Track record
Are they actually right? Or just lucky.
Every prediction market quotes a probability. Most never publish whether they were right. We compute the Brier score and calibration curve for 25 resolved markets — using the price 24 hours before close, not the post-resolution settlement.
Brier · 24h before close0.202Lower = better. Random = 0.250
Brier · at close0.151For comparison
Calibration error22.2%0% = perfect calibration
Markets analyzed25Top by volume
Calibration plot
Each bubble = one decile of forecasts. Bubble size = number of markets in that bucket. Dashed diagonal is perfect calibration. Bubbles below the line = overconfident; above = underconfident.
0–10% forecastn=1414.3% hit
20–30% forecastn=2100.0% hit
40–50% forecastn=333.3% hit
50–60% forecastn=2100.0% hit
60–70% forecastn=10.0% hit
70–80% forecastn=10.0% hit
90–100% forecastn=2100.0% hit
Patterns
High-confidence accuracy88%
Of 16 markets where the 24h forecast was ≥80% or ≤20%, 14 resolved that way.
Category gap0.29 Brier
Politics (0.291) was the hardest; Economy (0.000) the easiest.
Final-24h move8.1%
Average absolute price move in the last day before close. 2 of 25 markets flipped across 50% in that window.
Brier by category
| Category | Markets | Brier (24h) |
|---|---|---|
| 📈Economy | 6 | 0.000 |
| ⚽Sports | 2 | 0.168 |
| 🌍Other | 2 | 0.175 |
| 🏛️Politics | 15 | 0.291 |
Recent resolutions
Methodology
- Sample: Top 25 resolved binary prediction markets by lifetime USD volume, ranging from Nov 2024 to Jul 2026.
- Forecast: Mid-price of the YES token 24 hours beforethe market's end timestamp. We deliberately avoid the post-resolution clearing price (which would trivially equal the outcome).
- Brier score: mean of (forecast − outcome)². A perfect oracle scores 0; a coin-flip scores 0.25.
- Calibration error (MAE):mean absolute error between each decile's observed YES rate and the bucket's midpoint forecast — weighted by bucket size.
- Refresh: recomputed every hour from live prediction-market data and CLOB price-history feeds.