EthDeribitWeeklyTrendRiskReversalLS

ETH Deribit Trend-Aligned Weekly Risk Reversal, Long-Short — Financed Convex Directional Exposure That Sidesteps the Positive Crypto VRP (Long the ~25-Delta Wing in the Trend Direction, SHORT the ~25-Delta Opposite Wing to Pay for It, Direction Set by the Daily ETH Trend, Rolled Weekly, 3-Parameter)

A LONG-SHORT, single-underlying options structure on ETH Deribit weeklies that expresses the ONE mechanism this factory has repeatedly promoted — daily directional trend on a crypto major — but through a delta-positive/vega-neutral RISK REVERSAL rather than a linear perp or a long-premium call. Each Wednesday roll the strategy reads the sign of a slow daily ETH trend (50-day EMA slope / price-vs-EMA). In an UPTREND it BUYS the ~25-delta weekly CALL and SELLS the ~25-delta weekly PUT (~7 DTE); in a DOWNTREND it BUYS the ~25-delta PUT and SELLS the ~25-delta CALL. The short wing FINANCES the long wing, so the position is near zero-cost, combined delta ~±0.4 in the trend direction, net vega ~flat. This deliberately fixes the abandoned EthDeribitTrendSixtyDeltaOptionRollLS (long-premium 60-delta, ~12 biweekly trades, structurally short the positive VRP): here the short wing HARVESTS that VRP to offset the long wing's decay, and the cadence is WEEKLY producing ~100 rolls over ~2yr of ETH-Deribit weekly history. Fills three under-represented buckets: options venue (0.1% vs ≥15% target), Deribit (5.9%), long-short (14.5%).

Implements the hypothesis exactly: a single-underlying, long-short options structure on ETH Deribit weeklies that expresses daily directional trend through a delta-positive/vega-neutral risk reversal instead of a linear perp or a long-premium call. The trend engine is close/EMA(ema_days)-1 on the Deribit ETH perp's daily bars (continuous, returned every bar by calculate_signal, thresholded in should_enter), the wings are the ~25-delta strikes (offset +-0.06 is the ~25-delta moneyness for a 7-DTE ETH option at 55-75% IV, and sits well inside the liquid part of the chain), and the cadence is weekly with ~7 DTE. The short wing harvests the positive crypto VRP that made the abandoned long-premium 60-delta version structurally short carry. Direction is locked once per expiry at the entry window so both wings always agree and an open structure is never flipped mid-week; a week is skipped entirely unless BOTH a call and a put leg resolved and are quoting, so the book can never hold a naked single wing. There is no underlying/smoke fallback -- if the chain is unavailable nothing trades. Venue is DERIBIT (deepest ETH weekly chain, 0.03%/side) with leverage 1.0, which options venues cap at anyway, so sizing deliberately does not reference leverage. One caveat the Research Lead should weigh: the hypothesis asks for ~100 rolls, but each option bar type costs a roughly fixed ~9-10s catalog query against the 600s Layer-3 options budget, so rolling_options.lookback_days is capped at 154 (22 weekly cycles, 44 legs) -- an infrastructure ceiling on weekly multi-leg chains, not a tuning choice.

data_wall: unmeasurable / insufficient statistical power. This is NOT an expectancy or fee failure — the book is net-positive (PF 1.21, expectancy +$29.9/structure, +120% total_return) and the -4.19% avg_trade_return_pct is the wing-notional artifact QA flagged, not a real sub-cost edge. The problem is that the strategy produced only ~20 risk-reversal structures (40 legs, effective_daily_obs 20) over ~148 days, and the sample CANNOT be enlarged: rolling_options.lookback_days is hard-fixed at 154 and is not in the tunable set, and raising it blows the ~600s options wall-clock budget (~9-10s/leg, developer-flagged infrastructure ceiling). The metrics are non-diagnostic — Sharpe 0.646 inside a CI of [-2.75, 3.31] that straddles zero, PSR 0.56 barely above no-skill. A 3-phase run would hold ~6-7 structures per walk-forward window and a holdout below the hard 10-trade floor, curve-fitting noise. This is the same wall that abandoned the near-identical siblings EthDeribitTrendLongConvexityCheapIvRoll (12 trades) and BtcDeribitIvTermStructure (12 trades). Not worth 2 hours of optimization; revive if the ETH-Deribit weekly option series and the per-leg query cost ever permit a testable (>=30, ideally ~100) structure sample. abandon_class = unmeasurable.

Trend-aligned ~25-delta weekly RISK REVERSAL on Deribit ETH options (long one wing, short the opposite wing, 1:1). Each weekly roll the strategy reads the sign of a slow daily ETH trend (close vs a 50-day EMA of ETH-PERPETUAL.DERIBIT). In an uptrend it BUYS the ~6%-OTM (~25-delta) front-week CALL and SELLS the ~6%-OTM PUT; in a downtrend it buys the PUT and sells the CALL; inside a small deadband it sits out the week. The short wing finances the long wing, so the package is near zero-cost with combined delta ~+-0.4 in the trend direction and roughly flat net vega. Structure, roll and exit are owned by RollingOptionsStrategy: the historical weekly chain is resolved from parameters['rolling_options'], both wings are opened inside the week's entry window and closed 24h before European cash settlement, then rolled into the next week. Sizing is equity-relative (equity * risk_frac / spot contracts per wing, identical on both wings) and re-derived at each roll. Three tunables: ema_days, trend_deadband, risk_frac.

If the analyst wants statistical power, raise lookback_days for the full (non-sandbox) backtest, accepting the sandbox-budget tradeoff, or expose it as a bounded parameter. As-is, treat the run as a low-N smoke of the mechanism.

rolling_options.lookback_days=154 caps the strategy to ~22 weekly rolls over full history despite ~2yr of ETH-Deribit weekly option data; the hypothesis targets ~100 rolls. lookback_days is not exposed to the optimizer, so the sample cannot grow during optimization.

Acceptable given the framework resolves strikes by moneyset offset and the code documents the approximation. No change required; noted so the ~25-delta claim is read as approximate.

Wings are placed at a FIXED +-6% moneyness offset (rolling_options offset 0.06), not a solved ~25-delta strike. The hypothesis names ~25-delta; 6% OTM at 7-DTE/55-75% IV is an approximation of that, and true delta drifts with realized IV.

Intentional directional structure with documented tail; not a defect. Risk Officer should note the naked-short tail when sizing risk_frac in optimization.

The short wing of the risk reversal is a NAKED short option (unbounded on the adverse side). This is inherent to a risk reversal and is documented (worst-week ~ (move - wing_offset)*risk_frac of equity; ~6% at a 30% adverse gap, default risk_frac 0.25).

Two advisory concerns for the analyst at backtest_review, neither a correctness fault. (1) Sample-size ceiling: rolling_options.lookback_days is hard-fixed at 154 (~22 weekly cycles = ~20 risk-reversal structures / ~40 legs) and is NOT in the tunable set (_param_bounds only exposes ema_days/trend_deadband/risk_frac), so the FULL backtest — not just the sandbox — is capped at ~22 independent directional bets even though ETH-Deribit weekly option history spans ~2yr (~100 possible rolls). The developer explicitly flagged this as an infrastructure ceiling (~9-10s/leg vs the 600s options budget). ~20 structures is thin for a 3-window walk-forward; the near-identical sibling EthDeribitTrendLongConvexityCheapIvRoll was abandoned at backtest_review as unmeasurable on 12 trades (cf. L26). The data EXISTS (so this is not a not_measurable/data verdict), but the analyst should weigh whether ~22 rolls can support a promote decision, or whether lookback_days must be raised first. (2) Sandbox avg_trade_return_pct is -4.19% while total_return is +120%, profit_factor 1.21 and USD expectancy +$29.9/trade — the negative per-notional figure is largely an artifact of measuring option-structure PnL against the ~24.9% underlying notional per wing (short leg is naked, so notional denominator is large), compounded by real theta/fee drag. Per QA rules this is a performance question for the backtest, not a QA block; noted so it is not misread as a hard fee-viability fail at Layer 4.

Genuinely differentiated structure: a delta-positive/vega-neutral 25-delta risk reversal that harvests the positive crypto VRP on the short wing to finance the long wing — a real fix to the abandoned long-premium 60-delta sibling.

Not a losing or fee-fragile book: profit_factor 1.21, positive expectancy +$29.9/structure, total_return +120%. The avg_trade_return_pct -4.19% is the wing-notional measurement artifact QA identified (short leg's large naked notional as denominator), NOT a real edge-below-cost fail — commission is ~0% of gross.

Trades faithfully implement the hypothesis: 20 calls / 20 puts, avg_dte 7.0, avg_iv 0.62, near-zero net delta at entry, held to weekly roll — a coherent trend-aligned risk reversal.

Statistically non-diagnostic sample: only ~20 risk-reversal structures (40 legs). Sharpe 0.646 sits inside a bootstrap CI of [-2.75, 3.31] that hugely straddles zero; PSR 0.56 is barely above no-skill; effective_daily_obs = 20.

The sample cannot be enlarged: rolling_options.lookback_days is hard-fixed at 154 and is NOT tunable, and raising it blows the ~600s options wall-clock budget (~9-10s/leg). Optimization cannot change it.

A 3-window walk-forward would hold ~6-7 structures per window and a holdout of a handful — below the hard 10-trade holdout floor. Two hours of optimization would only curve-fit noise.

ETH-Deribit weekly option history itself spans only ~2yr and is largely one regime, so even removing the infra cap the mechanism cannot be validated across regimes.

EthDeribitWeeklyTrendRiskReversalLS-d28312415a

A weekly multi-leg options chain is capped by the sandbox per-leg query budget, so even an economically sound structure can be unmeasurable.

Abandoned at backtest-review for insufficient statistical power — a non-diagnostic ~20-structure sample that cannot be enlarged (lookback_days hard-fixed at 154 against the ~600s options budget).

Express the factory's daily ETH trend edge through a weekly ~25-delta risk reversal on Deribit ETH options — buying the wing in the trend direction and selling the opposite wing to finance it, harvesting the positive crypto variance risk premium while keeping convex directional exposure.

Over ~148 days it produced 40 trades (~20 risk-reversal structures) with a profit factor of 1.21, +$29.9 expectancy per structure, and +120% total return; Sharpe 0.65 but with a CI of [-2.75, 3.31] and PSR 0.56.