3-year Deribit backtest: does volatility term-structure backwardation predict direction?
Tracking 30/90/180d ATM IV: when short-term IV exceeds mid/long-term, 24h ±3% move probability jumps from 37% to 66% — what about direction?
01 / Volatility Term Structure (VTS) & Backwardation
Every 5 minutes our collector pulls Deribit public option summaries, selects the nearest-ATM strike for each expiry, and linearly interpolates fixed 30-day (1M), 90-day (3M), and 180-day (6M) implied volatility.
Normally, longer maturities carry higher IV (1M < 3M < 6M). When near-term demand pushes 1M IV above both 3M and 6M, the curve enters backwardation.
02 / 3-year backtest findings (2023-10 to 2026-11)
After deduplicating noisy flips within 24 hours, we isolated 38 independent backwardation events:
| Metric | After Backwardation | Random Baseline | Takeaway |
|---|---|---|---|
| 24h ±3% move probability | 66% | 37% | Statistically significant expansion |
| Median time to ±3% | 16.5 hours | 34.0 hours | ~2× faster expansion |
| First P95 1h range bar | 5 hours | 21 hours | Large candles arrive quickly |
| Directional hit rate (+3% first) | 63% | — | Not significant for direction |
03 / How we use it: volatility filter, not directional trigger
Backwardation is not a coin-flip direction predictor. In our system it serves strictly as a risk-regime switch: widen stops, reduce leverage, and avoid shorting volatility whenever 1M > 3M & 6M.