be.
Both Polymarket and Kalshi have said that trading on non-public information is strictly against their rules.
The document bases this risk on a specific case: the removal by the United States of Nicolás Maduro from power in Venezuela in January. In the days and hours before the operation, three newly created Polymarket accounts accumulated into a contract asking if Maduro would be removed. At the time, the market priced the odds at about 10%.
The new accounts placed unusually large bets, including orders for tens of thousands of shares, before the price moved. At the time of the raid, the accounts together earned more than $630,000. Two stopped trading completely soon after and the third remained virtually inactive. There is no evidence of irregularities in these accounts.
Insider trading, when it occurs, moves prices per dollar even more aggressively, about seven to 12 times more than typical insider trading. But they are rare and focus on a handful of events, not the daily pricing engine. Most of the time, market accuracy still depends on repeat traders who consistently get better results than one-time bets.
The findings challenge the idea that prediction markets are powered by crowds. They seem to work because of who is informed.
