We backtested 1,137 resolved Polymarket markets — taking each one’s price 24 hours before it closed and checking what really happened. The short answer: the price is honest. “Safe” is not the same as “profitable.”
Favorite-side price at 24h to close vs. the realized outcome. Negative EV = you lose money buying it.
| Market says | Actually happens | Calibration | EV after spread |
|---|---|---|---|
| 50–60% | 50.3% | -4.6% | -10.0% |
| 60–70% | 58.3% | -6.2% | -11.1% |
| 70–80% | 72.0% | -2.6% | -4.7% |
| 80–85% | 63.8% | -18.5% | -23.7% |
| 85–90% | 84.0% | -3.5% | -5.3% |
| 90–93% | 89.3% | -2.2% | -3.4% |
| 93–97% | 92.5% | -2.7% | -3.7% |
| 97–99% | 98.9% | -0.7% | -0.9% |
Every band loses money after spread. A 99¢ “lock” pays you ~1¢ when it wins and costs you 99¢ when it doesn’t — and it doesn’t about 1% of the time. Splitting your stake across many of them changes your variance, not the math.
Favorites priced ≥80%, by category. Some markets are far more reliable than others.
| Category | Avg price | Actually won | EV |
|---|---|---|---|
| Politics | 98% | 97.4% | -0.4% |
| Other | 97% | 95.0% | -1.9% |
| Crypto | 97% | 97.9% | +0.5% |
| Geopolitics | 97% | 86.5% | -10.8% |
| Econ/Fed | 99% | 100.0% | +1.2% |
| Sports | 92% | 80.4% | -13.7% |
The danger isn’t the number, it’s the path. A favorite that sat calmly at 95% for weeks usually holds. A 97% that spiked up from 22% yesterday on a headline reverts far more often than the price implies — that’s where the upsets cluster.
So across PolyQuant we tag those markets with a Unsettled flag. It doesn’t tell you to bet — there’s no systematic edge here (we checked: fading the “sure losers” doesn’t survive a walk-forward test either). It tells you when a price is less settled than it looks.