EthDeribitWeeklyDeltaHedgedShortStrangleVrp

ETH Deribit Short-Dated Variance Risk Premium Harvest — Delta-Hedged Short Strangle, IV-Rich Gated, Weekly Roll (Cross-Venue: DERIBIT options + BINANCE ETH perp hedge)

A market-NEUTRAL, options-selling strategy that harvests the persistent crypto Variance Risk Premium (VRP)... [delta-hedged short strangle, IV-rich gated, weekly roll]

Implements the hypothesis literally. SHORT STRANGLE + WEEKLY ROLL: parameters['rolling_options'] declares underlying ETH, weekly roll, dte_target 7, legs C +0.07 / P -0.07, and RollingOptionsStrategy owns entry, roll and close; verified against the live Deribit chain, the 84-day spec resolves 12 consecutive weekly expiries with BOTH legs present for every one (ETH-19JUN26-1800-C / -1550-P ... ETH-4SEP26-2600-C / -2250-P). offset 0.07 is the ~20-delta distance at ETH's typical weekly IV (0.84 x 0.65 x sqrt(7/365) = 7.5%) and stays inside the band where Deribit ETH weeklies actually have candles. IV-RICH GATED: the entry test is the sold structure's own inverted IV vs trailing realized vol (vrp_ratio), which is precisely the IV-RV gap the delta-hedged P&L identity (0.5 x gamma x S^2 x (IV^2-RV^2) x dt) pays out; DVOL is fallback only and a missing reading blocks rather than degrades to a price signal. DELTA-HEDGED, CROSS-VENUE: the hedge leg is ETHUSDT.BINANCE (also the primary instrument, so it is the roll clock and realized-vol source), re-hedged once per daily bar inside a band - the band is the classic gamma-hedging bandwidth knob that trades hedge fees against residual delta, hence the second tunable. calculate_signal returns the CONTINUOUS annualized IV-minus-RV spread every bar (the same quantity the gate thresholds), not a gated constant. TWO DELIBERATE CALLS worth flagging: (a) bar_type is 1-DAY, not weekly - 'weekly' in the title is the ROLL cadence, while the bar interval is set by the daily hedge cadence, so the Layer-1.5 timeframe warning is expected and intended. (b) lookback_days is 84 (12 cycles) rather than a year: the binding cost of a rolling-options backtest here is per-bar-type catalog loading (~9-10s per catalog.bars() call regardless of payload), so 24 legs + 1 perp = 25 queries ~ 240s sits safely inside the 600s options wall-clock while 52 cycles (~105 queries) would be killed. Twelve cycles is enough to show the mechanism prices and runs, but it cannot sample the tail that dominates short-vol risk - please read the metrics that way (and note avg_trade_return_pct is not meaningful on option legs, whose notional is premium, not underlying). leverage stays 1.0: Deribit caps options at 1x and no sizing path reads leverage.

data_wall: unmeasurable / data-wall: this delta-hedged short-strangle VRP book is structurally capped at ~12 weekly cycles (lookback_days=84 is a hard infrastructure ceiling that applies to the FULL backtest, not just the sandbox), which is far too few short-vol cycles to characterize the fat left tail that dominates the strategy's risk — Sharpe 1.64 with CI [-2.92, 7.26] straddling zero on effective_daily_obs=26 is non-diagnostic, and options cadence means iteration cannot raise the trade count (L27). The tail already fired within the sample: a -99.3% single-day loss on the 2026-08-21 expiry produced a >100% max_drawdown (1.038, ruin-level), and the +111% headline survives only on the single following +127% mark. This is not a validated edge, it is one bad-then-lucky expiry with sizing (risk_frac 0.20/leg) hot enough to blow through the capital base on a second adverse settlement. Spending 2 hours optimizing 3 parameters over ~9 cycles would curve-fit noise. The mechanism prices and hedges correctly (not broken), but the premise cannot be tested on the available data. abandon_class = unmeasurable.

ETH variance-risk-premium harvest: every weekly Deribit ETH expiry the strategy sells a ~7%-OTM short strangle (short 1 OTM call + short 1 OTM put on the front weekly expiry, symmetric moneyness so the structure is delta-neutral at inception), holds it through the week collecting theta, and closes 24h before European cash settlement. Two things gate and protect it. (1) IV-RICH GATE: the cycle is written only when the implied vol of the CONTRACTS BEING SOLD exceeds trailing 10-day realized vol by vrp_ratio. The IV is term- and strike-matched - each leg's own Deribit mark (quoted in ETH, scaled to USD by spot) is inverted to a Black-Scholes IV at its own strike and time-to-expiry, and the two are averaged; the ETH DVOL index is a degraded fallback and if neither implied reading exists the cycle is BLOCKED (no ungated path, no price-only proxy). (2) DAILY DELTA HEDGE: every 1-DAY bar the aggregate Black-Scholes delta of the live legs is computed from their own strikes/expiry/IV and the ETHUSDT.BINANCE perp is moved to the offsetting size whenever the residual exceeds hedge_band x the per-leg contract size, so the book harvests gamma/theta rather than direction; when no option leg is live the desired hedge is zero, so the perp is unwound with the structure. Cross-venue exactly as declared: short options on DERIBIT (0.03%/side), hedge on BINANCE USD-M (0.05% taker). Per-leg size is equity-relative (whole ETH contracts, equity x 20% / spot). The roll/entry/close schedule is framework-owned via the rolling_options spec - there is no underlying-only smoke path.

Have the analyst run the full-history backtest at the maximum cycle count the options wall-clock permits and weight tail/stress statistics accordingly; do not promote on a headline Sharpe from ~12 cycles.

The 84-day lookback caps the strategy at ~12 weekly strangle cycles for the entire backtest (self-imposed for the 600s options wall-clock, not a data-availability limit). Short-vol tail risk cannot be sampled at N=12; the strategy is inherently under-powered for measuring edge, as the developer explicitly acknowledges. Data exists further back — the ceiling is engine throughput, so this is not a not_measurable data verdict, but the analyst should treat any performance number as an existence proof, not a measurement.

No change required; noted so the warnings are not re-raised as defects downstream.

Static analysis flags three base_shadow overrides (min_bars_required, _leg_size, _enter_leg) and several division-by-variable anti-patterns. These are inspected and benign: the _leg_size/_enter_leg overrides are the documented RollingOptionsStrategy extension points (signatures match the base), min_bars_required is a deliberate rv_window-relative override, and every flagged division (n, spot, prev) has an explicit >0/finite guard preceding it. The realized-vol deque is trimmed to rv_window so the 'unbounded list growth' warning is a false positive.

Two performance/measurability caveats for the analyst (neither blocks): (1) lookback_days=84 is baked into the rolling_options spec as a hard wall-clock ceiling (~12 weekly cycles / 24 option legs), so the FULL backtest — not just the sandbox — can never sample more than ~12 short-vol cycles. That is enough to prove the structure prices and hedges but far too few to characterise the fat left tail that dominates short-strangle risk; a single bad expiry can dominate the result, and Sharpe CI [-3.06, 7.13] on 15 effective daily obs is non-diagnostic. Read any headline Sharpe/PF with heavy skepticism. (2) sandbox avg_trade_return_pct is -96% because it is measured against premium notional (a small fraction of underlying notional) on the option legs — as the developer notes, that metric is not meaningful for this book, so the usual fee-viability floor (checks 7/8) does NOT apply here; judge economics on net USD PnL and the IV-RV gate margin over Deribit 0.03%/side + Binance 0.05% hedge fees, not on avg_trade_return_pct. Delta-hedged premium selling is also not defined-risk, so L72's defined-risk-spread abandonment rule does not directly apply.

The structure runs, prices and hedges as designed: 18 option-leg trades across ~9 weekly strangle cycles plus perp re-hedges, VRP-gated entry, both-legs-or-neither, daily delta hedge. This is NOT a smoke/underlying-only degeneration and NOT the defined-risk options signature (L67) or the naked few-trade put-write signature (L27) — it is a genuine delta-hedged premium-selling book.

Headline economics look positive on the tiny window: total_return +111%, profit_factor 1.84, win_rate 79%, expectancy +$93/trade net of the $66 commission (commission only 1.35% of gross). Per QA's note the -96% avg_trade_return_pct is measured against premium notional and is correctly ignored here; economics are judged on USD PnL, which is positive.

Delta-hedge mechanism appears to work between cycles (book flattens to zero when no leg is live).

STRUCTURALLY UNMEASURABLE. lookback_days=84 is a hard wall-clock ceiling baked into the rolling_options spec, so the FULL backtest — not just the sandbox — can never sample more than ~12 weekly short-vol cycles (here ~9 strangles / effective_daily_obs=26). That is far too few to characterize the fat left tail that dominates short-strangle risk. A 2-hour 3-phase optimization over ~9-12 cycles would curve-fit noise, and options cadence means iteration cannot raise the trade count (L27).

The tail already fired inside this tiny sample: daily_returns show 2026-08-21 = -0.9932 (a near-total -99% single-day loss on one expiry) followed by +1.2703 the next day. max_drawdown = 1.038 (>100% — a ruin-level event). One short-strangle expiry nearly wiped the book; the headline PnL survives only because of a single subsequent +127% mark. This is precisely the picking-up-pennies-in-front-of-the-steamroller profile QA warned about.

Sharpe 1.64 is non-diagnostic: sharpe_ci = [-2.92, 7.26] straddles zero, probabilistic_sharpe 0.67, return_kurtosis 11.96, on 15 non-zero daily obs. The result is dominated by one bad-then-good expiry, not a demonstrated repeatable edge.

avg_position_pct/exposure and a 103% drawdown mean the sizing (risk_frac 0.20 per leg) is far too hot for an undefined-risk short-vol book — a second adverse expiry would blow through the capital base.

EthDeribitWeeklyDeltaHedgedShortStra-ea81cdd812

A short-vol structure whose data cadence caps the full backtest at ~12 cycles cannot demonstrate a repeatable edge — the tail that defines its risk goes unsampled.

Abandoned at backtest-review as structurally unmeasurable: lookback_days=84 caps the full backtest at ~12 weekly cycles, far too few to characterize the fat left tail, and the +111% survived only on one subsequent +127% mark after a ruin-level expiry.

Harvest the crypto variance risk premium by selling ~7%-OTM weekly ETH Deribit strangles only when implied vol is rich to realized vol, delta-hedging daily with a Binance ETH perp to strip out direction and keep the vol P&L.

Over ~9 weekly cycles (24 trades, 18 option legs) it showed total_return +111%, profit_factor 1.84, win_rate 79%, expectancy +$93/trade net of $66 commission, but Sharpe 1.64 with a CI of [-2.92, 7.26] straddling zero and a max_drawdown of 1.038 driven by a -99.3% single-day loss on the 2026-08-21 expiry.