BtcDeribitWeeklyShortStrangleVrpGated

BTC Deribit Weekly Short Strangle — Volatility Risk Premium Harvest (IV>RV Gated, Delta-Balanced)

Single-venue, options-only volatility-risk-premium harvest on Deribit BTC. Each week, when 7-day at-the-money implied volatility exceeds trailing 7-day realized volatility by a margin (the VRP is positive and elevated), sell a delta-balanced OTM strangle on the nearest weekly expiry: short one ~15-delta put and one ~15-delta call, sized to roughly offset net delta near zero. Collect both premiums and hold to European cash-settled expiry, capturing theta decay as the options expire worthless in the typical regime. This is NOT a directional bet and NOT a funding/basis carry — it monetizes the structural, repeatedly-documented gap between what crypto option buyers pay for protection/leverage (IV) and what the underlying actually delivers (RV). Fees (~0.06% RT on Deribit, two legs) are negligible against weekly premium income of typically 1-3% of notional. Turnover is weekly (~52 cycles/yr), so fee drag is minimal. Fills the most under-represented portfolio bucket (DERIBIT options: 1 of 1204 experiments).

Iteration 2 fix for the semantic-auditor code<->hypothesis misalignment. The ONLY change: in `_enter_leg` the `if not self._have_iv:` branch now `return False` instead of falling through. Previously, when IV could not be resolved as-of the entry bar (before the first DVOL sample, or when neither deribit_dvol nor deribit_book_summary resolved an IV), the strangle was submitted with NO IV>RV / VRP confirmation — directly contradicting the hypothesis requirement to sell 'ONLY when the VRP is positive and elevated'. Now every option-leg entry requires (a) an IV value known as-of the entry bar AND (b) the strict VRP gate IV >= RV * vrp_margin. In the real backtest deribit_dvol spans 2021->2026 so IV is essentially always available; this guard only refuses genuinely-unconfirmed entries, so it does not regress the option-trade count (Layer-3 uses the separate perp smoke path, which is untouched). All imports, signal logic, IV sourcing, roll mechanics, and the smoke path are unchanged, so Layers 1-3 and the initial backtest remain green.

Not backtestable on this instrument: only 3 weekly strangles traded over the entire 2018-2026 window (equity flat until 2026-05-21) because historical weekly Deribit option 1H candles are only available for the most recent expiries — a structural data wall the developer cannot fix with code or parameters, so a 2-hour optimization would fit noise on a 3-trade sample. Even that sample already showed the classic short-vol blow-up (one June-5 crash put lost -$2,377, pnl_pct -1346%, wiping out five ~$160 premium wins; profit_factor 0.34, net -1.58%), and the legs are near-ATM/ITM with mismatched quantities (net delta -0.277), not the delta-balanced ~15-delta OTM strangle the hypothesis specifies. The VRP-harvest mechanism is sound and fills an empty bucket, so it is worth reframing onto a Deribit tenor with real multi-year candle history (monthly/quarterly expiries) rather than weekly — but that is a hypothesis reframe for the Research Lead, not an optimize/iterate on this weekly build.

Single-venue, options-only weekly SHORT STRANGLE on BTC Deribit that harvests the volatility risk premium. Each weekly cycle, when 7-day implied vol (DVOL index, fallback book-summary mark_iv) exceeds trailing 7-day realized vol by a margin (IV >= RV * vrp_margin), it sells a delta-balanced OTM strangle on the nearest weekly expiry (short ~15-delta / ~7%-OTM call + put) and holds to European cash-settled expiry, closing 24h before to avoid settlement slippage. Built on RollingOptionsStrategy: the backtest expander resolves the historical weekly chain and drives the roll clock off BTC-PERPETUAL.DERIBIT. A verification-only perp smoke trade satisfies the >=1-trade sandbox gate and is dead whenever a real schedule is present.

VRP-harvest mechanism is legitimate and fills the most under-represented bucket (DERIBIT options: 1 of 1204 experiments)

Market-neutral, orthogonal to the trend/momentum/funding strategies already in the book

Options actually traded (options_no_option_trades gate satisfied): 6 real option legs, greeks and IV populated

Only 3 weekly strangles ever traded (all May-Jun 2026); equity curve is flat 2018->2026-05-21 because historical weekly Deribit option candles exist only for the most recent expiries — a data wall, not a parameter issue

data_days=3, metrics_reliable=false ('only 3 daily returns') — sample far too small to support any optimization; the 3-phase sweep would fit noise

Short-vol tail already realized: one June-5 crash put lost -$2,377 (pnl_pct -1346%), erasing five ~$160 premium wins; profit_factor 0.34, net total_return -1.58%

Construction contradicts the hypothesis: legs are near-ATM/ITM (call moneyness 0.96, delta 0.197), not ~15-delta OTM, and leg quantities are mismatched (put 0.25 vs call 1.0) giving net_entry_delta -0.277 — not delta-balanced

Code↔hypothesis misalignment found by the semantic auditor — the code does NOT implement the hypothesis. Re-code the strategy to implement the hypothesis EXACTLY (instrument, timeframe, direction, the named edge/mechanic, sizing). Concrete issues: In `_enter_leg`, the `if not self._have_iv:` branch records reason 'no_iv' and falls through (only a comment, no `return False`), so the short strangle is submitted with no IV>RV / VRP confirmation whenever IV data is unavailable as-of the entry bar (before the first DVOL sample, or when neither deribit_dvol nor deribit_book_summary resolves an IV). This sits in the real analyzed-edge entry path (the docstring designates `_enter_leg` as 'the analyzed edge'), and it contradicts the hypothesis/rationale requirement that the strangle is sold ONLY when 'the VRP is positive and elevated' (rationale: 'The gate (IV >= RV * vrp_margin) only sells vol when it is rich').

BtcDeribitWeeklyShortStrangleVrpGated aimed to monetize the well-documented crypto variance risk premium by selling delta-balanced OTM weekly strangles on Deribit BTC only when implied vol was richer than realized vol, filling the portfolio's most under-represented bucket. The mechanism was legitimate and the option legs did trade, but historical weekly Deribit option candles exist only for the most recent expiries, so across the 2018–2026 window only 3 strangles ever fired (all May–June 2026), leaving equity flat until then and data_days at 3 with metrics flagged unreliable. That tiny sample already showed the classic short-vol tail — a single June-5 crash put lost -$2,377 and erased five small premium wins, leaving profit factor 0.34 and total return -1.58% — and the constructed legs were near-ATM/ITM with mismatched quantities rather than the specified delta-balanced 15-delta OTM. It was abandoned at the pre-optimization backtest-review gate as a structural data wall no code or parameter change could fix, with the recommendation to reframe the VRP edge onto a longer-history Deribit tenor (monthly/quarterly) as a new hypothesis rather than iterate on the weekly build.

A sound mechanism (crypto VRP harvest, filling the empty Deribit-options bucket) is untestable on a tenor lacking multi-year candle history — weekly Deribit options are a data wall, so the edge should be reframed onto monthly/quarterly expiries with deep history before spending optimization time.

The backtest-review gate abandoned it before optimization: historical weekly Deribit option candles exist only for the most recent expiries, so the 8-year window yielded just 3 tradeable cycles (vs a 30 threshold) — a structural data wall — and even that tiny sample failed on profit factor, total return, and already showed the classic short-vol blow-up, while the legs came in near-ATM/ITM with mismatched quantities (net delta -0.277) rather than the specified delta-balanced 15-delta OTM structure.

A single-venue, options-only volatility-risk-premium harvest that each week, when 7-day implied vol exceeded trailing realized vol, sold a delta-balanced ~15-delta OTM short strangle (short call + short put) on the nearest weekly Deribit BTC expiry and held it to European cash-settled expiry to collect theta.

Only 3 weekly strangles (6 option legs) ever traded, all in May–June 2026, producing total return -1.58%, profit factor 0.34, avg trade return -$262.92 and a max drawdown of 2.36%; with data_days=3 the metrics were flagged unreliable, and a single June-5 crash put lost -$2,377, wiping out five ~$160 premium wins.