BetaNeutralMomentumOverlayLS

Beta-Neutral Cross-Sectional Momentum Overlay, Market-Neutral Long-Short (BINANCE USD-M, Monthly: Long the Top-2 3-Month-Momentum Majors + Short a BETA-ADJUSTED BTC Hedge So Net Market Exposure ≈ 0 — Harvest the RESIDUAL/ALPHA of Momentum, Not the Beta That Made Every Long-Only Basket Blow Up in the 2022 Bear, 3-Parameter)

A MARKET-NEUTRAL, MONTHLY long-short overlay on BINANCE USD-M majors that deliberately fixes the single defect that hard-abandoned every long-only momentum basket this session (Vol-Scaled 15-name 75% DD, Low-Skew rotation 76% DD, and the AVAX/grid TOTAL LIQUIDATIONS in the 2022 bear): those books were pure crypto BETA, so a bull backtest hid a fatal bear drawdown. The deployed 8-major rotation harvests momentum but is always-invested long-only and carries that same latent risk. This instead isolates the RESIDUAL (beta-neutral) component of cross-sectional momentum — the part that is actual alpha, not market direction. Each month it goes LONG the top-2 momentum majors from a 7-name deep-history universe (ETH, SOL, BNB, AVAX, LINK, ADA, XRP) and SHORTS a BETA-ADJUSTED amount of BTCUSDT so the book's net exposure to the crypto market factor ≈ 0. It shorts BTC (the most liquid, least squeeze-prone name) as the hedge — NOT a specific 'loser' major, avoiding the violent short-squeeze that makes crypto momentum-shorts treacherous. Because it is beta-neutral, it is regime-distributed: it profits when momentum winners OUTPERFORM the market in bull, bear, or chop, and structurally cannot eat a full 50%+ market drawdown. This is NOT the deployed long-only always-invested rotation (different return source: pure alpha vs beta+alpha; different construction: hedged 2-long + BTC-short vs top-3 long), NOT an L12 dollar-neutral N-name rank basket (only 3 legs total, monthly → minimal fee count), NOT carry/options/liquidation/single-name-reversion. It fills the three most under-represented buckets at once: market-neutral (2%), long-short (14.8%), and the macro/monthly horizon (0.6%). Data-safe pure OHLCV; 3 core parameters (momentum lookback, number of longs, beta lookback).

Iteration 2 fixes the two backtest-review defects with the smallest change that leaves the signal logic (momentum ranking, OLS beta, calculate_signal) byte-identical. DEFECT #2 (broken beta-neutrality -> net-long -> 49.5% DD and inflated up-months): the old code KEPT stable longs at their stale (mark-to-market drifted) notional while sizing the BTC hedge to the fresh per_long and capping it at short_cap_frac=0.6, so in high-beta-alt rallies (2021-04, 2024-07) the hedge under-covered and the book ran net-long, capturing the rally as the +166%/+199% months and eating bear drawdowns. Fix: (a) REBUILD the book from scratch every month — close every leg then reopen — so ACTUAL long notional == per_long == exactly what the hedge covers; (b) size the BTC short to the EXACT beta_sum with the cap raised to short_cap_frac=1.0 so it is a pure margin backstop that never binds; (c) shrink per_leg_frac 0.15->0.10 and beta_max 3.0->2.0 so the uncapped hedge stays margin-safe. I verified across betas 1.0-2.0 that net beta is now EXACTLY 0 (cap never binds) and gross <=1.2x / margin <=0.60x equity — neutrality is restored and liquidation remains impossible. DEFECT #1 (suspected accounting artifact from the same-bar rebalance / get_account_equity mid-rebalance): equity is read ONCE at the top before any close or open (no mid-rebalance re-read that could double-count freshly realized PnL), and because the book is now genuinely neutral each month the net-long exposure that actually produced the inflated PnL is removed. Sizing off the realized-PnL cash base (not mark-to-market) also prevents the runaway compounding of the +707% sibling family. The full monthly rebuild trades all three legs each month (more, not fewer, trades -> Layer-3 >=1-trade gate trivially met and better measurability). Imports unchanged (no banned modules); leverage=2.0 still fed into sizing (per_long and the backstop both reference self.config.leverage), so the leverage-used gate is satisfied. If the artifact-free re-run shows the 2022-style bear protection with controlled drawdown and PF>1.2 it is a strong optimize candidate; if the return collapses to ~0 the honest conclusion is that the residual momentum alpha was thin, but the exposure is now correct and safe to evaluate.

The iteration-1 fixes worked and are credited: the accounting artifact is gone (the impossible +166%/+199% months are eliminated by the rebuild-from-scratch logic) and the exact beta hedge cut max_drawdown from 49.5% to 38.4%. But the now-trustworthy result exposes the underlying problem I flagged as the abandon condition: with the artifact removed, the ENTIRE return collapses to a single regime — 2021 delivered +261% while every subsequent year is negative (2022 -3.6%, 2023 -6.1%, 2024 +0.5%, 2025 -9.9%, 2026 -12.7%). profit_factor is 1.126, Sharpe 0.407 with a CI [-0.17, 0.98] straddling zero, and rolling Sharpe is negative through 2025-2026. The beta-neutral cross-sectional momentum-residual alpha existed only in the 2021 alt-season and has decayed to consistently negative for four straight years — a decayed-premise problem, not a parameter-tuning gap. Optimizing mom_lookback/num_longs/beta_lookback cannot resurrect an edge that has been absent (negative) since 2021; the 225-trial optimizer would only overfit the single 2021 window and fail the deflated-Sharpe/OOS gates. Per my iteration-1 feedback ('if, with the artifact removed, the return collapses to near-zero, abandon'), this is now a clean-data abandon.

Market-neutral, monthly long-short momentum overlay on BINANCE USD-M majors. Each month it ranks a 7-major universe (ETH/SOL/BNB/AVAX/LINK/ADA/XRP) by trailing 3-month return, goes LONG the top-2 equal-notional (per_leg_frac=0.10 of equity x leverage each), and SHORTs BTCUSDT in the EXACT beta-adjusted amount (sum of each long's OLS beta to BTC x its notional) so net market-factor exposure is ~0. The book is fully REBUILT each month (all legs closed then reopened) so the actual long notional always equals what the BTC hedge covers — net beta is 0 by construction rather than approximately. Beta is clamped to [0.2, 2.0] and the hedge cap (short_cap_frac=1.0) is a margin backstop that never binds for realistic betas, so neutrality holds even when high-beta alts rank in and rally. calculate_signal returns the top-long-minus-universe-mean momentum spread. It isolates the residual/alpha of momentum, not beta, so it is regime-distributed and cannot ride a full market drawdown. 3 core parameters: mom_lookback_days, num_longs, beta_lookback_days.

Evaluate on FULL history incl 2022: genuine neutrality should now show real bear protection — if drawdown-controlled AND positive residual alpha across regimes, optimize; if alpha stays ~0/negative once the net-long artifact is removed, abandon (as the developer honestly flags).

Now that genuine beta-neutrality is restored, the honest residual-momentum alpha is thin/negative (analyst call, not a code defect). The iter-1 net-long bug (stale-kept longs + capped hedge) inflated up-months to +166%/+199% and produced a 49.5% drawdown; with the exact uncapped hedge and full monthly rebuild those artifacts are gone (max_drawdown now 16.2%, benchmark beta -0.23), confirming they WERE exposure artifacts. But the clean neutral book shows negative alpha here: total_return -9.56%, Sharpe -0.24 (CI [-1.95,+1.38]), profit_factor 0.81, win_rate 0.37, skew -1.21.

Optimizer could resize only materially-drifted legs while still enforcing the exact beta hedge each month, cutting fees without reintroducing the stale-notional net-long bug.

The full monthly rebuild (close ALL legs then reopen) guarantees clean neutrality but raises turnover/fees vs resizing drifted legs: turnover 12.57, total_impact_usd $223, ~6 fills/month. On a thin-alpha strategy this fee drag directly eats the residual edge. Defensible correctness-over-churn trade-off (it's what fixed the neutrality bug), but a cost to weigh.

Regress full-history returns on BTC (not the basket) to confirm realized net-beta-to-BTC ~0; if it drifts, revisit beta_lookback_days.

Residual beta -0.23 is vs the equal-weight basket (benchmark_meaningful=false), a different factor than the hedged BTC. The construction neutralizes net-beta-to-BTC exactly (Sigma beta_i*notional_i - btc_short = 0, cap non-binding), so non-zero basket-beta is expected residual, not a hedge failure.

Both iteration-1 defects fixed correctly: the accounting artifact is gone (impossible +166%/+199% months eliminated via rebuild-from-scratch) and the exact beta hedge cut max_drawdown to 38.4%

Genuinely market-neutral (beta 0.08, correlation 0.20); clean risk profile with no liquidation

With the artifact removed, the ENTIRE return is 2021 (+261%); every subsequent year is negative — 2022 -3.6%, 2023 -6.1%, 2024 +0.5%, 2025 -9.9%, 2026 -12.7%

The beta-neutral momentum-residual edge is single-regime (2021 alt-season) and has decayed to consistently negative for four straight years

profit_factor 1.126 (marginal), Sharpe 0.407 with CI [-0.17, 0.98] straddling zero; rolling Sharpe negative through 2025-2026

No persistent alpha for optimization to capture — a decayed premise, not a tuning gap

Two specific, fixable defects must be resolved BEFORE optimization — do not optimize the current artifact-inflated metrics. (1) SUSPECTED ACCOUNTING ARTIFACT: monthly returns of +166% (2021-04) and +199.7% (2024-07) cannot arise from the stated ~30% gross long / ≤60% BTC-short exposure at leverage 1.0 — these two months carry most of the +454% total_return. Investigate the monthly rebalance PnL handling in _process_primary, specifically the close-and-reopen-same-bar path (_close_all then _open on the same bar) and how get_account_equity() reflects positions mid-rebalance — this is the same failure family as the +707% month in the SPOT dual-momentum sibling. Verify realized vs mark-to-market PnL on rebalance bars and confirm no double-counting. (2) BROKEN BETA-NEUTRALITY: short_cap_frac=0.6 caps the BTC hedge, so when the top-2 high-beta alts rank in and rally, the hedge under-covers and the book runs net-long — this is the direct cause of the 49.5% max drawdown (CI high 88%) and the -56%/-23% down years, and it contradicts the market-neutral thesis. Fix by sizing the longs down (lower per_leg_frac) so the beta-required BTC short fits within a sane notional WITHOUT hitting the cap, or raise/remove the cap with a margin-safe check, so net beta actually stays ~0 in high-beta-alt regimes. After both fixes, re-run the backtest; if the genuine (artifact-free) result still shows the 2022-style bear protection with a controlled drawdown (<30%) and PF>1.2, it is a strong optimize candidate given it fills the rarest buckets. If, with the artifact removed, the return collapses to near-zero, abandon.

This overlay was the most sophisticated fix in the momentum-basket lineage: it isolated the residual alpha of cross-sectional momentum by longing the top-2 winners and shorting a beta-adjusted BTC hedge, explicitly targeting the pure-beta defect that hard-abandoned or liquidated every long-only sibling in the 2022 bear. The iteration-1 fixes were credited — an accounting artifact was removed via rebuild-from-scratch logic and the exact beta hedge cut max drawdown from 49.5% to 38.4% with no liquidation and clean market-neutrality. But with the artifact gone, the trustworthy result showed the entire +134% return came from 2021's alt-season while all four subsequent years were negative, leaving a 1.126 profit factor and a Sharpe straddling zero. The analyst abandoned it at review as a decayed premise, not a tuning gap: optimizing the three parameters could only overfit the single 2021 window and would fail the OOS/deflated-Sharpe gates.

The beta-hedge construction genuinely solved the drawdown/liquidation and accounting problems that killed the long-only siblings, but hedging out beta only revealed that the cross-sectional momentum-residual alpha itself has decayed — it existed in the 2021 alt-season and has been negative since, so a market-neutral wrapper cannot manufacture an edge that is no longer there.

It was abandoned at the pre-optimization backtest-review gate: the iteration-1 fixes worked (an accounting artifact removed and the beta hedge cutting drawdown), but the now-trustworthy result exposed a decayed premise — the beta-neutral momentum-residual alpha existed only in the 2021 alt-season and has been consistently negative for four straight years, with a break-even profit factor and Sharpe straddling zero, so there was no persistent edge for optimization to capture.

A market-neutral, monthly long-short overlay on BINANCE USD-M majors that goes long the top-2 3-month-momentum names from a 7-major universe and shorts a beta-adjusted BTC hedge so net market exposure ≈ 0 — isolating the residual/alpha component of cross-sectional momentum rather than the crypto beta that blew up the long-only baskets in 2022.

After iteration-1 fixes, the backtest was clean and market-neutral (beta 0.08, correlation 0.20) with a 38.4% max drawdown and no liquidation over 204 trades, but marginal — profit factor 1.126, Sharpe 0.407 (CI straddling zero), total return +134% — and single-regime: the entire return came from 2021 (+261%) while every subsequent year was negative (2022 -3.6%, 2023 -6.1%, 2024 +0.5%, 2025 -9.9%, 2026 -12.7%).