Strategies

Cross-Sectional 14-Day Momentum Rotation on Top Alt Perps (Weekly Rebalance)

Outcome: Abandoned

CrossSectionalMomentumRotationTopAltPerps

Outcome Summary

This strategy tested a long-only top-2-of-5 cross-sectional momentum rotation on liquid alt perps, framed as a more robust, always-invested alternative to time-series momentum. Its backtest produced a spectacular +2246% headline return, but that was carried almost entirely by the 2021 alt-mania, leaving a Sharpe of just 0.446 (CI through zero), profit factor 1.136, a 79% max drawdown, and kurtosis of 234. The backtest reviewer abandoned it at the pre-optimization gate as a high-beta, outlier-carried bet with a marginal, non-significant edge, so it never proceeded to optimization or analyst review. The decisive reasoning was structural rather than tunable: a long-only rotation cannot avoid bear-market bleed, and the validated cross-sectional edge already lives in the portfolio's net-directional long-short momentum baskets.

Strategy report

Hypothesis

A long-only multi-instrument cross-sectional momentum strategy on Binance USD-M perpetuals. Every Monday 00:00 UTC, rank a basket of 5 liquid perps (BTCUSDT, ETHUSDT, SOLUSDT, BNBUSDT, AVAXUSDT) by their trailing 14-day total return. Hold equal-weighted long positions in the top 2 ranked symbols for one week, then re-rank and rotate. Uses ONLY daily OHLCV bars — no funding, no L/S ratios, no liquidations. The cross-sectional structure (relative ranking) is structurally more robust than time-series momentum because the strategy is always taking SOME position regardless of absolute market direction, and the ranking mechanism naturally adapts to whichever assets are leading the current regime. Distinctly different from existing portfolio: every current strategy is single-instrument or pairs (2 symbols); this is the only 5-symbol rotation, and the only cross-sectional approach.

Implements the hypothesis exactly: a 5-symbol, weekly-rebalanced, top-2 cross-sectional momentum rotation on daily bars. The primary leg (BTCUSDT) plus four same-timeframe extra legs mean the base class's same-timeframe alignment barrier defers the signal until all 5 legs have a contemporaneous daily bar, so rankings are computed on synchronized prices. Ranking uses each symbol's OWN 14-day return (a scalar), so there is no cross-symbol spread/ratio and therefore no multi-instrument index-alignment hazard — the classic pitfall for basket strategies. All trading is executed inside calculate_signal via _submit_entry_instrument/_submit_exit_position across the basket, with the primary should_enter/should_exit/position_size hooks left inert so the base single-instrument flow does not double-trade BTC. Trading is gated on every basket symbol having full lookback history, which means Layer-2 synthetic runs (primary bar only, no extras) never submit orders yet calculate_signal still returns a varying, finite diagnostic (primary trailing return / 1-bar ROC) to satisfy the frozen-signal guard. Weekly cadence is detected with pure integer weekday math (Monday=0) plus a >=7-day catch-up so drift never skips a rebalance, and a >=min_gap guard prevents same-week duplicates. Venue is deliberately BINANCE futures (MARGIN), overriding the long-only-goes-to-spot rule, because get_account_equity() reads the stable-balance: on a spot CASH account that balance is consumed as the basket is purchased, collapsing equity-relative sizing mid-rotation, whereas a MARGIN account keeps balance ~constant so equal-weight leg sizing stays correct; deployed notional caps at invest_frac (0.90) so real leverage stays <= 1x and leverage is wired into sizing.

Not worth optimizing: a beta mirage with a marginal, non-significant risk-adjusted edge and catastrophic, structurally-unavoidable drawdown. The +2246% headline is driven almost entirely by the 2021 alt-mania (2021 +2319%, then -70.6% in 2022 and -46.7% in 2026), so despite the huge total return the Sharpe is only 0.446 with sharpe_ci_low -0.247 (CI through zero) and profit_factor 1.136 — not significant even before deflation. beta 0.705 / benchmark_correlation 0.752 confirm this is a levered long-alt beta bet (alpha only 0.171), not genuine cross-sectional alpha, and return_kurtosis 234 with a 79% max drawdown shows it is outlier/beta-carried with catastrophic risk. The weakness is structural: a long-only top-2-of-5 rotation on correlated majors cannot avoid bear-market bleed, and optimizing lookback_days/top_k/rebalance adds no downside protection — so a 2-hour optimization would predictably fail deflated Sharpe. The genuine cross-sectional-momentum edge already exists in the portfolio's promoted long-SHORT net-directional momentum baskets, which are net-directional precisely to earn the risk-adjusted edge and crash protection this long-only variant lacks; this is variant-selectivity, not a fixable tuning problem.

Implementation

Long-only cross-sectional momentum rotation over 5 liquid Binance USD-M perps (BTC, ETH, SOL, BNB, AVAX) on daily bars. Every Monday it ranks the basket by trailing 14-day total return and holds equal-weighted longs in the top 2, exiting symbols that fall out of the top 2 and buying new entrants (holding stayers to minimise turnover). Total deployed notional is 90% of equity split across the 2 legs, keeping real leverage <= 1x. Pure daily OHLCV — no funding, L/S ratios, or liquidations.

Backtest Review

Sharpe: 0.45
Total return: 2246.86%
Max drawdown: 79.02%
Trades: 251
Win rate: 47.8%
Profit factor: 1.14

Mechanically functional: 251 basket trades executed via the rotation logic, cross-sectional ranking is a validated mechanism class (promoted long-short momentum siblings exist)

Modest positive alpha (0.171) and clean same-timeframe/same-venue basket construction (no stale-leg artifact risk)

Headline +2246% return is a 2021-beta mirage: 2021 +2319% carries everything, then -70.6% (2022) and -46.7% (2026) — Sharpe only 0.446 despite the huge total return

High beta 0.705 / benchmark_correlation 0.752 — essentially a levered long-alt beta bet, not cross-sectional alpha; long-only gives no downside protection

Catastrophic risk: max_drawdown 79.0% (CI to 98.6%), return_kurtosis 234.4 (outlier-carried), exposure_pct 200% suggests sizing balloons past 1x as gains compound

Marginal, non-significant edge: profit_factor 1.136, sharpe_ci_low -0.247 (CI through 0) — fails significance before any 225-trial deflation

Structural: a long-only cross-sectional rotation cannot avoid bear-market bleed; optimizing lookback/top_k/rebalance adds no crash protection

Outcome Summary

A long-only cross-sectional rotation on correlated majors is effectively a levered long-alt beta bet (beta 0.705, correlation 0.752) with no downside protection, and tuning lookback/top_k/rebalance cannot add the crash protection that the portfolio's net-directional long-short momentum baskets earn structurally.

It was abandoned at the pre-optimization BACKTEST_REVIEW gate (iteration 1): the analyst judged the headline return a 2021-beta mirage (2021 +2319%, then -70.6% in 2022 and -46.7% in 2026), with a marginal non-significant edge and structurally unavoidable bear-market drawdown, so optimization and the later stages were never reached.

A long-only cross-sectional momentum rotation that each Monday ranked a 5-perp basket (BTC, ETH, SOL, BNB, AVAX) by trailing 14-day return and held equal-weighted the top 2 for a week, using only daily OHLCV on Binance USD-M futures.

The single backtest posted a +2246% total return (CAGR 71%) over 2084 days on 251 trades but only a Sharpe of 0.446 with sharpe_ci_low -0.247, profit factor 1.136, beta 0.705, a 79.0% max drawdown, and return kurtosis of 234.

Backtest and paper results are hypothetical. Trading involves risk of loss.