DualMomentumIntersectionNetDirectionalBasketLS

Dual-Momentum Long-Short Basket Across Majors (Relative-Strength Rank ∩ Absolute-Trend Filter, Binance USD-M Futures, Pure OHLCV, Weekly Rebalance)

A LONG-SHORT, MULTI-INSTRUMENT DUAL-MOMENTUM basket across ~10 confirmed-full-history Binance USD-M major perpetuals, combining the two momentum signals this session validated into their documented most-robust INTERSECTION. It holds a name only where RELATIVE strength (cross-sectional rank) and ABSOLUTE trend (own price trend) AGREE: LONG the top-K relative-momentum names that ALSO have positive absolute momentum; SHORT the bottom-K relative-momentum names that ALSO have negative absolute momentum. This is genuinely distinct from the two pure momentum baskets already in the portfolio: vs the CROSS-SECTIONAL (relative-only, dollar-neutral) basket it ADDS the absolute-trend filter (regime/crash protection that pure relative momentum lacks); vs the ABSOLUTE (own-trend-only) basket it ADDS relative-strength concentration into the top/bottom-K (conviction the pure absolute version lacks by holding every name). EVERY design choice respects this session's binding lessons: (1) MOMENTUM direction -- the only validated edge (cross-sectional momentum basket positive and in optimization; every fade anti-predictive). (2) FUTURES venue -- SPOT strategies repeatedly hit a recurring sizing/accounting bug this session (>100% position, impossible single-day returns); USD-M futures size cleanly as discrete trades, and allow the short leg. (3) NO funding/supplementary data -- funding_rates ends ~2024-12 (empty 2025-26 holdout) and all eight supplementary-stream-primary attempts died on data coverage; this is PURE OHLCV. (4) DIVERSIFIED BASKET, not single-asset -- single-asset momentum/funding edges do NOT transfer to higher-beta alts (funding-trend died on SOL/BNB); the basket form is what makes momentum robust. (5) DUAL FILTER = crash protection -- the absolute-trend gate means in a broad bull more longs qualify (net-long tilt) and in a broad bear more shorts qualify (net-short tilt), the regime awareness pure dollar-neutral cross-sectional momentum lacks. (6) DENSE firing -- weekly rotation across ~10 names is a distributed trade stream (no rare-conjunction trade-starvation, the Bollinger-squeeze death). (7) LET-TRENDS-RUN exits, not small-TP/big-SL (the losing-by-construction flush-bounce payoff). Intentionally minimal: 3 tunables (momentum_lookback, abs_mom_period, top_k).

Combines the two momentum signals validated this session into their documented most-robust intersection, using the proven centralized delta-to-target basket architecture. Relative strength is the cross-sectional momentum_lookback return rank; absolute momentum is the abs_mom_period own-trend return; a name is held only where they agree -- long the top_k relative names with positive absolute momentum, short the bottom_k with negative absolute momentum. This is distinct from the portfolio's pure cross-sectional basket (adds the absolute-trend crash filter) and pure absolute basket (adds top/bottom-K relative concentration). The intersection self-tilts the book net-directional: in a broad bull most short candidates fail the negative-absolute gate so the book runs net-long, and vice versa in a bear -- the regime awareness pure dollar-neutral cross-sectional momentum lacks -- and because each QUALIFYING leg is sized at a fixed equity*gross_exposure/(2*top_k), gross = (#qualifying/2*top_k)*gross_exposure stays <=gross_exposure (90%) and runs lighter when few names qualify. Exposure is bounded as required: each rebalance recomputes fresh targets and moves every leg to its signed target via a single netting delta order (the bounded full-flatten-and-reopen end-state), with an aggregate gross cap backstop, structurally preventing the uncappable held-winners balloon. All order management is centralized in calculate_signal, fired by the base only after the cross-leg alignment barrier confirms every same-daily-timeframe leg is contemporaneous; should_enter/should_exit/position_size are inert. min_bars_required = max(momentum_lookback, abs_mom_period)+2 and a bar counter gate the first rebalance; ranking waits internally until names have enough history (legs lacking history target 0 and are flattened), so no trade fires on partial data. The tanh log-return proxy is a continuous per-bar signal so Layer-2 frozen-signal detection passes while trades stay schedule-gated. Three edge tunables keep the optimization surface small; all 10 constituents are 2020-listed liquid majors with full daily history so no leg risks data_unavailable. leverage stays 1.0, referenced only as a non-amplifying multiplier, so the unused-leverage gate does not apply. Pure OHLCV avoids the funding/supplementary/tick/options walls. If the multi-instrument daily-return aggregation artifact flagged elsewhere this session recurs, this strategy's own gross is correctly capped, so that is the harness accounting bug to escalate, not a strategy sizing error.

Overfit with negative forward edge AND a degenerate/blown-up optimized config. Walk-forward is_overfitted=TRUE: avg in-sample Sharpe 1.066 collapses to avg OOS -0.892 with two of three OOS windows deeply negative (-1.06, -3.36) — the strategy loses money out-of-sample. The multiple-testing gates concur: deflated_sharpe=0.0 (vs 0.95), is_significant=false, PBO=0.6799 (>0.5), sharpe_ci_low=-0.3634 straddling 0, optimized Sharpe below the expected-max luck bar 1.696. The holdout 'pass' is spurious (ratio 0 computed against a negative WF-OOS). Decisively, the optimizer DESTABILIZED the strategy: it pushed gross_exposure to 1.087 and max_gross to 1.004 (above the design's 1.0 cap) while concentrating to top_k=2, producing a degenerate book — avg_position_pct 156.96% (levered beyond equity), impact_cost_pct 104.1% (market impact exceeds gross PnL), annualized_volatility 3306%, return_kurtosis 1580, profit_factor 0.9998 (<1), expectancy -0.82 (negative), CAGR -100%, recovery_factor -1.0, and physically-impossible single-day returns (+3329% on 2023-12-16, +8776% on 2024-11-30) that are over-leverage artifacts, not edge. capacity_usd is just $92K (toy scale). Not iterate (attempt 1 of 2): sensitivity is flat with 0 cliffs (no robust region to tune toward) yet the walk-forward is overfit with negative average OOS, and the optimizer reliably concentrates and levers the intersection signal into a blow-up — a second best-of-225 sweep re-finds the same over-levered top_k=2 degenerate config. Not revise_hypothesis: the net-directional momentum-basket structure is already validated on the promoted Absolute Momentum sibling; this dual-momentum INTERSECTION variant fails deflation and blows up on the same 10-major universe (variant-selectivity), not a sound mechanism stranded on a dead target. FAILURE PATTERN: intersecting relative-rank and absolute-trend momentum into a top-K=concentrated net-directional basket is fragile under optimization — the small qualifying set (top_k=2) plus an optimizer that levers gross above the design cap produces an over-concentrated, over-levered book that blows up (avg position 157% of equity, impact >100% of PnL, impossible MTM spikes, negative profit factor and CAGR -100%) while failing every forward gate (is_overfitted, avg OOS -0.89, DSR 0, PBO 0.68). Concentration removes the diversification that made the always-on Absolute basket promotable; the intersection filter thins the book into instability rather than improving it.

Long-short dual-momentum basket across 10 full-history Binance USD-M majors (BTC, ETH, SOL, BNB, XRP, ADA, DOGE, AVAX, LINK, LTC), daily, weekly rebalance. It longs the top-3 cross-sectional 60-day momentum names that also have positive 90-day absolute momentum, and shorts the bottom-3 that also have negative absolute momentum (the intersection of relative-rank and absolute-trend). Each qualifying leg is sized equal-weight at gross_exposure/(2*top_k), so the absolute-trend gate makes the book net-directional (net-long in bulls, net-short in bears) with gross <=90%. Every rebalance moves each leg to its fresh signed target via a single netting delta order with an aggregate gross-cap backstop. Pure OHLCV, futures, leverage 1.0. Three tunables: momentum_lookback, abs_mom_period, top_k.

Strongest basket consistency: profit_factor 1.180, positive in 5 of 7 years (2021/2022/2024/2025/2026), win_rate 48.2%, max_drawdown 35.7% (lowest of the family)

Crash-protection thesis VALIDATED: +18% in the 2022 bear (vs Donchian -30%, dual-TF -24%) — the absolute-trend gate genuinely tilted net-short and protected

Less 2021-concentrated (2021 +130%, not +346%) with strong non-mania years 2024 (+49%) / 2025 (+55%); least MTM-inflated (end_unrealized_pct 116.9%)

Dual-momentum intersection is the documented most-robust momentum construction; validated diversified net-directional structure, deep capacity ($124M), low fees (1.35%)

Modest Sharpe 0.447 with CI straddling 0 (sharpe_ci_low -0.30) — robustness must be proven by deflated Sharpe

One severe down year (2023 -49.8%); return_kurtosis 121 indicates some outlier dependence

Profit factor 1.180 still below the 1.2 promotion guideline

5 of 7

majority

no single-leg domination

Sensitivity grid clean (0 cliffs) at the base config; sound diversification rationale (relative ∩ absolute momentum intersection)

Base (pre-optimization) backtest was coherent (Sharpe 0.45, max_dd 35.7%) before the optimizer destabilized it

Walk-forward is_overfitted=TRUE: avg IS 1.066 collapses to avg OOS -0.892, with two of three OOS windows deeply negative (-1.06, -3.36)

FAILS deflated Sharpe: DSR=0.0 (vs 0.95), is_significant=false, PBO=0.6799 (>0.5), sharpe_ci_low=-0.3634 straddling 0; optimized Sharpe below the expected-max luck bar 1.696

Optimized config is DEGENERATE/BLOWN-UP: optimizer pushed gross_exposure 1.087 and max_gross 1.004 above 1.0 with top_k=2 → avg_position_pct 156.96%, impact_cost_pct 104.1%, annualized_volatility 3306%, return_kurtosis 1580, profit_factor 0.9998 (<1), expectancy -0.82, CAGR -100%, recovery_factor -1.0

Physically-impossible single-day returns (+3329% on 2023-12-16, +8776% on 2024-11-30) — artifacts of over-levered over-concentration, not edge

Holdout 'pass' is spurious (ratio 0 vs negative WF-OOS); capacity_usd just $92K (toy scale)

DualMomentumIntersectionNetDirectionalBasketLS combined this session's two validated momentum signals into their documented most-robust intersection, holding names only where relative strength and absolute trend agreed, and its base backtest was genuinely the family's most consistent (Sharpe 0.447, profit factor 1.18, lowest drawdown 35.7%, positive in 5 of 7 years, and real +18% crash protection in 2022). The analyst gated it to optimization, but the robustness suite was damning: deflated Sharpe 0.0, PBO 0.68, an is_overfitted walk-forward whose average OOS Sharpe was -0.892 with two deeply negative windows. Worse, the optimizer destabilized the strategy entirely — pushing gross above the 1.0 cap and concentrating to top_k=2 produced a degenerate book with 157%-of-equity positions, market impact exceeding PnL, profit factor below 1, CAGR -100%, and physically-impossible MTM spikes. After one iteration it was abandoned: concentration thinned the book into instability rather than improving it, leaving the promoted Absolute Momentum basket as the configuration where this structure still clears the bar.

Intersecting relative-rank and absolute-trend momentum into a concentrated top-K net-directional book is fragile under optimization — thinning the basket removes the diversification that made the always-on Absolute Momentum sibling promotable, and an optimizer that levers gross past the cap turns the small qualifying set into an over-levered blow-up.

After the backtest-review 'optimize' verdict, the analyst abandoned it at the post-optimization ANALYZING stage: it failed deflated Sharpe (0.0 vs 0.95), was is_overfitted=true with negative average OOS, and the optimizer levered gross above the design's 1.0 cap while concentrating to top_k=2, blowing the book up into impossible single-day returns (+3329%, +8776%) and a toy $92K capacity.

A long-short dual-momentum basket across ~10 Binance USD-M majors that held a name only where relative-strength rank and absolute-trend agreed — going long top-K names with positive own-trend and short bottom-K names with negative own-trend — to add crash-protective net-directional tilt to pure cross-sectional momentum.

The base backtest was the family's most consistent: Sharpe 0.447, profit factor 1.18, max drawdown 35.7%, +23.1% CAGR, positive in 5 of 7 years, and a validated +18% in the 2022 bear from the absolute-trend gate. Optimization then destabilized it — deflated Sharpe 0.0, PBO 0.68, walk-forward avg OOS Sharpe -0.892 (windows -1.06 and -3.36), and a degenerate optimized config with avg position 157% of equity, profit factor 0.9998, expectancy -0.82, and CAGR -100%.