A new academic study challenges one of the most common claims made about prediction markets: that their accuracy primarily reflects the wisdom of a large and diverse crowd. In the case of Polymarket, the researchers argue that price discovery is instead driven by a very small minority of informed and consistently skilled traders.
The paper, titled Prediction Market Accuracy: Crowd Wisdom or Informed Minority?, was posted to SSRN on April 20, 2026 and revised on April 25, 2026. It was authored by Roberto Gomez-Cram, Yunhan Guo, and Howard Kung of London Business School, along with Theis Ingerslev Jensen of Yale University. Using Polymarket’s full transaction history, the researchers examined one of the broadest data sets yet assembled for a crypto-native prediction market.
The scope of the analysis was substantial. According to the study, the sample covered 98,906 events, 210,322 markets, 1.72 million accounts, and $13.76 billion in aggregate trading volume. With that data, the authors attempted to determine whether profitable trading on the platform reflected genuine skill, random luck, or some other structural feature of market participation.
A Small Skilled Minority, Not the Crowd, Appears to Matter Most
To separate skill from chance, the authors applied a statistical approach known as a sign-randomization test. This allowed them to classify traders into groups such as skilled winners, lucky winners, and unskilled losers. Their central result was striking: only 3.14% of Polymarket accounts qualified as genuinely skilled winners.
Those accounts did not merely post positive returns. The researchers found that they traded across an average of 79 markets, repeatedly positioned themselves in the direction of eventual outcomes, and produced profits that persisted outside the original sample used to classify them. In other words, these traders did not simply get lucky on a few bets; they appeared to contribute systematically to market efficiency.
The study also found measurable predictive power in the order flow of this group. A one-percentage-point increase in skilled net buying was associated with an 8-basis-point increase in the probability of the correct final outcome. By contrast, “lucky” profitable traders did not show meaningful forecasting power for either subsequent price movements or final market resolution. That distinction is central to the paper’s argument: profitability alone is not evidence of informational value.
These findings run against a widely repeated industry narrative. Prediction market operators, including Polymarket and other platforms such as Kalshi, often describe market accuracy as the emergent product of broad participation and collective intelligence. The paper does not deny that many users take part, but it argues that the actual mechanism of price formation is far more concentrated than that framing suggests.
Rapid Growth Has Not Reduced Skill Concentration
The study arrives after a period of dramatic expansion for Polymarket. Monthly trading volume reportedly grew from $3.3 million in December 2023 to $1.98 billion in December 2025, representing an increase of nearly 600 times in about two years. Over the same period, active accounts rose from roughly 1,600 to more than 519,000.
Despite that growth, the concentration of informational skill remained narrow. The evidence suggests that simply adding more participants did not make market accuracy meaningfully more decentralized. Instead, a very small subset of traders continued to account for much of the useful signal embedded in prices.
The paper also tested whether trading skill persisted over time. Researchers randomly split events into training and test sets, then checked whether traders classified as skilled in one sample remained skilled in the other. Among those labeled skilled in the training set, 44% retained that classification out of sample. For unskilled losers, 51% stayed in the same category. As a benchmark, the paper notes that skilled mutual funds in a parallel test retained their classification only 10% of the time.
That comparison led the authors to describe prediction markets as showing unusually high persistence not only of skill, but also of “anti-skill.” In practical terms, some traders consistently help prices converge toward the right outcome, while others repeatedly lose money without adding equivalent informational value.
Reaction to News and the Role of Informed Trading
Another part of the study examined how different trader groups responded to scheduled information releases, including Federal Open Market Committee (FOMC) announcements and corporate earnings reports. The authors found that only the skilled group adjusted order imbalance in the direction of the news surprise within a narrow time window around those releases.
Other trader categories did not display a consistent pattern. That result reinforces the broader claim that informed traders, rather than the median participant, are the group most responsible for integrating new information into prices. In essence, the market may be socially broad, but informationally narrow.
The paper also considered insider trading as a possible explanation for Polymarket’s forecasting performance. Researchers identified 1,950 accounts that met timing and conviction thresholds suggesting they may have traded on non-public information. These accounts averaged around $15,000 in profits each and appeared to have outsized price impact when active.
One case highlighted by the researchers involved three accounts that traded a contract tied to the removal of Venezuelan President Nicolas Maduro from power just hours before a secret U.S. military operation on January 3, 2026. According to the paper, those accounts collectively earned more than $630,000.
CFTC Complaint Highlights Regulatory Risk
The question of insider trading has already drawn regulatory attention. On April 23, 2026, the U.S. Commodity Futures Trading Commission filed a complaint alleging that an active-duty U.S. Army service member engaged in insider trading using one of the accounts linked to the Maduro-related market.
Even so, the authors conclude that insider activity, while important in specific episodes, is too concentrated in isolated events to explain price discovery across the platform as a whole. In other words, the study does not argue that Polymarket’s overall accuracy is primarily an insider-trading phenomenon. Rather, it suggests that accuracy is generated by a broader but still highly concentrated set of informed traders whose advantages appear persistent.
The distribution of gains and losses within the platform adds another layer to that conclusion. The paper says that unlucky and unskilled losers accounted for 67% of all accounts and absorbed the entirety of aggregate losses. Meanwhile, market makers and skilled takers together made up fewer than 3.5% of accounts but captured more than 30% of total gains.
That framing leads to one of the paper’s most provocative claims: the majority of participants may be financing market accuracy rather than creating it. Prices still become informative, but not because all users contribute equally to forecasting power. Instead, the informed minority appears to do most of the heavy lifting, while the broader user base provides liquidity, opposing flow, and loss absorption.
For the prediction market industry, the implications are significant. If market quality depends disproportionately on a tiny cohort of sophisticated traders, then platform design, fees, access, and regulatory treatment could all affect price accuracy more than commonly assumed. The paper leaves open an important question for future research: whether these highly skilled traders will continue to participate as platforms scale and operating costs change.
For now, the study offers a sharp reframing of how prediction markets may work in practice. Rather than a pure expression of decentralized crowd wisdom, Polymarket may function more like a venue where a small number of informed traders identify and monetize mispricing—while the crowd makes that process possible.