Crypto Cross-Sectional Momentum Top-Bottom Tercile (10-Symbol Universe, 60-Day Lookback, Monthly Rebalance, Daily Bars, Long-Short Multi-Instrument)

Multi-instrument long-short cross-sectional momentum strategy that EXPLICITLY addresses the post-mortem recommendation from the abandoned `ETH-BTC Ratio Momentum Spread on Hyperliquid` strategy. That hypothesis failed because it applied the 12-1 month cross-sectional momentum effect (Jegadeesh-Titman 1993) at the WRONG horizon — 4H bars with 3.3-day lookback on a pair of highly-correlated assets, where mean-reversion dominates. The post-mortem explicitly recommended: 'Move to monthly horizons on daily bars to actually test the 12-1 cross-sectional momentum effect from equities literature.' This proposal does exactly that, on a DIVERSIFIED universe (10 crypto perps, not 2 correlated ones). Universe: BTCUSDT, ETHUSDT, SOLUSDT, BNBUSDT, XRPUSDT, ADAUSDT, DOGEUSDT, AVAXUSDT, LINKUSDT, DOTUSDT — all on BINANCE USD-M futures with multi-year daily bar history (safe data). Mechanism: every 30 days, rank all 10 by trailing 60-day return; go long the top 3 performers (recent winners) and short the bottom 3 performers (recent losers), equal-weight within each leg. Hold 30 days, then rebalance. This is the canonical academic implementation of cross-sectional momentum that earned 12% annualized in equities (Jegadeesh-Titman 1993), 8% in commodities (Erb-Harvey 2006), and has been documented in crypto (Liu-Tsyvinski 2021, Bianchi-Dickerson 2023 showing ~15% annualized in crypto top-30 universe with monthly rebalance). Fills FOUR under-represented quotas SIMULTANEOUSLY: MULTI-instrument scope (8.1% vs target), LONG-SHORT direction (9.7% vs ~45% target), and uses a horizon dimension (30-day hold + daily bars) that's distinct from the over-represented 4H-12H bucket. NOT a classical chart pattern. NOT a same-asset funding extreme (dead). NOT cross-venue basis. NOT a correlated-pair trade. A genuinely different mechanism class than every previous hypothesis.

Implements the post-mortem fix for the abandoned ETH-BTC 4H ratio momentum: it tests the Jegadeesh-Titman 12-1 cross-sectional momentum effect at the CORRECT horizon (60-day lookback, 30-day hold, daily bars) on a DIVERSIFIED 10-name universe rather than a correlated pair at a mean-reverting 4H/3-day horizon. Architecture follows the landed AbsoluteMomentumLSBasketDaily pattern: all leg management is centralized in calculate_signal, which the base fires only after the cross-leg alignment barrier confirms every same-daily-timeframe leg has a bar at the primary timestamp, so all 10 closes are ranked on synchronized data with no stale-leg lag; should_enter/should_exit/position_size are inert because the base entry path manages only the primary instrument. Reconciliation uses single delta orders per symbol (delta = target_signed_qty - current_signed_qty): a name flipping from a winner to a loser tercile is handled in one sell on the NETTING account, and names leaving both terciles are flattened via _submit_exit_position, so the book is always dollar-neutral and gross never transiently doubles up during a rotation (the failure mode that liquidated earlier rotation strategies). Sizing is equal-weight: per_leg = equity*gross_target/(2*top_n) = 15% each across 3 longs + 3 shorts -> 90% gross, ~0 net, comfortably within the leverage-1.0 futures margin ceiling (so leverage is not set >1 and the unused-leverage gate does not apply); each ~15k-notional leg clears the $20 min-notional by a wide margin. min_bars_required is 2 so the lockstep buffers start filling immediately, and ranking waits internally until ALL 10 names have lookback+1 aligned closes (so no trade fires on partial history); the first rebalance auto-fires once history is ready (counter seeded at rebalance_days). The returned cross-sectional return dispersion (max-min trailing return) is a genuinely continuous per-bar signal so Layer-2 frozen-signal detection passes while trades remain schedule-gated. Three edge parameters keep the optimization surface small per the low-parameter guidance; all 10 symbols are liquid majors with multi-year daily history so no leg risks data_unavailable.

Do not optimize — the backtest is corrupted by the same runaway-exposure bug that has now recurred across this developer's multi-instrument long-short rebalancers, and the cross-sectional momentum mechanism has no demonstrated edge after repeated attempts. Evidence of the blow-up: exposure_pct 608.95% (~6x equity vs the intended gross_target 0.90), avg_position_pct 140%, physically impossible daily returns for a dollar-neutral book (2021-07-20 +929%, 2021-02-20 -201%, 2026-03-26 +349%, 2024-12-01 -150%), max_drawdown_ci_high 627%, annualized_volatility 369%, return_kurtosis 1293, CAGR -100%. The root cause is identical to the 12-instrument WEEKLY cross-sectional version flagged earlier this session (exposure 604%) and the 4H version (3073%): the _signed_qty->delta netting in _rebalance does not cap gross, so positions accumulate to ~6x equity instead of netting to the signed target. That exact netting/gross-cap fix was already provided as iterate feedback and the defect recurred unchanged here, indicating a systemic multi-instrument-rebalance problem rather than a one-off. Separately, the cross-sectional momentum mechanism itself has been comprehensively tested and failed — the prior 5-coin basket at profit_factor ~0.99, and every multi-instrument variant blows up before producing an evaluable edge — so even a correctly-sized backtest has no demonstrated alpha to optimize. Required before ANY further multi-instrument L/S work (framework-level, not per-strategy): (1) after each rebalance, verify via the cache that every leg's net signed notional equals its signed target (per_leg long / -per_leg short) rather than stacking; (2) add a hard cap so sum of |leg notional| <= gross_target * equity, scaling orders down if exceeded; (3) guard equity<=0 and never size against a collapsed denominator. Until that shared defect is fixed AND a properly-sized cross-sectional momentum backtest shows a positive net-of-cost edge across years, this mechanism class should not consume further optimization or iteration cycles — abandon this run.