EthDeribitTrendSixtyDeltaOptionRollLS

ETH Deribit Directional Trend via 30-45D ~0.6-Delta Options (Long-Short, Calls in Uptrend / Puts in Downtrend, Trend-Timed Roll) — Convexity Expression of a Confirmed Daily+Weekly Trend, NOT Vol Selling

A LONG-SHORT, low-turnover DIRECTIONAL trend strategy that expresses a confirmed medium-term ETH trend by BUYING Deribit options rather than trading a perp: buy a moderately in-the-money (delta ~0.60) CALL with 30-45 days to expiry when the trend is up, buy a delta ~0.60 PUT when the trend is down, hold while the trend persists, and roll to a fresh 30-45D ~0.60-delta contract when time-to-expiry falls below ~14 days. This is a deliberate, quota-driven move into the near-empty options bucket (BINANCE_OPTIONS 0.0%, DERIBIT 5.8%; combined ~5.9% vs the >=15% target) and into long_short (14.4% vs push). It is NOT a member of any dead class: it is the OPPOSITE of the VRP / short-strangle / covered-call vol-SELLING graveyard (no_edge/risk_reject) — we BUY convexity, so our worst case is capped at premium and gap risk works FOR us. It is also distinct from the existing EthDeribitTrendLongConvexityCheapIvRoll pipeline item: that one is LONG-ONLY and gated on CHEAP IV rank (a vol-timing bet); this one has NO IV/vol-timing gate at all — the edge is pure directional trend capture, taken in BOTH directions, and it uses ~0.60-delta (not far-OTM) contracts specifically so option legs are liquid and actually trade (avoiding the options_no_option_trades degeneration that just killed the Deribit covered-call) and so theta dependence is low. The trend signal is computed on the liquid ETHUSDT.BINANCE perp (cross-venue: signal on BINANCE, execution on DERIBIT), so the trigger never depends on illiquid option-chain bars.

Implements the hypothesis literally. (1) LONG-only-in-premium, both directions: legs are always BOUGHT (leg_sides forced to BUY in on_start), calls in confirmed uptrends and puts in confirmed downtrends — the opposite of the short-vol/VRP graveyard, with loss capped at the premium paid and gaps working for us. (2) ~0.60 delta is coded as a moneyness offset: a 30-45D ETH option struck ~3% in-the-money prices at N(d1)~0.60 under any plausible IV (60% IV, 36 DTE => ln(S/K)~0.030), so legs are {C: -0.03, P: +0.03}. Moderately-ITM strikes are the liquid part of the chain, which is what avoids the options_no_option_trades degeneration and keeps theta dependence low relative to delta. (3) 30-45D with a <14D roll: monthly Deribit expiries, dte_target=40 for entry, roll_out_dte_days=14 for the close, giving 26-40 DTE holds and disjoint, near-seamless periods (~4 flat days per roll). (4) The trigger never touches option-chain liquidity: the trend is measured entirely on the ETHUSDT.BINANCE perp (primary instrument, 1-DAY bars, years of history, also the roll clock); all ORDERS go to the DERIBIT option legs via _submit_entry_instrument. (5) NO IV/vol gate exists in the file, which is the stated distinction from the cheap-IV-rank long-call pipeline item. The chain is resolved by the framework (parameters['rolling_options'] -> backtest_agent.expand_rolling_options_into_config, also run by the Layer-3 sandbox) and driven by RollingOptionsStrategy; on_bar is overridden only to gate entry on trend direction and to flatten when the trend stops confirming. There is deliberately NO smoke path and NO underlying fallback — if the chain is unavailable the run fails loud rather than reporting delta-1 perp metrics. ETH-PERPETUAL.DERIBIT is carried as a non-traded reference leg: Deribit quotes option premium in COIN, and the Deribit ETH price is the correct denominator for converting it to USD before sizing (falls back to the Binance close if that leg has no bar). leverage stays 1.0 — a bought option is already convex and no margin is used, so the code never references leverage in sizing. Sizing is capital-relative (premium_frac of get_account_equity()), and fees (0.03%/side on Deribit) are immaterial against option premia that move tens of percent.

analyzing_stuck_3_timeouts

Long-short DIRECTIONAL trend strategy on ETH expressed with BOUGHT Deribit options instead of a perp. A vol-normalised daily+weekly trend confluence is computed on ETHUSDT.BINANCE 1-DAY bars: z_daily (25-day log return / sigma*sqrt(N)) and z_weekly (10-week log return / sigma*sqrt(N)); the returned signal is their mean. When both legs are positive and the score clears entry_threshold the strategy BUYS the live monthly Deribit ~0.60-delta CALL (strike ~3% ITM); when both are negative it BUYS the ~0.60-delta PUT (strike ~3% above spot). Each contract is opened ~40 days before its expiry and closed/rolled 14 days before expiry, so every position is a fresh 26-40 DTE option. The position is held while the trend persists (hysteresis band = 40% of the entry threshold) and closed on a flip, on trend fade, or at the roll deadline; a flip re-enters the opposite leg on the same bar. Size = premium_frac (5%) of equity spent on premium, which is also the maximum loss of the position. No IV / vol-regime gate anywhere: this buys convexity to express direction, it does not trade volatility.

If the base uses its own min_bars for option-period activation, ensure the override does not delay or advance period entry relative to the schedule.

min_bars_required(self) overrides a framework method the base RollingOptionsStrategy may rely on for its own warm-up/period gating. Here it is a deliberate signal warm-up (weekly_lookback_days + 2 daily closes) and is internally consistent, but confirm the base does not need its own value for chain-period readiness.

No change needed for correctness; consider a comment clarifying SELL here maps to buying the put leg.

should_enter returns OrderSide.SELL for the downtrend case. This is only a signal token consumed by the Layer-2 synthetic harness; real trading in on_bar always BUYs the matching PUT leg via _enter_long_option (OrderSide.BUY). No naked short is ever submitted. Flagged only so it is not mistaken for spot/short-open logic.

Sandbox is a trending-window smoke test (Sharpe 1.71, +55%, 51 trades all registered as long option legs) and cannot confirm the PUT side actually gets selected/traded in real downtrends — short_trades=0 is expected (bought puts are long the put instrument) but leaves the downtrend leg unverified on live history. Over full 3y history the analyst should confirm both leg kinds trade and that per-position premium capture nets out across roll cycles; the ~40-DTE monthly roll cadence yields a moderate trade count (~17/yr in sandbox), measurable but on the lower side.

Trades faithfully implement the hypothesis: 22 calls bought in confirmed uptrends (e.g. ETH-29MAR24-2800-C +$18.3k into the Mar-2024 rally, ETH-29NOV24-2600-C +$10.0k into the post-election surge) and 29 puts bought in confirmed downtrends (ETH-28MAR25-2800-P +$12.2k, ETH-25APR25-1900-P +$9.2k during the early-2025 crash). QA's put-side concern is RESOLVED — both leg kinds trade and the put leg profits in real downtrends over full history.

Real Deribit option legs actually trade (no smoke path, no far-OTM options_no_option_trades degeneration) — the ~0.60-delta moneyness choice worked as intended.

Strongly positive, fee-viable realized edge: expectancy +$2,168/trade, PF 4.06, win-rate 57% with a ~3:1 win/loss payoff — the genuine long-convexity profile (bounded losses at premium, fat right tail). This is the OPPOSITE of the negative-per-notional credit-spread graveyard.

Low, capped drawdown (max DD 2.09%, premium risk capped at 5%/trade) and positive in every calendar year (2023 +3.5%, 2024 +22.8%, 2025 +17.3%, 2026 +4.1%); realized-trades Sharpe 1.71, Sortino 1.95.

Differentiated: fills the near-empty options / long_short buckets with a long-convexity directional mechanism, not a member of the dead vol-selling or defined-risk-spread classes.

Moderate trade count (51 over ~3y, ~17/yr) is on the lower side for a 3-window walk-forward — IS windows will hold ~15-17 trades and the holdout ~10, so parameter fits risk noise sensitivity.

Edge is right-tail-driven — a handful of large winners (+$18k, +$12k, +$12k, +$11k, +$11k) dominate the return; which window a big winner falls in could swing WF-OOS Sharpe.

avg_trade_return_pct (46.77%) is premium-notional and not directly fee-comparable; judge on realized USD P&L (which is strongly positive) rather than that number.

Strategy trailed the equal-weight/buy-hold basket for much of the sample and only pulled ahead recently — its value is the risk-adjusted (low-DD, always-positive-year) profile, not raw return.

realized-trades basis; CI [0.99, 2.35]

22 calls + 29 puts; both legs trade