nyasekai

Graveyard / Options and volatility

Trading options skew on Deribit

Does Deribit options skew predict the next week's Bitcoin or Ether return?

Family
Options and volatility
Verdict
Dead
Cause
Overfit
Buried
23 November 2026

The idea

Skew is the price of puts relative to calls. When traders overpay for puts, the market may be too fearful, so buy the coin. When they overpay for calls, sell it.

How we tested it

DataFirst pass: Deribit weekly option files (one expiry per file), with a moneyness proxy for the risk reversal (puts at 0.88 to 0.97 of spot minus calls at 1.03 to 1.12). Second pass: Deribit public option trades with a per-trade implied volatility, sampled on Fridays 02:00 to 08:00 UTC, delta from Black-Scholes, and the true 25-delta risk reversal from a delta-weighted regression of traded volatilities, plus a term-structure variant. Weekly outright perpetual trades for BTC and ETH.
Period2021 to 2026, holdout from January 2024 (about 151 weeks).
Costs5 bp per side plus actual funding.
HoldoutTwelve looks in total after the second pass, so the pass bar was raised to t of 2.87 (Bonferroni).
ControlA random twin. The notes also report the deflated Sharpe ratio.

Result

Proxy, ETH: holdout t2.31 (near miss at the time)
True 25-delta, ETH: development t−2.18 (18 active weeks, sign flips)
True 25-delta, ETH: holdout t1.41 (47 active weeks)
True 25-delta, ETH: holdout halves+0.36% / +0.65%
True 25-delta, ETH: twin p / deflated Sharpe0.07 / 0.48
True 25-delta, BTC: holdout t0.75
Term-structure variant, ETH / BTC: holdout t1.05 / −0.48
Pass bar over 12 lookst of 2.87

Why it failed

The proxy result did not survive the true measure. With a real 25-delta risk reversal the ETH holdout t falls to 1.41 and the development t flips sign to −2.18, so the earlier near miss was noise plus a post-2024 regime. The proxy and the true measure correlated at 0.86, so the proxy was decent; the difference came from which weeks and strikes qualified. The data uses traded volatilities, not quotes, so the 25-delta point is noisy in thin early weeks.

What would make it worth retesting

A second venue with multi-year options history (Binance has only 2023-05 to 2023-10, OKX only snapshots).

Past results of a backtest say nothing certain about the future.