Daily Cross-Sectional 60-Day Drawdown-from-Peak Reversal: Long Deepest-Drawdown + Short Shallowest-Drawdown Across 6 Major Binance Perps

A dollar-neutral, cross-sectional mean-reversion strategy on the same fixed universe of 6 major Binance USD-M perpetuals (BTC, ETH, SOL, BNB, AVAX, LINK — all confirmed clean OHLCV data) that captures the DRAWDOWN REVERSAL anomaly first documented by DeBondt & Thaler (1985) 'Does the Stock Market Overreact?'. Every 24 hours, the strategy ranks the 6 assets by their current drawdown from their 60-day rolling high (current_price - rolling_60d_high) / rolling_60d_high × 100, goes LONG the 2 assets with the DEEPEST drawdown (most oversold from peak) and SHORT the 2 assets with the SHALLOWEST drawdown (least pullback, most-stretched at highs), equal-weighted within each leg, dollar-neutral. This is the third in the cross-sectional series — DailyCrossSectional7DReturnReversal uses 7-day return ranking (recent direction), DailyCrossSectionalLowVolAnomaly uses 30-day vol ranking (recent magnitude), and this strategy uses 60-day drawdown ranking (longer-horizon peak-relative position). Three structurally orthogonal signal axes on the same universe = three independent edges that can stack without overlap. Returns, volatility, and drawdown-from-peak are EMPIRICALLY UNCORRELATED in major crypto baskets — an asset can have a positive 7-day return WHILE deeply drawn down from peak (recovery phase), or a negative 7-day return WHILE near all-time highs (early pullback). The strategies will hold DIFFERENT positions on most days. Uses ONLY BINANCE USD-M perp daily bars — no supplementary data, no spot, no options, no Hyperliquid. Avoids EVERY failure mode confirmed this session. Fills FIVE under-represented buckets: long-short direction (14.1% → toward 45%), multi-instrument scope (7.8% → grow), cross-sectional mechanism class (the only confirmed winner), portfolio diversifier vs directional strategies, and OHLCV-only data dependency (the only reliably-clean dependency type). Position sizing: 12% per leg × 4 legs = 48% gross exposure, dollar-neutral. Risk per trade: ~1% from cross-sectional spread volatility.

Implements the DeBondt-Thaler drawdown-reversal anomaly as the third orthogonal axis in the cross-sectional series (return-rank, vol-rank, now drawdown-from-peak), reusing the proven centralized multi-leg architecture from the landed cross-sectional momentum and absolute-momentum baskets: all leg management lives 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 6 names 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 name (delta = target_signed_qty - current_signed_qty): a name flipping from the deepest-drawdown long set to the shallowest short set is handled in one order on the NETTING account, names leaving both sets are flattened via _submit_exit_position, so the book stays dollar-neutral and gross never transiently exceeds the 48% target during a rotation. Drawdown is computed from the rolling 60-bar high using each leg's highs, ascending-sorted so the most-negative (deepest) drawdowns go long and the shallowest go short. Sizing is equal-weight (per_leg = 12% equity) across 2 longs + 2 shorts -> 48% gross, ~0 net, comfortably within the leverage-1.0 futures margin ceiling (leverage referenced only as a non-amplifying multiplier, so the unused-leverage gate does not apply); each ~12k leg clears the $5 min-notional. min_bars_required is 2 so the lockstep close/high buffers start filling immediately, and ranking waits internally until ALL 6 names have drawdown_lookback+ aligned bars (no trade on partial history); the first rebalance fires once history is ready (counter seeded at rebalance_days). The returned cross-sectional drawdown dispersion (max-min) is a continuous per-bar signal so Layer-2 frozen-signal detection passes while trades stay schedule-gated. All 6 symbols are liquid majors with multi-year daily history so no leg risks data_unavailable; three edge parameters keep the optimization surface small.

Cross-sectional 60-day drawdown-reversal LS on 6 majors has a negative edge: PF 0.91, Sharpe -0.32 (sharpe_ci_low -0.99), expectancy -$38/trade, total_return -44.6%, negative in 6 of 7 years (only 2022 positive) over a healthy 957-trade sample, with max_DD 46.8% (ci_high 73.7%) and a 1,986-day drawdown. Root cause is an under-diversified cross-section: long-deepest-drawdown / short-shallowest-drawdown among 6 highly-correlated majors is essentially 'long the laggard, short the leader' on names that move together, so after shared crypto-beta cancels there is too little idiosyncratic dispersion for the reversal anomaly to overcome costs — the same under-diversification that left the cross-sectional 7-day-momentum sibling at a no-edge Sharpe ~0.22. The hypothesis's 'cross-sectional = the only confirmed winner' premise does not hold on a 6-name correlated basket. Not optimize: net-losing with negative 6/7 years — optimizing would curve-fit the lone 2022 positive and fail forward. Not iterate: the negative edge is structural to the thin/correlated cross-section, not a parameter miss; the exposure_pct 400% anomaly (vs intended ~48% gross, same as the 5-major momentum sibling) is worth verifying but is leverage-invariant in PF terms, so fixing it only scales the loss toward zero, not positive. FAILURE PATTERN: cross-sectional reversal/momentum on a small basket of highly-correlated crypto majors (BTC/ETH/SOL/BNB/AVAX/LINK) does not have a robust edge on ANY ranking axis (7-day return, 30-day vol, 60-day drawdown all reviewed) — the 6-name cross-section is too correlated to extract a tradeable long-short spread after costs, producing negative-to-near-zero risk-adjusted returns. The cross-sectional mechanism needs a much broader, more diverse universe (15+ names) to generate genuine dispersion; on the majors-only basket it is just costly noise. The Research Lead should stop proposing additional ranking-axis variants on the same 6-major universe and either widen the universe substantially or retire the lane.