Prediction Market

A prediction market is a marketplace for questions about the future. What’s traded is not company shares or commodities, but a contract on an event. An example: “Will candidate X win the election in November?” Whoever buys this contract receives one euro in the end if X wins, and nothing if she loses. Because the contract will ultimately be worth either one euro or zero, its price sits somewhere in between: somewhere between 0 and 100 cents. If it costs 63 cents, the participants are collectively saying that they consider victory to be about 63 percent likely.

Why prices often estimate better than polls

In a poll, a wrong answer costs nothing. You can claim whatever you believe to be true, or whatever sounds good. In a prediction market, a wrong assessment costs real money. This difference noticeably changes participants' behavior. Those who are unsure hold back; those with good information bet more.

On top of that comes a second effect. The price condenses the knowledge of very many people into a single number. Some know internal figures, others have local observations, still others run through models. None of them knows everything, but together they cover a lot of ground. Economists call this convergence of scattered knowledge into a price the information function of markets.

For media outlets and analysts, such prices are therefore attractive: they update within seconds, not every two weeks like a poll. Still, one caveat is important. A market price of 63 percent is not a prophecy. It is a probability, and events with a 63 percent chance simply fail to happen in roughly four out of ten cases.

From purchase to payout

Technically, a prediction market functions like a small exchange. For every question there are Yes shares and No shares, whose prices always add up to one euro. Buyers and sellers submit bids, and a trading system matches compatible bids. If demand for Yes rises, its price rises, and No automatically becomes cheaper. You don’t have to wait until the end: shares can be resold at any time, whether at a profit or a loss.

In the end, someone has to determine what actually happened. This step is called resolution. For this, a source is defined in advance, such as an official election result or a government announcement. The platform then pays out to the holders of the correct side. This is exactly where most disputes arise, because questions are sometimes worded imprecisely.

A well-known problem is thin markets. If only a few people are trading, a single large bet can shift the price significantly. The number then looks like collective wisdom, but actually reflects only one opinion. That’s why it’s always worth checking the trading volume, that is, how much money has actually been moved.

From election nights to bets on AI models

Prediction markets are most visible around elections. Platforms like Polymarket or Kalshi continuously publish prices that are cited in news broadcasts and market reports. There are also markets on central bank interest rate decisions, on commodity prices, and on sports results. The tech industry shows up too: people trade on which company will release a new AI model next, or whether an IPO will actually happen.

Some companies use the idea internally. Employees trade there on questions like “Will the product be finished in the third quarter?” Such internal markets are considered more honest than official status reports, because no one needs to please their boss. Google and Ford have experimented with this in the past.

Legally, the topic is delicate, since the line to gambling is thin. In the US, the regulatory agency CFTC has approved some platforms and banned others. In Germany, there is no broad offering for private individuals. So anyone who comes across a number from a prediction market in the news should be able to put it in context: as a snapshot of an expectation, not as a fact.