EthDeribitWeeklyDeltaNeutralShortStrangleVrpBinanceHedged

ETH Weekly Volatility-Risk-Premium Harvest via Delta-Neutral Short Strangle (Deribit Options, IV-RV-Gated Entry, Delta-Hedged with Binance ETH Perp, Roll/Exit at Expiry or 50% Decay, Tail Stop)

A DELTA-NEUTRAL, options-based volatility-risk-premium (VRP) harvest on ETH Deribit weekly options, hedged with an ETHUSDT.BINANCE perp leg. This is a DELIBERATELY DIFFERENT FAMILY from the momentum/trend/reversion mechanisms that dominate (and mostly die in) the corpus: the edge is theta + vega decay of the volatility risk premium, not directional price prediction. Crypto options implied vol persistently exceeds subsequently-realized vol (a well-documented VRP), so systematically SELLING an out-of-the-money call + out-of-the-money put (a strangle) on the nearest weekly expiry, while holding the net delta near zero with a small perp hedge, collects premium that on average exceeds the realized movement. The corpus has ZERO options strategies of any kind (options quota target >=15%, current ~0%), so this is not a variant of anything in the failure graveyard: it is not funding cash-and-carry (L87), not premium-index fade (L85), not a basket (L68/L77/L88), not sub-hourly reversion (L89), not OI/funding-gated directional (L70/L71). It is also market-neutral (direction bucket, long_short/neutral under-represented) and cross-venue (Deribit + Binance, cross-venue quota >=15%). Entry is gated on IV being RICH relative to trailing realized vol so we only sell when the premium is actually there, which is the single most important knob and is deliberately the ONLY discretionary filter to keep the parameter surface small.

Implements the hypothesis exactly: short OTM strangle on Deribit ETH weeklies (theta/vega decay of the VRP, not direction), IV-vs-RV richness as the SINGLE discretionary entry gate (deribit_dvol vs trailing realized vol — a missing DVOL sample degrades to ungated rather than swapping in a price-only proxy, so the edge mechanism is never silently substituted), delta-hedged with the ETHUSDT.BINANCE perp declared in the hypothesis (added as an extra instrument so the engine funds/fees it on the Binance USD-M venue), roll/exit at the expiry buffer or 50% premium decay, plus the tail stop the hypothesis asks for. It extends RollingOptionsStrategy so the framework owns chain resolution, entry, roll and close; the strategy only adds the VRP gate, the paired-leg readiness guard, the hedge and the two early exits. calculate_signal returns the CONTINUOUS IV-minus-RV spread every bar (the decision variable in natural annualized-vol units); entries are schedule-driven, so the single-instrument entry hooks stay inert and there is no underlying-only trading path. Venue choice: primary ETH-PERPETUAL.DERIBIT drives the weekly roll clock and is the index the options settle against; leverage stays 1.0 because Deribit options cap at 1x and no sizing path reads leverage (declaring more would be inert). rolling_options.lookback_days is 180 (~26 weekly cycles, 52 legs, loaded in one batched catalog query) to fit the Deribit 600s Layer-3 budget. One honest caveat for the analyst: ~26 cycles cannot sample the tail that dominates a short-vol distribution, the engine funds each venue separately so headline percentage returns sit on roughly double the capital the options book uses, and avg_trade_return_pct is not meaningful for option legs (premium notional is a small fraction of underlying notional) — read per-cycle premium capture, gate hit rate and the hedge's residual delta instead.

data_wall: Insufficient data to validate the premise, not an overfit/fee-edge death. The Deribit weekly ETH option chain only covers ~2026-02 to 2026-09, so the entire backtest is a ~26-weekly-cycle / 56-trade / ~6-month sample despite the 7-year primary span (no options existed to sell before then). A short-strangle VRP payoff is defined by its tail — the rare week where realized vol blows through the strangle — and 26 benign cycles structurally cannot sample it. The evidence that the tail is real and unmodeled is already present: a 67.3% max drawdown, negative skew, and avg_loss > avg_win even in this short window. The flattering headline (Sharpe 5.22, win_rate 0.82, avg_trade_return_pct 8.09%) is non-diagnostic — 8.09% is premium-notional not fee-comparable, and the return is booked over ~2x the options book's capital. Spending 2 hours on a 3-phase walk-forward (~8 cycles/window) + one-shot holdout would only curve-fit noise; no parameter setting can establish robustness for a tail-dominated distribution the data does not contain. This is a DATA verdict, not an edge verdict: re-run the identical mechanism once the Deribit option-chain / DVOL history is deep enough (~1.5-2 years of weekly cycles) to sample the short-vol tail. abandon_class = data_wall / unmeasurable.

Delta-neutral weekly volatility-risk-premium harvest on ETH Deribit options, hedged with the Binance ETH perpetual. Each weekly Deribit expiry cycle the strategy SELLS one ~6% OTM call and one ~6% OTM put on the front (7-DTE) expiry, but only when the Deribit ETH DVOL implied-vol index is trading at least vrp_ratio (1.15x) above trailing rv_window-day realized vol — i.e. only when the premium is actually there. Both legs enter together or neither does, so the book is never left holding a naked directional short. Net delta is held near zero by trading ETHUSDT.BINANCE once per 1-DAY bar whenever the aggregate Black-Scholes delta of the live legs drifts more than hedge_band x per-leg size away from flat; with no option leg live the desired hedge is zero, so the perp is fully unwound between cycles. The structure exits on whichever comes first: the schedule close 24h before European cash settlement, a profit take once profit_take_frac (50%) of the collected premium has decayed, or a tail stop once the combined open loss reaches stop_loss_mult (2x) the premium collected. The chain is resolved by the framework's rolling-options expander (no per-bar scanning, no underlying/smoke fallback), per-leg size is equity-relative (equity * risk_frac / spot, floored at the Deribit ETH minimum), and all per-bar state is O(1) incremental or O(log n) searchsorted.

Optionally reword the hypothesis title to 'weekly-expiry, daily-hedged' so the automated timeframe consistency check does not keep flagging future variants.

Layer-1.5 auto-check flags 'hypothesis says Weekly but bar_type is 1-DAY'. On inspection this is NOT a genuine mismatch: 'weekly' in the hypothesis refers to the OPTION EXPIRY cadence (correctly implemented via rolling_options roll='weekly', dte_target=7), while the 1-DAY bar_type is the daily delta-rehedge and realized-vol cadence the hypothesis explicitly requires ('delta-hedged', 'small perp hedge'). The hypothesis states no weekly BAR timeframe. Faithful implementation.

For the full backtest, ensure lookback_days is widened (or removed) so the strategy samples the maximum available Deribit weekly history; otherwise expect an underpowered / not-measurable outcome downstream.

rolling_options.lookback_days=180 caps the resolvable weekly chain to ~26 cycles / 52 legs. This is stated as a Layer-3 budget fit, but if it also bounds the full backtest, the sample stays too small to distinguish edge from noise for a tail-driven short-vol payoff (cf. L26/L16 on options trade-count).

Short-vol profile with the usual flattering headline metrics (sandbox Sharpe 5.24, win_rate 0.82, but avg_loss $282 > avg_win $223 and return_skew -1.62 / kurtosis 9.75 — losers are larger and fat-tailed, exactly the short-strangle shape L72 warns about). The developer honestly flags that ~26 weekly cycles (lookback_days=180) cannot sample the tail that dominates a short-vol distribution, and avg_trade_return_pct (8.09%) is not meaningful for option legs (premium notional << underlying notional). The real read for the analyst is per-cycle premium capture, DVOL-gate hit rate, and residual hedge delta — NOT the headline percentage return, which the engine computes over ~2x the capital the options book consumes (separate per-venue balances). Also confirm DVOL history depth over the full backtest window: the gate degrades to ungated when DVOL has no sample, so a sparse index would silently convert this to an untimed short-vol harvest. None of this blocks — it is context for the performance decision.

Genuinely orthogonal family — a delta-neutral short-strangle VRP harvest; the corpus has ~0 options strategies, so this is not a variant of the trend/reversion/funding graveyard.

Mechanism is correctly implemented and honest: paired-leg entry (no naked one-legged fills), VRP gate (IV>=RV*ratio), daily banded Black-Scholes delta hedge on the Binance perp, and a bounded tail stop. It executed 50 option trades + a hedge leg as designed.

Front-month structure, 25 calls / 25 puts, net entry delta ~-6.6 on ~410 vega — the book behaved delta-light at inception as intended.

DATA WALL — the tradeable sample is ~6 months, not the 7-year primary span. The Deribit option chain only exists from 2026-02 to 2026-09 (~26 weekly cycles / 56 trades); every legged trade falls in that window. The 2019-2026 ETH-PERPETUAL history is irrelevant because no options existed to sell.

This window cannot sample the tail that DEFINES a short-vol payoff — the developer and QA both concede this. Yet even inside this short, benign window the strategy already printed a 67.3% max drawdown with negative skew and avg_loss ($284) > avg_win ($223): the uncompensated left tail is already visible, not hypothetical.

Headline metrics are non-diagnostic: avg_trade_return_pct 8.09% is premium-notional (premium << underlying notional — not fee-comparable), Sharpe 5.22 / win_rate 0.82 / tail_ratio 610 are the classic flattering short-strangle shape, and total_return is computed over ~2x the capital the options book consumes (separate per-venue balances).

Optimization cannot fix this. A 3-phase walk-forward (3 windows) over ~26 cycles is ~8 cycles/window — pure noise-fitting — and the holdout would be a single-quarter one-shot. There is no history to establish parameter robustness for a distribution whose mean is dominated by rare tail weeks the data does not contain.

EthDeribitWeeklyDeltaNeutralShortStr-952b6438e7

This strategy was a deliberately novel, options-based bet: a delta-neutral short strangle on weekly Deribit ETH options that monetizes the volatility risk premium, hedged with a Binance ETH perp and gated to sell only when implied vol was rich to realized vol. In its ~6-month tradeable window it executed cleanly — 56 trades, paired-leg entries, a working daily banded delta hedge — and posted flattering numbers (Sharpe 5.22, 82% win rate), but the uncompensated left tail was already visible as a 67.3% drawdown with negative skew and losses larger than wins. The backtest-review analyst abandoned it before optimization on a data_wall basis: the Deribit weekly chain only spans 2026-02 to 2026-09, roughly 26 benign cycles that cannot sample the tail a short-vol payoff is defined by, so no walk-forward or holdout could establish robustness. The recommendation was to re-run the identical mechanism once ~1.5–2 years of weekly option-chain and DVOL history exist.

A short-vol/VRP strategy cannot be validated on a chain history too shallow to contain the rare tail weeks that dominate its P&L distribution — flattering headline metrics on a benign sample are non-diagnostic.

The backtest-review gate returned an abandon verdict (data_wall): the Deribit weekly ETH option chain only exists from ~2026-02 to 2026-09, so despite the 7-year ETH-PERPETUAL span the entire test is ~26 benign weekly cycles that structurally cannot sample the tail that defines a short-vol payoff. Optimization and analyst stages were not reached.

Harvest the persistent crypto volatility-risk-premium by selling a delta-neutral OTM short strangle on weekly Deribit ETH options — gated on implied vol being rich to trailing realized vol — while delta-hedging with the Binance ETH perp.

Over a ~6-month tradeable window it executed 56 trades (50 option legs, 25 calls / 25 puts) with a Sharpe of 5.22, 82% win rate and avg_trade_return_pct of 8.09% (premium-notional, not fee-comparable), but already showed a 67.3% max drawdown with negative skew and avg_loss ($284) exceeding avg_win ($223).