Prediction Markets Don't Reflect the Wisdom of the Crowd: Research Shows Only About 3% of Traders Determine Price Trends

This working paper, written by Roberto Gómez-Cram, Yunhan Guo, Theis Ingerslev Jensen, and Howard Kung, scholars from the London Business School and Yale University, was officially released this week. It systematically examines the core proposition upon which prediction markets rely for self-justification: that price accuracy stems from the dispersed information and judgment of numerous participants. The research team collected and analyzed data from all 1.72 million accounts and approximately $13.76 billion in transaction volume on Polymarket. However, the results point in the opposite direction: it is a small number of consistently profitable "informed" traders who truly move prices toward the correct answer.

The study shows that these few accounts have a decisive impact on price formation in event contracts. They often consistently build positions in the "correct direction" and drive price corrections, while the remaining approximately 97% of participants play a greater role as liquidity providers and counter-parties. Overall, most ordinary traders lose money when playing against this "informed minority," and the latter's profits come directly from the losses of these counterparties.

Distinguishing "true skill" from "luck" in a large number of transactions was a difficult problem that this research attempted to overcome. The research team pointed out that in a market with millions of traders, sheer chance alone can create several "legendary accounts." To eliminate the element of pure luck, the authors constructed a counterfactual world for each trader: under completely identical time points, contracts, and bet amounts, they randomly replaced the decision to buy or sell with a "coin flip" and simulated 10,000 times for each account.

Through this method, the research established a "baseline for returns under a no-information-advantage scenario" for each trading account. If an account's performance in the real world consistently and significantly exceeds this baseline, it can be inferred that the account possesses stable predictive ability; otherwise, it is more likely to be simply a matter of luck. Among the group screened for "top winners" based on absolute profit, only about 12% were able to significantly outperform this benchmark, demonstrating genuine "technical content." At the same time, the study found that approximately 60% of "lucky winners" turned from winners to losers when their performance was tested against another set of independent events, exposing the fact that their previous achievements were largely due to chance.

Despite the limited number of "high-level" players, their positive contribution to the overall information efficiency of the market is quite significant. When the proportion of these accounts in trading volume increases, contract prices converge to the true outcome faster and more accurately, especially in the final stages near the event settlement. The study also found that, compared to most participants, traders with advantageous information are often the first to react to new information, such as when the Federal Reserve announces its policy or when large companies release their earnings reports. They quickly adjust their positions, while the behavior of other accounts at these nodes does not exhibit consistency or foresight.

However, this information advantage also raises a more difficult question: how to balance market efficiency and compliance boundaries when the relevant information has not yet been made public, or even should not have been available to traders. Polymarket and another regulated prediction market platform, Kalshi, have repeatedly emphasized that their platforms prohibit trading based on non-public information. However, this research, combined with recent cases, points out that this risk is no longer just a theoretical assumption in reality.

The article illustrates this risk with the example of the United States' actions against Venezuelan leader Nicolás Maduro in January of this year. In the days and hours before the United States implemented measures to remove Maduro from power, three newly registered accounts suddenly appeared on Polymarket, heavily betting on the contract "Will Maduro be removed?" At the time, the market gave a success probability of only about 10%. These few new accounts placed unusually large orders before the price fluctuated significantly, with single buy orders often reaching tens of thousands of contracts. After the action was officially launched and the event occurred, these three traders collectively profited more than $630,000. Subsequently, two of the accounts almost completely stopped trading, and the other also slowed down significantly, approaching a "dormant" state. The research also emphasizes that there is currently no evidence that these accounts engaged in illegal or non-compliant behavior, but the case itself is enough to demonstrate the vulnerability of prediction markets to insider information.

This case also echoes the recent "special forces betting incident" that has attracted public attention. Previously, a member of a U.S. special forces unit involved in the operation against Maduro was arrested by U.S. authorities for betting on the related classified operation on Polymarket. This seemingly isolated scandal is now seen as an extreme sample in a larger picture within the framework of the paper: a small number of traders with sensitive information not only deeply influence price trends but also pose a potential impact on platform compliance.

The research further points out that in a small number of events that can be clearly identified as "suspected insider trading," the price impact of these transactions is even more severe: with the same capital scale, the driving force of insider funds on prices is about 7 to 12 times that of ordinary "high-level trading." However, the authors also emphasize that such transactions are extremely concentrated, appearing only in a few events, and are not the usual driving force behind daily price discovery. Most of the time, the accuracy of market pricing still relies mainly on those "repeat" skilled traders who consistently outperform the benchmark across multiple events, rather than one-time deals.

From a broader perspective, this research poses a fundamental challenge to the traditional narrative of "prediction markets as products of collective wisdom." The conclusion of the paper is that these markets appear to give prices close to the true probability not because of "strength in numbers," but because a small number of participants with real information advantages and judgment are continuously correcting prices, while others are merely providing liquidity and chips. With the regulatory framework still being formed and the compliance boundaries still unclear, this finding is sure to spark a new round of debate about the positioning and rule design of prediction markets: if price accuracy depends not on the "masses" but on the "informed," how to ensure both information efficiency and prevent insider trading and unfair profits will be a problem that regulators and platform operators cannot avoid.