CryptoLowVolAnomalyRotationLS1D

Crypto Low-Volatility Anomaly Rotation Long-Short on 5 Major Perps (30-Day Vol Window, Daily Decision, Analyst-Sanctioned Replacement for Failed Cross-Sectional Momentum)

Market-neutral long-short rotation strategy on a fixed universe of 5 Binance USD-M perpetuals: BTCUSDT, ETHUSDT, SOLUSDT, BNBUSDT, AVAXUSDT. Each DAILY close, compute 30-day realized volatility for every asset, rank them, then LONG the lowest-vol asset (the 'safety bid' leg) and SHORT the highest-vol asset (the 'lottery bid' leg). This is the EXPLICIT analyst-sanctioned replacement for the abandoned 4H Cross-Sectional Momentum Rotation: per the abandon-feedback quote, 'The cross-sectional bucket can be filled by an alternative approach: LOW-VOLATILITY ANOMALY rotation (long lowest-vol asset, short highest-vol asset over 30+ day window) — captures a documented anomaly with very different statistical properties and is empirically less susceptible to regime decay.' Why this is STRUCTURALLY DISTINCT from the failed cross-sectional momentum rotation (which collapsed at PF 0.99 over 2994 trades) and the failed pairs ratio MR (which collapsed at PF 0.75 over 556 trades): (1) SIGNAL TYPE — VOL RANKINGS are empirically far more persistent on monthly horizons than RETURN RANKINGS. Once an asset enters a high-vol regime it tends to stay there for 30+ days (vol clustering is one of the most robust stylized facts of asset prices, well-documented since Engle 1982 ARCH). The analyst's diagnosis 'short-horizon relative-performance prediction on crypto is dominated by noise' applies to RETURN ranks, NOT VOL ranks. (2) MECHANISM DIRECTION — rotate INTO safety, not INTO recent winners. Avoids the 'buy after peak / sell after trough' whipsaw trap that doomed momentum rotation. (3) ACADEMIC SUPPORT — Frazzini-Pedersen 'Betting Against Beta' (2014) and Baker-Bradley-Wurgler 'Low-Risk Anomaly' (2011) document this across asset classes; microstructural roots (lottery preference, leverage constraints) are mechanistically different from momentum and fail under different conditions. (4) WINNER-TAKE-ALL ALIGNMENT — BTC is typically lowest-vol, AVAX/SOL typically highest-vol; strategy systematically tilts toward the structural long-term winner (BTC) and away from structural laggards. This is ALIGNED with crypto winner-take-all per the analyst's #1 lesson, not fighting it (which is the failure axis shared by pairs MR and momentum rotation). Only 3 parameters (well below the 5-param overfit ceiling). Trade frequency ~25-50 rotations/year per leg × 2 = 50-100 trades/year over 7+ years of BINANCE data → 350-700 trades total → 23-47 OOS trades per walk-forward window, well above the 5-trade floor.

The backtest reached Layer 3 and liquidated (2021-02-11, total_return -100%, impossible -207.9%/+378% daily prints, exposure 200% vs 80% intended); signal/alignment logic already passed earlier layers, so the changes are confined to risk/sizing and the short leg per the analyst's three required fixes and decision rule. Fix 1 (sizing/leverage, no transient double-up): the old _rotate closed the outgoing leg and opened the incoming leg in the SAME bar, so on rotation days the netting book carried the old + new legs simultaneously (the 200% exposure). The new _rotate sets closed_any whenever it closes anything and then RETURNS before opening, so a new long is only established on the next signal bar — transitions can never stack. A single long leg sized at long_fraction (0.80) of equity bounds gross at ~80%. Fix 2 (hard stops): each held long tracks its entry price (reconstructed from avg_px_open if state is lost) and is closed when price <= entry*(1-stop_pct); a stopped asset gets a cooldown_bars re-entry lockout so it can't immediately re-arm and re-stop. Fix 3 (short leg, structural): the analyst judged shorting the highest-vol crypto asset unbounded-loss by construction and sanctioned a long-only low-vol tilt as the replacement; the short side is removed entirely (no desired=-1, and any stray short is defensively flattened), which is what makes liquidation impossible — a long leg can lose at most ~stop_pct of 80% notional. The vol-ranking signal, lockstep multi-asset close alignment, dispersion diagnostic return, and inert abstract hooks are unchanged, keeping every earlier layer green. Class name and hypothesis lineage are preserved; leverage stays 1.0 (referenced nowhere as a multiplier, so no unused-leverage gate issue). Net: a correctly-sized, stop-protected, structurally non-liquidating implementation that still tests the documented low-volatility anomaly on its long (safety-bid) side.

Do not optimize — the strategy has abandoned its own thesis and what remains is a marginal, decaying long-beta tilt. The low-volatility anomaly is a market-neutral long-short premium, but the developer (correctly) removed the short leg because shorting the highest-vol crypto asset mid-parabola is unbounded-loss and liquidated the prior version. The long-only salvage that holds the lowest-vol major is therefore just directional crypto beta — benchmark_correlation 0.71, beta 0.43, alpha only 0.045 over the equal-weight basket — not the anomaly the hypothesis set out to capture. Its edge is marginal (Sharpe 0.50 with sharpe_ci_low -0.19, CI straddling zero; PF 1.11) and carries a 47.7% max drawdown (CI high 96%), with a severe recent collapse (2025 -11.8%, 2026 -44.8%) that will very likely fail the chronological holdout; the +423% headline is also half open-position mark-to-market (end_unrealized_pct 212%). This is a structural dead-end, not a parameter problem: re-adding the short leg re-introduces the blow-up, while long-only cannot capture the anomaly (it is beta), so there is no clean iteration path and best-of-N optimization of a CI-straddling-zero, 48%-DD beta tilt will fail the deflated-Sharpe and holdout gates. Abandon rather than optimize; if the cross-sectional/low-vol bucket is still wanted, it needs a genuinely market-neutral construction whose short side is risk-bounded (a fresh hypothesis), not this long-only beta proxy.

Long-only low-volatility-anomaly tilt on 5 Binance USD-M perps (BTC, ETH, SOL, BNB, AVAX), daily. Each daily close it ranks 30-day realized volatility across the universe and holds a SINGLE long in the lowest-vol asset (the safety bid), cash otherwise. Iteration 2 converts the previously liquidating long-short version to the analyst-sanctioned long-only salvage: it removes the unbounded short-highest-vol leg, sizes one long leg to 80% of equity (gross <= ~80%), defers opening a new leg until the bar after any close so rotations can never transiently double up exposure, and adds a per-leg hard percentage stop (15%) with a 5-bar re-entry cooldown. Long-only with leverage 1.0 means the account cannot liquidate. Three tunable parameters: vol_window, long_fraction, stop_pct.

Prior long-short liquidation bug is fixed (deferred opens, long-only, hard stop): liquidated=false, avg_position_pct 116% (sane)

Healthy absolute return over the full sample (+423%) with 92 trades and no correctness defect

Long-only salvage abandons the market-neutral low-vol ANOMALY thesis — it is now directional crypto beta (benchmark_correlation 0.71, beta 0.43, alpha only 0.045), mostly just holding BTC

Marginal edge: Sharpe 0.50 with sharpe_ci_low -0.19 (CI straddles 0), PF 1.11

High drawdown and recent collapse: max_DD 47.7% (CI high 96%), 2025 -11.8%, 2026 -44.8% — the chronological holdout will likely fail

Headline inflated by mark-to-market: end_unrealized_pct 212% (much of +423% is an un-exited open position)

Structural dead-end: the genuine anomaly needs the short leg, which the developer correctly found is unviable on crypto (unbounded loss shorting highest-vol mid-parabola); long-only = beta, re-adding the short re-blows-up

Do NOT optimize — the account LIQUIDATED on 2021-02-11 (total_return -100%) and produced impossible daily returns (-207.9%, +378%), so nothing can be evaluated. Fixes required: 1. FIX SIZING / LEVERAGE. exposure_pct is 200% vs the intended 0.40/leg (80% gross). A -208% single-day return is impossible for a correctly-sized dollar-neutral book — there is a position-accounting/sizing bug. Ensure long notional == short notional, gross <= ~80% equity, and that the rotation's close-then-open transitions don't transiently double up exposure. Verify post-fix that no single-day return exceeds plausible bounds and the account never liquidates. 2. ADD HARD STOPS ON BOTH LEGS. There is currently no stop-loss. The SHORT leg (highest-vol asset) gets squeezed catastrophically in crypto bull runs (2021). Add a per-leg ATR/percent hard stop and re-rotate only on the next signal. 3. RECONSIDER THE SHORT LEG (structural). In crypto, the highest-vol asset is frequently the one in a parabolic pump — shorting it daily is unbounded-loss by construction, unlike the equity low-beta studies cited. Consider testing a LONG-ONLY low-vol tilt (long lowest-vol, hold cash instead of shorting), or cap/■collar the short leg. DECISION RULE: re-run only after (1) and (2); if a stop-protected, correctly-sized version still relies on a short-highest-vol leg that bleeds in bull runs, abandon — the short side of the low-vol anomaly may be structurally unviable in crypto.

CryptoLowVolAnomalyRotationLS1D was the analyst-sanctioned replacement for the abandoned cross-sectional momentum rotation, betting that volatility ranks are more persistent than return ranks and tilting into safety (structurally toward BTC). Its original long-short form liquidated when the short-highest-vol leg blew up mid-parabola, so this iteration removed the short and held only the lowest-vol major — which turned the strategy into directional crypto beta (correlation 0.71, beta 0.43, alpha 0.045) rather than the intended anomaly. The +423% headline was half unrealized, the Sharpe was 0.50 with a CI straddling zero, and a 47.7% drawdown with a 2026 collapse of -44.8% pointed to holdout failure. Recognizing a structural dead-end — re-adding the short re-blows-up while long-only cannot capture the anomaly — the analyst abandoned the arm at the backtest-review gate, noting a genuinely risk-bounded market-neutral construction would require a fresh hypothesis.

The low-volatility anomaly is intrinsically a market-neutral long-short premium — stripping the short leg (correctly removed here because shorting the highest-vol crypto asset mid-parabola is unbounded-loss) leaves only directional beta, so there is no clean iteration path between a blow-up short and a beta-proxy long.

The analyst abandoned it at the backtest-review gate before optimization because the long-only salvage abandoned its own market-neutral anomaly thesis — with benchmark correlation 0.71, beta 0.43, and alpha only 0.045, it had become directional crypto beta (mostly holding BTC) rather than the low-vol premium, with a marginal CI-straddling-zero edge and a severe recent collapse likely to fail the chronological holdout.

An OHLCV-only low-volatility-anomaly rotation on a fixed universe of 5 Binance majors (BTC/ETH/SOL/BNB/AVAX) — originally a market-neutral long-short (long the lowest-30-day-vol asset, short the highest) as the analyst-sanctioned replacement for failed cross-sectional momentum, but by this iteration reduced to a long-only tilt holding only the lowest-vol major with a per-leg hard stop, after the short leg liquidated the prior version.

With the prior liquidation bug fixed (no liquidation, sane sizing), the base backtest showed +423% total return (CAGR 22.4%) over 92 trades with a 57.6% win rate, but only Sharpe 0.50 (CI low -0.19, straddling zero), profit factor 1.11, and a 47.7% max drawdown (CI high 96%); much of the headline was unrealized (end_unrealized_pct 212%), and recent years collapsed (2025 -11.8%, 2026 -44.8%).