DualMomentumLongRotation8Majors

Dual-Momentum Long-Only Rotation over 8 USD-M Majors (BINANCE Perps, Weekly, Hold Top-3 by Relative Strength ONLY If Their Own 12-Week Return Is Also Positive — Absolute-Momentum Cash-Out Overlay, Pure OHLCV, 2-Parameter)

A LONG-ONLY, weekly-rebalanced DUAL-MOMENTUM rotation over the same 8 data-verified USD-M majors that already produced a paper-stage winner (BTCUSDT, ETHUSDT, SOLUSDT, BNBUSDT, AVAXUSDT, LINKUSDT, ADAUSDT, XRPUSDT). It deliberately extends the promoted 'Cross-Sectional Momentum 8-major LONG' (Sharpe 2.6) with the one well-documented, mechanically-simple robustness overlay it lacks: an ABSOLUTE-momentum (time-series) gate. Pure relative-strength rotation stays fully invested even in bear markets, holding 'the best of a falling universe'; dual momentum (Antonacci) fixes exactly this by requiring a name's OWN trailing return to be positive before it can be held — otherwise that capital goes to cash. This is a distinct mechanism from the existing regime-gated rotation (which uses a single market-wide BTC-SMA switch): here each slot is gated INDEPENDENTLY on its own absolute momentum, so the book scales its net exposure smoothly from fully-invested (broad bull) to fully-cash (broad bear). It is engineered against this session's lessons: PURE OHLCV only (best-surviving tag, 5%; no OI/liquidation/funding data walls), MULTI-ASSET (escapes the 0/213 single-symbol no-edge class that just killed the SOL squeeze), STANDARD BINANCE USD-M venue and a plain rank-and-hold structure (avoids the venue-mismatch and 10-parameter implementation failures that sank my COIN-M and taker-flow proposals at Layers 2/4). Only 2 tunable parameters (momentum lookback, top-K) to resist overfit (202x killer).

Iteration 2 makes the smallest change that fixes the Layer-4 QA block (venue misroute) while preserving every earlier-passing layer. The strategy is long-only, leverage 1.0, no shorts, with a pure-OHLCV momentum edge (not funding), so per the factory venue rule it must run on BINANCE_SPOT, not USD-M perps. On USD-M the backtest got perp-cheap fees AND zero funding — a combination that doesn't exist live (real USD-M longs pay ~0.03%/day funding in bull markets, tens of % over the span), making the +2078%/Sharpe 3.8 result optimistic and non-deployable. Two changes only: (1) all instruments/bar-types rerouted from .BINANCE to .BINANCE_SPOT (CASH account, leverage forced to 1, no funding applied — costs now match reality); (2) _portfolio_value switched from the futures assumption (equity is a stable base) to the spot-correct computation (a CASH account's get_account_equity() returns only the USDT balance, which shrinks as coins are bought, so I add the mark value of held coins to recover true equity and keep the equal-weight target stable). The base already blocks any spot short at the submission layer, and all order flow here is BUYs plus strict long-reducing SELL trims/exits, so it is long-only-safe on the CASH venue. The dual-momentum signal, weekly-cadence rebalance, per-slot absolute-momentum gate, fixed-denominator equal-weight cash scaling, and exits-before-buys ordering — everything that passed Layers 1-3 — are unchanged. The QA's secondary note (84-day gate too slow to dodge fast crashes; 70.8% drawdown) is a thesis-quality concern the QA explicitly deferred to the analyst post-reroute; I kept the hypothesis's specified 12-week absolute gate and 2-parameter design rather than altering the thesis, so the analyst can judge the overlay on the now-realistic spot backtest.

The hypothesis's central claim is falsified by its own backtest: the per-slot absolute-momentum cash-out overlay is meant to fix rotation drawdown, but it produced max_drawdown 76.9% (CI high 93.2%) — decisively over the 50% hard-abandon line (L19) and FAR WORSE than the sibling regime-gated SPOT rotation (33.7% DD, Sharpe 4.83) and the promoted perp rotation (46% DD). Root cause: an 84-day per-slot trailing-return gate on ~0.6-correlated crypto majors is a slow, noisy substitute for a market-wide gate — the majors crash together, so all slots' 12-week returns stay positive deep into a decline and flip late/whipsaw (2022 -45%, a -33.6% single week in 2020-03). Iteration cannot rescue this: shortening the lookback or adding a market-wide gate makes it converge to the already-promoted regime-gated rotation (a duplicate), while keeping the per-slot slow gate keeps posting >50% drawdowns. It is also strictly dominated by that sibling on Sharpe and drawdown, with sharpe_ci_low just 0.104. Per L19, >50%-DD momentum baskets have burned 5-7 iterations here without surviving. Recommend the Research Lead treat the market-wide BTC-SMA regime-gated rotation (fcb46ca9) as the canonical drawdown-controlled expression of this edge and not pursue per-slot absolute-momentum overlays on correlated crypto majors — the correlation structure makes per-slot gating strictly inferior to a market-wide switch.

Long-only, weekly-rebalanced dual-momentum rotation over 8 BINANCE_SPOT majors (BTC, ETH, SOL, BNB, AVAX, LINK, ADA, XRP) on daily bars, pure OHLCV. Each week it ranks the universe by trailing 84-day (12-week) return and takes the top 3 by relative strength, but holds a name only if its OWN 12-week return is also positive (Antonacci absolute-momentum gate); names failing the absolute gate go to cash. Held names are equal-weighted at gross/top_n with a FIXED denominator, so net exposure scales smoothly from fully invested in broad bull markets down to fully cash in broad bear markets. Retained winners are re-equalized each cycle against total portfolio value; losers are sold before winners are bought. Spot, long-only, no leverage, no shorts, no supplementary data. 2 tuned parameters (lookback_days, top_n).

Verification failed (Layer 4 — QA review): - VENUE MISROUTE + OPTIMISTIC FUNDING-FREE BACKTEST. Long-only, leverage 1.0, no shorts, pure-OHLCV momentum edge (not funding) → per the factory rule it must run on BINANCE_SPOT (futures only when leverage>1, shorts, OR funding-edge). It's on USD-M perps. funding_events_available=0 means the backtest applied ZERO funding, but real USD-M longs PAY funding (~0.03%/day in bull), tens of % over the span — dwarfing the 0.05-vs-0.10% taker saving cited. USD-M backtest gets perp-cheap fees AND zero funding, a combo that doesn't exist live → +2078%/Sharpe 3.8 is optimistic and non-deployable. Same 'backtest doesn't reflect real venue costs' principle that blocked the COIN-M iterations. - The absolute-momentum overlay did NOT deliver its promised drawdown protection: max_drawdown 70.79% (vs dip basket 19.7%, carry basket 31.7%), beta 0.46 — MORE directional, not less. An 84-day gate is too slow to dodge fast crashes. Sharpe 3.8 is bull-beta-driven, not proof the overlay works. Undercuts the thesis; analyst should scrutinize after reroute. - Non-blocking: data_days=109 under-reports true span; exposure_pct 211% is futures time-in-market not a leverage breach; frozen-signal Layer-2 warnings benign (L3 made 133 trades).

Likely dominated by the promoted regime-gated sibling; strong abandon candidate unless a shorter gate lookback cuts the DD. Don't optimize a 77%-DD, CI-near-zero series as-is.

Absolute-momentum overlay STILL fails its core promise on the honest spot backtest: max_drawdown 76.95% (worse than iter-1's 70.79%), beta 0.50. Decisively, the EXISTING regime-gated spot rotation achieves 33.74% DD on the same universe — less than HALF. The hypothesis's central bet (per-slot 12-week absolute gate beats a market-wide BTC-SMA switch for drawdown control) is empirically FALSIFIED: the 84-day gate is too slow to de-risk in a crash. Sharpe 2.55 is bull-beta (CI lower bound 0.10). The thesis-quality call I deferred last time — now judge on realistic spot data.

Non-blocking: exposure_pct 198% is a non-gross artifact (unlevered CASH, avg_position 37.8%/name ~ 0.90x gross); data_days under-reports span; frozen-signal warnings benign; heavy overlap with existing regime-gated rotation is a PM dedup matter.

No data wall: pure OHLCV, 163 trades over a full 2017→2026 window, metrics_reliable=true; extends a proven promoted family

Positive headline in bull regimes (2019 +139%, 2021 +780%, 2024 +154%), PF 1.49, avg_trade_return_pct 6.36% above fee floor

Falsifies its own premise: the absolute-momentum overlay is meant to CONTROL drawdown, yet max_drawdown is 76.9% (CI high 93.2%) — far worse than the regime-gated sibling's 33.7% and the promoted perp sibling's 46%

max_drawdown 76.9% is decisively over the 50% hard-abandon line (L19); DD-CI high 93.2%, max_consecutive_losses 13

Strictly worse than the already-promoted/in-pipeline regime-gated rotation on both Sharpe (2.55 vs 4.83) and drawdown (77% vs 34%); sharpe_ci_low just 0.104

Mechanism is structurally inferior for correlated crypto: a per-slot 84-day absolute-momentum gate on ~0.6-correlated majors is a slow, noisy version of a market gate — it can't diversify the cash-out (majors crash together and the gate flips late)

DualMomentumLongRotation8Majors extended the factory's promoted cross-sectional momentum rotation with the one robustness overlay it lacked — an Antonacci-style per-slot absolute-momentum gate that sends a name's capital to cash unless its own trailing return is positive — built cleanly against the session's lessons (pure OHLCV, multi-asset, standard venue, just two parameters). But the backtest falsified the premise it was designed around: rather than controlling drawdown, the overlay posted a 76.9% max drawdown (CI high 93.2%), because on tightly-correlated majors that crash together the 12-week per-slot gate flips late and whipsaws, leaving it strictly worse than the market-wide regime-gated sibling on both Sharpe and drawdown. The reviewer abandoned it over the 50% hard-abandon line, noting that any fix either converges to the existing regime-gated rotation (a duplicate) or keeps posting >50% drawdowns. The recommendation was explicit: treat the market-wide BTC-SMA regime-gated rotation as canonical and stop pursuing per-slot absolute-momentum overlays on correlated crypto majors.

On ~0.6-correlated crypto majors a per-slot absolute-momentum gate is a slow, noisy substitute for a market-wide switch — the majors crash together so every slot's 12-week return stays positive deep into a decline and flips late, making per-slot gating structurally inferior; the market-wide BTC-SMA regime-gated rotation is the canonical drawdown-controlled expression of this edge.

The backtest reviewer returned 'abandon' at the pre-optimization gate: the hypothesis's central claim was falsified by its own backtest — the drawdown-control overlay produced a 76.9% drawdown, decisively over the 50% hard-abandon line and strictly dominated by the already-promoted regime-gated rotation on both Sharpe (2.55 vs 4.83) and drawdown (77% vs 34%).

A long-only, weekly-rebalanced dual-momentum rotation over 8 Binance majors that held the top-3 by relative strength but only if each name's own 12-week trailing return was also positive — an Antonacci-style per-slot absolute-momentum cash-out overlay (pure OHLCV, 2 parameters) meant to extend the promoted cross-sectional momentum rotation with independent bear-market de-risking.

Over a full 2017-2026 window with 163 trades (metrics_reliable=true) the headline was strong in bull regimes — Sharpe 2.55, profit factor 1.49, avg_trade_return_pct 6.36%, positive in 2019/2021/2024 — but the absolute-momentum overlay failed at its core purpose: max drawdown was 76.9% (CI high 93.2%) with 13 consecutive losses, far worse than the regime-gated sibling's 33.7% and the promoted perp rotation's 46%. No optimization stage ran.