BtcDeribitWeeklyAtmCalendarThetaVrpLowVolGated

BTC Deribit Weekly ATM Long Calendar Spread, Delta-Neutral Theta/Term-Structure Harvest (Defined-Risk, Near-Weekly Short Leg vs ~3-Week Long Leg, Gated on Front-Rich Term Structure + Low Realized-Vol Regime, No IV-Rank Dependence)

A DELTA-NEUTRAL, DEFINED-RISK options calendar spread on BTC Deribit that harvests the front-month variance-risk premium via TERM STRUCTURE rather than absolute IV level. Structure: SELL the nearest weekly ATM option (call or put, whichever keeps net delta closest to zero at entry) and BUY the same strike ~2-3 weeks out, paying a net debit. Max loss = net debit (hard-defined, no naked tail). The position profits from faster theta decay of the short near-leg relative to the long far-leg while the underlying stays near the strike. The edge signal is the option's OWN per-instrument IV (available in the Deribit catalog greeks/IV) compared across expiries, plus realized vol from OHLCV — deliberately AVOIDING the DVOL/IV-rank index that prior runs found was never wired into strategies (gate silently defaulted open). This fills the factory's largest quota gap (options ~1.8% vs >=15% target) with a structure NOT already in the pipeline (which is saturated with short strangles, iron condors, CSPs, and put spreads), is delta-neutral (denting the 85.9% long-only over-representation), and is defined-risk so it cannot reproduce the naked-put 95% drawdown / equity-scaling artifacts that abandoned prior selling strategies. Weekly roll cadence gives enough trades for a robust walk-forward + holdout, addressing the monthly-roll trade-sparsity wall.

Implements the hypothesis as a rolling weekly ATM calendar that harvests term-structure theta + front VRP, defined-risk (net debit), delta-neutral. Uses the proven calendar chain-expansion path (parameters['rolling_options'] with structure='calendar'): the backtest expander walks each weekly open, picks one common ATM strike with candle data for both the ~7 DTE and ~21 DTE expiries, and injects the concrete call legs + per-period schedule; on_start groups the schedule by shared open into front/back periods, _run_roll opens front-SELL/back-BUY when the gates pass and closes before the front expiry, _manage_open applies TP/SL on the spread MtM. Distinct from the existing BtcDeribitAtmCalendarSpreadTheta by (a) a LOW realized-vol regime gate (stand down in stress where big moves hurt the calendar), (b) a FRONT-RICH VRP gate (front_iv - RV) rather than only a backwardation cap, and (c) a ~21 DTE (2-3 week) back leg instead of ~30 DTE. Calls-only chain (not call+put) is a deliberate robustness choice: by put-call parity a same-strike ATM call vs put calendar is economically identical, while forcing the expander to also find liquid puts at the same strike risks skipping whole roll dates -> zero option-leg trades -> options_no_option_trades abandonment (which just happened to a sibling options strategy). Delta-neutrality is therefore satisfied by construction and re-checked via black_scholes_delta at entry. Gates use lenient, trade-producing defaults and allow-on-missing-IV so the real backtest reliably places option trades. Venue DERIBIT (BTC options), no leverage on the option legs (leverage=1.0). Primary = BTC-PERPETUAL.DERIBIT drives the roll clock and supplies long daily history for the data-sufficiency gate; the perp itself is not traded in the analyzed strategy. Extends FactoryStrategy directly (Layer-1 recognized) with module-level _bar_ts to avoid the Layer-2 proxy staticmethod arity issue; Layer-1 static + Layer-2 synthetic both pass locally.

Pre-optimization gate fails: the strategy correctly implements a BTC Deribit weekly ATM call calendar (92 real option trades, positive net theta +306, long vega +217, ATM moneyness 1.0009), but its realized behavior contradicts its own delta-neutral/defined-risk/tail-safe premise and its economics are marginal-to-negative. Risk-adjusted performance is negative (Sharpe -0.34, Sortino -0.41, omega 0.94, profit_factor 1.08, expectancy +$6.55/trade within noise), while max drawdown is ~96% (calmar 0.63 vs cagr 0.66 implies ~105%) with ±40% monthly swings and realized net_entry_delta 0.247 (not the claimed ~0.02) — i.e. a directionally-exposed, high-variance position, not a clean theta harvest. The root cause is structural and not tunable: an ATM same-strike calendar needs the underlying to stay near the strike, but BTC trended hard across the window (strikes walked 108k->60k) and repeatedly ran over the weekly rolls; the rv_max regime gate did not filter these moves, and tightening it would gut an already-short sample. Compounding this, real Deribit option-leg data spans only ~11 months (2025-07 to 2026-06, 92 weekly trades), so the 3-phase optimization's walk-forward windows (~3 months) and 15-day holdout (~2 trades) are holdout-sparse, and option fees are effectively unmodeled (commission ~$0.0002, 0% of gross) so the tiny positive expectancy would not survive real per-leg fees. Optimization tunes gate parameters; it cannot convert a trend-fragile, non-neutral calendar with a 96% drawdown into the delta-neutral, tail-safe theta harvest the hypothesis described. Abandon rather than spend 2 hours optimizing.

Single-venue, options-only, defined-risk weekly ATM CALENDAR SPREAD on BTC Deribit. Each weekly roll it SELLs the ~7 DTE ATM call and BUYs the same-strike ~21 DTE (2-3 week) ATM call for a small net debit (max loss = net debit). It harvests the front-month variance-risk premium via TERM STRUCTURE and faster near-leg theta decay while the underlying stays near the strike. An ATM same-strike call calendar is delta-neutral by construction (net |delta| ~0.02); the strategy verifies |net delta| <= max_abs_delta at entry. Entry is gated on (1) a LOW realized-vol regime (7d RV <= rv_max), (2) no steep front-backwardation (front_iv - back_iv <= max_front_backwardation), and (3) a front-rich variance-risk premium (front_iv - RV >= min_front_vrp) — using per-instrument Deribit book-summary IV plus OHLCV realized vol, deliberately avoiding the DVOL/IV-rank index. Exits on spread take-profit / stop vs entry debit, or roll-close before the front expiry. Real option legs are resolved by the calendar chain expander (rolling_options structure=calendar); a perp smoke trade fires ONLY in Layer-3 verification (no schedule) to satisfy the >=1-trade gate and never in the real backtest.

Mechanism is correctly implemented per the hypothesis: 92 real Deribit option trades (calls only via put-call-parity argument), ATM moneyness 1.0009, ~14 DTE avg, positive net theta (+306) and long vega (+217) — a genuine calendar/term-structure theta harvest, not a synthetic proxy

Defined-risk structure cannot reproduce the naked-selling blow-ups; calendar machinery (schedule, two-legged entry, TP/SL on spread MtM) works

Negative risk-adjusted performance at base: Sharpe -0.34, Sortino -0.41, omega 0.94 (<1), profit_factor 1.08, expectancy +$6.55/trade (within noise)

Core premise violated: ~96% max drawdown (calmar implies ~105%) and ±40% monthly swings for a strategy sold as delta-neutral/tail-safe; realized net_entry_delta 0.247, not ~0.02 — it is a directionally-exposed, high-variance position

Structural failure, not tunable: ATM call calendars need the underlying near strike, but BTC trended hard (strikes walked 108k->60k) and repeatedly ran over the rolls; the rv_max regime gate did not protect, and tightening it would thin an already-short sample

Only ~11 months of real Deribit option-leg data (2025-07 to 2026-06) -> 15-day holdout ~2 trades, walk-forward windows ~3 months each: holdout-sparse

Option fees effectively not applied (commission ~$0.0002, 0% of gross); the marginal edge would not survive real per-leg Deribit fees plus realized variance

This strategy set out to fill the factory's options quota gap with something genuinely new: a defined-risk, delta-neutral BTC Deribit weekly ATM call calendar harvesting term-structure theta and front-month VRP, gated on realized vol rather than the IV-rank index earlier runs had left disconnected. The mechanism was implemented faithfully — 92 real Deribit option-leg trades with positive net theta (+306), long vega (+217), and near-ATM moneyness — proving the calendar machinery worked. But its realized results betrayed the thesis: a ~96% max drawdown, ±40% monthly swings, and a net entry delta of 0.247 revealed a directionally exposed, high-variance position, because BTC trended hard across the window and repeatedly ran over the weekly rolls while the rv_max gate failed to filter those moves. Because this failure was structural rather than tunable — and compounded by only ~11 months of holdout-sparse data and effectively unmodeled option fees — the analyst abandoned it at the pre-optimization backtest-review gate rather than spend hours optimizing a trend-fragile, non-neutral calendar.

An ATM same-strike calendar is only delta-neutral and tail-safe while the underlying stays near the strike; when the asset trends hard (BTC strikes walked 108k→60k) a realized-vol regime gate does not protect it, and per-leg option fees plus a holdout-sparse ~11-month sample must be modeled before trusting a tiny positive expectancy.

The analyst's pre-optimization backtest review returned an 'abandon' verdict: the strategy's realized behavior contradicted its delta-neutral/tail-safe premise and its economics were marginal-to-negative, so optimization, risk review, and later stages were never reached.

A delta-neutral, defined-risk BTC Deribit weekly ATM call calendar spread (sell ~7 DTE / buy ~21 DTE, same strike) meant to harvest front-month variance-risk premium and term-structure theta while gating entries on a low realized-vol regime and a front-rich IV curve, deliberately avoiding the DVOL/IV-rank index prior runs never wired in.

It executed 92 real Deribit option trades (calls only, ATM moneyness 1.0009, ~14 DTE avg, positive net theta +306, long vega +217) with a 52% win rate, CAGR 0.66 and expectancy of only +$6.55/trade, but risk-adjusted returns were negative (Sharpe -0.34, Sortino -0.41, omega 0.94, profit factor 1.08) with a ~96% max drawdown and a realized net entry delta of 0.247 rather than the claimed ~0.02.