The Wisdom of Crowds Is the Wrong Slogan for a Market 3% of Traders Price

Key takeaways

• A working paper covering every Polymarket trade from 2023 to 2025 found roughly 3% of traders account for most of the price discovery.
• Only about 12% of the biggest winners beat a coin-flip benchmark once their bets were simulated ten thousand times, and around 60% of apparent winners turned into losers when tested on a separate set of events.
• The top 1% of profitable users captured 76.5% of all profits in a study covering $67B of Polymarket volume.
• Whale, high-frequency and power-trader tiers held 81.4% of notional across 12.6% of addresses.
• Dr. Robin Hanson's response is that this describes financial markets generally, and that prediction markets should be compared to other information institutions rather than to an idealised crowd.
• Dr. Rajiv Sethi found that 87% to 90% of traders in the 2012 election markets never once switched the direction of their bets, leaving roughly 5% who moved with the news.

What the trade tape says about the industry's favourite phrase

Prediction markets are sold on information aggregation, the idea that dispersed knowledge from many participants gets priced into a probability that beats any single expert. In 2026 researchers finally had enough of the onchain record to test that story at the level of individual trades. The picture that came back is closer to how any financial market works, and very far from the egalitarian framing in the marketing. This piece covers the evidence and the two most serious responses to it, from the economist who founded the field and from the economist who has studied these markets longest. Both interviews were conducted for the Hodl Up Prediction Markets Report 2026.

The concentration evidence, in one place

Four separate studies using different methods point the same way.

The mechanism behind the winning is mundane. Winners tend to provide liquidity with limit orders that resolve well, losers tend to take liquidity with market orders, and the ground-truth record shows high-spread niche markets are where informed specialists cluster rather than where noise traders congregate.

One finding cuts against the loudest critics. The same $67B study reports that insider activity is unlikely to explain the performance of the largest winners, who mostly made money by providing liquidity rather than by trading on private information. For why the volume figures underneath these studies deserve caution of their own, see Nobody Knows How Big Prediction Markets Are.

Hanson: you are comparing it to the wrong thing

Dr. Robin Hanson, whose market scoring-rule work founded the field, does not dispute the data. His argument is that it describes markets generally.

"A lot of what many people don't like about prediction markets is what they don't like about ordinary financial markets, but maybe just don't know. Ordinary financial markets have also had this feature for a long time, which is that most traders lose on average in their trades."

His reframing is the substantive part. If a prediction market is an information institution, the comparison should be to other information institutions rather than to an idealised crowd.

"If we think about prediction markets as an information institution, I think the comparison should be to other information institutions, like journalism or gossip or academic journals. And then in those institutions, we're okay with the idea that there's net resources going into those institutions, because the information coming out is the value."

He concedes the funding model is unattractive, noting that he would rather these markets were paid for some way other than by taxing fools, while observing that no other funding model has materialised and the fools have volunteered. His practical advice to individuals is blunt: speculate only if you are world class, and test that belief at small size before scaling it. He goes further on using markets to make decisions rather than just forecast in our futarchy episode.

Sethi: whether concentration hurts accuracy is an empirical question

Dr. Rajiv Sethi's contribution is to refuse both easy answers. The argument that markets must be accurate because money is at stake fails. So does the argument that they cannot be accurate because the trading population is young, male and unrepresentative. Neither can be settled by reasoning, only by data.

His own evidence complicates the crowd story further. In work with David Rothschild of Microsoft Research on the 2012 presidential election, using transaction-level InTrade data, he found that roughly 87% to 90% of traders never once switched the direction of their bets. They were Romney bettors or Obama bettors rather than updaters, which leaves perhaps 5% who moved with the news.

"Those few people who are really sensitive to the biases of others are going to play a really pivotal role in determining where the price goes. Whether that leads to good or bad forecasting is ultimately an empirical question."

That is the honest resting place. A price set by a thin layer of attentive traders exploiting everyone else's entrenched positions might be an excellent forecast or a poor one. What it is not is the wisdom of a crowd. Where the accuracy evidence is strongest, see What the Fed Found When It Tested Kalshi Against Its Own Decisions and our piece on prediction markets vs polls.

The full evidence, including the dataset-level figures from the published onchain archives, is in the Prediction Markets Report 2026. Hanson's complete interview is in our futarchy episode, Sethi's is here, and the report summary carries the headline findings.

None of this is investment advice.