Skip to content
TwapTrade

Are prediction markets accurate?

Updated · 5 min read

Prediction markets are often accurate in a specific sense: across many markets, outcomes priced at 70% tend to happen about 70% of the time. They work because traders have money at stake and react quickly to news. They can be wrong when few people trade, when the rules are unclear or when real surprises happen, and a single price is never a certainty.

  • For probabilities, accuracy means calibration: 70% events should happen about 70% of the time.
  • Markets combine many people's information, and money at stake rewards getting it right.
  • Studies of election and other markets have generally found them competitive with polls and expert forecasts.
  • Thin markets, unclear rules and long-shot pricing are common weak spots.
  • A market can be right about the odds and still see the less likely outcome happen.

What “accurate” means for odds

A forecast that says 70% isn't wrong just because the event didn't happen. A 70% chance means the other result happens about three times in ten. So you can't judge a prediction market from one result. You judge it over many.

The right test is [calibration](/glossary#calibration). Take every market that traded around 70¢ at a given point, then check how often Yes won. If it's close to 70%, the markets were well calibrated at that level. Repeat for 10%, 30%, 50%, 90% and so on. In a well-calibrated set of markets, low prices win rarely and high prices win often, in roughly the proportions the prices suggested.

Example · Calibration in a hypothetical set of markets

Imagine 100 separate markets, each priced at 70¢ the day before it settled.

If about 70 settle Yes, the prices were well calibrated at that level.

If only 50 settle Yes, the markets were overconfident: 70¢ was too high.

If 90 settle Yes, they were underconfident: 70¢ was too low.

Any single one of those markets tells you very little on its own.

Why markets tend to forecast well

  • The [wisdom of crowds](/glossary#wisdom-of-crowds). Many people with different information each know a piece of the picture. A single price combines all of it.
  • Money at stake. Traders lose money when they're wrong and gain when they're right. That rewards careful thinking and punishes wishful thinking, at least among the people who keep trading.
  • Fast updating. Prices move within minutes of news. A poll or report is fixed once it's published.
  • Self-correction. If a price looks too high or too low, anyone who notices has a reason to trade against it, which pulls it back.

What research has found

Prediction markets have been studied for decades. Studies of election markets have generally found market prices to be competitive with polls as forecasts of who wins, and often better than any single poll. Companies that have run internal markets, where staff traded on questions such as whether a project would ship on time, have also reported useful forecasts. Research on betting markets more broadly has found prices to be reasonably well calibrated, with some known biases.

None of this means markets are always right. It means that, across many questions, their prices have usually been a reasonable guide to the odds. For a direct comparison, see prediction markets vs polls.

When prediction markets get it wrong

  • Thin markets. With few traders and little money, one person can move the price, and errors can sit uncorrected. Check the volume and the spread.
  • Long shots. Research on betting markets has often found that very unlikely outcomes are priced a little too high. A 3¢ price may overstate the real chance.
  • Shared blind spots. If most traders read the same sources and share the same assumptions, the crowd can be confidently wrong together.
  • Unclear rules. If traders read a question differently, the price mixes two different bets. Reading the rules helps.
  • Genuine surprises. Some events are hard for anyone to forecast. A market can price them sensibly and still be surprised.
  • Long time frames. Markets that settle far in the future can trade at prices that partly reflect traders' reluctance to tie up money for a long time, not only the odds.

Can someone manipulate a prediction market?

Someone with enough money can push a price for a while. But moving a price away from where others think it belongs means buying shares that other traders consider overpriced, and those traders have a reason to take the other side. In an active market, attempts to move the price tend to be costly and short-lived. Studies of attempted manipulation have generally found the effects to fade quickly once other traders respond.

In a thin market, a push can last longer because fewer people are watching. That's another reason to look at volume, depth and the spread before relying on a price.

How to use market odds sensibly

  • Treat a price as an estimate with uncertainty, not a forecast that will come true.
  • Give more weight to markets with steady trading, tight spreads and clear rules.
  • Look at how the price has moved over time, not just today's number.
  • Compare it with other sources, such as polls, models or expert views, and ask why they differ.
  • Remember that the price reflects what traders believe right now, which can change quickly.

How prediction market prices work explains where the number comes from, and how to read election odds applies all of this to political markets.

Compare markets and polls

Questions

Are prediction markets more accurate than polls?

They measure different things. Polls estimate how people will vote; markets estimate the chance of an outcome. Studies of election markets have generally found prices competitive with, and often better than, single polls as forecasts of who wins, but neither is right every time.

If the favorite loses, was the market wrong?

Not necessarily. A 70% chance means the other result happens about three times in ten. Accuracy is judged over many markets, by checking whether 70% events happen about 70% of the time.

Can prediction markets be manipulated?

Prices can be pushed for a while, especially in thin markets. In active markets, other traders have a reason to trade against a price that's been pushed out of line, so the effect tends to be costly and short-lived.

Why are some prediction markets inaccurate?

Common reasons are few traders, wide spreads, unclear rules, long time frames and shared blind spots among traders. Very low-priced long shots are also often a little overpriced.

What is calibration in forecasting?

It's how closely forecast probabilities match what happens. If events forecast at 30% happen about 30% of the time, and the same holds at other levels, the forecasts are well calibrated.

Keep reading