The 3% Who Make Polymarket Accurate Face a Squeeze

Just 3.14% of Polymarket accounts show a real trading edge, and that group, with a sliver of market makers, takes over 30% of the gains. Yale economist Theis Ingerslev Jensen said he expects competition to cut the skilled share below 1%.

The working paper covers 1.72 million accounts, 98,906 events, and $13.76 billion of Polymarket volume from 2023 to 2025. It finds the crowd does not set the prices. It funds the people who do.

The Crowd Was Never the Engine

Prediction-market founders still sell the usual story: lots of small bets, skin in the game, a crowd that is right in the aggregate. Jensen, an assistant professor of finance at the Yale School of Management, said that story feels reasonable. He and three London Business School co-authors tested it anyway against the way ordinary financial markets work, where a thin skilled layer does the real pricing.

The paper is Crowd Wisdom or Informed Minority, written by Roberto Gómez-Cram, Yunhan Guo, Howard Kung, and Jensen. Using Polymarket’s public blockchain record, they conclude that a small group of informed traders drive prices and take a large share of the winnings. Accuracy, they write, comes from that minority, not from the crowd and not from insiders.

Skilled accounts do three things that look like a newsroom desk, not a mob.

HOW SKILLED ACCOUNTS TRADE

• Public news: They buy and sell in the direction of scheduled releases such as Federal Reserve decisions and company earnings, and they do it faster than the rest of the book.
• Broken prices: They close gaps when related contracts disagree, the same law-of-one-price work a cash-equity desk does all day.
• Crowd errors: They take the other side of behavioral mistakes, which is where most of the volume still sits.

They also trade widely. Skilled winners in the paper’s cut entered 79 markets on average, a breadth the authors use to separate them from one-off insiders. Jensen said insider trades do show up, but they are sporadic and small, and he does not think they systematically drive price discovery.

Everyone has skin in the game, and a very strong incentive to state their true beliefs. It feels reasonable when people say, ‘Oh, they work because of the wisdom of crowds.’

Theis Ingerslev Jensen, assistant professor of finance, Yale School of Management

The remaining majority, the authors conclude, does not produce accuracy. It funds it.

A Coin-Toss Test Against 1.72 Million Accounts

Raw profit is a bad skill meter. The authors kept each trader’s markets, timing, prices, and size, then flipped the buy-or-sell direction at random and reran that history 10,000 times. The test asks whether the real profit-and-loss beat a coin toss. Most accounts failed it.

THE FIVE TRADER GROUPS

Only 12% of the biggest winners by raw dollars also clear the skill test. About 60% of the lucky-winner group flipped to losses on a second sample of events. Skill that does clear the test tends to stick: 44% of accounts tagged as skilled in one half of their events stayed skilled in the other half, against about 10% in a parallel check on active mutual funds.

Jensen put it in plain language. Most people treat a hot streak as luck that will not repeat. On Polymarket, a skilled tag in one period made the next period far more likely to look skilled too.

Who Pays When 3 Percent Keep Winning

A Polymarket trade is a yes-or-no share. Users buy and sell event contracts with binary outcomes priced from $0 to $1. A winning share pays $1. A 40-cent yes is a 40% implied chance. That structure makes the transfer easy to see, because someone has to hold the losing side.

WHERE THE MONEY GOES

• The skilled slice: 3.14% of accounts, plus 0.1% who mainly make markets, together under 3.5% of the book and over 30% of total gains.
• The lucky majority of wins: 29.0% of accounts made money that the coin-toss test could not tell from chance, and they took the remaining 69% of gains.
• The loss pool: 61.4% unlucky losers and 6.4% unskilled losers, about twice the skilled group on the unskilled side, absorbed every aggregate loss on the platform.

That split is harsher than a whale-caste headline. The people who move prices are a small skilled class. Most of the dollars that were won still went to accounts that did not pass the skill test. The people who lost, on net, paid both groups.

Retail playbooks still read like 2025. Buy yes and no when the two prices sum to less than $1, copy a hot wallet, or run a bot that polls the book every few seconds. Those gaps are exactly the thing professional flow compresses. A dashboard that advertises a straight equity line at a 49% win rate is marketing, not a method that survives a 10,000-run test.

ICE’s $1.6 Billion Bet on Tighter Spreads

The paper’s sample ends in 2025. The capital that followed is the point. Intercontinental Exchange, the parent of the New York Stock Exchange, put real cash behind the order book that those skilled accounts had been trading.

THE CASH THAT FOLLOWED THE BOOK

1. October 2025: ICE makes an initial $1 billion direct investment in Polymarket as part of a larger arrangement.
2. March 27, 2026: ICE completes a new $600 million direct cash investment in a Polymarket equity raise and says it expects to buy up to $40 million of shares from existing holders.
3. April 20, 2026: The Yale and London Business School working paper is posted, covering the $13.76 billion sample that predates this wave of cash.
4. April 2026: Combined monthly volume on Kalshi and Polymarket International reaches about $24 billion in Pew Research Center’s tally.

The two ICE cheques are $1.6 billion of completed direct cash. ICE said the stake is not expected to move its own results or capital-return plans in a material way. For Polymarket, it is a different kind of signal: the people who run the world’s largest stock exchange now have a reason to care whether these books look like real markets.

Kalshi, the CFTC-regulated rival, pulled in its own Wall Street money over the same stretch and took the lead on monthly volume. Prediction markets stopped being a crypto side bet and started looking like a product an exchange group would own.

Jensen Sees the Skilled Share Falling Below 1%

Jensen’s forecast is the mechanical result of that cash. If more skilled people show up, they compete, and prices get more correct. He said he expects the skilled share to fall from roughly 3% to below 1%. He also said only the very, very best, hedge funds in his example, will still beat these markets.

Julie Hoover, a Bank of America equity analyst, described the same squeeze from the trading desk. As markets get more efficient and spreads get tighter, she said, mispricing and arbitrage get harder to find. The 3-cent yes-plus-no lock that still circulates as a starter bot is the first thing that dies in a tighter book.

If you have a lot of skilled people, then they compete, and in doing so, they make prices more correct.

Theis Ingerslev Jensen, assistant professor of finance, Yale School of Management

He added that an efficient market makes it harder to be wrong in a consistent way. Frequent traders can still lose after fees. Quoted prices, though, should sit closer to the risk they are taking. That is a better forecast tool. It is a worse hunt for easy money.

Federal Reserve Board economist Anthony M. Diercks, with Jared Dean Katz of Northwestern’s Kellogg School and Jonathan H. Wright of Johns Hopkins, ran a separate check on Kalshi’s macro contracts. Headline consumer-price forecasts from that book beat the Bloomberg consensus by a statistically meaningful margin. Core inflation and unemployment looked about the same as the consensus. Rate-decision odds sat in line with professional surveys. Better prices have a use that does not require a retail winner.

Thin Markets Still Shelter Specialists

Hoover did not write off every small account. She said specialists can still hold an edge in niche markets, because the contract list is wide enough to let someone become the expert, and even the market maker, in a quiet corner. Jensen made the same point from the other side of the size divide. Large firms tend to skip low-liquidity books, he said, because a small order can chew up their own edge.

That leftover is narrow. It is weather in a mid-size city, a mention market on a single word in a speech, a contract too small to bother a hedge-fund allocator. It is not the liquid sports and politics books that now carry most of the volume. And it is not a promise that a copied whale wallet will keep paying.

The CFTC has also opened insider-trading inquiries on Polymarket contracts, including a complaint tied to a market on Nicolás Maduro’s hold on power. Jensen’s own read of the 2023-to-2025 tape is that insider flow is not the engine. The engine is public news, relative-value, and fading the crowd, done over and over across dozens of markets. That is the work a funded desk can scale. It is also the work that desk will compete away.

Platforms Collect Fees Either Way

Pew Research Center, using figures from The Block, found combined monthly trading on Kalshi and Polymarket International rose from less than $5 billion to about $24 billion between September 2025 and April 2026. Legal U.S. sportsbooks handled around $14 billion a month in 2025, on average. In April 2026, Polymarket’s U.S. book did $1.3 billion; the international book did $9 billion.

WHERE THE VOLUME SITS

• Sports: 80% of Kalshi volume and 39% of Polymarket volume since July 2024.
• Politics: 4% on Kalshi and 32% on Polymarket over the same span.
• Crypto: 7% on Kalshi and 20% on Polymarket.
• The three together: 91% of Kalshi volume and 90% of Polymarket volume.

Fees come off that stack whether the skilled share is 3.14% or under 1%. A tighter book is a better hedge and a cleaner data feed. It is also a worse place to show up with a tutorial bot and a few hundred dollars. Jensen’s line is the one the new backers are already underwriting. He said the people who still get paid to be right will be hedge funds.

Disclaimer: This article is news reporting and analysis of a working paper and of public figures on prediction-market platforms. It is for information only and is not investment, trading, or betting advice, and it is not a recommendation to buy or sell any event contract, share, or token. Readers who are considering a trade should speak with a licensed financial adviser who can review their own situation, including tax treatment and local rules on event contracts. The figures and statuses here reflect the cited papers, company statements, and data sets as they stood on the dates given in the piece and can change as markets move and as the working paper is revised.