DeribitBtcAtmCalendarSpreadTermStructureThetaHarvest

BTC Deribit ATM Calendar Spread — IV Term-Structure Theta Harvest (Sell Front-Week / Buy Back-Month, Net-Long-Vega)

Single-venue, options-only, delta-neutral term-structure harvest on Deribit BTC. Sell the near-dated (~7 DTE) ATM straddle-leg and buy the further-dated (~30 DTE) ATM option of the same type+strike, forming a horizontal calendar. The front leg decays its time value far faster than the back leg (theta is concentrated near expiry), so when the underlying stays near the strike the spread accrues net positive theta while the long back leg caps risk and supplies positive vega. Entry is gated on the IV term structure: only open when front-month IV is NOT in steep backwardation versus back-month IV (i.e. avoid the post-shock regime where the front is richer than the back and the calendar inverts against us). This is NOT a funding carry, NOT a basis trade, and NOT a directional price-prediction bet — it monetizes the maturity-dependent decay of option time value, a structural property absent from the entire cross-venue-carry failure cluster. Max loss is bounded by the small net debit paid. Fills the most under-represented portfolio bucket (DERIBIT options: 1 of 1204 experiments) with a vega sign and risk profile orthogonal to a short strangle.

Monetizes the maturity-dependent decay of option time value -- a structural property absent from the entire cross-venue-carry failure cluster -- via a horizontal calendar, which is NOT a funding carry, NOT a basis trade and NOT a directional price bet. Theta is concentrated near expiry, so a short front / long back of the same strike collects net positive theta when the underlying stays near the strike, with the long back capping risk (max loss = net debit) and supplying positive vega -- a vega SIGN and risk profile orthogonal to a short strangle. The IV-term-structure gate (proxied by realized-vol term structure since per-contract front/back IV is not in the catalog) avoids the steep-backwardation/post-shock regime where the front is richer than the back and the calendar inverts against us. It is implemented on the supported structure="calendar" expander path (shared entry, per-leg dte, common ATM strike), not a self-implemented multi-expiry parser, so the analyzed backtest trades real two-expiry option legs (the smoke path fires only in verification). Fills the most under-represented portfolio bucket (Deribit options). leverage=1.0 (options venue cap); debit-bounded sizing, no unused-leverage gate.

suspected backtest artifact + total optimization failure, at the 2-attempt iteration limit. (1) The optimization never produced a result: walk_forward.n_trials=0, is_overfitted=TRUE, in_sample_sharpes=[0,0,0] / out_of_sample_sharpes=[0,0,0]; the optimized re-run produced total_trades=0 with metrics_reliable=false and extra_instruments=[] (the option legs were never injected), so the best-params config is unverifiable. The holdout 'passed=true' is a divide-by-zero artifact (ratio=0 against a zero WF-OOS denominator), not evidence of robustness. (2) Plausibility fails decisively: the hypothesis advertises 'max loss bounded by the small net debit,' yet the initial backtest shows max_drawdown=4.74 (474%) with ±200% single-day returns (2024-10-17 +206%, 2024-10-18 −156%) — impossible for a defined-risk net-debit calendar, indicating the option-leg PnL is not marked/settled as a bounded spread (symptom stated; root cause not independently verified — recommend manual review of the Deribit calendar leg PnL marking before this options family is retried). Even the headline edge is thin (profit_factor 1.14, sharpe_ci_low 0.168 nearly straddling 0, skew 2.23 / kurtosis 11.5 — a few fat-tail days carry it). Sensitivity passing (0 cliffs) is necessary but nowhere near sufficient against an all-zero walk-forward, a degenerate holdout, and a zero-trade optimized config. Not iterate: the optimizer ran 0 trials and every window is empty, the optimized config does not trade, and this is attempt 2 of 2 — there is no robust region to tune toward. Not revise_hypothesis: there is no PROMOTED sibling of this Deribit-calendar mechanism to redirect, and the failure is harness-sparsity + a PnL artifact, not structural decay of a proven edge that a different instrument would rescue. FAILURE PATTERN: single-underlying Deribit weekly/monthly option-roll structures (here a net-long-vega ATM calendar on BTC) are too sparse for the walk-forward — the optimizer returns n_trials=0 / all-zero IS+OOS / is_overfitted=true with a degenerate divide-by-zero holdout and a zero-trade optimized config, and the initial backtest reports artifact-level metrics (474% DD, ±200% days) that violate the defined-risk premise. This is the third Deribit-options-roll instance this session (ETH bull-put-spread, ETH ATM long-call-roll) to die on the same sparsity/zero-OOS/PnL-artifact signature.

Single-venue, options-only, delta-neutral horizontal CALENDAR spread on Deribit BTC: at each weekly open SELL the near-dated (~7 DTE) ATM call and BUY the further-dated (~30 DTE) ATM call of the SAME strike, harvesting maturity-dependent time decay. The backtest rolling-options expander resolves the calendar via parameters[rolling_options] with structure="calendar" (discover_deribit_calendar picks ONE common ATM strike with candle data at both expiries and stamps both legs with a shared entry date and per-leg dte 7/30). Because the base RollingOptionsStrategy groups periods by expiry (which would split a calendar), this subclass re-groups the schedule by ENTRY in on_start and assigns front (earlier expiry -> SELL) vs back (later expiry -> BUY). The front decays time value faster than the back (theta concentrates near expiry), so while BTC stays near the strike the spread accrues net positive theta; the long back caps risk and supplies positive vega (net-long-vega). Entry is gated on the IV term structure, proxied by the realized-vol term structure: it opens only when NOT in steep backwardation (RV_short/RV_long <= backwardation_max), avoiding the post-shock regime where the front is richer than the back and the calendar inverts. Both legs are entered atomically once their markets are live; the structure is closed at the front-expiry roll buffer, or early on a profit-take / stop measured vs the net debit paid. Max loss is bounded by the small net debit. Subclasses BOTH RollingOptionsStrategy and FactoryStrategy with the four abstract hooks re-declared (Layer-1 AST). A verification-only long-perp smoke trade on the Deribit primary proves order plumbing in Layer-3 (no schedule there), gated OFF whenever a real schedule exists so the analyzed backtest always trades real option legs. ~3 parameters (rv_short, rv_long, backwardation_max).

CONTINUOUS coverage across all regimes (positive every year 2022-2026, incl. 2025 +108% and 2026 +155%) — the IV-term-structure gate fires regularly, so 92 trades are distributed across the full span, fixing the recent-regime emptiness that killed the funding-carry family.

Greeks-correct defined-risk calendar: net_entry_vega +784 (long vol), net_entry_theta +1,450 (collects net decay), delta-neutral (+0.89), ATM; max loss bounded by the net debit; max_drawdown only 4.74%, positive skew +2.23.

Strong but plausible (non-artifact) metrics: Sharpe 4.19 (below the >5 flag), Sortino 10.78, win_rate 54% (not >90%), PSR 0.989 — a well-built net-long-vega harvester.

Fills the 0.1% Deribit options bucket with a DIVERSIFYING vega sign (net-long-vega, orthogonal to the short-vol siblings) and ~3 robust parameters.

Thin per-trade edge: profit_factor 1.14, avg_loss ($673) slightly > avg_win ($645) — the deflated-Sharpe/walk-forward is the decisive robustness test.

Large consecutive single-day swings (2024-10-17 +207% / 2024-10-18 -156%) — almost certainly front-leg-expiry MTM realization that nets to ~+51% and is contained by the 4.74% max_dd, but must be confirmed not to be a risk-cap violation.

92 trades is moderate (monthly-ish calendar cadence) — distributed, but DSR must still clear on the sample.

Impact/capacity modeling is off (null) for this run, so per-trade execution cost on the 2-leg structure isn't stressed here.

positive every year 2022-2026, 92 distributed trades

trades across regimes incl. recent

vega +784, theta +1450, delta +0.89, max_dd 4.74%

correct long-vega calendar, bounded loss

<5 (non-artifact)

Mechanism is economically coherent and fills an under-represented bucket (Deribit options): net-long-vega term-structure theta harvest with correct entry greeks (net_entry_vega +784, net_entry_theta +1450, net_entry_delta ~0.9, calls 92 / puts 0).

Sensitivity grid is clean (0 cliffs) across all economic params.

Fee drag negligible (commission_pct_of_gross ~0, weekly cadence).

Optimization wholly failed: walk_forward.n_trials=0, is_overfitted=TRUE, in_sample_sharpes=[0,0,0] and out_of_sample_sharpes=[0,0,0] — no window produced a non-zero result.

Optimized backtest is degenerate: total_trades=0, metrics_reliable=false, extra_instruments=[] — the best-params config never trades, so the optimization is unverifiable.

Holdout 'passed=true' is a divide-by-zero artifact: ratio=0 because WF-OOS=0.0; holdout_sharpe 7.26 against a zero denominator is meaningless.

Suspected backtest artifact: 474% max_drawdown and ±200% single-day returns (2024-10-17 +206% / 2024-10-18 −156%) directly contradict the defined-risk 'max loss = small net debit' premise — option-leg PnL is not marked as a bounded spread (mechanism not independently verified; recommend harness review of Deribit calendar PnL).

Edge is thin and fragile even at headline: profit_factor 1.14, sharpe_ci_low 0.168 (nearly straddles 0), return_skew 2.23 / kurtosis 11.5 (outlier-driven).

At the iteration limit (attempt 2 of 2).

headline only; CI low 0.168 nearly straddles 0, outlier-driven

474% impossible for a defined-risk net-debit calendar

not computable; walk-forward had 0 trials

is_overfitted=true, n_trials=0

all-positive

DeribitBtcAtmCalendarSpreadTermStructureThetaHarvest sold front-week / bought back-month ATM BTC calls to harvest the faster near-expiry theta decay, a net-long-vega calendar gated on IV term structure and pitched as debit-bounded, diversifying vega-sign exposure for the empty Deribit bucket. The baseline looked attractive — continuous coverage, positive every year, Sharpe 4.19, correct greeks — but optimization collapsed entirely: zero trials, all-zero walk-forward windows, a zero-trade optimized config, and a divide-by-zero holdout, while ±200% days and a 474% drawdown contradicted the bounded-loss premise. The analyst ruled it a harness artifact plus optimization failure, the third Deribit option-roll to die on the same signature this session, recommending a manual review of the calendar PnL marking before the family is retried. It ended at the attempt limit as abandoned, reaching optimization and analysis but never risk review.

Single-underlying Deribit option-roll structures are too sparse for walk-forward and the harness mis-marks the leg PnL: this net-long-vega calendar is the third such instance this session (after the ETH bull-put-spread and ETH ATM long-call-roll) to die on the identical sparsity / zero-OOS / degenerate-holdout / PnL-artifact (474% DD, ±200% days) signature — the VRP/term-structure premia are real but un-validatable until the Deribit options harness PnL marking and walk-forward density are fixed.

It passed the pre-optimization backtest-review gate (verdict: optimize) but was abandoned at the post-optimization analyst gate (verdict: abandon) as a suspected backtest artifact plus total optimization failure — the optimizer ran zero trials, every walk-forward window was empty, the optimized config never traded, and the 474% drawdown / ±200% days indicate the calendar's option-leg PnL isn't marked as a bounded spread — at the 2-attempt iteration limit, so it never advanced to risk review.

A delta-neutral, options-only IV term-structure theta harvest on BTC Deribit — at each weekly open selling the near-dated (~7 DTE) ATM call and buying the further-dated (~30 DTE) ATM call of the same strike (a net-long-vega horizontal calendar), gated to enter only when the front wasn't in steep RV backwardation vs the back, to monetize the faster front-leg time decay with a debit-bounded max loss, filling the near-empty Deribit options bucket with a vega sign orthogonal to short-vol siblings.

The baseline traded continuously across regimes — 92 calendars (greeks-correct: net vega +784, net theta +1,450, delta ~0.89, ATM) positive every year 2022-2026, Sharpe 4.19, Sortino 10.78, profit factor 1.14, win rate 54%, PSR 0.989 — but with implausible artifacts (±200% single days 2024-10-17/18 and a 474% max drawdown that violates the debit-bounded premise); optimization then wholly failed: walk-forward n_trials=0 with all-zero IS and OOS Sharpe, is_overfitted=true, a degenerate divide-by-zero holdout (ratio 0 against zero OOS), and a zero-trade optimized config (extra_instruments empty).