Probability Atlas
See what prices are plausible—and where the evidence is thin.
Q reflects option pricing, including risk premiums. P is a historical statistical forecast. Neither promises what happens next.How to read P, Q and the intervals
P uses historical returns adjusted for volatility. Q extracts risk-neutral probabilities from option prices, including risk premiums. Q baseline uses the forward and near-ATM volatility without the full smile. Complexity is not proof of accuracy.
The median splits probability in half. Central 80% means 10% below and 10% above; central 95% means 2.5% in each tail. These are ending-price ranges, not promises or probabilities of touching a level along the way. Area under the curve is probability; its height is density.
UTC + 8 hours = Hong Kong time. Check the exact target, currency and settlement reference before comparing curves. P is uncalibrated research; inspect the historical scoring below.
Q versus real-world probabilities · Bank of England · Static-arbitrage constraints · Gatheral–Jacquier
Why trust—or question—this curve?
Intervals describe the fitted distribution, not certainty about the model. Market quotes can be sparse or stale; historical models can miss a new regime. USD, USDT and USDC are kept separate.
Prediction scoring
Lower CRPS is better. Samples below 30 per horizon remain inconclusive; different assets and horizons are not independent.
| Asset | Days | Model | OOS n | CRPS / price | 80% coverage |
|---|
Prediction markets
Market disagreement ≠ model-dependent value ≠ matched arbitrage. Missing fees, exact event rules or executable legs keep the decision on HOLD.
Available target dates
This public edition serves the frozen targets in the selector above. Other dates require a new verified research calculation; there is no public calculation API.
Public research, not personalized investment advice. No account connections or order execution.