How accurate are prediction markets?
Understand the probability—and its limits. Historical outcomes reveal more than a single right or wrong call.
Historical performance reported by Foresee
Source page snapshot · September 5, 202698.5%
Accuracy 4 hours before resolution
90.0%
Accuracy 1 month before resolution
0.0627
Brier score · Lower is better
More information becomes available near resolution
Near-resolution performance is not a substitute for early forecasting skill. Compare forecasts at the same lead time.
A 70% probability is not a promise
If many comparable events are forecast at 70%, a well-calibrated system should see roughly seven in ten happen. One event failing to happen does not by itself discredit the forecast.
Calibration illustration · Not an actual sample
Try it: score a forecast
The Brier score is the squared difference between a forecast probability and the outcome: 1 if it happens, 0 if it does not. Lower is better.
Single-forecast Brier score
0.0900
(0.70 − 1)²
An educational example, excluded from all historical metrics. System-level evaluation averages scores across the sample.
Data & methods
Foresee describes using resolved markets and price snapshots at multiple lead times. It also reports a Yes / No resolution split of 26.2% / 73.8%. Outcome balance, sampling time and sample coverage all matter when interpreting accuracy.
Foresee presents prediction-market information; historical accuracy does not guarantee future outcomes. We have not independently recomputed the full upstream sample or plotted calibration and volume-score curves without their underlying data.
Frequently asked questions
Does high accuracy guarantee a profit?
No. Forecast accuracy, entry price, transaction costs and realized returns are different questions.
Why do probabilities change?
New information and trading change prices. Upstream refresh timing and caching can also affect what Foresee displays.