London, 27 July 2026 — EBM Weekend Read — By Brad Adams
In the 2024 US election, Polymarket called Trump’s win while the pollsters dithered over a coin-flip, and a legend was born: the crowd, with money on the line, sees the truth that experts miss. Eighteen months later the platform is valued around $9 billion, the parent company of the New York Stock Exchange has committed $2 billion to it, and prediction markets are being sold as the future of forecasting. There is only one problem with the oracle. Between 70 and 84 per cent of the people consulting it lose money, and almost all the winnings go to a tiny sliver of automated traders who are not predicting anything at all
This is the contradiction at the centre of a booming industry, and it is worth understanding before the model crosses fully into Europe. Prediction markets market themselves as truth engines powered by the wisdom of crowds. The evidence increasingly shows something closer to a very well-dressed casino, where the house edge has simply been renamed “price discovery” and handed to a handful of bots. Both things are described with the same word — market — and the gap between them is where the money is.
Join The European Business Briefing
New subscribers this quarter are entered into a draw to win a Rolex Submariner. Join 40,000+ founders, investors and executives who read EBM every day.
SubscribeThe claim: crowds beat experts
Start with the strongest version of the bull case, because it is not nonsense.
The theory is old and respectable. James Surowiecki popularised “the wisdom of crowds”; the Iowa Electronic Markets have run since 1988 and have a decent record of beating polls. The logic is clean: pollsters ask people who they want to win, while a market asks people to put money on who they think will win, and money concentrates the mind. A trader who is merely loyal loses to a trader who is correct.
And 2024 gave the theory its poster night. National polls showed a statistical tie. Polymarket had Trump above 60 cents — better than a 60 per cent implied probability — for weeks. When Trump won 312 electoral votes, the market looked like a prophet and the pollsters looked like fools. Vitalik Buterin has defended these platforms as better truth-seeking mechanisms than social media. Nate Silver joined Polymarket as an adviser. The narrative wrote itself.
If you stopped there, you would conclude the crowd had beaten the experts. Almost everyone did stop there.
The evidence: accuracy fell as the hype rose
Here is what the celebration skipped. When academics actually measured it, the flagship market was the least accurate of the lot.
A Vanderbilt study by Josh Clinton and TzuFeng Huang examined more than 2,500 political markets across four platforms in the final weeks of 2024, covering over $2 billion in trades. The results invert the story. PredictIt, the small, boring, heavily capped academic-style platform, called outcomes correctly 93 per cent of the time. Kalshi managed 78 per cent. Polymarket — the biggest, flashiest, most-quoted “oracle” of them all — came in at 67 per cent.
The market everyone cites as proof that prediction markets work was the worst performer measured. And the pattern is not random. PredictIt caps positions at $850, which keeps out the whales and attracts a crowd trading on information rather than firepower. Polymarket has no such cap, and in 2024 a single French trader — the “Trump whale” — put roughly $30 million on Trump and reportedly moved the price by several cents on his own. No manipulation was found. That is precisely the problem: one rich, opinionated person can bend the “probability” without breaking a single rule, because a thin market is not a mind, it is an order book.
The same study found the markets weren’t even efficient among themselves. Identical contracts traded at different prices on different platforms. Arbitrage gaps widened rather than closed in the final fortnight. An efficient oracle does not disagree with itself about the same fact.
Who actually profits
Now follow the money, because this is where the oracle story fully collapses.
An analysis of Polymarket’s on-chain data found that roughly 70 per cent of traders lose money; other analyses put it as high as 84 per cent. Profits are not spread among a wise crowd. They are hoarded by a rounding error: fewer than 1 per cent of wallets captured about half of all profit across the main political markets, and by one measure the top 0.04 per cent took the overwhelming majority.
And who are those winners? Mostly not clever forecasters. A study by Spain’s IMDEA Networks Institute traced 86 million transactions and found the biggest, most consistent profits went to wallets running automated strategies — arbitrage bots, market-making algorithms, high-frequency systems. Arbitrage traders alone extracted around $40 million purely from price gaps between contracts. Separately, coders have reportedly built bots earning $200,000 a month on Polymarket without ever predicting an outcome. They aren’t forecasting the future. They are harvesting the spread between a manual retail trader and the correct price, in the moments before the human catches up.
Strip the language away and the shape is unmistakable. A small number of professional, automated players make money at the expense of a large number of retail participants who think their opinion is an edge. That is not a description of an oracle. It is a description of a casino, or a poker table where the amateurs are the product.
Polymarket, to be fair, does not profit when you lose — it is a peer-to-peer exchange, not the house, and it charged near-zero fees for most of its life. But that distinction, while real, is also the trap. Because Polymarket isn’t the house, it can honestly say it isn’t gambling. The losing still happens. It just happens to the other traders, and disproportionately to the bots.
Why Wall Street is buying anyway
If most users lose, why did Intercontinental Exchange — owner of the NYSE, one of the most serious financial institutions on earth — commit $2 billion, at a valuation that has since climbed toward $9 billion and beyond?
Because ICE isn’t betting on the punters. Its stated thesis is data, not gambling. A live, liquid, real-money market generates a continuous stream of probabilities on world events that can be licensed to banks, hedge funds, media and traders. The value to ICE isn’t the €10-a-go election flutter. It is owning the pipe that produces the numbers, and stitching event-probability data into mainstream financial infrastructure. Polymarket is now the official prediction-market partner of X and Stocktwits. The bet is that “what’s the probability of X” becomes a data product as normal as a stock ticker.
This is the same move EBM has traced elsewhere: Wall Street annexing a category and relabelling it. The genius is in the vocabulary. Call it gambling and it faces gambling regulators, gambling taxes and gambling stigma. Call it a prediction market trading event contracts and it becomes a CFTC-adjacent financial instrument with a $9 billion valuation and the NYSE’s parent on the cap table. Same wager. Different word. Wildly different regulatory bill.
The gambling question Europe is already answering
That relabelling is being tested hardest not in America but here.
While US regulators dithered and Kalshi won court fights to run event contracts nationwide, European authorities looked at the same product and saw a bookmaker. From late 2024 into 2025, regulators in France, Belgium, Poland, Switzerland and Singapore, along with Australia, ordered platform-level or internet restrictions, enforcing national gambling laws against what they classed as unlicensed wagering. Europe’s fragmented gambling regimes, so often a weakness, here act as a filter the American “it’s a market” framing has struggled to pass.
The sports contracts make the blur explicit. Kalshi runs an in-house team that often trades against its own users — which is, functionally, exactly what a sportsbook does. When the platform is the counterparty and the punter loses to it, the word “market” is doing an enormous amount of work.
The verdict
Prediction markets are not useless. As a fast, liquid aggregator of breaking information they genuinely can beat a static poll, and on big, deep, heavily traded questions the price carries real signal. The 2024 headline call was correct.
But the story sold on top of that — the infallible oracle, the wisdom of the crowd, the death of the expert — does not survive the data. The most-quoted market was the least accurate. The prices move on whale money and thin liquidity. Most participants lose, and the winners are mostly machines running spreads. What is being built and sold to the public as a truth engine functions, for the ordinary user, as a casino that has borrowed the vocabulary of a stock exchange to escape the rules of one.
The oracle, in other words, is real — but it is speaking to the bots, and it is charging everyone else for the privilege of being wrong in public. Before Europe waves the model through in the name of financial innovation, it is worth being honest about which of the two things it actually is. Wall Street has already picked its word. The regulators should pick theirs with the data in front of them.
Related Analysis
