Polymarket’s core sales pitch has always been simple: markets work because crowds are smart. A new academic paper says that story is incomplete. Researchers from London Business School and Yale found that just 3.14% of Polymarket accounts qualify as “skilled winners,” while the rest either lose, break even by luck, or fail to add meaningful price discovery. That matters because it shifts the platform’s edge away from collective wisdom and toward a tiny, persistent minority of informed traders.
What the new study actually found
A working paper highlighted on April 26, 2026, by The Block and Bitcoin News analyzed Polymarket’s full transaction history from 2023 through 2025 and reached a blunt conclusion: the platform’s predictive power comes from an informed minority, not the crowd at large. The paper, titled Prediction Market Accuracy: Crowd Wisdom or Informed Minority?, was published to SSRN on April 20, 2026, and revised on April 25, 2026. The authors are Roberto Gomez-Cram, Yunhan Guo, and Howard Kung of London Business School, along with Theis Ingerslev Jensen of Yale University.
The dataset is big enough to matter. According to the paper details summarized in coverage of the study, the researchers examined 98,906 events, 210,322 markets, $13.76 billion in trading volume, and 1.72 million accounts. Out of that entire user base, only 3.14% of accounts met the paper’s threshold for “skilled winners.” That means roughly 53,900 accounts out of 1.72 million generated the kind of repeatable performance that held up under statistical testing, while about 1.67 million did not.
That ratio is the real headline. It is not just that a small group wins. It is that the overwhelming majority does not appear to possess durable forecasting skill. The study used a sign-randomization test to separate genuine skill from random positive outcomes, an important distinction in markets where a lucky streak can look like expertise for a while. In plain English: some traders made money, but only a narrow slice did so in a way that consistently predicted future price moves and final outcomes.
Why the 3.14% figure changes the “wisdom of crowds” narrative
Prediction markets are often marketed as a cleaner, incentive-driven version of public opinion. The idea is that when enough people put money behind a view, prices converge toward truth. This paper complicates that. It suggests Polymarket prices are not mainly the product of broad crowd intelligence. They are shaped disproportionately by a small cadre of traders whose order flow contains real information.
That is a very different mechanism. If only 3.14% of users are consistently informed, then the crowd is less a forecasting engine and more the liquidity pool those informed traders trade against. The Block’s summary of the paper put it even more sharply: everyone else funds their gains. That framing may sound harsh, but it fits the economics. In any market, informed participants need counterparties. On Polymarket, the study suggests the counterparties are mostly less-skilled users.
There is another number that stands out. A one-percentage-point increase in skilled net buying corresponded to an 8 basis point increase in the probability of the correct final outcome, according to the paper summary reported by Bitcoin News. That is not noise. It means the trades of this minority were measurably informative in real time. By contrast, “lucky winners” showed no meaningful predictive power when tested out of sample.
That distinction matters for anyone trying to copy wallets, chase leaderboard profits, or treat headline PnL as proof of insight. A profitable account is not automatically a smart account. The paper argues that persistence and predictive value matter more than one-off gains.
The overlooked angle: Polymarket may be efficient for prices, but not for users
Most coverage of prediction markets focuses on whether market odds beat polls, pundits, or analysts. The more interesting takeaway here is different: a market can still produce useful prices even if most participants lose. That is the gap many headlines miss.
In other words, Polymarket may be informationally efficient at the top layer while being economically unforgiving at the user layer. Those are not contradictory facts. They can both be true. Skilled traders can push prices toward accurate outcomes precisely because less-skilled traders provide the volume, mispricings, and emotional flow needed to exploit.
This is where the study becomes more than a curiosity. It turns the platform’s social narrative on its head. The crowd may still help by creating liquidity and surfacing diverse opinions, but the final pricing signal appears to come from a narrow elite. That is closer to “wisdom of a filtered minority” than wisdom of crowds.
It also lines up with other research around Polymarket’s structure. Separate reporting in 2025 cited a Columbia University study alleging that a meaningful share of Polymarket activity was inflated by artificial or wash-like trading behavior. Another 2025 study, reported by Yahoo Finance, found users lost millions to “bot-like” bettors exploiting arbitrage opportunities. Those studies addressed different questions, but together they point to the same broader reality: raw activity on Polymarket does not always equal broad-based, healthy, informed participation.
What this means for traders, researchers, and regulators
For traders, the message is uncomfortable but useful. If you are not in that 3.14%, you are probably not the one extracting edge. You may be donating spread, paying for conviction, or confusing event knowledge with market skill. Those are not the same thing. Knowing politics, sports, or crypto helps, but timing, sizing, and price sensitivity matter just as much in a prediction market.
For researchers, the paper offers a cleaner framework for evaluating market quality. Instead of asking whether Polymarket prices are accurate in aggregate, it asks who makes them accurate. That is a better question. It gets at market microstructure, not just outcomes.
For regulators, the findings may add fuel to ongoing scrutiny. The Block noted that the paper also found suspected insider activity moved prices roughly 7 to 12 times more aggressively than skilled trades, even if that activity was too concentrated in isolated events to explain overall market accuracy. That does not mean insider trading drives the whole platform. It does mean market integrity questions are not theoretical.
There is also a legal backdrop. Polymarket has faced regulatory pressure before, including its 2022 settlement with the CFTC that barred U.S. users from the platform. So when a study shows a tiny minority drives price discovery and potentially informed actors can move odds sharply, policymakers are unlikely to ignore it.
Bottom line
The new Polymarket study does not prove prediction markets are broken. It proves something subtler, and maybe more important. Their accuracy may depend less on mass intelligence than on a small group of repeat winners who know how to identify and trade mispriced probabilities. Only 3.14% of users appear to do that consistently. Everyone else? They are mostly noise, liquidity, or exit liquidity.
That is not the romantic version of market wisdom. It is the realistic one. And for anyone treating Polymarket as a democratic forecasting machine, this paper is a needed correction.
Frequently Asked Questions
What does “only 3% of users beat the crowd” mean?
It means a new study found that just 3.14% of Polymarket accounts qualified as “skilled winners,” meaning their trades consistently predicted short-term price moves and final event outcomes. The remaining users either lost money, broke even by luck, or did not show repeatable forecasting skill.
Who conducted the Polymarket smart money study?
The paper was written by Roberto Gomez-Cram, Yunhan Guo, and Howard Kung of London Business School, plus Theis Ingerslev Jensen of Yale University. It was published on SSRN on April 20, 2026, and revised on April 25, 2026.
How large was the dataset in the study?
According to summaries of the paper, the researchers analyzed 98,906 events, 210,322 markets, $13.76 billion in trading volume, and 1.72 million accounts across Polymarket’s 2023-2025 transaction history.
Does this mean Polymarket is inaccurate?
No. The study suggests Polymarket can still produce informative prices. The key point is that the accuracy appears to come mostly from a small informed minority, not from the average user or the crowd as a whole.
Why is this important for ordinary users?
Because it challenges the idea that simply participating in a prediction market gives you an edge. If only a tiny share of users show persistent skill, most traders are likely competing without a durable advantage and may be providing liquidity to better-informed participants.
Leave a comment