BinanceTopThreeCrossSectionalReversalLongShort

BINANCE Top-3-Major Short-Horizon Cross-Sectional Mean Reversion (BTC/ETH/SOL Weekly Buy-Loser-Sell-Winner Long-Short, OHLCV-Only)

Short-horizon cross-sectional MEAN REVERSION strategy across the three system-explicitly-confirmed-safe BINANCE perps (BTCUSDT, ETHUSDT, SOLUSDT). Each week, rank the three by trailing 7-DAY return; LONG the WORST performer (recent losers tend to bounce in 1-2 week horizons) and SHORT the BEST performer (recent winners tend to give back gains short-term). This is the academic counterpart of cross-sectional momentum — Lehmann (1990) 'Fads, Martingales, and Market Efficiency' and Jegadeesh (1990) 'Evidence of Predictable Behavior of Security Returns' both document that returns at 1-WEEK horizons exhibit STRONG NEGATIVE serial correlation (mean reversion), while 1-MONTH+ horizons show positive serial correlation (momentum). The reversal effect is driven by short-term liquidity provision: assets that move sharply attract market-maker hedging flows that mechanically push prices back. This hypothesis DELIBERATELY PAIRS with my in-pipeline BinanceTopThreeCrossSectionalMomentumLongShort (which uses 30-day lookback for MOMENTUM) — if BOTH validate positive expectancy, the factor zoo (momentum + reversal at different horizons) is established for crypto majors and creates a complementary portfolio. The mechanism is FUNDAMENTALLY DIFFERENT from momentum because: (1) opposite signal direction, (2) shorter lookback (7d vs 30d), (3) economic driver is liquidity provision rather than under-reaction. Fills the SAME under-represented buckets as the momentum version: (a) MULTI scope (6.8%), (b) market-neutral direction (9.7% → reducing 90.3% long-only), (c) safe-instrument-only data path. Distinct from PAIRS ratio MR (in pipeline) because this is 3-asset cross-sectional RANKING, not 2-asset ratio z-score.

Implements the hypothesis literally: a 1-week-horizon cross-sectional reversal (Lehmann 1990 / Jegadeesh 1990) on the three explicitly-safe majors. (1) Opposite sign to the in-pipeline 30-day momentum strategy — long loser / short winner. (2) Shorter 7-day lookback. (3) Market-neutral MULTI-scope book (fills the under-represented market-neutral + multi buckets). Venue is BINANCE futures because the strategy MUST short (spot CASH cannot), and the edge is not funding-based; leverage stays 1.0 (long+short gross ~0.9x equity fits futures margin) so no leverage gate applies and sizing does not reference it. Robustness: on ~5-7 years of daily data the weekly rebalance produces ~260-360 rebalances each flipping up to two legs (~hundreds of trades), giving the walk-forward and holdout enough sample — avoiding the sparse-trade failure mode. The signal is the continuous best-minus-worst return spread (varies every bar, so Layer-2 frozen-signal detection passes) while the rebalance action is gated weekly. Cross-sectional ranking is done on timestamp-aligned closes: all three legs share the daily timeframe, so the base's same-timeframe alignment barrier defers primary processing until every leg has a contemporaneous bar, eliminating stale-leg lag in the ranking. Multi-instrument buffers are read from the base-maintained self._bars / self._extra_bars and snapshotted into a single synchronized history list (index-aligned as one unit), never trimmed per-instrument, per the multi-instrument alignment rule.

Strategy liquidated to -100% on 2021-02-25 and carries negative expectancy across the entire sample (profit_factor 0.347, expectancy -$247/trade, negative in 5 of 7 years). The 7-day cross-sectional reversal signal on BTC/ETH/SOL has no positive edge for optimization to tune toward, and the short-the-best-major leg has structural catastrophic tail risk in crypto bull markets (a known 'shorts on crypto majors refuted' failure mode) that risk-control fixes can only shrink, not remove. This is a bankrupt, negative-edge mechanism — not a parameter problem — so it is not worth 2 hours of optimization. FAILURE PATTERN: weekly cross-sectional MEAN-REVERSION (long loser / short winner) on 3 crypto majors is the inverse of the momentum sibling and inherits crypto's persistent short-horizon MOMENTUM (not reversal) at the leg level — shorting the strongest major weekly gets liquidated in bull phases while the long-loser leg fails to compensate, producing a sub-1 profit factor and account blow-up. The academic 1-week reversal result (Lehmann/Jegadeesh, equities) does not transfer to trending crypto majors.

Market-neutral cross-sectional mean-reversion long/short across BTC/ETH/SOL Binance USD-M perpetuals on 1-DAY bars. Every 7 days it ranks the three assets by trailing 7-day return, goes LONG the worst performer (short-horizon reversal bounce) and SHORT the best performer (short-horizon give-back), leaving the middle asset flat. Each leg is sized to per_leg_frac of equity so the book is roughly dollar-neutral (long ~0.45x, short ~0.45x, net ~0). Pure OHLCV — no supplementary data. The whole 3-leg rebalance is driven from calculate_signal (which fires once per timestamp-aligned bar); the primary single-instrument hooks are inert so the base engine never double-trades.

Adequate trade count (392: 201 long / 191 short) and the code faithfully implements the intended 3-asset weekly cross-sectional reversal (long 7d-loser, short 7d-winner)

Market-neutral construction is a genuinely under-represented bucket; the mechanism is well-specified and distinct from the momentum sibling

Account was LIQUIDATED on 2021-02-25 — total_return -100%, max_drawdown 100%. The short-winner leg has catastrophic tail risk in crypto bull markets (daily returns -73%, -80% during the blowup; kurtosis 652, skew -23)

No positive edge to optimize toward: profit_factor 0.347, expectancy -$247/trade, Sharpe -0.526, Sortino -0.37, omega 0.47

Negative in 5 of 7 years (2020 -25%, 2021 -95%, 2023 -14%, 2025 -2.5%) — a persistently losing signal, not a single bad regime

probabilistic_sharpe 0.009, information_ratio -2.50 vs the equal-weight basket; exposure_pct 200% suggests ~2x intended gross (sizing concern), but moot given the negative edge

BinanceTopThreeCrossSectionalReversalLongShort was designed as the reversal complement to a momentum sibling, ranking BTC/ETH/SOL weekly by 7-day return to buy the loser and short the winner in a market-neutral book. The mechanism was faithfully implemented with an adequate 392-trade sample, but it had deeply negative edge: profit factor 0.347, expectancy -$247/trade, negative in five of seven years, and — decisively — the short-winner leg blew up in the early-2021 bull market, liquidating the account on 2021-02-25 for a -100% total return. The analyst abandoned it at backtest review on iteration 1 as a bankrupt, negative-edge mechanism not worth optimizing, concluding that the equities 1-week reversal effect does not transfer to trending crypto majors whose short-horizon behavior is momentum, not reversal. Shorting the strongest major is a known refuted failure mode, and no parameter tuning removes that structural tail risk.

The academic equity 1-week reversal result does not transfer to trending crypto majors — at the leg level crypto majors exhibit short-horizon momentum, so shorting the strongest major weekly gets liquidated in bull phases while the long-loser leg fails to compensate, and shorting crypto majors is a known refuted failure mode.

The analyst abandoned it at the pre-optimization backtest-review gate: a liquidated, negative-edge mechanism (PF 0.347, expectancy negative, negative in 5 of 7 years) has no positive signal for optimization to tune toward, and the short-the-strongest-major leg carries structural catastrophic tail risk that risk controls can shrink but not remove — so optimization was not run.

A market-neutral, OHLCV-only long-short across BTC/ETH/SOL BINANCE perps that each week ranked the three by trailing 7-day return, going long the worst performer (expecting a bounce) and short the best (expecting give-back) — the short-horizon mean-reversion counterpart (Lehmann/Jegadeesh) to its 30-day cross-sectional momentum sibling.

It was catastrophic: over 392 trades (201 long, 191 short) the account was liquidated on 2021-02-25 for total return -100% (max drawdown 100%), with Sharpe -0.526, profit factor 0.347, expectancy -$247/trade, PSR 0.009, information ratio -2.50 vs the equal-weight basket, extreme tails (kurtosis 652, skew -23), ~200% gross exposure, and losses in 5 of 7 years despite a 54.8% win rate.