What is a prediction market?

A prediction market is a market in contracts whose value depends on a verifiable future event, built so that the price reads directly as a probability. The standard form is binary: the contract pays one unit if the event happens and zero if it does not, so a price of 60 cents amounts to saying the market assigns the outcome about a 60% probability. On every trade the price updates, and with it the estimate. Every contract carries a settlement date and a verification source declared in advance, because without an indisputable outcome there would be no way to close the positions, and therefore no reason to price them carefully. What separates this from a survey is not technology but incentives: declaring a conviction costs money in proportion to the conviction itself, and whoever is wrong pays immediately, instead of imposing the cost on whoever used the forecast to decide.

See also: Wisdom of the crowd · Aggregating many independent estimates to get a result more accurate than nearly every single estimate, expert ones included.

The value is in the number, not the fee

That the informational output is worth more than the trading activity producing it is not a theoretical claim. In October 2025 ICE, the company that owns the New York Stock Exchange, announced an investment of up to $2 billion in Polymarket, stating its goal of becoming the global distributor of its event-driven data, to provide sentiment indicators to institutional customers. At the time the platform charged no trading fees. The price was set on a stream of probabilities, not on an income statement.

An enterprise example

Internal prediction markets are documented, and under hostile conditions. Bo Cowgill and Eric Zitzewitz studied those at Google, Ford and a large private company in basic materials for the Review of Economic Studies: few participants, weak incentives, limited entry, and people with excellent reasons to lie about their own project's delivery date. Despite that, the three markets improved on internal experts' forecasts by as much as a 25% reduction in mean squared error. Two results matter more than the headline: the most visible distortion was an optimism bias in Google's markets, exactly the one you would expect from people betting on the project they work on; and the inefficiencies shrank over time, because traders with better track records traded against the distortions while less skilled ones exited the market.

Why it matters for decision makers

That last point is the useful part for anyone running a company, and it is also the empirical answer to a question usually settled by seniority: how do you tell someone who can forecast from someone who merely sounds certain. You do not decide it in advance, you discover it by letting forecasts be recorded and checked across many iterations. Opening an internal market remains disproportionate for almost every mid-sized company, though, and the transferable part is not the platform: it is the constraint that makes it useful. Every forecast that matters gets written with a probability and a date ("60%: we do not reach production by March"), and after that date it gets scored. This produces a decision maker's calibration, a measurable property almost no organization measures, and it closes the missing link in decision intelligence: from data to a recorded decision, and from the decision to the check. Two limits have to be kept in mind, though, because the mechanism promises more than it delivers. A market is mute over long horizons, since a contract settling in ten years ties up capital and attracts no volume, so it is reliable precisely where feedback is fast and mute where the decisions matter most. And it compresses everything into a binary outcome: it prices a delivery date well and does not price a priority at all, because the first has a true value to discover while the second is a choice to make. In Europe, finally, the constraint is regulatory too: the same instrument that operates under the CFTC in the United States is classified case by case, often as gambling. It should also be remembered that objective resolution is a governance property rather than a structural one: somebody has to declare that the event happened, and on decentralized platforms that depends on an oracle with a dispute mechanism, which is where the best-grounded criticism concentrates. The statistical foundation of the mechanism, and the conditions it has to respect, are those of the wisdom of the crowd, and what keeps the price honest is the skin in the game of whoever bets: being wrong costs you something of your own, and that weights opinions in a way no survey can.