What Are Prediction Markets For?

Are prediction markets, as their name suggests, markets that predict outcomes, or are they barometers of public sentiment with some predictive power, whether for an election, an interest rate decision, or whether the McDonald's ice cream machine will be broken tomorrow? The last one probably does not need a market.

Before asking what prediction markets are for, it helps to establish what they actually do.

At their simplest, prediction markets let people trade contracts on future events. A contract might pay $1 if a candidate wins an election and nothing if they lose. People buy and sell the contract, and those trades give it a price.

So far, nothing has been predicted. We have established what people are willing to pay for a bet.

Prediction markets are designed in a way that makes the next step feel almost automatic. To a layperson, 60 cents out of a possible dollar looks a lot like 60 out of 100, and therefore like a 60 percent chance.

But the market has produced a price. Whether that price behaves like a probability is something we can test.

If events priced around 60 cents happen roughly 60 percent of the time, the interpretation holds up. This is calibration, and prediction markets have generally performed reasonably well by this standard.

There is an important qualification: time.

Page and Clemen studied more than 500,000 transactions across 1,787 prediction markets. They found that prices were reasonably well calibrated close to expiration but significantly biased for events farther into the future. In particular, distant prices showed a favorite-longshot bias, with unlikely events priced too high and likely events priced too low.

More recent evidence points in the same general direction. A 2026 study of Kalshi examined more than 2.2 million resolved markets and found that its Brier score improved from roughly 0.08 to 0.09 three months before resolution to around 0.02 at close. The study was produced by researchers affiliated with Kalshi, so its results deserve independent replication, but the pattern itself is unsurprising. Markets know more as there becomes more to know.

Polls arrive. Economic data is released. Candidates drop out. Injuries happen. Eventually the event itself begins to reveal the answer.

A market becoming extremely accurate near resolution is therefore not quite the same accomplishment as predicting an uncertain future. Knowing who won an election while votes are being counted may be useful information, but it is not much of a prediction.

The more interesting question is what these markets know while there is still time to be wrong.

Why should betting improve a forecast?

The usual answer is information aggregation.

Information is scattered across people, and the person who knows something useful may not be someone we would think to ask. A market does not need to decide beforehand who counts as an expert.

Someone who thinks the market is wrong can trade against it. If they actually know more, being right can be profitable. That gives people an incentive to act on information and potentially to find information nobody else has.

The market can discover expertise instead of appointing it.

Money seems central to this explanation. Profit rewards people for being better informed.

But does it need to be money?

In a study of 208 NFL games, Servan-Schreiber and coauthors compared a real-money prediction market with a play-money one. Both performed well, but they found no statistically significant difference in predictive accuracy between them.

The authors suggested an interesting explanation. Real money may create stronger incentives to discover information, while play money may allow information to aggregate more efficiently. The study is old and narrow, so it does not establish that money is unnecessary. But it complicates the idea that financial risk is what makes the forecasting mechanism work.

If much of the information can be produced without large amounts of real money changing hands, what exactly is the money paying for?

Who pays for the information?

An informed trader needs someone on the other side.

Some liquidity comes from market makers, hedgers, and sophisticated traders who genuinely disagree. Some may also come from people who are simply wrong.

A retail trader might see a contract and think the outcome is obvious. Maybe they have followed the issue casually. Maybe they are partisan. Maybe they are simply more confident than informed. Maybe the Magic Conch told them otherwise.

A better-informed trader takes the other side.

One loses. The other profits. The price gets better.

This is not necessarily a failure of the market. It may be part of how the market works. Mistakes create profit opportunities for people who can correct them.

The phrase "wisdom of crowds" can obscure this. The whole crowd does not need to be wise. A smaller number of informed traders may only need enough incentive and liquidity to trade against everyone else.

The result is a peculiar arrangement. Some participants provide liquidity and make mistakes. Better-informed participants can profit from correcting those mistakes. Everyone else gets to observe the resulting price for free.

Perhaps that is an efficient way to produce information.

But it is not free.

What does it cost?

If one trader loses $100 and another gains $100, society has not lost $100. The money has mostly been transferred rather than destroyed.

The system surrounding that transfer still has costs. People can lose money they cannot afford. Some will treat prediction markets as gambling. Platforms benefit from more trading. Resources go into fees and finding exploitable mistakes. Markets can also create opportunities for manipulation.

There are benefits on the other side. People enjoy betting. Some contracts provide genuine hedges. Large financial rewards may uncover information that nobody would otherwise bother finding.

The balance will not be the same for every market. A contract that helps businesses hedge an economically important risk is different from one predicting an event nobody can meaningfully act upon. A market that uncovers private information is different from one that mostly repackages polls, models, and public news.

Which leaves the harder question.

What is a better forecast worth?

Prediction markets do not compete with ignorance.

We already have polls, statistical models, expert surveys, financial markets, sportsbooks, and forecasting tournaments. Much of the information reflected in a prediction market may already exist elsewhere.

The value created by the market is whatever it adds.

Suppose a prediction market produces a forecast that is measurably better than the best available alternative. That establishes something important. The market has produced additional information.

It still does not tell us how valuable that information is.

A small improvement can matter enormously when somebody can act on it. Better information about inflation, weather, or demand could change decisions about investment, insurance, inventories, or public resources.

Other forecasts are harder to act on. An election happens whether or not there is a market predicting it. A slightly better probability may help campaigns, journalists, businesses, or financial firms. It may also mostly give everyone a more accurate number to watch while waiting for something that was going to happen anyway.

So perhaps accuracy is only the beginning.

How early did the market know? What did it tell us that we could not learn elsewhere? Did anyone make a better decision because of it? How much real money was necessary to produce that information? Who ultimately paid for it?

Prediction markets may be very good at turning disagreement into information. But producing information and producing value are not quite the same thing.

We can measure whether the odds are getting better.

The harder question is whether they are worth it.

Is it worth the odds?