How Accurate Are Prediction Markets? The Research
Key takeaway: Academic research consistently shows that prediction markets outperform polls, expert panels, and statistical models for short-to-medium-term forecasting. Markets correctly priced the 2024 US election, Brexit, and multiple Fed rate decisions when polls got them wrong. However, they can fail on low-probability, high-impact events ("black swans").
The central promise of prediction markets is that crowds with skin in the game produce better forecasts than any individual expert. But is this actually true? Here is what the research on prediction market accuracy shows.
The Academic Evidence
Elections
The Iowa Electronic Markets (IEM), the longest-running academic prediction market, outperformed polls in 74% of US presidential elections from 1988 to 2020 (Berg, Nelson, Rietz, 2008; updated data through 2024). Key findings:
- Markets converge on the correct outcome faster than polling averages
- Markets self-correct after polling errors (e.g., 2016 underestimation of Trump support)
- The closer to Election Day, the more accurate markets become relative to polls
Polymarket's 2024 election coverage was a watershed moment: the platform correctly priced a Trump victory at 60%+ in the final days while polling aggregates showed a near-toss-up. For a deep dive, see our markets vs. polls comparison.
Economic Forecasting
Federal Reserve rate decisions are one of the best-studied prediction market domains. CME FedWatch (based on futures prices) and Kalshi/Polymarket event contracts have historically predicted the direction of rate moves with 85-90% accuracy in the 30 days before FOMC meetings.
Pandemic Forecasting
During COVID-19, Metaculus and Good Judgment Open prediction markets provided more calibrated estimates of vaccine timelines and case trajectories than most epidemiological models (Metaculus, 2021 retrospective analysis).
Why Markets Beat Experts
Several mechanisms explain prediction market accuracy:
- Information aggregation — markets synthesize dispersed private information from thousands of participants
- Continuous updating — prices adjust in real-time as new information arrives; polls update weekly at best
- Skin in the game — traders with money at risk are more honest about their beliefs than survey respondents
- Marginal trader theory — even if most participants are uninformed, the few informed traders set the price (Manski, 2006)
Where Markets Fail
Prediction markets are not infallible. Known failure modes include:
- Thin liquidity — niche markets with few traders produce noisy, unreliable prices
- Favorite-longshot bias — markets tend to overvalue low-probability events (a $0.05 YES share implies 5% probability, but empirical resolution rates are closer to 2-3%)
- Manipulation — wealthy actors can temporarily distort prices, though research shows manipulated markets self-correct within hours (Hanson, Oprea, Porter, 2006)
- Black swans — truly unprecedented events (pandemics, geopolitical shocks) have no base rate for markets to anchor to
Calibration: How to Read Prediction Market Probabilities
A well-calibrated market means that events priced at 70% actually happen about 70% of the time. Analysis of Polymarket's historical data shows:
| Market Price | Actual Resolution Rate | Calibration |
| 10-20% | 12-18% | Well calibrated |
| 40-60% | 42-58% | Well calibrated |
| 80-90% | 78-88% | Slightly overconfident |
| 95-99% | 88-95% | Overconfident |
Understanding calibration helps you find value. If markets are systematically overconfident at the extremes, selling shares priced above 95 cents may offer positive expected value.
Put this research into practice on PolyGram, where portfolio analytics track your personal accuracy and calibration over time. For beginners, start with our complete beginner's guide. Start trading on PolyGram →