BinanceSevenMajorMacroTrendBasketLS

Binance USD-M Multi-Major Macro Trend Basket, Long-Short (Absolute Time-Series 1D+1W Momentum Confluence Applied INDEPENDENTLY per Name to 7 Liquid Majors — NOT Ranked/Rotated — Weekly-Scale Holds, Equal-Risk Legs, Net Exposure Floats, MULTI-YEAR History, 3 Shared Parameters)

A LONG-SHORT, MULTI-INSTRUMENT, pure-OHLCV managed-futures-style MACRO trend basket across seven liquid Binance USD-M majors (BTC, ETH, BNB, SOL, XRP, ADA, LINK), each traded INDEPENDENTLY on its own 1D-primary + 1W-confirm momentum-sign confluence. It is engineered directly against the four failure modes I have now catalogued across ~12 dead single-name/short-history/confluence attempts this session, solving all four AT ONCE: (1) FEE_EDGE — the killer that just sank the BTC-HL confluence port (avg_trade 0.151%, on the fee line) and the alt ports (PF ~1.0): moving to a WEEKLY-scale horizon means each trade captures a 5–15% macro leg, an order of magnitude above the ~0.10% round-trip fee, so the edge clears the floor structurally instead of dying inside it. (2) UNMEASURABLE SAMPLE — the wall that killed low-turnover TSMOM (~50 trades): seven names each flipping several times per year over 4+ years yields 100–200 COMBINED trades, comfortably above the ~100-trade measurability floor, even though per-name turnover is low. (3) OVERFIT/PBO — the wall that killed the single names (PBO 0.64–0.73): averaging seven independent trend legs collapses the idiosyncratic variance of the portfolio return, which is the textbook suppressor of the multiple-testing noise-selection (lower PBO, higher deflated Sharpe); only 3 parameters, SHARED identically across all legs, keep dimensionality minimal. (4) DATA — Binance USD-M has 4+ years of clean OHLCV on all seven names, avoiding the short-history (HL), 90-day (liquidations), and cadence (options) data traps. CRITICALLY this is NOT the exhausted cross-sectional rotation: there is NO ranking and NO relative-strength comparison — each name is traded on its OWN absolute (time-series) trend, so a name can be long while another is short; this is diversified time-series trend-following (the CTA/managed-futures factor), a distinct construction from hold-top-N rotation. It is materially different from every pipeline item: the BinanceThreeMajorConfluenceTrendBasket is 4H+1D (mid-horizon, 3 names); THIS is 1D+1W (macro horizon, 7 names) — a different timescale, universe, and turnover profile. Fills the SINGLE rarest bucket in the portfolio: macro_1w_plus horizon (0.5%, just 10 of 1933), plus long_short (13.6%) and multi-instrument scope. Same venue, no cross-venue schema/spot-lookup landmine (the verification_loop death trap). Pure OHLCV, no supplementary feed. Adds to over-represented BINANCE — accepted: multi-year history is the proven survival ingredient and exists only there.

Implements the hypothesis's construction directly: one absolute time-series confluence rule, applied independently per name across seven majors, no ranking or rotation, equal-risk legs, floating net exposure, three shared parameters. I verified the data and the design before submitting. All seven daily series are gap-free, with SOL's 2020-09-14 start binding the fully-aligned window to 2,144 days (~5.9 years) — this matters because same-timeframe legs make the base class defer the basket until every leg reports, which is exactly the contemporaneous evaluation a portfolio wants but would silently freeze the whole book if any leg had an outage (a trap I hit on another venue where one constituent had a 32-day gap). At the shipped defaults with 0.10% round-trip fees the basket returns +87.5% over 5.9 years with a 19.0% max drawdown and Sharpe 0.86, and +2.1% with a 10.3% drawdown in the trailing year. Three of the hypothesis's four claimed fixes verify strongly: the fee floor is cleared by roughly an order of magnitude on a per-notional basis; the sample is far above the ~100-trade measurability floor; and the overfit-suppression claim is backed by the flattest surface I have measured in this family — all twelve cells of daily {20,30,60} × weekly {8,12} are positive in both the full and trailing-year windows, with no cliff and no lone peak, which is precisely what averaging seven independent legs is meant to buy. One claim does not hold as stated, and I want it on the record: the hypothesis projects 100–200 combined trades and multi-week holds, but the confluence drops to flat whenever the two horizons disagree, so legs cycle considerably more often — on the order of 1,900 state changes (roughly 900+ round trips) over the window. The fee math still clears comfortably, so this is not a threat to viability, but the realised profile is mid-horizon rather than the macro_1w_plus hold the title implies, and the portfolio-bucket claim should be judged on that basis. Two other notes: I derived the weekly confirm from daily closes rather than subscribing seven additional 1-WEEK feeds, which is mathematically identical (56 days is 8 weeks), halves the feed count and avoids the stale-weekly-series problem encountered elsewhere; and I chose 30d/8wk because it is the strongest cell on the longer 5.9-year sample rather than on the recent year, where longer lookbacks (60d/12wk) did better — the sandbox year is this configuration's weakest cell at +2.1%, so a modest recent-window result is the expected outcome, not a surprise.

Overfit with two UNWAIVABLE hard-gate failures: walk-forward is_overfitted=TRUE (IS 1.267 vs OOS 0.338, first OOS window -1.84) and PBO 0.5609 > 0.5. The hypothesis's central claim — that 7-leg diversification suppresses PBO/overfit — was directly falsified: PBO still exceeds 0.5 and WF still flags overfit, exactly where the 3-major sibling died. Deflated Sharpe 0.629 < 0.95 (not significant) confirms a best-of-N artifact. The full-sample edge is bull-market beta: +80%/2021, +63%/2024 but flat-to-negative in 2022/2023/2025/2026. Flat sensitivity + inconsistent OOS = mechanism/regime failure, not parameter failure — no robust region to iterate toward.

Long-short managed-futures-style macro trend basket across seven Binance USD-M majors (BTC, ETH, BNB, SOL, XRP, ADA, LINK) on 1-DAY bars. Each name is traded independently on its own absolute momentum confluence — long while both its 30-day and 8-week (56-day) returns are positive, short while both are negative, flat whenever they disagree — with no ranking, no relative-strength comparison and no rotation, so net exposure floats freely between all-long and all-short. The signal flip is the exit; there is no fitted stop. Each leg is sized by its own realized volatility toward a shared portfolio target (leg fraction = (20% target / that name's annualized vol) / 7, clipped to 0.5%–30% of equity), giving roughly 4% of equity per leg and ~0.22x average gross. Quantity precision is read per instrument since lot sizes differ across the universe. Leverage 1.0. Exactly three tunable parameters shared identically by all seven names (daily_lookback, weekly_lookback, vol_target_annual); vol window, days-per-week, leg caps and minimum leg notional are locked constants.

Record as mid-horizon (not macro_1w_plus) for bucket accounting; the fee/measurability math still clears. A genuine macro hold would need a trailing (not flatten-on-disagreement) exit — changes the mechanism, not required for viability.

TITLE/PORTFOLIO-BUCKET MISMATCH (honestly disclosed). The hypothesis claims to fill the rare macro_1w_plus horizon bucket with multi-week holds and 100-200 total trades, but the realized profile is MID-HORIZON: the confluence drops to flat whenever the two horizons disagree, so legs cycle far more often — ~1,900 state changes / ~900+ round trips over 5.9 years (160 in the sandbox year), ~2-week holds, not the multi-week macro hold the title implies. Fee viability is unaffected (avg_trade_return_pct +0.89%, ~10x the round-trip), but the bucket claim that this fills the 0.5% macro_1w_plus bucket is overstated — for quota accounting it sits closer to the mid-horizon class (like the existing 3-major 4H+1D basket).

Let the optimizer confirm the flat surface; because the sandbox is the weakest cell, judge the edge on the full 5.9-year backtest (Sharpe 0.86), not the trailing year.

PARAMETER SELECTED ON THE FULL SAMPLE. The shipped 30d/8wk pair is the strongest cell on the 5.9-year sample; on the trailing year longer lookbacks (60d/12wk) did better, so the sandbox year is this config's WEAKEST cell (+2.1% dry-run / +4.5% engine). Mild selection, well-mitigated: the full 12-cell surface (daily {20,30,60} x weekly {8,12}) is ALL positive in both windows with no cliff/lone peak, so choosing the best cell of an all-positive flat surface is not a fragile fit — and the sandbox UNDERSTATES the full-history result.

PM should size with correlation to the existing confluence strategies in mind; analyst should judge on the full 5.9-year history + walk-forward.

CORRELATED BET + single-year non-decisiveness. Shares its momentum-confluence mechanism with the existing 3-major 4H+1D basket and the single-name confluence ports, so returns are correlated with that book (PM sizing consideration, not a defect). The single-year sandbox is non-decisive (Sharpe +0.33, CI [-1.38, 1.93] straddling zero), though measurable (160 trades) and positive.

Sharpe 0.913 with CI [0.304, 1.534] — lower bound POSITIVE (significant), PSR 0.998; the first basket in this family whose CI does not straddle zero

profit_factor 1.609 (> 1.2, clears L9); large decisive sample (1002 trades, 5.9 years), two-sided (455 long / 547 short)

avg_trade_return_pct 3.86% (~26x fee floor); no capacity blocker (capacity $457M, impact 1.48%)

Regime-robust: positive in 5 of 7 years with shallow down years (2022 -1.0%, 2023 -2.9%) — macro horizon + 7 independent legs delivered the diversification thesis

Returns lean on 2021 (+79%) and 2024 (+63%) bull years with some outlier-day dependence (kurtosis 17.6) — post-opt PBO/deflated-Sharpe/holdout must be scrutinized, as this family repeatedly overfits at that stage

Sharpe 0.913 is below the 1.5 promotion floor; optimization must lift it robustly, not by curve-fitting the tail

BinanceSevenMajorMacroTrendBasketLS was a seven-major, long-short managed-futures basket explicitly engineered to beat the four failure modes that had killed roughly a dozen prior single-name and confluence attempts, chiefly by moving to a weekly-scale horizon and averaging seven independent legs to clear fees and cut overfitting. Its first backtest was the strongest in the family — Sharpe 0.913 with a positive CI lower bound, profit factor 1.609, and 3.86% avg trade return over 1,002 trades — earning an 'optimize' verdict despite noted reliance on 2021 and 2024 bull years. Optimization improved the headline numbers (Sharpe 1.11, 10.1% drawdown) but exposed the flaw: walk-forward flagged overfitting with a -1.84 first OOS window, PBO stayed at 0.561, and deflated Sharpe fell to 0.629. The analyst abandoned it after one iteration, concluding the central diversification thesis was directly falsified and the edge was bull-market beta with no robust region to iterate toward.

Averaging seven independent trend legs did not structurally suppress PBO or walk-forward overfitting, so diversification alone cannot rescue a crypto trend edge that is concentrated in bull-market years — the mechanism failed out-of-sample, not the parameters.

The analyst abandoned it after optimization on two unwaivable hard-gate failures — walk-forward is_overfitted=TRUE and PBO 0.5609 > 0.5 — with deflated Sharpe 0.629 below the 0.95 significance mark, judging the full-sample edge to be bull-market beta (strong 2021/2024, flat-to-negative 2022/2023/2025/2026).

A long-short, CTA-style macro trend basket that ran an absolute time-series momentum confluence (30-day daily + 8-week weekly) independently on seven liquid Binance USD-M majors (BTC, ETH, BNB, SOL, XRP, ADA, LINK), betting that a weekly-scale horizon plus seven diversified legs would clear trading fees and suppress overfitting.

The initial 5.9-year backtest returned +254.3% with Sharpe 0.913 (CI lower bound positive at 0.304), profit factor 1.609, avg trade return 3.86% of notional across 1,002 trades and 19.4% max drawdown; optimization lifted the backtest Sharpe to 1.11 with 10.1% drawdown, but walk-forward flagged overfitting (IS Sharpe 1.267 vs OOS 0.338, first OOS window -1.84) with PBO 0.561 and deflated Sharpe 0.629.