Graveyard / Options and volatility / Binance perps
Trading Bitcoin volatility spikes (DVOL)
Does buying or fading Bitcoin after a spike in implied volatility work?
The idea
DVOL is Deribit's Bitcoin implied-volatility index. In stocks, buying after a VIX spike has worked in the past. The first test copied that: fade a big 24-hour move when DVOL spikes. Two later tests flipped the direction and traded with the spike.
How we tested it
| Data | Deribit DVOL index and BTCUSDT perpetual prices. Event entries when DVOL is high relative to its recent history and the 24-hour realised return passes a threshold. Version 1: fade, 54 configurations. Version 2: with the spike, 54 configurations. Version 3: with the spike plus a funding-rate regime filter, 108 configurations. |
|---|---|
| Period | June 2024 to June 2026 (two years, all three market regimes). |
| Costs | 0.11% round trip. |
| Holdout | Full period, in-sample and out-of-sample slices (last 20%, about 150 days) each had to pass on their own. |
| Control | None stated. Filters were tested against unfiltered pairs. |
Result
| V1 (fade): configurations passing | 0 of 54 |
|---|---|
| V1 best Sharpe | −0.15 |
| V1 win rate | 10% to 45% |
| V2 (with spike): entry passing | 12 or more of 54 configurations |
| V2 example: trades / Sharpe | 78 / 1.39 |
| V3: configurations passing the corrected gate | 0 of 108 |
Why it failed
The first test was on the wrong side: DVOL spikes marked continuation, not exhaustion. Flipping the direction produced a signal that passed on its own. It was buried anyway: the one filter that appeared to help could not be shown to help, and in version 3 the signal fires only about 0.1 to 0.3 times a day, so the last 20% of a two-year window holds too few trades to pass on its own. Too few trades, not a proven loss.
What would make it worth retesting
A much longer backtest window (about 4.5 years is estimated) or a design that raises trade frequency, such as pooling several symbols. The entry rule and the funding filter do not need re-checking on their own.
Past results of a backtest say nothing certain about the future.