Get early access

Markets4 min read

How accurate are prediction markets? What the evidence shows

Prediction markets are often accurate on big, liquid questions and beat polls most of the time, but miss on long shots and thin markets. What research shows.

By Kristian FagerlieFounder & CEO, Sikt Intelligence

Short answer: prediction markets are often accurate, especially on big, heavily traded questions, and they have beaten polls most of the time. But they are not uniformly reliable. They are weaker on thin markets, they tend to overprice long shots, and they can be confidently wrong, as they were on Brexit in 2016.

Here is what decades of research, and the 2024 US election, actually show.

What "accurate" means for a probability

A market that gives an event 70% is not "wrong" if it doesn't happen. That is what a 70% forecast expects three times in ten. Accuracy for probabilities is measured across many questions, in two ways:

  • Calibration: do events priced at 70% happen about 70% of the time? See what a calibrated forecast is.
  • Scoring: how close were the prices to what happened, measured with a proper score such as the Brier score?

Keep this in mind whenever someone says a market "got it right" or "got it wrong" from a single result.

The case for accuracy

Iowa Electronic Markets, 1988–2004. Researchers compared the prices of these small election markets with 964 polls across five presidential elections. The market was closer to the final result 74% of the time, and its edge held even months before election day (Berg, Nelson and Rietz, 2008).

The broad review. Surveying the evidence in 2004, economists Justin Wolfers and Eric Zitzewitz found that market forecasts are typically fairly accurate (Journal of Economic Perspectives).

Kalshi, 2021–2025. A study of over 300,000 contract prices on Kalshi found that prices are informative and become more accurate as markets approach closing (Bürgi, Deng and Whelan, 2026).

The 2024 presidential election. A Vanderbilt study found that Polymarket outperformed polling aggregates in forecasting the 2024 US presidential election, on a race where about $3.7 billion was wagered on that platform alone (Cutting et al., Vanderbilt).

The case for caution

Long shots are overpriced. On Kalshi, contracts costing under 10¢ won far less often than their price implied: buyers lost over 60% of their money on average (Bürgi, Deng and Whelan). This is the favorite-longshot bias, and it shows up in dramatic questions like aliens or an invasion of Taiwan.

Accuracy varies by platform and market. A Vanderbilt study of the final weeks of the 2024 election found that the share of markets that predicted better than chance was 93% on PredictIt, 78% on Kalshi and 67% on Polymarket, and that Polymarket's price moves were barely correlated with those on the other platforms, even on the same races (Clinton and Huang, via DL News).

Markets can be confidently wrong. Before the 2016 Brexit vote, betting markets strongly favored Remain. Betfair put Remain at around 88% shortly before the polls closed, and Leave won (Cambridge Judge Business School, The Conversation).

Accuracy depends on a few traders. Research suggests much of a market's accuracy comes from a relatively small number of skilled traders (Yale Insights).

When to trust market odds more, and less

Trust moreTrust less
High volume, many tradersThin markets, few trades
Prices between about 20% and 80%Long shots under 10%
Close to the resolution dateYears before the event
Clear, simple resolution rulesFine print that differs from the headline
Two platforms agreePlatforms disagree by many points

You can check most of these directly on our odds pages, which show volume, rules and both platforms side by side.

Markets and AI forecasts together

Markets fail in specific, predictable ways: long shots, thin trading, herding. An evidence-based AI superforecaster fails in different ways. Putting the two numbers next to each other, as Sikt does, is more informative than trusting either alone. Nothing here is financial advice.

Key takeaways

  • On big, liquid questions, prediction markets are often accurate and have beaten polls most of the time.
  • The Iowa Electronic Markets were closer to the result than polls in 74% of 964 comparisons.
  • Kalshi prices are informative and improve as markets near their close.
  • Long shots are overpriced; thin markets and cross-platform disagreement are warning signs.
  • Judge accuracy across many questions with calibration and proper scores, never from one result.

FAQ

Are prediction markets more accurate than polls?

Often, yes. The Iowa Electronic Markets beat polls 74% of the time over five presidential elections, and Polymarket outperformed polling aggregates in the 2024 US presidential race. But they are not always right, as Brexit showed.

Is Polymarket or Kalshi more accurate?

It depends on the market. One study of the 2024 election found a higher share of markets that beat chance on Kalshi than on Polymarket, but results vary by question and liquidity. Compare both where you can.

Why were prediction markets wrong about Brexit?

Traders strongly expected Remain, and prices stayed confident until results came in. It is a reminder that a high probability is not a certainty, and that markets can share the same blind spot.

Sources

Keep reading