Polymarket Got Trump Right. But the Data Exposes the Limits of Prediction Markets

The platform’s 2024 election call helped turn prediction markets into an $8 billion financial phenomenon. New research suggests the signal can be valuable — but the profits flow overwhelmingly to a small group of sophisticated traders.

In the 2024 US presidential election, Polymarket appeared to see something that many pollsters could not.

While national polling showed Donald Trump and Kamala Harris locked in an exceptionally close contest, Trump’s probability of victory on Polymarket had risen to approximately 62 per cent by late October. He went on to win all seven swing states and secure 312 electoral votes.

A legend was born: when people are required to put money behind their views, the argument went, markets reveal truths that experts, pollsters and commentators miss.

That reputation has since become extraordinarily valuable. In October 2025, Intercontinental Exchange — the owner of the New York Stock Exchange — agreed to invest up to $2 billion in Polymarket, valuing the company at approximately $8 billion before the investment. ICE also agreed to distribute Polymarket’s event-based data to institutional investors around the world.

The bet is no longer simply that people enjoy wagering on elections, sport and world events. It is that prediction-market probabilities will become a financial data product: a continuously updating measure of what the world believes will happen next.

There is a compelling idea here. There is also a large gap between saying that prediction markets contain useful information and saying that they are reliable “truth engines”.

New research into Polymarket and its competitors suggests that the signal is real, but inconsistent. It also suggests that the people producing useful prices are not necessarily the people making the money.

The case for prediction markets

The intellectual case for prediction markets is neither new nor foolish.

A conventional poll asks people what they intend to do or what they believe. A prediction market asks participants to buy and sell contracts whose value depends on an outcome occurring. Prices adjust as traders respond to polling, breaking news, economic data and the behaviour of other participants.

Because money is at risk, traders have an incentive to separate what they want to happen from what they believe is likely to happen.

Markets also move faster than polls. A survey may take several days to conduct and publish. A prediction-market price can change within seconds of a political announcement, court judgment, military development or economic release.

Some research supports the central claim. A 2025 study comparing Polymarket with traditional polling concluded that the platform performed better in predicting the 2024 presidential result, particularly across the swing states.

That does not prove that every price is accurate. It does show why journalists, investors and political analysts have become interested in prediction markets as another source of information.

The mistake is not taking the markets seriously.

The mistake is treating a market price as an objective verdict produced by a single, all-knowing crowd.

One correct election call is not enough

The strongest challenge to the oracle narrative comes from research by Vanderbilt University academics Joshua Clinton and TzuFeng Huang.

They examined more than 2,500 political markets across several platforms during the final five weeks of the 2024 US election campaign, involving approximately $2.4 billion in transactions.

Using the researchers’ measure of whether markets predicted outcomes better than chance, PredictIt achieved an accuracy rate of 93 per cent. Kalshi recorded 78 per cent. Polymarket, despite being the largest and most frequently quoted platform, recorded 67 per cent.

This does not mean Polymarket failed to predict the presidential winner. It plainly did not. It means the platform’s correct headline call should not be confused with consistently superior forecasting across thousands of individual political contracts.

The most famous market was, under the Vanderbilt study’s broader measurement, the least accurate of the three major commercial platforms examined.

Prediction markets may sometimes outperform experts and polls. They can also be wrong, inefficient or poorly calibrated, particularly where liquidity is limited or the wording of a contract is ambiguous.

The lesson is less exciting than the marketing, but more useful: a market price is evidence, not revelation.

The crowd is weighted by money

Prediction markets are often described as representing the wisdom of the crowd. But they are not democratic surveys.

A thousand participants do not receive one vote each. The influence of a trader depends on how much money that person is prepared to commit and how much liquidity is available on the other side of the trade.

That distinction became visible during the 2024 election.

Polymarket investigated several accounts placing large wagers on a Trump victory and concluded that they were controlled by one French trader with extensive financial experience. The company said it found no evidence that the trader was attempting to manipulate the market.

At the time, the trader’s positions were associated with a potential payout of roughly $46 million, while Trump’s quoted probability on the platform had risen to around 62 per cent.

There is no basis for alleging wrongdoing merely because a participant makes a large and ultimately successful trade. The episode nevertheless illustrates an important limitation.

A prediction-market price is not simply the average belief of a large population. It is a capital-weighted price produced by participants with very different levels of wealth, information, confidence and technical sophistication.

In a deep and liquid market, the effect of one trader may be absorbed by other participants. In a thinner market, a large position can have a much greater effect on the published probability.

The “crowd” may therefore contain thousands of people, but some voices are considerably louder than others.

Who actually makes the money?

The more difficult question is not whether prediction markets sometimes produce good forecasts.

It is who profits from them.

A recent working paper by Pat Akey, Vincent Grégoire, Nicolas Harvie and Charles Martineau analysed 588 million Polymarket trades representing approximately $67 billion in volume.

The researchers found an extraordinary concentration of gains: the most profitable 1 per cent of users captured 76.5 per cent of total profits.

That does not mean every other participant lost heavily. Many users trade in small amounts, and some may regard their losses as the cost of entertainment. It does, however, challenge the image of a broad crowd collectively earning money from superior insight.

A separate working paper by Joshua Della Vedova examined 222 million resolved Polymarket trades and attempted to distinguish forecasting skill from trading execution.

Its conclusion was striking. Automated accounts with an accuracy rate close to a coin toss — approximately 49.9 per cent — earned an estimated $133 million. Retail traders who selected the correct outcome slightly more frequently — approximately 51.3 per cent of the time — collectively lost an estimated $79 million.

The difference was not primarily who predicted the future correctly. It was who entered at the right price, provided liquidity and executed trades more efficiently.

These figures come from working papers and should be treated as emerging research rather than unquestionable final judgments. But the findings point in the same direction.

In prediction markets, being right about the eventual outcome is not enough. A trader must also be right relative to the price paid.

A person who correctly predicts a result but buys at an excessively high price may still make little money or lose it. An automated trader can be indifferent to the ultimate event and instead earn small, repeated profits from spreads, temporary mispricing and the behaviour of less efficient participants.

The machines are not necessarily predicting anything

Research from Spain’s IMDEA Networks Institute provides another view of the same market structure.

The researchers studied approximately 86 million Polymarket bids and identified two forms of arbitrage created when related contracts were temporarily priced inconsistently.

They estimated that traders had extracted approximately $40 million in realised arbitrage profits from those pricing differences.

Arbitrage can make markets more efficient by correcting inconsistent prices. It is not inherently improper and is a standard feature of financial markets.

But an arbitrage strategy is not necessarily making a forecast about politics, economics or world events. It may simply identify two contracts that cannot both be correctly priced and trade the difference before other participants react.

That distinction matters because it changes the story of who is benefiting from the supposed wisdom of the crowd.

The market may generate an informative public probability. But some of its most sophisticated participants can profit without possessing a superior view of the event itself. Their advantage lies in speed, technology, liquidity provision and execution.

The retail participant may be trying to predict the future.

The professional participant may be trading the retail participant.

Polymarket is not a conventional bookmaker

Polymarket can reasonably object to being described as a traditional casino or sportsbook.

Its principal platform operates as a peer-to-peer market. Users buy and sell outcome contracts from one another, with trades matched through smart contracts. Polymarket does not simply set fixed odds and automatically collect every customer loss in the way a conventional bookmaker might.

That is a genuine distinction.

It does not eliminate the consumer question.

A peer-to-peer structure can still produce an environment in which a large number of casual participants compete against professional traders, automated systems and market makers with significant advantages in data, execution and technology.

The platform may not be “the house” in the traditional sense. But the economic experience for an ordinary participant can still resemble sitting at a poker table dominated by professionals.

The safest conclusion is not that Polymarket is legally or structurally identical to a casino. It is that the boundary between financial trading and gambling becomes difficult to maintain when contracts are bought primarily to win money from uncertain political, sporting and cultural events.

Why Wall Street is buying

ICE’s investment makes more sense when viewed through the value of the data rather than the fortunes of individual traders.

Under its agreement, ICE became a global distributor of Polymarket’s event-driven information. It can package changing probabilities into institutional data feeds that banks, hedge funds, asset managers and professional traders may use alongside conventional financial information.

For ICE, the strategic asset is not simply the person placing a small election wager.

It is the continuously updating stream of market expectations produced by all those trades.

A probability attached to an election, interest-rate decision, government shutdown, corporate event or geopolitical development can become another indicator on an institutional trading screen.

The crowd takes the risk. The infrastructure owner sells the signal.

That is why the language matters. A gambling company attracts one type of regulation and valuation. A financial-data platform connected to global capital markets attracts another.

Polymarket may contain elements of both.

Europe is less persuaded by the label

European authorities have generally shown less willingness than parts of the United States to accept that calling a product a “prediction market” necessarily takes it outside gambling law.

France’s National Gambling Authority has described prediction-market platforms such as Polymarket as illegal gambling services in France. On 16 July 2026, the authority ordered French internet service providers to block access to the Polymarket website.

In Great Britain, the Gambling Commission said in February 2026 that the current generation of prediction-market products would likely fall within the legal definition of a betting intermediary, similar to a betting exchange. It said an operator would not be able simply to classify itself as offering a non-gambling product.

The precise treatment will depend on each platform’s structure and the law of the country concerned. But the regulatory message is clear: the label chosen by the company is not decisive. Authorities will examine what the product actually does.

That is a more useful approach than arguing indefinitely over vocabulary.

A contract can generate information and still function as a wager. A market can assist price discovery while exposing retail participants to the same behavioural and financial risks found in gambling. Financial sophistication does not automatically remove those risks; sometimes it merely changes who is best equipped to exploit them.

The verdict

Prediction markets are not useless, fraudulent or inherently irrational.

On large, liquid and clearly worded questions, they can aggregate information rapidly and produce valuable signals. Their performance during the 2024 US election demonstrated why they deserve to be taken seriously.

But “useful signal” is not the same as “infallible oracle”.

The wider evidence shows inconsistent accuracy across platforms and contracts. Prices can be influenced by differences in liquidity and the concentration of capital. Profits are heavily concentrated among a relatively small group of sophisticated participants. Automated and professional traders can succeed through pricing and execution rather than superior forecasting.

For the ordinary user, this is not simply a competition to predict the future correctly. It is a competition against people and machines that may be faster, better informed and structurally better positioned.

Polymarket’s great commercial achievement is that it has created two products at once.

For the public, it offers the opportunity to trade opinions about the future.

For Wall Street, it offers the data generated when the public does so.

The first product carries the excitement. The second may contain the lasting value.

European regulators should therefore resist both extremes. Prediction markets should not be dismissed as meaningless gambling merely because money is involved. Nor should they be waved into the financial system as neutral truth machines simply because their probabilities sometimes prove correct.

The right question is not whether they should be called markets or betting platforms.

It is who takes the risk, who earns the profits, how reliable the prices really are and what protections apply to the people providing the money.

The oracle exists. But it does not speak equally to everyone.