Are prediction markets accurate?
By the Callquake team · Published · 1 min read
Last reviewed . Laws and platform rules change; check the sources linked below.
Short answer: on average, yes, in the sense that things priced at 70% tend to happen about 70% of the time. But averages hide real weaknesses, especially in thin or emotional markets.
The case for: well calibrated on average
A 2026 paper in PNAS (2026) looked at about one million resolved contracts and questions from Polymarket, Kalshi, Manifold and Metaculus, across twelve domains from business to elections, with horizons from hours to years. It found the forecasts were highly calibrated overall: events given a probability of p% happened roughly p% of the time. The authors also found systematic over- or underestimation in some important areas, and noted that calibration does not make any single forecast certain (summary on Marginal Revolution).
The case against: noisy and herd-driven at times
Researchers Joshua Clinton and TzuFeng Huang of Vanderbilt University studied about 2,500 political markets with $2.5 billion in volume during the last five weeks of the 2024 US election. By their measure, Polymarket was right in 67% of markets, Kalshi in 78% and PredictIt in 93%. They found similar markets priced inconsistently, price spikes that often reversed the next day, and signs of herd behavior. Kalshi disputed the method, arguing that calibration is the right yardstick (DL News, December 5, 2025).
What this means when you read the odds
- Treat prices as probabilities, not predictions. A 30% outcome happening is not a market failure.
- Weight liquid markets more. Heavily traded markets are harder to move and tend to be better calibrated.
- Be wary of sudden spikes without news. If we cannot find a headline behind a move, it may reverse.
- Compare venues. When Kalshi and Polymarket disagree, the truth is often somewhere in between.