ShortTermCrossSectionalReversalDollarNeutralBasketLS1D

Binance USD-M Short-Term Cross-Sectional Reversal Dollar-Neutral Basket (Established Majors, 5-Day Vol-Normalized Rank, Weekly Rebalance, No-Leverage, Long-Short)

A dollar-neutral, daily cross-sectional REVERSAL basket across ~10 established, confirmed-full-history Binance USD-M majors (BTC, ETH, SOL, BNB, XRP, DOGE, ADA, LTC, LINK, BCH) — the natural diversifier to the portfolio's momentum-heavy book. Each rebalance it ranks majors by their recent (≈5-day) VOLATILITY-NORMALIZED return, goes LONG the biggest relative LOSERS and SHORT the biggest relative WINNERS, in equal dollar notional, betting on the short-term overreaction reverting. This is deliberately the STRUCTURALLY-SAFE form of mean-reversion: every time-series MR this factory tried (15m z-fade, 4H volume-climax) blew up catastrophically (-99.96%, -97.7%) on fee drag + single-asset gap-through tail losses, but a CROSS-SECTIONAL, dollar-neutral, RANK-based reversal is fundamentally different — it is self-hedging (long-loser/short-winner nets out market moves), has NO single-asset stop to gap through, and rebalances WEEKLY (not intraday) to control fees. There is positive evidence this class works here: CrossSectional7DayReversalV3 is in the portfolio and is NOT in the failure list (unlike all the time-series MR). It stays data-safe (established majors only — broad-alt/HL baskets die on data), market-neutral (improving the 87%-long skew), and is engineered to dodge BOTH recurring sizing failures: NO vol-targeting leverage scalar and NO uncapped inverse-vol weights — equal-ish capped sizing with gross hard-capped at 1.0× (the leverage-explosion fix), and a single frozen lookback (the overfit fix). It complements rather than duplicates the existing 7-day reversal by using a shorter ≈5-day vol-normalized formation and weekly rebalance. It is genuinely orthogonal to the factory's validated momentum factor (opposite sign), the highest-value diversifying sleeve available on the validatable surface.

The structurally-SAFE form of mean-reversion and the highest-value diversifier to the portfolio momentum-heavy, 87%-long book. Every TIME-SERIES MR the factory tried (15m z-fade, 4H volume-climax) blew up (-99.96%, -97.7%) on fee drag + single-asset gap-through tail losses; a CROSS-SECTIONAL, dollar-neutral, RANK-based reversal is fundamentally different -- self-hedging (long-loser/short-winner nets out market moves), no single-asset stop to gap through, weekly (not intraday) rebalance to control fees. Positive in-factory evidence: CrossSectional7DayReversal is in the portfolio and NOT in the failure list. This complements rather than duplicates it via a shorter ~5-day vol-normalized formation and weekly rebalance, and is orthogonal (opposite sign) to the validated momentum factor. Data-safe: established majors only (broad-alt/HL baskets die on data). Both sizing-explosion fixes are baked in (gross hard-capped 1.0x, equal capped weights, no vol-target scalar). All legs BINANCE USD-M futures (genuine long+short); leverage=1.0 and sizing uses gross_cap_frac not leverage, so no unused-leverage gate triggers.

Wrong-sign edge — a premise failure, not a code or sizing problem. The engineering is actually correct (the recurring leverage-explosion failure is fixed: gross hard-capped at ~1x, annualized_vol 34.7%, no liquidation, dollar-neutral beta -0.016), but the strategy bets the wrong direction: short-term cross-sectional REVERSAL (long the biggest relative losers / short the biggest relative winners) loses -90.9% with alpha -0.18 and is negative in 6 of 7 years (only 2021 positive), bleeding steadily (rolling Sharpe negative from 2022 on, 2024 -70%) rather than blowing up. Crypto majors exhibit cross-sectional MOMENTUM, not reversal: the mirror-image cross-sectional momentum basket (hypothesis 8040a98b) returned +233% with alpha +0.14, so long-loser/short-winner systematically fights the factor that works and loses. The hypothesis's claim that a 7-day reversal sibling works is empirically refuted by this 5-day vol-normalized version. Not iterate: the loss is pervasive across every regime and no formation-lookback or k tuning can flip a wrong-sign edge to positive; the only 'fix' is to invert the signal into momentum, which merely duplicates the existing positive cross-sectional momentum basket and the promoted weekly momentum sleeve. Not optimize: a strategy losing 6 of 7 years has no robust positive region to tune toward. FAILURE PATTERN: dollar-neutral short-horizon cross-sectional REVERSAL (long losers / short winners) on crypto majors is the losing inverse of the validated cross-sectional momentum factor — majors trend cross-sectionally (winners persist), so reversal systematically loses regardless of correct sizing/fees; the reversal sign is refuted on this surface and the momentum sign is already captured.

Dollar-neutral, daily cross-sectional REVERSAL basket across 10 established Binance USD-M majors (BTC, ETH, SOL, BNB, XRP, DOGE, ADA, LTC, LINK, BCH) on 1-DAY bars, rebalanced WEEKLY. Each rebalance ranks the majors by their recent ~5-day VOLATILITY-NORMALIZED return (cumulative formation return divided by daily-return std), goes LONG the top_k biggest relative LOSERS and SHORT the top_k biggest relative WINNERS in equal dollar notional, betting the short-term overreaction reverts. Sizing is engineered against both recurring failure modes: NO vol-target leverage scalar and NO uncapped inverse-vol weights -- equal capped notional per name = gross_cap_frac x equity / (2 x k), so gross is hard-capped at gross_cap_frac (1.0x, no leverage) and the book is dollar-neutral (long notional == short notional). A single frozen formation lookback is the overfit fix. Ranking is driven from the sync-gated calculate_signal so all legs are ranked on contemporaneous same-timestamp closes; a turnover-minimizing side-diff keeps same-side holdings, flattens only dropped/flipped names and opens only new selections to control weekly fees. Orders use direct per-leg submission; base single-position hooks are inert. calculate_signal returns the primary asset vol-normalized score (continuous). 3 core parameters (formation_days, rebalance_days, top_k); asset list and caps frozen.

Engineering is sound and the recurring sizing failures are genuinely fixed: gross hard-capped at ~1x (avg_position_pct 15.3% x 6 ~= 92%), annualized_vol 34.7% (not a leverage explosion), no liquidation, weekly rebalance keeps fees low (1.79% of gross). Dollar-neutral (beta -0.016).

Data-safe (established majors only) and genuinely market-neutral — the construction is the structurally-safe form of the primitive, as intended.

Wrong-sign edge: total_return -90.9%, negative in 6 of 7 years (only 2021 positive), alpha -0.18, profit_factor 0.88, expectancy -$69.68/trade, max_drawdown 91.8% from a slow persistent bleed (rolling Sharpe negative from 2022 on; 2024 -70%).

Cross-sectional reversal is the losing mirror of cross-sectional MOMENTUM, which is what actually works on crypto majors (the momentum sibling 8040a98b returned +233% with alpha +0.14) — long-loser/short-winner systematically fights the dominant factor.

The hypothesis's premise (short-term cross-sectional reversal works on majors; '7-day reversal sibling not in failure list') is empirically refuted by this 5-day version's -91%.

Not fixable by tuning: the loss is pervasive across all regimes and lookback can't flip a wrong-sign edge; flipping to momentum would just duplicate existing promoted strategies.

positive

negative 6 of 7 years

positive across regimes

ShortTermCrossSectionalReversalDollarNeutralBasketLS1D applied a weekly-rebalanced, dollar-neutral long-loser/short-winner reversal across ten established majors, carefully engineered to dodge the leverage-explosion and overfit failures of prior MR attempts. The construction worked exactly as intended — market-neutral, no blow-up, low fees — but the signal was pointed the wrong way: it lost 90.9% with alpha -0.18, negative in 6 of 7 years, bleeding steadily as winners kept winning. The analyst ruled it a wrong-sign premise failure — crypto majors trend cross-sectionally, the mirror momentum basket made +233%, and inverting the signal would just duplicate an existing promoted strategy. It ended after one iteration as abandoned, never advancing to optimization or risk review.

Short-horizon cross-sectional reversal is the losing inverse of the validated cross-sectional momentum factor on crypto majors — winners persist, so reversal loses regardless of correct sizing and low fees — and the claim that a 7-day reversal sibling works is refuted by this 5-day version's -91%; the only 'fix' (invert to momentum) merely duplicates the already-promoted momentum sleeve.

It was abandoned at the pre-optimization backtest-review gate (verdict: abandon) as a premise failure, not a code or sizing problem — crypto majors exhibit cross-sectional momentum, not reversal, so longing losers / shorting winners systematically fights the factor that works (its mirror momentum basket returned +233% with alpha +0.14), a wrong-sign edge no lookback or k tuning can flip — so optimization and all later stages were never reached.

A dollar-neutral, daily cross-sectional reversal basket across ~10 established Binance USD-M majors (BTC, ETH, SOL, BNB, XRP, DOGE, ADA, LTC, LINK, BCH) — each weekly rebalance ranking names by their ~5-day volatility-normalized return, longing the biggest relative losers and shorting the biggest relative winners in equal capped notional with gross hard-capped at 1.0x — as a structurally-safe, self-hedging mean-reversion sleeve to diversify the momentum-heavy book and improve the long skew.

Over ~6 years and 1,238 trades (634 long / 604 short) the engineering held (dollar-neutral beta -0.016, annualized vol 34.7%, no liquidation, commission 1.79% of gross), but the edge was wrong-signed: total return -90.9%, Sharpe -0.53 (CI [-1.30, 0.16]), alpha -0.18, profit factor 0.88, expectancy -$69.68/trade, max drawdown 91.8%, negative in 6 of 7 years (only 2021 positive) as a slow persistent bleed (2024 -70%).