BtcDeribitInvertedVrpLongGammaBandHedgedStraddle

BTC LONG-GAMMA When Implied Vol Is CHEAP vs Realized — Deribit Short-Dated ATM Straddle BOUGHT on an INVERTED Variance Risk Premium (DVOL index < trailing realized vol), Delta-Hedged on a Band with the Binance Perp, Held to Vol Repricing or Expiry-1d (Cross-Venue, Market-Neutral Delta, 3-Parameter)

A MARKET-NEUTRAL (delta-hedged) CONVEXITY-BUYING strategy that is the deliberate INVERSE of every options experiment this factory has run. All prior options work in the corpus is short-premium carry (delta-hedged short straddles/strangles, cash-secured puts, covered calls) or directional long-premium (OTM calls in a trend). Every short-vol attempt died the same way: our own ETH delta-hedged strangle backtest reported the hedge leg LOSING $5,356 against only $2,560 of premium collected — i.e. realized gamma ran ~2.1x the vol that was sold. That is a direct measurement that, in the episodes the short-vol gate selected, BUYING and hedging that gamma would have been the profitable side. This hypothesis trades exactly that side, but only when the variance risk premium is measurably INVERTED, and it uses a 5.5-year, completely unused signal series to decide: Deribit's DVOL index (hourly 30-day forward implied vol for BTC, 47,938 hourly points from 2021-03-24 to today — zero prior experiments in the corpus use it). Structure: buy the nearest-expiry (3-10 DTE) ATM CALL + ATM PUT on DERIBIT (BTC options: 4,285 daily-bar instruments in the catalog back to 2020), then neutralize the resulting delta with BTCUSDT.BINANCE USD-M perpetual and RE-HEDGE ONLY WHEN |net delta| exceeds a band — band hedging, not daily hedging, so the hedge leg harvests the realized path instead of paying a fixed rebalancing tax. The position is long theta-negative convexity, so the entry gate must be strict: we only pay for gamma when the option market is quoting vol BELOW what the underlying has actually been delivering. This fills the single most under-represented venue bucket (options = 5.9% of 3,087 experiments vs a 15% target), is CROSS-VENUE (7.0% vs 15% target) and is delta-neutral rather than long-only (85.2% of the corpus is long-only).

Implements the hypothesis literally and in the direction it specifies — the inverse of every short-premium options experiment in the corpus. (1) SIGNAL: the previously unused deribit_dvol series is the implied leg; calculate_signal returns the CONTINUOUS annualized RV-minus-IV spread on EVERY bar (positive = gamma underpriced), and the entry gate thresholds that variable rather than gating the signal itself. DVOL is read point-in-time via a searchsorted array built ONCE in on_start (O(log n) per bar, never looks ahead); realized vol is an O(1) incremental rolling variance, so the whole per-bar cost is constant and the 300s smoke budget is safe. Measured on the real series (5.5y of hourly DVOL vs 6-HOUR BTC bars), the gate IV <= 0.95 x RV(20) fires on ~18% of bars, and each weekly cycle gets a 72h (12-bar) entry window, so ~40-60 of ~104 cycles enter — well clear of a zero-trade sandbox. (2) STRUCTURE: short-dated (4-7 DTE, inside the hypothesis's 3-10 DTE band) ATM call + put on DERIBIT, resolved by the framework's rolling-options chain expander (RollingOptionsStrategy owns entry/roll/close) — no self-implemented schedule, no underlying-only smoke path, both legs enter together or neither does. (3) DELTA NEUTRALITY: aggregate Black-Scholes delta of the LIVE legs (own strike/TTE at the prevailing DVOL IV) is offset on BTCUSDT.BINANCE, re-hedged only outside the band — the band, not a fixed daily rebalance, is what the hypothesis asks for and is the second tunable. (4) EXIT: vol repricing (latched per cycle, with a 1-day minimum hold so a single RV wobble cannot churn the structure) or expiry-1d. If DVOL has no sample the cycle is BLOCKED, never entered on a price-only proxy — the edge IS implied-vs-realized, so no substitute mechanism is allowed. Venue choice follows the hypothesis: Deribit for the options legs (declared) and Binance USD-M for the hedge (declared); leverage stays 1.0 because the hedge notional never exceeds the straddle's own delta, so a higher leverage would be inert. One caveat stated for the record: buying gamma is the negative-carry side of a structurally positive VRP (measured mean DVOL 0.60 vs mean RV 0.49 over 5.5y), so the whole result hinges on whether the inverted-VRP subset is genuinely selective — which is exactly what the backtest is for.

Negative expectancy on a reliable sample, with the loss located precisely in the mechanism the hypothesis is about. Across 38 gated weekly cycles the LONG straddle legs lost $8,447 and only 5 of 38 cycles were profitable on the option leg, while the band delta hedge was fine (+$1,588 net of fees) — so this is not a hedging or fee artifact. The gate does not select weeks whose subsequent realized vol exceeds the implied vol paid: PF 0.663, Sharpe -1.65, expectancy -$66.59/trade, negative in all three calendar years. Contributing design flaw: a 30-day implied index compared against ~5-day trailing realized vol to buy 4-7 DTE gamma fires after vol bursts that mean-revert. Optimization cannot rescue a structure whose core leg wins 5 of 38 cycles while paying positive-VRP carry every cycle.

BTC LONG GAMMA on an INVERTED variance risk premium. Each weekly Deribit BTC cycle the strategy BUYS the front-week ATM straddle (long ATM call + long ATM put, ~7 DTE) — but only when Deribit's DVOL index (30-day forward implied vol) prints at or below iv_rv_entry_ratio x trailing rv_window-bar realized vol, i.e. only when the option market is quoting vol BELOW what BTC has actually been delivering. The resulting delta is neutralized with the BTCUSDT.BINANCE USD-M perpetual and RE-HEDGED ONLY WHEN the residual delta breaches hedge_band x per-leg size (band hedging, so the long-gamma book harvests the realized path instead of paying a fixed rebalancing tax). The structure is closed on whichever comes first: (a) VOL REPRICING — IV has climbed back to RV x (iv_rv_entry_ratio + 0.10), so the dislocation bought has corrected; or (b) the schedule's close 24h before European cash settlement (expiry-1d). With no live option leg the desired hedge is zero with a zero band, so the perp fully unwinds and the book is flat between cycles. Cross-venue by construction: options on DERIBIT, hedge + roll clock on BINANCE USD-M 6-HOUR bars. Risk per cycle is bounded by the premium paid (~1.1% of equity at the structural 0.20 sizing fraction). Exactly 3 tunables: iv_rv_entry_ratio, rv_window, hedge_band.

Mechanism implemented as described: 38 gated weekly cycles, both straddle legs entered together (38 calls / 38 puts), 27 band-hedge round trips, no underlying-only fallback.

The delta hedge is not the problem — the perp leg was mildly profitable (+$1,588 net of $306 commissions).

Genuinely differentiated: delta-neutral long gamma on an under-represented venue using the previously unused DVOL series.

Negative expectancy on a reliable sample: 103 trades / 38 cycles, profit_factor 0.663, Sharpe -1.65, expectancy -$66.59/trade, total_return -3.43%.

The gamma leg IS the loss: option legs -$8,447 with only 5 of 38 cycles profitable — the 'DVOL < trailing RV' gate has no forward selectivity.

Losses are broad: all three calendar years negative, omega 0.36, tail_ratio 0.0, 551-day max drawdown duration, avg_loss > avg_win at a 41.7% win rate.

Horizon mismatch: 30-day DVOL vs ~5-day trailing RV, then buying 4-7 DTE gamma — fires after short-horizon vol bursts that mean-revert (adverse selection for a gamma buyer).

avg_trade_return_pct 2.89% is premium-notional and not fee-comparable for a long-premium structure; the honest read is the realized $ P&L.

BtcDeribitInvertedVrpLongGammaBandHe-68aad297b4

The strategy deliberately inverted the factory's short-premium options work: buy 4-7 DTE Deribit ATM straddles only in weeks where DVOL implied vol was below trailing realized vol, and delta-hedge on a band with the Binance perp so the hedge harvests the realized path instead of paying a rebalancing tax. Mechanically it worked as designed — 38 gated cycles, both legs always entered together, 27 band-hedge round trips, and a hedge leg that was actually mildly profitable at +$1,588. But the edge itself did not exist: the long straddle legs lost $8,447 with only 5 of 38 cycles profitable, producing Sharpe -1.65, profit factor 0.663 and -3.43% total return, negative in all three calendar years. The analyst abandoned it at backtest review rather than spending optimization hours, attributing the failure to a horizon mismatch — a 30-day implied index versus ~5-day trailing realized vol — that made the gate fire after mean-reverting vol bursts while paying positive-VRP carry every cycle.

A 30-day implied-vol index compared against ~5-day trailing realized vol has no forward selectivity for 4-7 DTE gamma — it fires after short-horizon vol bursts that mean-revert, so horizon-matched IV/RV measurement is a prerequisite for any long-gamma gate.

The backtest-review gate returned 'abandon' before optimization: negative expectancy on a reliable sample with Sharpe, profit factor, total return and winning-cycle count all failing, and the loss located precisely in the mechanism being tested. Optimization, analyst and risk stages were never reached.

Buy short-dated ATM BTC straddles on Deribit only when Deribit's DVOL implied-vol index printed below trailing realized vol (an inverted variance risk premium), delta-neutralizing on a band with the BTCUSDT.BINANCE perp to harvest realized gamma.

Over 669 days of data and 103 trades across 38 gated weekly cycles it returned -3.43% with Sharpe -1.65, profit factor 0.663, a 41.7% win rate and expectancy of -$66.59 per trade. The loss sat in the option legs (-$8,447, only 5 of 38 cycles profitable) while the band delta hedge on the perp was mildly positive (+$1,588 net of $306 commissions).