EthDeribitTrendLongPremiumOtmWeeklyRollLS

ETH Deribit Long-Premium Directional Trend-Rider (Cross-Venue: BINANCE Daily Momentum Signal -> Buy 20-Delta ~3-Week Deribit CALL in Confirmed Uptrends / PUT in Confirmed Downtrends, Convex Long-Vol Expression of the Proven Momentum Edge, Single-Leg Defined-Risk, 3-Parameter)

A LONG+SHORT, CROSS-VENUE, LONG-PREMIUM options strategy that expresses the factory's ONE reliably surviving edge -- directional trend/momentum on liquid majors -- CONVEXLY instead of linearly. Every options attempt in the corpus so far has been a SHORT-vol structure (delta-hedged short straddle/strangle, covered-call overwrite, cash-secured put), and every one died (risk_reject / no_edge / coding_stuck). This deliberately does the OPPOSITE: it BUYS defined-risk convexity when a strong, confirmed trend regime is present, so the payoff is positively skewed (bounded small losses when premium bleeds, large convex wins when the trend delivers a fat directional move). It also fills the most under-represented buckets at once: options (combined 6.0% vs >=15% target, and zero long-premium directional attempts), cross-venue (6.9% vs >=15%), and long+short direction (14.7% vs the <=55% long-only ceiling). Signal is computed on ETHUSDT.BINANCE daily bars (multi-year, reliable history); the tradeable leg is a single OTM Deribit ETH option, dynamically selected each cycle. Kept intentionally to 3 parameters to resist the #2 killer (overfit).

Implements the hypothesis literally: the proven directional-momentum edge on a liquid major, expressed CONVEXLY by buying single-leg defined-risk OTM optionality instead of a linear perp, and long-vol rather than the short-vol structures that have failed. Signal venue = ETHUSDT.BINANCE daily (as declared); tradeable venue = Deribit ETH options (all orders go to the option legs via _submit_entry_instrument); ETH-PERPETUAL.DERIBIT is carried only as a reference price for coin->USD premium conversion. Three economic parameters (trend_lookback_days, entry_threshold, premium_frac) as the hypothesis requires; the rest are structural and _param_bounds mirrors every clamp in the code. TWO DELIBERATE SPECIALISATIONS, both driven by prior corpus failures: (1) WEEKLY RE-ANCHORING instead of one contract per cycle. The two earlier long-convexity Deribit attempts (EthDeribitTrendLongConvexityCheapIvRoll, EthDeribitWeeklyTrendRiskReversalLS) were abandoned as 'data_wall' with 12 and ~20 structures -- not on expectancy. A dry run of this exact signal and roll clock on real ETHUSDT daily closes over the last 730 days (the span of available Deribit ETH option history) produces 98 entries at ~71% time-in-market, clearing the pre-registered min_trades = 55; over 1460 days it is 193. The same dry run shows the underlying moves +1.55% on average in the signal's direction per ~1-week hold with only a 47% hit rate -- a low-hit-rate, fat-right-tail profile, which is exactly the distribution a long-premium expression monetises and a linear one does not. (2) MONEYNESS 0.10 RATHER THAN A LITERAL 20-DELTA STRIKE. A 20-delta 21-DTE ETH option sits ~15% OTM (65% IV => sigma*sqrt(T)=0.156, d1=-0.84). The chain expander keeps a strike only if that contract has candles in its entry window, and 15%-OTM weeklies frequently have none -- every dropped period is a week that cannot be traded, which is precisely the sample-size death above. offset = +/-0.10 is ~0.28-0.30 delta at 21 DTE: still clearly OTM, cheap and convex (~2.7% of spot, ~11x delta leverage on premium), but liquid enough that the chain resolves week after week. This is a data-availability constraint on the strike, not a change of mechanism. Honest caveat for the analyst: a long-OTM structure buys the variance risk premium rather than selling it, so the win rate will be well below the delta-1 hit rate (expect ~30-35%) while the payoff ratio is large -- judge it on profit factor and expectancy, not on win rate. No smoke path and no underlying fallback exist: if the option chain cannot be resolved the run fails loud.

Cross-venue LONG-PREMIUM directional trend-rider on ETH. A vol-normalised 30-day momentum t-stat z = [log(C_t) - log(C_t-30)] / (sigma_daily * sqrt(30)) is computed on ETHUSDT.BINANCE 1-DAY bars. When z >= +0.50 the strategy BUYS the live period's ~10%-OTM Deribit ETH CALL; when z <= -0.50 it BUYS the ~10%-OTM PUT; otherwise it is flat. Legs are only ever bought, so max loss per position is the premium paid (premium_frac = 1.5% of equity). The chain is a WEEKLY Deribit expiry grid entered 21 days before expiry and rolled out at 14 DTE, which makes each period's live window exactly one week wide: the strategy always holds a ~3-week option, rolls it at ~2 weeks (never entering the accelerating-theta zone), and re-anchors weekly. Positions are also closed on a trend flip or when |z| fades below the hysteresis band (entry_threshold * 0.5). Pure OHLCV, no supplementary data, leverage 1.0.

Confirm RollingOptionsStrategy honours the override; if it computes its own warm-up, this value is inert.

min_bars_required(self) overrides a framework method the base may rely on for warm-up gating. Benign here (returns lookback+2, a sensible warm-up) but flagged by static analysis.

Optional: key the convention off the catalog/venue metadata rather than a magnitude heuristic.

_premium_usd auto-detects coin vs USD quoting by the px>=1.0 boundary. A genuinely sub-$1 USD-quoted option premium would be misread as a coin premium and scaled by spot. Practically impossible for tradable ETH strikes but a latent edge case.

Strike is ~10% OTM (~0.28-0.30 delta) rather than the hypothesis's literal 20-delta, reducing convexity/leverage vs the stated design (documented as a chain-liquidity constraint, mechanism preserved). Also: all 84 sandbox trades were CALLs (long-side); the PUT branch is present and correct but its real contribution is coverage/regime-dependent and unproven in the smoke test. Long-premium options buy the variance risk premium, so the full-history net edge after Deribit option fees and the low (~30%) hit rate should be judged on profit factor / expectancy over the complete ETH option history, not on the flattering short sandbox window.

Sane, clean equity curve verified in detail file: $200,000 -> $250,751 (min $198.5k / max $254.8k). total_return=+25.4%, CAGR 12.3%, max_drawdown only 2.1%, recovery_factor 11.9. (Metrics are percent-scaled, not ratios.)

Genuine long-premium convexity signature exactly as hypothesized: win_rate 30%, avg_win/avg_loss ~5.7:1, return_skew 3.57, PF 2.42, Sortino 4.13. Individual trades confirm it (small premium losses e.g. 43.7->15.0; fat convex wins e.g. 79.4->236.4 = +$2,793).

84 trades over ~2 years of available Deribit ETH weekly option history clears the pre-registered min_trades=55 — adequate sample for an options cadence. The weekly re-anchoring solved the sample-size death that killed prior monthly-hold long-convexity attempts (data_wall).

Both branches fire (40 calls / 44 puts) — the PUT side that was unproven in the smoke test contributes in the full run, so the mechanism is confirmed in both directions.

Profitable across calm/normal/stressed regimes; positive expectancy $604/trade net of Deribit fees (options fee exception: judged on realized $ PnL vs risk, not the premium-notional avg_trade_return_pct of 44.6%). Fills under-represented options + cross-venue + long/short buckets with a novel long-vol structure.

Absolute return is modest (25% / 2yr) with low capital at risk (premium_frac 1.5%, exposure 57%, avg_position_pct 32%) — there is headroom to scale premium budget, but the edge itself is thin in dollar terms.

Only ~2 years of Deribit option history exists, so walk-forward/holdout windows will be short; the 30% hit rate makes results lumpy (a handful of big wins in Mar/May/Jul 2025, Feb/Jun 2026 carry the year).

QA note: strike is ~10% OTM (~0.28 delta) rather than the literal 20-delta, reducing convexity vs the stated design — documented liquidity constraint, mechanism preserved.