CrossSectionalMomentumWeeklyRotationEightMajorLong

Cross-Sectional Momentum Weekly Rotation across Eight Major Crypto Perpetuals Long

A cross-sectional (relative-strength) momentum rotation strategy over a fixed universe of 8 major Binance USD-M perpetuals: BTCUSDT, ETHUSDT, SOLUSDT, BNBUSDT, AVAXUSDT, LINKUSDT, ADAUSDT, XRPUSDT. Every 7 daily bars (weekly rebalance) the strategy ranks all 8 symbols by their trailing 30-day total return, then holds equal-weight long positions in the TOP 3 ranked symbols only. On each weekly rebalance: exit any currently-held symbol that is no longer in the top 3, and enter any new top-3 symbol that is not currently held. This is structurally orthogonal to every strategy currently in the pipeline — none use cross-sectional ranking. It is also structurally orthogonal to every failure pattern in the log: it is NOT a single-instrument strategy (so it cannot suffer rare-event regime concentration), it is NOT a single-bar pattern (the 30-day return signal aggregates 30 bars of information), it is NOT a mean-reversion-with-technical-gates (it is pure relative momentum with no oscillator filters), and it does NOT depend on supplementary data (avoiding the verification framework's no-trades failure mode on OI/L-S-ratio strategies). Two free parameters total — the lookback window (30 days) and the top-K cutoff (3) — which is roughly half the parameter count of the average failed strategy in the log, reducing overfitting surface. Trade frequency: 8 symbols × 52 weekly rebalances/year × ~25% probability that a given slot rotates = ~100 position-events per year, well above the walk-forward statistical threshold. Cross-sectional momentum has 70+ years of academic evidence in equities (Jegadeesh-Titman 1993, Asness 1994) and has been shown to extend to crypto (Hubrich 2017, Liu-Tsyvinski-Wu 2020) with comparable or stronger effect sizes than equities. Risk profile: 3 concurrent long positions, 30% notional per leg, 90% gross long, 10% cash buffer for slippage and fees; no leverage, no stop-loss (positions exit purely on rotation), drawdown bounded by basket diversification across 3 of 8 majors.

Implements the cross-sectional relative-strength rotation directly: a single timestamp-keyed close buffer per instrument, a weekly (every 7 bars) rank of trailing 30-day returns, and equal-weight long exposure to the top 3 with rotation-only exits. Multi-instrument handling follows the proven pattern — on_bar buffers primary bars and triggers _rebalance, on_extra_bar records the other 7 symbols, and returns are computed from each instrument's OWN latest bar so a same-timestamp bar-ordering race cannot zero out the universe (the alignment bug that produces no-trades). The rebalance waits until all 8 symbols have full lookback history before ranking, then exits drop-outs and buys new entrants sized at equity*leg_pct/price with per-symbol quantization and a min-notional guard. Per the coder venue-routing rule the strategy is routed to BINANCE_SPOT rather than the hypothesis's USD-M perps: it is strictly long-only, uses no leverage, and its edge is price momentum (not funding), so spot is the correct venue — it avoids funding drag and inert leverage, and leverage stays 1.0 (spot CASH cap). calculate_signal returns a continuous primary 30-day-momentum monitor (tanh) so Layer-2 sees a varying signal, while the actual rotation trades are driven by _rebalance; should_enter/should_exit/position_size are inert because this is a basket strategy. The two free parameters (lookback, top_k) plus rebalance cadence keep the overfitting surface small.

No active edge over the correct benchmark: information_ratio -0.60 vs a meaningful equal-weight 8-major basket means the rotation underperforms simply holding the majors — the premise (rotate into winners to beat the basket) is falsified. The +569% return is concentrated in the 2021 bull (+434% that year) while the strategy loses in 2022 (-30.7%) and 2026 (-25.5%), and rolling Sharpe collapses to -8.7 in mid-2026, so momentum is decaying in exactly the walk-forward OOS / holdout window. With Sharpe 0.73 (CI low 0.18) on OHLCV-only cross-sectional momentum, optimizing the two params would curve-fit to the 2021 regime rather than find a robust edge. Matches the concentrated basket-momentum overfit pattern (L36) and the negative-IR-vs-basket signal. Not worth 2 hours of optimization.

Long-only cross-sectional momentum rotation over 8 major Binance SPOT majors (BTC, ETH, SOL, BNB, AVAX, LINK, ADA, XRP), daily bars, pure OHLCV. Every 7 daily bars it ranks all 8 by trailing 30-day return and holds equal-weight long the top 3 (30% equity notional each, ~90% gross, ~10% cash); on each rebalance it exits holdings that drop out of the top 3 and enters new top-3 names. No stop-loss — positions exit purely on rotation. Each symbol's return uses its own latest daily bar (scale-free ranking), avoiding cross-instrument timestamp-alignment issues. Routed to BINANCE_SPOT (CASH, leverage 1.0, long-only).

Legitimate, differentiated mechanism (cross-sectional relative-strength momentum) that trades match exactly: 281 long trades, 0 shorts, ~30-day holds, weekly rotation into top-K

Multi-regime, multi-instrument sample (2020-2026, 281 trades) — not a low-count or single-instrument concentration case

avg_trade_return_pct 11.2% is well above the fee floor; PF 1.53 and DD 36.9% (below the 50% hard line)

information_ratio -0.60 vs a meaningful equal-weight basket benchmark — the rotation adds negative active value over simply holding the 8 majors; the premise is falsified

Return concentrated in the 2021 bull (+434% of the +569% total); losing years in 2022 (-30.7%) and 2026 (-25.5%)

Rolling Sharpe collapses to -8.7 in mid-2026 — momentum decaying precisely in the walk-forward OOS / holdout window

Headline Sharpe only 0.73 with CI low 0.18 (barely significant); OHLCV-only momentum class that optimization tends to overfit to the bull regime

CrossSectionalMomentumWeeklyRotationEightMajorLong applied a well-evidenced relative-strength momentum concept, rotating weekly into the top-ranked 3-4 of eight major crypto majors, and the backtest traded exactly as designed: 281 long trades, ~30-day holds, +569% total return with a healthy 1.53 profit factor and sub-50% drawdown. But its information ratio was -0.60 against an equal-weight basket of the same eight majors, meaning the rotation actually underperformed just holding them, and the gains were almost entirely a 2021 bull-market artifact (+434% that year) with losses in 2022 and 2026 as momentum decayed. The analyst abandoned it at the pre-optimization backtest-review gate on the falsified premise and marginal Sharpe (0.73), a call corroborated by the optimization run's failed holdout and overfit flag, and the strategy never reached promotion, risk review, or paper trading.

A large absolute return can still mask no real edge — a momentum rotation must be measured against the correct basket benchmark (information ratio), because beating cash is not the same as beating buy-and-hold of the same universe.

It was abandoned at the backtest-review gate on a negative information ratio (-0.60): the rotation added negative active value over simply holding the eight majors, falsifying the core premise, and with sub-threshold Sharpe and 2021-concentrated returns the analyst judged optimization would only curve-fit the bull regime.

A long-only cross-sectional (relative-strength) momentum rotation that every ~7-10 days ranked 8 major Binance spot majors by trailing ~21-30-day return and held equal-weight the top-K (3-4), rotating out of names that fell out of the top and into new leaders, with no leverage and no stop-loss.

Over a 2017-2026, 281-trade sample it returned +569% with profit factor 1.53, 48.4% win rate, 36.9% max drawdown, and avg_trade_return_pct 11.2%, but headline Sharpe was only 0.73 (CI low 0.18) and information_ratio was -0.60 versus an equal-weight 8-major basket. Returns were concentrated in the 2021 bull (+434% of the +569% total), with losing years in 2022 (-30.7%) and 2026 (-25.5%); the optimization run's holdout also failed (ratio -0.008, PBO 0.59, flagged overfitted).