SpotMajorsRebalancingPremiumHarvestBasket

Volatility-Harvesting Rebalancing-Premium Basket, Long-Only Spot (6 Imperfectly-Correlated BINANCE_SPOT Majors: BTC/ETH/SOL/BNB/XRP/DOGE — Hold Fixed Risk-Parity Weights and Systematically Rebalance Back to Target on a Drift BAND, Mechanically Selling Relative Winners / Buying Relative Losers to Harvest the Diversification Return; a Structural Non-Predictive Edge, 2-Parameter)

A LONG-ONLY, NO-LEVERAGE, MULTI-INSTRUMENT basket on 6 liquid BINANCE_SPOT majors (BTC, ETH, SOL, BNB, XRP, DOGE) — placed on the under-represented SPOT venue (11.9% vs the >=15% target), OHLCV-only so no supplementary-feed risk, same-venue so no L44 problem. It is a genuinely DIFFERENT edge family from everything now saturated and dying here: it is NOT cross-sectional momentum/reversal (those are collapsing overfit/fee_edge), NOT a positioning FADE (all falsified), NOT carry, and critically NOT a directional price-PREDICTION at all — so it cannot be 'premise-falsified' the way the signal strategies keep being. The edge is the REBALANCING PREMIUM (a.k.a. volatility harvesting / the diversification return / Shannon's demon): a basket of volatile, imperfectly-correlated assets held at FIXED target weights and periodically rebalanced back to those weights mechanically SELLS whichever names have risen in relative terms and BUYS whichever have fallen, capturing a return ABOVE the weighted buy-and-hold average that grows with the constituents' variance and shrinks with their correlation. Crypto is close to the ideal environment for it — very high idiosyncratic volatility and imperfect (and time-varying) pairwise correlations among majors — which makes the harvest materially larger than in equities. This is distinct from the deployed momentum survivor and from my in-flight books (regime-switch, mid-cap residual momentum, dual-TF momentum, vol-target passive long): those are directional/relative-strength or single-asset risk-scaling; this is a purely mechanical contrarian-rebalancing harvest across a diversified basket. Threshold-BAND rebalancing (only rebalance when a weight drifts materially) keeps turnover — and the ~0.20% spot round-trip — low. Just 2 parameters.

Implements the hypothesis as a purely mechanical, non-predictive harvest. WEIGHTS: risk parity via inverse realized volatility, normalized to sum to 1, computed from a deliberately LONG 90-day window so the targets are slow-moving and effectively fixed - they are risk-based, never return-based, so the book never expresses a relative-strength or directional view; equal risk contribution also maximizes the harvest, which scales with the weighted variance actually held. TRIGGER: band, not schedule - the book is only traded when max_i |actual_i - target_i| exceeds drift_band, and then every leg is restored in one pass with sells submitted before buys so proceeds fund the purchases. calculate_signal returns that max drift every bar, a continuous decision variable that varies bar to bar and is thresholded against drift_band; before the first allocation it equals the largest target weight, which is what triggers the initial buy pass (no special-cased entry path). VENUE: long-only, no leverage, at most fully invested, so per the routing rule this belongs on BINANCE_SPOT (CASH, leverage 1.0); all six legs are on the same venue so there is no cross-venue leg problem, and only OHLCV is used so there is no supplementary-feed risk. FEES: I measured the actual behaviour on the aligned 6-major BINANCE_SPOT daily panel (2,181 aligned days, 2020-08-11 to 2026-07-31). At vol_lookback=90 / drift_band=0.05 the band fires 82 times over the full history and 9 times in the last 365 days, with ~2.1x book turnover per year - so at the 0.10% spot one-way fee the strategy pays on the order of 0.2% per year in commissions, a small fraction of the harvest, and the 365-day sandbox window (92-bar warmup, ~273 tradable days) still produces roughly 7 rebalance events and ~40 leg orders. PARAMETERS: exactly two searched knobs, vol_lookback and drift_band; the cash buffer and minimum order notional are fixed design constants. SAFETY: alignment is a single synchronized panel trimmed as one unit so index -k is the same calendar day for every leg; flat legs fall back to equal weight instead of dividing by ~0; sells are floored to 6 dp and clamped to the holding so a rebalance can never become a phantom spot short.

Not worth optimizing. (1) The core risk/diversification thesis is falsified: max_drawdown is 73.6% (CI high 89.3%) — a hard-floor violation and far worse than a diversified, rebalanced book is supposed to deliver — and the payoff test that matters, beating the equal-weight buy-and-hold basket (benchmark_meaningful=true), FAILS with information_ratio -0.32 (negative). The rebalancing premium is real but sub-1%/yr and is swamped here; the strategy just holds 6 volatile majors and eats their full drawdown. (2) It is a closet long-basket beta play (beta 0.42, correlation 0.61) — the 2883% headline is the majors appreciating over 9 years, not harvest; judged on alpha/IR vs its basket benchmark it underperforms. (3) It is structurally unmeasurable: positions never close (6 hold-forever trades, 2090-day average hold) and the daily-return series collapsed to a single end-of-backtest point across a 9-year window, so the 3-phase optimization's IS/OOS windows and 15-day holdout cannot meaningfully populate and Sharpe/DD are fragile. Optimizing 2 parameters cannot turn a 74%-drawdown, negative-IR closet-beta basket into a promotable risk-management edge, nor fix the degenerate measurement. Abandon at BACKTEST_REVIEW rather than spend 2 hours.

Long-only, unlevered 6-asset BINANCE_SPOT basket (BTC primary, plus ETH, SOL, BNB, XRP, DOGE) on 1-DAY bars, held at risk-parity (inverse-volatility) target weights computed from a slow 90-day trailing window and restored to those weights whenever any legs actual weight drifts more than drift_band (5% absolute) from target. Restoring fixed weights mechanically sells whatever has risen in relative terms and buys whatever has fallen, which is the rebalancing premium / volatility harvest: a structural return above the weighted buy-and-hold average that grows with constituent variance and shrinks with correlation. Nothing in the code forecasts direction. Every SELL only reduces an existing holding (quantities floored and clamped to the holding), the basket is at most fully invested with a 1% USDT cash buffer, and the rebalance trigger is a pure function of current prices and holdings so a restart re-derives the same decision.

Evaluate risk-adjusted vs weighted buy-and-hold over the full multi-year history, not one down-market window.

Sandbox metrics (win_rate 0.0, PF 0.0, total_trades 6, avg_trade_return_pct -29.07%, total_return -31.65%) are NOT the L17 code-defect signature. trades_count is 30 fills across ~7 rebalance events (matches the developer's estimate), end_unrealized_pct is positive, avg_holding_period 182d — an actively-rebalanced long-only basket held through a falling market, so the 6 closed long legs are losers by construction. |Sharpe|=0.93 (<5), spot venue (no inverse-contract bug), polarity structurally correct → beta/edge outcome, not a defect.

Backtest engineer/analyst: confirm the strategy P&L series reflects the ~99% invested basket and the benchmark is computed on the same constituents/window.

Anomaly to sanity-check (not a demonstrable defect): benchmark_correlation 0.1286 / beta 0.069 vs the equal-weight basket of the SAME six constituents is surprisingly low for a fully-invested long basket. The 30 fills confirm active investment (not stuck-in-cash), and risk-parity + drift noise decorrelates somewhat, but the magnitude warrants a quick verification.

No code change; analyst must judge whether the harvest net of costs is real and large enough given full long-crypto beta.

Efficacy caveat (not a fault): the rebalancing premium does NOT hedge the dominant common crypto beta — the six majors are highly correlated, so a long-only fully-invested basket carries full market drawdown (35.8%). The 'materially larger harvest' claim must be weighed against that unhedged beta.

Correct venue routing (long-only/no-leverage → BINANCE_SPOT, under-represented bucket); OHLCV-only, same-venue, no data risk

Genuinely different, non-predictive edge family (rebalancing premium) — cannot be premise-falsified like signal strategies; only 2 parameters, huge capacity

Clean risk-parity + drift-band construction

max_drawdown 73.6% (CI high 89.3%) — catastrophic and a hard-floor violation; the diversification/risk thesis is falsified (the ride was not smoothed)

information_ratio -0.32 (negative) vs the equal-weight basket it must beat — the rebalancing premium did not beat weighted buy-and-hold on a risk-adjusted basis

Closet long-basket beta play: beta 0.42, correlation 0.61; the 2883% return is the 6 majors appreciating, not harvest

Structurally unmeasurable: 6 hold-forever positions (2090-day holds), daily_returns collapsed to a single point over a 9-year backtest — walk-forward/holdout cannot populate

return_kurtosis 38.6 — reported metrics rest on a degenerate return series

This strategy reached for a genuinely different, non-predictive edge — the rebalancing premium (Shannon's demon): hold a diversified risk-parity basket of crypto majors and mechanically rebalance back to target on a drift band, capturing the diversification return that grows with variance and shrinks with correlation. The construction was clean and correctly venue-routed, but the backtest falsified the thesis on every axis that matters: a catastrophic 73.6% drawdown showed the ride was not smoothed, and a negative information ratio meant the harvest did not beat the equal-weight buy-and-hold basket it was designed to improve — the 2883% headline was simply the six majors appreciating at partial exposure (beta 0.42). The analyst abandoned it at backtest review as a closet long-basket beta play that is also structurally unmeasurable (six hold-forever positions collapsing the return series to a single point over 9 years), concluding that the real but sub-1%/year premium was swamped and that optimizing two parameters could neither create the edge nor fix the degenerate measurement.

The rebalancing premium is real but sub-1%/year and is swamped by holding six highly-volatile, still-substantially-correlated majors that inflict their full ~74% drawdown — so a mechanical harvest that neither smooths the ride nor beats its own buy-and-hold basket risk-adjusted is a closet-beta long, and hold-forever positions also leave the result unmeasurable for the downstream gates.

It was abandoned at the BACKTEST_REVIEW gate for three reasons: the core risk thesis is falsified (74% drawdown, far worse than a diversified rebalanced book should deliver), it fails on a risk-adjusted basis against the buy-and-hold basket it must beat (negative IR), and it is structurally unmeasurable (positions never close, so the 3-phase optimization's IS/OOS windows and 15-day holdout cannot populate). Optimizing 2 parameters cannot turn a 74%-drawdown, negative-IR closet-beta basket into a promotable edge, so it never reached optimization.

A long-only, no-leverage, multi-instrument spot basket of 6 imperfectly-correlated BINANCE_SPOT majors (BTC, ETH, SOL, BNB, XRP, DOGE) held at fixed risk-parity target weights and rebalanced back to target on a drift band — mechanically selling relative winners and buying relative losers to harvest the rebalancing premium (volatility harvesting / diversification return), a structural non-predictive edge rather than a directional signal.

The diversification/risk thesis was falsified and the result was structurally unmeasurable: max drawdown 73.6% (CI to 89.3%, a hard-floor violation), and it failed the test that matters — beating its equal-weight buy-and-hold basket — with information ratio -0.32 (negative). It was a closet long-basket beta play (beta 0.42, correlation 0.61) whose 2883% headline is the majors appreciating over 9 years, not harvest, and the metrics rest on a degenerate series: 6 hold-forever positions (~2090-day holds) with the daily-return series collapsed to a single end-of-backtest point.