MidCapResidualMomentumDollarNeutralLS

Idiosyncratic (Beta-Adjusted) Cross-Sectional Momentum, Dollar-Neutral Long-Short — 8 Liquid MID-CAP BINANCE USD-M Alts Ranked by BTC-RESIDUAL Momentum (Long Strongest / Short Weakest ALPHA, Not Raw Return), Volatility-Scaled Legs, Bi-Weekly Rebalance, Drawdown-Stopped, 3-Parameter)

A DOLLAR-NEUTRAL, MULTI-INSTRUMENT cross-sectional momentum strategy on 8 liquid MID-CAP BINANCE USD-M perps (LINK, LTC, DOT, ATOM, NEAR, FIL, AVAX, UNI — SAME venue, OHLCV-only, so no supplementary-feed data-ceiling risk and no L44 cross-venue problem; BTCUSDT.BINANCE is carried as a non-traded beta benchmark only). It is deliberately distinct from my five in-flight books (raw cross-sectional momentum on MAJORS, OI-confirmed momentum, regime-switched factor, the ETH/BTC funding pair, intermediate momentum) on TWO axes at once: (1) UNIVERSE — mid-cap alts, whose larger cross-sectional dispersion produces bigger relative spreads that clear fees more easily than the tightly-correlated majors; and (2) SIGNAL — the ranking is IDIOSYNCRATIC (RESIDUAL) momentum, not raw return: each name's returns are regressed on BTC over the formation window and names are ranked by their BTC-residual cumulative alpha. Residual momentum is a documented improvement over raw momentum — it strips the shared market-beta component that drives momentum CRASHES (when the whole complex violently reverses, raw-momentum books blow up because their long/short legs are really a hidden beta bet; residualized legs are not), giving a more crash-resistant, more stationary relative-strength signal. This directly attacks the risk_reject death of the earlier LONG-ONLY mid-cap momentum rotation (81% DD) by (a) removing beta via dollar-neutral construction, (b) removing beta AGAIN via residual ranking, (c) inverse-volatility scaling each leg, and (d) a hard drawdown stop. Momentum is the ONLY edge family this factory has ever gotten to survive (deployed Sharpe 2.6; dual-TF Sharpe 2), so leaning into a more-robust, crash-resistant formulation of it — on a fresh universe with a genuinely different ranking — is a principled bet. Fills the under-represented long-short bucket (13.4% vs the <=55% long-only target).

Implements the hypothesis on all four of its axes. (1) UNIVERSE: 8 liquid mid-cap USD-M alts, all on one venue and OHLCV-only, so there is no cross-venue leg-availability risk (L4/L44) and no supplementary-feed dependency (L35); I confirmed all 8 legs plus the BTC benchmark have current 1-DAY catalog data through 2026-07-30. (2) SIGNAL: the ranking is BTC-residual cumulative alpha, not raw return. Beta is estimated by centered covariance over the formation window and the residual series is summed, which equals cumulative return minus beta times benchmark cumulative return, and is non-zero by construction -- unlike an intercept-inclusive OLS residual sum, which is identically zero and would produce a dead ranking. (3) BETA REMOVED TWICE: dollar-neutral equal-notional sides plus a residualized ranking, so a complex-wide reversal does not hit both legs the same way, the failure mode that killed the long-only mid-cap rotation at 81% DD. (4) RISK: legs are inverse-idiosyncratic-vol weighted (capped at 3x the side's smallest weight so one quiet name cannot become the whole leg), gross exposure is a FIXED 0.8x of current equity rather than a compounding add-on (L15), and a hard 20% drawdown stop measured on equity PLUS open-leg unrealized PnL flattens the book and re-anchors the peak. Exactly 3 edge parameters are exposed: mom_lookback, rebalance_days, max_drawdown_pct. Fees: 26 rebalances/year holding 4 names with netting delta orders and a min-notional churn guard is low turnover, and a 30-day residual-alpha spread across mid-caps is typically several percent, well clear of the ~0.10% round-trip taker cost. Venue is BINANCE USD-M because the book is long AND short; leverage stays 1.0 since gross notional is 0.8x equity, and sizing still reads self.config.leverage explicitly so there is no unused-leverage mismatch. The rebalance grid is calendar-anchored off bar.ts_event (day index modulo rebalance_days), never a process-relative bar counter, so it is identical across restarts, backtests and paper. calculate_signal returns a continuous value EVERY bar (the primary leg's cross-sectional alpha z-score, degrading to its own residual-vs-BTC momentum and then to raw momentum when the cross-section is unavailable, e.g. single-instrument synthetic verification), and all order management is centralized in the rebalance path. One caveat worth flagging: with 8 names the quartile is only 2 long / 2 short, so the book is concentrated; if the analyst prefers more diversification, min_universe and the quantile divisor are the natural knobs.

Overfit-prone, regime-decayed member of the dollar-neutral cross-sectional momentum family that will not clear post-optimization robustness gates. Sharpe 0.409 with 95% CI [-0.280, 1.037] straddling zero, profit_factor 1.12 (below the 1.2 meaningful line), information_ratio -0.20. The 155.6% total return comes entirely from 2021 (+76%), 2022 (+24%) and 2024 (+33%); the last two years are NEGATIVE (2025 -13.4%, 2026 -4.9%) and rolling Sharpe collapsed from ~3.5-4 (2025) to -4.7 (mid-2026) — the edge has decayed, so the 15-day holdout (in the negative 2026 regime) would near-certainly fail. Every sibling in this family has been abandoned: the near-identical residual/beta-adjusted momentum on majors, and MajorsCrossSectionalMomentumDollarNeutralLS which was optimized then died in ANALYZING (PBO 0.67, DSR 0.49, holdout FAILED). Reframing the same residual-momentum mechanism onto a fresh mid-cap universe is the L41 re-overfit trap — it enters weaker (Sharpe 0.41/PF 1.12) than the sibling that already failed, so best-of-225 selection would inflate its Sharpe on noise and repeat the overfit death. return_kurtosis 43.8 shows the edge depends on a few outlier fortnights, and max_drawdown 31.6% (CI high 77.1%) shows the crash-resistance thesis did not deliver. avg_trade_return_pct 1.04% clears the fee floor, so this is overfit/regime-fragility, not fee-fragility, and there is no robust parameter region to tune toward. Abandon at BACKTEST_REVIEW rather than spend 2 hours.

Dollar-neutral cross-sectional momentum on 8 liquid mid-cap Binance USD-M perps (LINK primary, plus LTC, DOT, ATOM, NEAR, FIL, AVAX, UNI) on 1-DAY bars, ranked by BTC-RESIDUAL (idiosyncratic) momentum instead of raw return. Over the last mom_lookback daily log returns on a benchmark-aligned timestamp grid, each name gets beta = cov(r_i, r_btc)/var(r_btc), residual returns r_i - beta*r_btc, cumulative alpha = sum(residuals), and idiosyncratic vol = std(residuals). Alphas are z-scored cross-sectionally; every 14 calendar days the top-quartile alpha names are bought and the bottom-quartile sold, inverse-idiosyncratic-vol weighted within each side and with equal dollar notional per side (structurally dollar-neutral). Each leg is moved to target with a single netting delta order. A hard drawdown stop on mark-to-market book equity (account equity plus open-leg unrealized PnL) flattens every leg if the book gives back max_drawdown_pct from its running peak, then re-anchors the peak so trading resumes at the next scheduled rebalance. BTCUSDT.BINANCE is loaded as a NON-TRADED beta benchmark only.

Expand the universe (>8 names) or lower the quantile divisor so each side holds >=3 names if more diversification is desired.

With only 8 basket names the quartile is k=2, so the book is 2 long / 2 short — a highly concentrated 4-name book. Inverse-vol weighting cannot diversify away single-name risk at this count (largest_win 9083 / largest_loss -7552 on 24% avg position confirm single-leg dominance). Developer already flagged this. Not a correctness defect.

Optimizer/analyst should treat per-window Sharpe cautiously; evaluate on a longer history than 362 days if available.

Sample-size caution: 46 round-trip trades over 362 days (~26 bi-weekly rebalances/yr). This is a low-frequency scheduled rebalancer, so trades occur reliably (not a rare-gate risk), but 46 trades is borderline for distinguishing edge from noise across walk-forward + 15-day holdout — each WF window holds only 2-3 rebalances, so single-window Sharpe CIs will be wide.

Route to backtest_review for optimize/iterate/abandon, not back to the developer as a code fix.

Informational, NOT a code defect: sandbox is net-negative (total_return -11.06%, Sharpe -0.33, PF 0.90, win_rate 0.478). This is not a bug signature (win_rate not 0.0, |Sharpe| not >5, PF not 0.0; long/short win rates 0.43/0.52 both plausible) and avg_trade_return_pct ~2.46% clears the fee floor — so it is neither a polarity/sizing bug nor a fee-viability failure. It is genuine mild underperformance of the edge on this window.

Genuinely market-neutral (beta 0.046, benchmark_correlation 0.12) with double beta-stripping (residual ranking + dollar-neutral), inverse-vol legs, and a hard drawdown stop — thoughtful crash-resistant construction

OHLCV-only, single-venue, adequate trade count (302) and capacity ($44M)

avg_trade_return_pct 1.04% clears the futures fee floor, so failure is not fee-fragility

Sharpe 0.409 with 95% CI [-0.280, 1.037] straddles zero; profit_factor 1.12 below the 1.2 meaningful line

Entire return is 2021/2022/2024; the last two years are negative (2025 -13.4%, 2026 -4.9%) and rolling Sharpe collapsed from ~3.5-4 to -4.7 — the edge has decayed in the modern regime

Dollar-neutral cross-sectional momentum family: every sibling abandoned, including the near-identical residual-momentum-on-majors variant and MajorsCrossSectionalMomentumDollarNeutralLS which failed ANALYZING (PBO 0.67, DSR 0.49, holdout FAILED)

return_kurtosis 43.8, skew 2.19 — edge depends on a few outlier fortnights, not a stationary spread

max_drawdown 31.6% (CI high 77.1%) — the crash-resistance thesis did not deliver steadiness

Reframing the same overfit mechanism onto a fresh mid-cap universe is the L41 re-overfit trap; 15-day holdout falls in the negative 2026 regime

This strategy made a principled bet on the factory's only surviving edge family — cross-sectional momentum — with a more crash-resistant formulation: rank mid-cap alts by BTC-residual idiosyncratic momentum rather than raw return, dollar-neutral and inverse-vol scaled with a hard drawdown stop, on a fresh higher-dispersion universe. The construction delivered genuine neutrality (beta 0.046) and a +155.6% headline, but the risk-adjusted edge was hollow: a Sharpe of 0.409 with a CI straddling zero, profit factor 1.12, a 31.6% drawdown that belied the crash-resistance claim, and returns confined to 2021/2022/2024 while 2025 and 2026 turned negative and rolling Sharpe collapsed from ~4 to -4.7. The analyst abandoned it at backtest review as an overfit-prone, regime-decayed member of the dollar-neutral cross-sectional momentum family — every sibling of which had already died, including the near-identical residual-momentum-on-majors — judging that reframing the same mechanism onto mid-caps was the re-overfit trap and would only repeat the deflated-Sharpe/holdout death in optimization.

Double beta-stripping (residual ranking plus dollar-neutrality) and a drawdown stop do not confer crash-resistance or a durable edge if the underlying signal has decayed — moving the same cross-sectional-momentum mechanism to a fresh mid-cap universe enters weaker than siblings that already died overfit, and an edge concentrated in a few outlier fortnights that turns negative in recent years has no robust region to optimize toward.

It was abandoned at the BACKTEST_REVIEW gate: a Sharpe straddling zero, profit factor below 1.2, a 31.6% drawdown that undercut the crash-resistance thesis, and an edge decayed to negative in the last two years with the 15-day holdout falling in the negative 2026 regime. As a weaker-entering member of the dollar-neutral cross-sectional momentum family whose every sibling was abandoned (the near-identical residual-momentum-on-majors, and MajorsCrossSectionalMomentumDollarNeutralLS which failed ANALYZING with PBO 0.67 / DSR 0.49 / holdout failed), reframing the same mechanism onto a fresh mid-cap universe is the L41 re-overfit trap, so it never reached optimization.

A dollar-neutral, multi-instrument cross-sectional momentum strategy on 8 liquid mid-cap BINANCE USD-M alt perps that ranks names by BTC-residual (idiosyncratic) momentum rather than raw return — regressing each name on BTC over the formation window and going long the strongest / short the weakest alpha — with inverse-volatility-scaled legs, a bi-weekly rebalance, and a hard drawdown stop, using residual ranking plus dollar-neutrality to double-strip beta and resist momentum crashes.

The construction was genuinely market-neutral (beta 0.046, correlation 0.12) with total return +155.6% and avg_trade_return_pct 1.04% over 302 trades, but the edge was thin and decayed: Sharpe 0.409 (CI [-0.280, 1.037] straddling zero), profit factor 1.12 (below the 1.2 line), information ratio -0.20, max drawdown 31.6% (CI to 77.1%). The return came entirely from 2021/2022/2024 while 2025 (-13.4%) and 2026 (-4.9%) were negative and rolling Sharpe collapsed from ~3.5-4 to -4.7, with return kurtosis 43.8 showing the edge rests on a few outlier fortnights.