BinanceCrossSectionalRelativeStrengthMomentumDollarNeutralLSBasketDaily

Cross-Sectional Relative-Strength Momentum, Dollar-and-Beta-Neutral Basket of Binance USD-M Majors (Long Winners / Short Losers, Inverse-Vol Weighted, Weekly Rebalance, Pure-OHLCV Dispersion Edge — NOT Time-Series Momentum, NOT Funding)

A DOLLAR-NEUTRAL, BETA-STRIPPED, MULTI-INSTRUMENT CROSS-SECTIONAL MOMENTUM strategy across ~10 liquid, deep-history Binance USD-M major perpetuals (BTC, ETH, SOL, BNB, XRP, ADA, DOGE, LINK, LTC, AVAX). Each rebalance it ranks the universe by trailing relative-strength (price momentum) and goes LONG the top tercile (strongest recent performers) and SHORT the bottom tercile (weakest), sized to EQUAL DOLLAR exposure long vs short. This is mechanistically DISTINCT from the factory's promoted ABSOLUTE/time-series momentum basket (which is net-directional — it can be wholly long or wholly short the market and profits when the market trends): this strategy is MARKET-BETA-STRIPPED and profits from the DISPERSION between winners and losers, an orthogonal return stream that is largely independent of overall market direction. It is also distinct from the abandoned cross-sectional FUNDING-carry basket: that one relied on per-symbol funding (a data/crediting trap) and used equal leg COUNTS that failed to neutralize beta when one alt ripped 7x in 2021. This strategy uses PURE OHLCV (no funding data) and explicitly engineers neutrality via INVERSE-VOLATILITY leg weighting plus a per-name weight cap, so no single ripping name can dominate and the book stays genuinely dollar-and-beta neutral. The basket construction diffuses the single-asset/single-regime concentration that abandoned the single-alt directional strategies. Kept to ~4 parameters at robust round values to resist overfitting.

Implements the cross-sectional dispersion edge as an orthogonal, market-beta-stripped return stream. Distinct from the promoted ABSOLUTE/time-series momentum basket, which is net-directional (own-trend sign, can be wholly long or short the market): this ranks RELATIVE strength and is dollar-neutral by construction, so it harvests winner-minus-loser spread rather than market trend. Distinct from the abandoned cross-sectional FUNDING basket, which relied on a per-symbol funding crediting trap and used equal leg COUNTS that failed to neutralize beta when one alt ripped 7x: here neutrality is engineered via equal-dollar legs plus inverse-vol weighting and a per-name cap, and the edge is pure OHLCV. Basket construction diffuses the single-asset/single-regime concentration that abandoned the single-alt directional strategies. Kept to ~4 tunables at robust round values (30d momentum, weekly rebalance, 20d vol, 0.40 cap) to resist overfitting. Routed to BINANCE USD-M futures because both legs require shorting; leverage=1.0 with gross capped at equity, sizing uses gross_exposure not leverage, so no unused-leverage gate triggers. Engine-safe: discrete weekly netting rebalance, no continuous vol-scaling or per-bar churn.

Cross-sectional momentum edge on crypto majors is real but too WEAK to survive multiple-testing deflation — a clean statistical failure across every decisive gate, exactly as flagged at pre-optimization ('weak base Sharpe 0.40, fat-tail-driven, may well fail DSR'). After 225-trial selection: deflated_sharpe 0.0528 (<<0.95), is_significant FALSE, sharpe_ci_low -0.24 (CI straddles 0), pbo 0.5542 (>0.5, overfitting more likely than not), and the optimized Sharpe 0.478 is BELOW the expected-max-under-noise bar (0.996). The holdout FAILED (ratio 0.331, holdout Sharpe 0.13 vs WF-OOS 0.39) and the walk-forward is_overfitted=TRUE with a negative OOS window (-0.19). This is NOT the funding-carry sparsity case (where a real edge couldn't be statistically validated on clustered episodes and lowering a threshold could densify) — here the parameter space is fully explored and the sensitivity grid is Sharpe ~0.3-0.6 EVERYWHERE, i.e. there is no robust under-explored region to iterate toward; the edge is simply marginal. The headline returns are fat-tail-driven (skew 9.1 / kurtosis 164 baseline), so deflation correctly discounts them. Not iterate: a second parameter-optimization attempt would re-find the same sub-luck-bar edge across the same weak grid; strengthening it would require a different signal construction, universe, or horizon (a hypothesis change / revise, not a parameter retune). The mechanism is genuinely market-neutral (beta 0.03, alpha +0.09) and the construction is sound, so the Research Lead may revisit cross-sectional dispersion with a stronger signal (e.g. residual/idiosyncratic momentum, longer formation, or a wider universe), but THIS implementation does not clear the deflated-Sharpe bar and cannot be promoted. FAILURE PATTERN: dollar-neutral cross-sectional relative-strength momentum on ~10 Binance majors produces a genuine but marginal ~0.4-0.5 Sharpe whose headline is carried by a few fat-tail dispersion weeks; it fails DSR (0.05), is_significant=false with a 0-straddling CI, fails the holdout, and is_overfitted=true after best-of-225 selection — the cross-sectional momentum factor on a small crypto-major universe is too weak to deflate, regardless of parameter tuning.

Dollar-neutral CROSS-SECTIONAL relative-strength momentum across 10 liquid Binance USD-M major perpetuals on 1-DAY bars: BTCUSDT (primary), ETHUSDT, SOLUSDT, BNBUSDT, XRPUSDT, ADAUSDT, DOGEUSDT, LINKUSDT, LTCUSDT, AVAXUSDT. Every rebalance_days (weekly) it ranks the universe by trailing mom_lookback price return, goes LONG the top tercile and SHORT the bottom tercile, and sizes each leg to EQUAL DOLLAR exposure (structurally dollar-neutral: long gross == short gross == equity*gross_exposure/2). Within each leg names are INVERSE-VOLATILITY weighted (1/vol over vol_lookback) and normalized to sum 1, then a per_name_cap is enforced by iteratively capping and redistributing excess to uncapped names so no single ripping name can dominate. Profits from the winner/loser DISPERSION, a return stream largely independent of overall market direction. Pure OHLCV, no funding data. All order management is centralized in calculate_signal via single netting delta orders per leg (a trend flip executes as one signed order; sub-min-notional deltas are skipped to avoid churn); should_enter/should_exit/position_size are inert. ~4 edge tunables: mom_lookback, rebalance_days, vol_lookback, per_name_cap.

Genuinely market-neutral with substantial positive alpha: beta 0.026, benchmark_correlation 0.07, alpha +0.14 (14% annualized) — an orthogonal, beta-stripped return stream the portfolio under-represents, distinct from the promoted absolute/time-series momentum basket.

Legitimate, academically robust mechanism (cross-sectional winner/loser dispersion) with careful neutral construction (inverse-vol weighting, per-name cap, dollar-neutral); full ~6yr history, 602 trades, balanced long/short (284/318), reasonable capacity ($26M).

Not uniformly decayed: 2024 +41.7% and 2025 +11.9% are positive in recent regimes, so the edge isn't purely a 2021 artifact.

Weak risk-adjusted edge: Sharpe 0.40, profit_factor 1.08, annualized vol 43.8% — the +233% headline is fat-tail-driven (return_skew 9.1, kurtosis 164) and heavily concentrated in 2021 (+153%); a few huge weeks (2021-10-12 +53.6%, 2021-02-23 +34.1%) carry it.

Recent regime mixed-to-poor: 2023 -22.0%, 2026 -20.7% YTD — the dispersion edge is unreliable in some recent years, a DSR/walk-forward risk after 225-trial selection on a 0.40 base Sharpe.

Sizing concern: exposure_pct 602% with gross_exposure=1.0 / leverage=1.0 suggests the book runs far more gross than intended (~100%), inflating the 43.8% vol and the fat tails.

max_drawdown 30.2%, downside_deviation 26 — high for a 'neutral' strategy.

beta 0.026, alpha +0.14

beta~0, positive alpha

2021 +153% dominant; 2023 -22%, 2026 -20.7%

edge spread across regimes

Genuinely market-neutral with positive alpha (optimized beta 0.03, alpha +0.092) and a legitimate, academically-grounded cross-sectional momentum mechanism — the construction (inverse-vol weighting, per-name cap, dollar-neutral) is sound.

Walk-forward is not catastrophically overfit in magnitude (avg IS 1.13 -> avg OOS 0.39) and one OOS window is decent (0.95); the edge is real, just weak.

Sensitivity essentially clean (1 mild mom_lookback cliff); huge capacity ($80M); the optimized config improved regime distribution (2022 +32.6, 2024 +37.4) over the 2021-concentrated baseline.

All decisive robustness gates fail: deflated_sharpe 0.0528 (<<0.95), is_significant FALSE, sharpe_ci_low -0.24 (straddles 0), pbo 0.5542 (>0.5). Optimized Sharpe 0.478 is BELOW the expected-max-under-noise bar (0.996).

Holdout FAILED: ratio 0.331 (holdout Sharpe 0.13 vs WF-OOS 0.39), and walk-forward is_overfitted=TRUE with a negative OOS window (-0.19).

Weak edge, not a tunable miss: the sensitivity grid is Sharpe ~0.3-0.6 across the entire parameter space, and 225 trials found no robust region — the cross-sectional momentum factor on 10 crypto majors is marginal.

Returns are fat-tail-driven (baseline skew 9.1 / kurtosis 164; optimized skew 6.0 / kurtosis 92) — the edge depends on a few big dispersion weeks, which is precisely what deflation penalizes.

BinanceCrossSectionalRelativeStrengthMomentumDollarNeutralLSBasketDaily traded winner/loser dispersion across ten majors as a beta-stripped, inverse-vol-weighted long-short basket, a legitimate academic factor and a genuinely orthogonal sleeve to the portfolio's directional momentum. The baseline was genuinely neutral with +0.14 alpha and a +233% headline, earning a full optimization, but that headline was fat-tail-driven and concentrated in 2021, and every decisive robustness gate failed: DSR 0.053 below the 0.996 luck bar, PBO 0.554, a failed holdout, and an overfitted walk-forward with a negative OOS window. The analyst ruled the edge real but too weak to deflate — not a tunable miss since the whole parameter grid sits at ~0.3-0.6 Sharpe — and suggested a stronger signal construction rather than a retune. It ended after one iteration as abandoned, reaching optimization and analysis but never risk review.

A genuinely market-neutral edge with positive alpha can still be un-promotable: cross-sectional relative-strength momentum on ~10 crypto majors yields only a marginal ~0.4-0.5 Sharpe carried by a few fat-tail dispersion weeks, which deflation correctly discounts (DSR 0.05) — strengthening it requires a different signal construction, longer formation, or wider universe (a hypothesis change), not parameter retuning of this implementation.

It passed the pre-optimization backtest-review gate (verdict: optimize) but was abandoned at the post-optimization analyst gate (verdict: abandon): the cross-sectional momentum edge on a small crypto-major universe is real but too weak to survive multiple-testing deflation — the optimized Sharpe sits below the expected-max-under-noise bar with a CI straddling zero and the sensitivity grid is ~0.3-0.6 everywhere (no robust region to tune toward) — so it never advanced to risk review.

abandoned

A dollar-neutral, beta-stripped cross-sectional relative-strength momentum basket across ~10 deep-history Binance USD-M majors — each weekly rebalance ranking the universe by trailing price momentum, longing the strongest top tercile and shorting the weakest bottom tercile in equal dollar exposure with inverse-volatility leg weighting and a per-name cap — a pure-OHLCV winner/loser dispersion edge engineered to be genuinely beta-neutral and distinct from the promoted absolute-momentum and abandoned funding-carry baskets.

The baseline over ~6 years and 602 trades (284 long / 318 short) was genuinely market-neutral with real alpha — beta 0.026, alpha +0.14, total return +233%, capacity $26M — but weak risk-adjusted and fat-tail-driven (Sharpe 0.40, profit factor 1.08, annualized vol 43.8%, skew 9.1 / kurtosis 164, heavily concentrated in 2021 +153%, with 2023 -22% and 2026 -20.7% YTD); optimization improved the regime spread (Sharpe 0.478, alpha +0.092) but failed every robustness gate — DSR 0.053, not significant, PBO 0.554, sharpe_ci_low -0.24, holdout ratio 0.331, walk-forward is_overfitted=true with a negative OOS window.