ETH and SOL Daily Multi-Lookback Time-Series Momentum Long Portfolio

A long-only multi-symbol time-series momentum (TSMOM) strategy on the two safest non-BTC majors — ETHUSDT.BINANCE and SOLUSDT.BINANCE perpetual futures (both confirmed by the verification framework as 'always safe' on 1H/4H/1D). Uses DAILY bars. The strategy applies the Moskowitz-Ooi-Pedersen 2012 'Time Series Momentum' framework adapted to a 2-symbol crypto portfolio with multi-lookback voting — structurally distinct from every other multi-symbol strategy in the pipeline (which uses either price-level Donchian breakouts or cross-sectional ranking momentum). The mechanism: for each symbol independently, on every 7th daily bar (weekly rebalance day), compute three trailing total returns over lookback windows of 30, 60, and 90 daily bars. Each lookback votes +1 if its trailing return > 0, else 0. The symbol receives a score in {0, 1, 2, 3}. A symbol with score >= 2 is in 'positive momentum regime' and held long for the next week at 40% equity notional. A symbol with score < 2 is flat (not held). The portfolio can hold 0, 1, or 2 positions depending on regime — when both symbols are in positive momentum, gross long exposure is 80% (40% each); when only one, 40%; when neither, 100% cash. This hypothesis is decisively orthogonal to every confirmed failure pattern in the log AND to every strategy already in the pipeline: (1) NOT BTC (avoiding the 'BTC-2024+-microstructure-regime-decay' that killed 4 BTC strategies and is now session-confirmed); (2) NOT INTRADAY (the latest analyst declaration explicitly closes the intraday pattern class for the remainder of this session — TSMOM operates on DAILY bars with weekly rebalance, putting it in the 'daily multi-symbol trend-following' class that the analyst explicitly endorsed as productive); (3) NOT cross-sectional ranking (CrossSectionalMomentumWeeklyRotation8MajorsLong uses RELATIVE strength ranking — TSMOM uses ABSOLUTE trailing-return sign, a mathematically distinct mechanism that can hold ZERO positions during crypto-wide bear markets, providing structural drawdown protection that ranking-based CS-Mom cannot); (4) NOT a Donchian price-level breakout (MultiSymbolMajorCryptoFourHourDonchianBreakoutPortfolioLong tracks N-bar HIGH thresholds — TSMOM aggregates 30-90 days of return-sign information into a multi-lookback ensemble, completely different signal architecture); (5) NOT single-instrument trend-following (the failed ETH Daily Trend-Following used a single 100-bar lookback and produced 2.7 trades/year — TSMOM's multi-lookback ensemble + 2-symbol portfolio + weekly rebalance lifts trade frequency to ~30-60 per year); (6) NOT a pullback strategy; (7) NOT a mean-reversion strategy; (8) NOT a single-bar pattern; (9) Does NOT depend on supplementary data (pure OHLCV from the explicitly-safe ETH/SOL.BINANCE daily timeframe). Three free parameters total: lookback_short (30), lookback_med (60), lookback_long (90) — three integers chosen as classic TSMOM convention (Moskowitz-Ooi-Pedersen 2012 used 1, 3, 6, 9, 12-month lookbacks on equity-futures, so 30/60/90 days is the crypto-scaled equivalent of the shorter end of the academic specification). The vote threshold (2 of 3) is hardcoded geometric majority — not a free parameter. The position size (40% per symbol) is hardcoded conservative — not a free parameter. The rebalance frequency (7 daily bars) is hardcoded weekly — not a free parameter. This minimal parameter surface (3 integers) is the lowest in any hypothesis this session and structurally defends against the curve-fit-during-optimization failure mode that killed every BTC strategy. Expected trade frequency: each symbol's score transitions across the 2-threshold ~4-8 times per year (once per major regime shift), so 2 symbols × ~6 regime changes × 2 entry/exit events = ~24-30 portfolio events per year, well above the walk-forward viability floor while being statistically meaningful per-event because the underlying signal is a 90-day trailing-return aggregate (not a single-bar noise-prone trigger).

Implements the Moskowitz-Ooi-Pedersen 2012 TSMOM framework as a 2-symbol crypto portfolio with multi-lookback voting. (1) Absolute own-trend momentum (sign of each symbol's trailing return) is mathematically distinct from cross-sectional ranking — it can hold ZERO positions when both symbols' 30/60/90-day returns are negative, giving structural drawdown protection. (2) Uses the proven multi-symbol pattern from AbsoluteMomentumLSBasketDaily: all order management for both legs is centralized in calculate_signal (which the base only fires once the same-timeframe ETH/SOL daily bars are timestamp-aligned via _sync_extra_iids, so both legs are read contemporaneously, no stale-leg lag), while should_enter/should_exit/position_size are inert because the base's entry path manages only the primary instrument. (3) Weekly cadence is enforced by a bar counter ((count-1) % 7 == 0), acting on the first post-warmup bar then every 7th, giving ~24-30 portfolio events/year. (4) Sizing is capital-relative via get_account_equity()*0.40, well above the $10 min-notional and the ~0.10% round-trip fee given 90-day-aggregate signals. (5) min_bars_required = lookback_long+2 so all three trailing returns are valid before trading; _score returns -1 when history is insufficient so no premature trades. (6) A continuous per-bar proxy (pct change) is returned from calculate_signal so Layer-2 frozen-signal detection passes even though entries are gated to rebalance bars. Three integer free parameters (the lookbacks) keep the optimization surface minimal, defending against curve-fitting. Leverage stays 1.0 (max 80% gross, long-only) so no leverage-sizing gate applies.

Do not optimize — the +1664% headline is disqualified on three independent axes that the 3 free lookback parameters cannot fix. (1) NO RISK-ADJUSTED EDGE: Sharpe 0.46 with sharpe_ci_low -0.82 (CI straddles 0), annualized_volatility 264.8%, and max_drawdown 63.6% (CI high 98.8%) — this is levered bull-beta, not a validated edge, and DSR will reject it at optimization. (2) TWO-TRADE CONCENTRATION: return_kurtosis 833, skew 27.3, tail_ratio 0.136; the entire result rests on a couple of bull-run legs (2021-06-21 +445%, 2024-06-17 +212%; 2021 annual +810%, 2024 +252%), so it is a fat-tail lottery, not a distributed edge. (3) FALSIFIED PREMISE: the hypothesis claims absolute-momentum provides structural drawdown protection, but the record shows 2022 -55.4%, 2026 -25.8%, and a 63.6% max drawdown — no protection. Additional flags: the headline is heavily mark-to-market (end_unrealized_pct 364.7%, i.e. most of the return is an un-exited open position, not realized), trade frequency is ~7/yr (42 trades) vs the claimed 24-30/yr, and single-period magnitudes like +445% on an 80%-max-gross unlevered long warrant a PnL/sizing sanity check. Root cause is structural and not tunable: a long-only TSMOM on two highly-correlated crypto majors (ETH/SOL, benchmark_correlation 0.67) collapses into one bull-beta bet, so the academic TSMOM diversification that produces its Sharpe never materializes. If this mechanism is to be pursued it requires a premise change to a broad, low-correlation, long/short instrument universe (a Research Lead reframe), not parameter optimization of the 2-asset long-only version — abandon this arm.