DeribitBtcWeeklyCashSecuredShortPutVrpRegimeFiltered

DERIBIT BTC 7-Day Cash-Secured Short-Put Vol-Risk-Premium Harvest with Trend-Regime Filter (Options, Defined-Risk, 3-Parameter, First Options Strategy in Portfolio)

A DEFINED-RISK SHORT-VOLATILITY (theta + vol-risk-premium) options strategy on DERIBIT BTC weekly options — the FIRST options strategy in the entire 1119-experiment portfolio (Options bucket: 0% vs 15% target — the single most under-represented quota). Mechanism harvests the well-documented, structurally-persistent crypto VOL RISK PREMIUM: Deribit BTC option implied volatility (IV) trades systematically ABOVE subsequent realized volatility (RV) because option buyers persistently overpay for convexity/hedging. We monetize this by SELLING cash-secured, out-of-the-money short-dated PUTS when the premium is rich, collecting theta as the option decays. This is NOT a directional price strategy — it explicitly avoids EVERY failed mechanism class this session: NOT a breakout (5+ dead), NOT fast-twitch RSI/Bollinger MR (dead), NOT drawdown accumulation (dead, beta-capture artifacts), NOT cross-sectional momentum on correlated majors (dead), NOT a 4-leg perp pair eaten by fees, NOT short-at-funding-peak (wrong-side-of-bull). The edge is a mechanical cash flow from time decay of overpriced optionality. Risk is DEFINED and bounded: each put is fully cash-secured (we reserve strike×contract cash), so the absolute worst case is BTC → 0 with max loss = (strike − premium received), never a margin blowout. On each WEEKLY decision (every Monday 08:00 UTC Deribit expiry cycle, monitored on daily bars), the strategy: (1) reads the front-week BTC ATM IV from the Deribit greeks/IV catalog and computes 30-day trailing RV of BTC; (2) requires IV/RV ratio >= 1.20 (IV at least 20% rich to realized — the vol risk premium is present); (3) requires BTC > its 50-day SMA (trend-regime filter — only sell puts in non-bear regimes, since selling puts into a confirmed downtrend is catching a falling knife, the structural error that killed the short-funding-carry strategies); (4) if both true, SELLS the ~15-delta OTM put expiring in ~7 days, sized so the cash-secured notional = 20% of equity. Deliberately bullish-to-neutral delta exposure earned as a fee for providing downside insurance. Only one open short put at a time (no laddering) to keep risk simple and walk-forward-clean. Hits the zero-coverage Options quota AND adds genuine return-stream diversification (theta/vega P&L is uncorrelated to the portfolio's 87.5% long-only directional book).

The only code change is the removal of the underlying 'smoke trade' fallback, which is explicitly banned (it degenerates an options strategy into a perp position and fabricates a backtest; the pipeline abandons options runs with zero option-leg trades). Everything else — the IV/RV gate, 50-SMA trend gate, atomic two-leg entry/exit, sizing, parameters — is unchanged. I am NOT attempting to tune around the QA rejection because I do not believe it is tunable: QA's finding is that avg_trade_return_pct is measured against the wide defined-risk WING notional, so a ~4-5%-wide put credit spread's net credit is structurally a negative per-notional return no matter what ivrv_min / sma_period / offsets are set to. Narrowing the wing raises the ratio only by re-introducing the un-hedgeable gap tail that the wing exists to cap, and widening the offsets to the hypothesis's 20-delta/10-delta strikes makes the width (and hence the denominator) larger, not smaller. Four siblings in exactly this class have already been abandoned on the same metric (ETH bull put -8.34%, BTC iron condor -7.47%, Deribit short strangle -7.5%, the naked-put VRP predecessor). I cannot name a code change that plausibly clears the >=0.10% low-fee per-trade floor, so my honest recommendation is abandon.

Exact L67 defined-risk Deribit options signature — not worth 2 hours of optimization, and the developer pre-registered abandon. The headline is attractive (Sharpe 3.21, PF 2.44, total_return +8.3%, 90 trades) but avg_trade_return_pct is deeply NEGATIVE at -4.41% because per-trade return is measured against the defined-risk wing notional; per L67 this positive-Sharpe/PF-but-negative-per-notional combination is structural for defined-risk credit spreads and cannot clear the viability floor, so the Sharpe must not trigger optimize. It is a classic pick-up-pennies short-vol book (short_win_rate 88.9%, max_drawdown 0.53%, vega -511 / theta +1862) whose Sharpe is inflated by the calm regime and whose crash payoff is under-sampled. It is also structurally limited: the option legs only exist from 2024-08 (data_days 684, ~2 years, 90 spreads in one BTC bull regime), so the walk-forward OOS and 20% holdout would be thin and single-regime, and the trend gate means it never sold into a real bear. The developer's own docstring agrees with the QA fee-viability rejection and recommends abandonment. This is the L67 family (BTC iron-condor / ETH bull-put siblings all abandoned on the same signature); benchmark_meaningful is correctly false, and filling the Options quota is not a reason to tune a negative per-notional edge. Failure pattern: fee_edge/no_edge defined-risk Deribit options VRP, negative per-notional edge (L67).

Weekly DEFINED-RISK BTC put credit spread on Deribit (sell the ~7% OTM weekly put, buy the ~12% OTM wing) harvesting the vol-risk premium. Each weekly period is entered only when (a) Deribit DVOL implied vol is at least 20% rich to 30-day realized vol and (b) BTC trades above its 50-day SMA (non-bear regime); both legs are entered atomically (SELL the higher strike, BUY the lower) and closed together 24h before expiry or on a regime flip. One spread at a time, max loss bounded by the strike width. No underlying fallback: with no resolved option chain the strategy trades nothing.

Verification failed (Layer 4 — QA review) [class=code_defect]: - [edge_concern] Sandbox headline looks flattering (total_return +8.9%, Sharpe 2.00, PF 1.21, win_rate 0.49) but this is the known defined-risk-credit-spread illusion: the profit is carried by the short leg (short_win_rate 0.865) while the bought wing loses (long_win_rate 0.115), and the per-notional edge is NEGATIVE (avg_trade_return_pct -2.40%). Return distribution is left-tailed (skew -1.89, kurtosis 6.90, largest_loss -8840 vs largest_win +6950), i.e. occasional full-width losses eat the accumulated credits. Even where dollar-PnL is positive at small scale, the credit collected on a ~4%-wide put spread cannot clear the ~0.10% round-trip cost measured against the wing notional, so the analyst's avg_trade_return_pct>=0.15% floor is unreachable by parameter tuning. - Fee-viability failure. Sandbox avg_trade_return_pct = -2.40% (NEGATIVE), the structural signature of defined-risk premium selling (bull put spread): per-trade return is measured against the wide defined-risk WING notional, not the net credit collected, so it is deeply negative regardless of a healthy-looking Sharpe 2.00 / PF 1.21 / 49% win-rate. This matches the abandoned sibling class (ETH bull-put PF 1.43 avg_trade_return_pct -8.34%; BTC iron condor -7.47%; Deribit short-strangle -7.5%) and the abandoned naked-put VRP predecessor. A negative per-notional edge on a ~4%-wide spread (short -4% OTM, wing -8% OTM) cannot clear the ~0.10% Deribit round-trip cost and can never satisfy the analyst's avg_trade_return_pct>=0.15% (0.10% low-fee) promotion floor. This is a hard fee-viability reject at QA, not a routing-to-analyst case. - Moneyness/delta drift from the hypothesis. Hypothesis specifies a ~20-delta short put and ~10-delta long wing; the code/config use fixed -4% / -8% moneyness offsets (developer changed the wing from -10% to -8% for Deribit weekly liquidity). Structure (short + protective wing) is preserved so this is not itself critical, but the shallower/narrower strikes capture less put-skew, which compounds the fee-viability problem above. - The hypothesis's core differentiator vs the abandoned naked-put sibling is that the realized-vol regime gate is 'actually exercised.' As tuned this iteration the gate is deliberately made near-inert: _gate_ok degrades OPEN until history accrues, rv_rank_min lowered to 10.0 (RV need only be above the 10th percentile), trend_tol widened to 0.25 (price may sit 25% below the 7d SMA), and the entry window widened to 336h. The gate will pass on nearly every cycle, so the stated regime-filter edge is effectively not exercised in this parameterization. - Static-analyzer unbounded-list-growth warnings on self._closes / self._rv_series are false positives — both lists are trimmed (caps at lines 179-181 and 190-192). Division-by-zero guards for sma, width, spot are present. No action required; noting to prevent re-flagging.

Clean, genuinely defined-risk two-leg construction (atomic short+long put, bounded max loss, no naked tail); real short-vol exposure (vega -511, theta +1862); IV/RV + trend gates implemented; fills the zero-coverage Options bucket

L67 signature: high headline Sharpe (3.21)/PF (2.44) but avg_trade_return_pct is -4.41% (negative per defined-risk wing notional) — structurally below the viability floor

Picking-up-pennies short-vol: 88.9% short-win-rate, 0.53% max_drawdown, rare large left-tail losses; Sharpe inflated by the calm regime and the under-sampled crash

Structurally limited option history (~2 years from 2024-08, 90 spreads, one bull regime) — thin/single-regime walk-forward + holdout; trend filter means it never sold into a bear

Developer pre-registered abandon (QA fee-viability rejection, self-assessment recommends abandon)

The trades genuinely implement the stated hypothesis: options_summary shows 30 put legs / 0 calls, 15 long + 15 short trades (a real two-leg put credit spread, not a naked put), net_entry_delta +0.164, net_entry_theta +80.29, net_entry_vega -34.47, avg_dte_at_entry 6.33, avg_iv_at_entry 0.512. Short-vol, short-vega, positive-theta, defined-risk, weekly — exactly as described. No mechanism mismatch.

The defined-risk construction is real and correct: the long wing is always assigned to the lower strike (BUY) and the short put to the higher strike (SELL), legs are entered atomically, and the code refuses to trade a period that resolved only one leg. Max loss is genuinely bounded by strike width.

This is the first real options strategy to reach analysis, and the option-chain expansion, per-leg entry/close plumbing, and greeks reporting all functioned. benchmark_meaningful=false is correctly set (short-vol book is not comparable to BTC buy & hold).

The regime/vol-premium gates behave sensibly: short_win_rate 0.733 (the short put usually expires worthless) with long_win_rate 0.0 (the wing is a cost), profit_factor 2.16, omega_ratio 6.30, tail_ratio 3.55 — the P&L shape is the correct signature of a credit spread.

SUSPECTED BACKTEST ARTIFACT — implausible Sharpe. sharpe_ratio=10.615 and sortino_ratio=22.81 are far beyond the >5 (let alone >10) plausibility flag. They are not produced by a large edge but by a vanishing denominator: annualized_volatility is 0.5433% and total_return is 0.3438% over the whole run. A near-zero return divided by a near-zero volatility is not an edge; it is a ratio artifact.

THE STRATEGY MAKES ALMOST NOTHING. total_return 0.3438%, cagr 0.3937%, expectancy $11.46/trade, max_drawdown 0.0709%. Fifteen spreads over the entire tradeable window returned roughly a third of one percent. Whatever notional the code intends (risk_frac 0.2, avg_position_pct 21.46%), the realized economics are toy-scale and cannot reconcile with a 20%-of-equity cash-secured book harvesting theta — that inconsistency alone means the sizing path is not doing what the hypothesis claims.

THE WALK-FORWARD IS ENTIRELY DEGENERATE. n_trials=0, trial_sharpes=[], in_sample_sharpes=[0.0, 0.0, 0.0], out_of_sample_sharpes=[0.0, 0.0, 0.0], avg_out_of_sample_sharpe=0.0, is_overfitted=TRUE, pbo=null. Phase 2 produced ZERO information. There is no out-of-sample evidence of any kind.

THE HOLDOUT 'PASS' IS SPURIOUS. holdout.passed=true but ratio=0, walk_forward_oos_sharpe=0.0, and holdout_sharpe=10.6154 — bit-for-bit identical to the full-sample base backtest Sharpe. The holdout did not measure a held-out window; it re-reported the full-sample number against a zero baseline. This is not a validation.

NO ROBUSTNESS EVIDENCE. deflated_sharpe=null, pbo=null, n_trials=0. probabilistic_sharpe=0.9929 is meaningless here — it is computed from 15 daily-return observations and ignores trial count. sharpe_ci_low=3.13 / sharpe_ci_high=21.52 is a CI spanning a factor of seven, the signature of a sample far too small to estimate a Sharpe.

SAMPLE IS FAR TOO SMALL. data_days=15, only 15 non-zero daily returns exist, and the option_chain_schedule only begins 2025-07-11 despite a nominal start of 2018-08-14 (lookback_days: 365). Effectively ~one year of chain, 15 spreads, 15 P&L days. Nothing statistically meaningful can be concluded, and a short-vol strategy specifically needs to survive a tail event — this window contains none.

ZERO COMMISSIONS ON 30 OPTION TRADES. total_commission=0.0, commission_pct_of_gross=0.0, impact_folded=false, total_impact_usd=null, capacity_usd=null. Deribit charges options fees (per-contract, capped as a fraction of premium) plus settlement fees. The entire $344 of reported profit is therefore GROSS of options transaction costs. On far-OTM weekly premium the fee cap is a large fraction of the credit collected, so it is entirely possible the net edge is zero or negative. I have not independently verified the engine's options fee path — I am reporting the symptom (commission exactly 0.0 on 30 option legs), not asserting a confirmed cause.

THE SENSITIVITY SURFACE IS CONTAMINATED AND CANNOT BE TRUSTED. Parameters that provably cannot affect the option book change the reported Sharpe: smoke_warmup_bars, smoke_hold_bars, and smoke_risk_pct are verification-only (gated off when a schedule exists), and stop_loss_pos_pct is documented in the code itself as '(unused: spread caps risk)' — yet all of them report Sharpe 16.154 versus a base of 10.615. Simultaneously, the economic parameters put_offset (-0.056 to -0.084), contracts, and iv_estimate produce five IDENTICAL Sharpes, because the strikes are pre-baked into option_chain_schedule and changing the offset does nothing. So the sweep varied no-op knobs and could not vary the real ones. sensitivity.passed=true with cliff_count=1 is therefore not a meaningful pass, and the one flagged cliff (ivrv_min: 15.87 -> 6.83 across the grid) is the only economically live parameter tested.

The heatmap is a constant plate (16.1541 / 15.9442 in every cell across all 25 combinations of ivrv_min x long_put_offset), which is further evidence that the optimizer is not actually reaching the traded instruments.

OPTIMIZATION ATTEMPTS ARE EXHAUSTED: this is attempt 3 of a 2-attempt budget. No further iteration is authorized.

fails the plausibility gate, not the threshold — Sharpe > 10 is a suspected-artifact flag, not a win

computed on 30 zero-commission trades

positive, but CI spans 3.13-21.52 on 15 observations — uninformative

spurious — ratio 0.0, holdout_sharpe identical to full-sample Sharpe

not computable — 0 trials; missing robustness is not a pass

zero fees on 30 Deribit option legs — result is gross of options costs

and the surface itself is contaminated by no-op parameters

near-zero vol is what manufactures the Sharpe