Prediction Market

A prediction market is a trading venue for questions about the future. There you don’t buy stocks or commodities, but contracts tied to an event. Such a contract might read: “The unemployment rate in Germany will be below five percent in December.” If the event occurs, the contract pays out a fixed amount, say one euro. If it doesn’t occur, it’s worthless. Because of this, the current price expresses something: if the contract costs 70 cents, traders consider the event roughly 70 percent likely.

Why prices often estimate better than polls

In a poll, an answer costs nothing. Someone who is wrong faces no downside, and someone with only a vague hunch still gives an answer. On a prediction market, it’s different. Every opinion costs money, and wrong opinions cost that money permanently. This filters out people who are loud but know little.

There is also a second effect. A market condenses the knowledge of many individual people into a single number. One person knows polling data, another knows the industry, a third read a piece of news earlier. As soon as someone trades, that knowledge flows into the price. The price also updates within seconds, whereas a good poll takes days.

That’s why companies and researchers are also interested in this. Firms have run internal markets to estimate whether a product would be finished on time. Such markets were sometimes closer to the truth than the official project reports. One reason: someone putting money on the line doesn’t have to worry about their boss.

How bets turn into a probability

Technically, a prediction market works like a small exchange. There are buy offers and sell offers, and where the two meet, a price emerges. Because each contract is ultimately worth either one euro or zero euros, the price can only lie between these values. It is converted directly into a percentage. A price of 0.12 euros thus corresponds to 12 percent.

Anyone who believes the market is wrong has an incentive to step in. If someone thinks 12 percent is too low, they buy and drive the price up. It is precisely this self-interest that ensures obvious mispricings disappear quickly. The market is not a vote where every voice counts equally. Whoever puts in more money moves the price more strongly — and correspondingly risks more.

The system isn’t perfect. For very unlikely events, prices are often too high because people overestimate small chances. If few traders are active, the price fluctuates wildly and says little. And the question must be phrased unambiguously, otherwise people end up arguing over the payout. A prediction market also only provides a probability, not certainty: an event with a 90 percent chance fails to occur in roughly one out of every ten cases.

From election nights to bets on AI models

Prediction markets are best known in connection with elections. Platforms like Polymarket or Kalshi display continuously updated percentages during an election night. News sites now cite these figures alongside traditional polls. They also turn up in the financial press around central bank interest rate decisions.

In the tech world, AI is traded in a similar way. Typical questions include which company will release a new language model next, or whether a model will reach a certain benchmark score by a given deadline. Such markets are small but are readily cited as a mood barometer for the industry. They should be read with caution, especially when only a few thousand euros are actually in play.

Legally, the topic is tricky because the line to gambling is blurry. In the US, a securities regulator has allowed some markets and banned others. In Germany, there are hardly any regulated offerings, which is why trading mostly takes place via foreign platforms. Anyone who reads such figures in a news report should therefore always keep in mind how much money was actually traded there.