Polymarket’s “What price will Bitcoin hit?” and “Bitcoin above X on date” markets trade tens of millions of dollars, much of it retail buying dramatic targets. Unlike most prediction markets, their fair value can be computed: the options market prices crypto volatility professionally. The idea was to sell longshot “Yes” only where it was overpriced against options-implied odds.
−1.0%
+3.8%
−1.8%
| Discovery | Holdout | |
|---|---|---|
| Overpriced vs options | +10.3% | −1.0% (t = −0.3) |
| Model-only rule | +24.5% | +3.8% (t = 0.6) |
| Model accuracy vs market | model better | tie |
In discovery the model even beat the market’s own prices on accuracy, and held up in a month when BTC rallied 25%. But the discovery months were a calm regime, in which bitcoin moved less than options expected — that is what the model was really harvesting. Over fourteen months that included a sharp sell-off, the “BTC drops to X” tickets paid out.
Without the pre-registered test, this would have been built on.