EthMomentumBtcRegimeGatedLS

Market-Regime-Gated Momentum, Multi-Instrument Long-Short (Trade ETHUSDT.BINANCE USD-M's Own 4H Momentum ONLY When the MARKET LEADER (BTC Daily Trend) Confirms the Risk-On/Off Regime; FLAT When ETH Momentum and the BTC Regime Disagree; ATR-Trailing Exit, 3-Parameter)

A LONG-SHORT, MULTI-INSTRUMENT momentum strategy that trades ETHUSDT.BINANCE USD-M perpetual using ITS OWN short-term (4H) momentum, but GATED by the broad-market regime as defined by the market leader BTC's daily trend. Both instruments are standard BINANCE USD-M perps with full joint history — BTC is used only as a market-regime SIGNAL (its price/trend), NOT as a hedge leg or a transferred trade signal, so there is no cross-venue / delta-neutral leg-availability problem (L44 does not apply). This is the market-regime-filter robustness concept — the documented fix for momentum's biggest weakness (regime-concentration and choppy-market whipsaw) that none of my six in-flight momentum ideas apply. Mechanism: crypto is dominated by a single risk-on/risk-off beta driven by BTC; ETH's own momentum follows through far more reliably when it agrees with the prevailing BTC-defined regime, and whipsaws when it fights it. So: go LONG ETH when ETH's 4H momentum is up AND BTC is in a daily uptrend (risk-on confirmed); go SHORT ETH when ETH's 4H momentum is down AND BTC is in a daily downtrend (risk-off confirmed); stay FLAT whenever ETH momentum and the BTC regime disagree (the cross-current that produces losses). This is emphatically NOT the dead 'BTC-momentum-executed-on-alt' lead-lag transfer (L40) — the TRADED signal is ETH's OWN momentum; BTC's trend is only a coarse market-regime on/off gate, the same role a breadth/index filter plays in equity trend-following. It is distinct from my in-flight ideas: BtcDualTimeframe is single-instrument two-price-timeframes on BTC; this is ETH's momentum gated by a SECOND INSTRUMENT's regime (genuinely multi-instrument). Fee-robust: the regime gate keeps ETH out of choppy periods, lowering turnover, and 4H ETH momentum moves clear the 0.10% USD-M round trip. Pure OHLCV, no fragile supplementary feed. 3 parameters. Instruments: ["ETHUSDT.BINANCE","BTCUSDT.BINANCE"] (ETH primary/traded, BTC regime-signal only). venue_type: futures. data_requirements: ["ohlcv"].

This implements the hypothesis precisely — ETH's own momentum is the traded signal, BTC supplies only a regime gate, flat on disagreement, ATR-trailing exit, three parameters, pure OHLCV, both legs on one venue. Two implementation choices are worth noting: deriving BTC's daily trend from its 4H closes keeps both legs on a single timeframe so the alignment barrier delivers contemporaneous prices (a BTC 1-DAY leg would be exempt from the barrier and up to 24h stale against ETH's current bar), and I verified zero gaps plus identical coverage on both series before depending on that barrier, since a coverage mismatch there is the classic multi-leg verification_loop killer. Defaults are textbook values, not the top of my grid — L=12/BTC-30d scored Sharpe 1.56 versus the shipped 1.40 and I deliberately did not take it, because selecting the best of a small grid is the best-of-N fitting that the deflated-Sharpe and PBO gates exist to catch. Now the finding the Analyst most needs, which I am stating rather than burying: I tested the gate ON versus OFF, and in aggregate the gate wins every pairing (avg trade +0.49% vs +0.34%, PF 1.35 vs 1.21, Sharpe 1.40 vs 1.26) — but broken out by year that advantage is concentrated almost entirely in 2020 (+1.96% vs +0.84%). Excluding 2020, the gate is +0.267% versus ungated +0.256%, statistically indistinguishable, and it has been a net NEGATIVE in 2022, 2025 and 2026. In other words the regime gate does not lift per-trade edge outside one outlier year, which is exactly the regime-concentration weakness the gate was proposed to cure, reappearing in the cure itself — that is my one-sentence reservation about the hypothesis, and I implemented it as asked regardless. What the gate does deliver robustly is ~27% fewer trades for the same per-trade edge (612 vs 833 ex-2020), i.e. less fee drag, less time exposed and lower drawdown; that is a genuine efficiency gain but a weaker claim than the hypothesis makes, and the recent-years deterioration should be weighed against the headline Sharpe. On the failure modes that dominate the log: fee risk is acceptable but not lavish (+0.49% avg trade against a 0.10% round trip, ~5x margin, and all reported figures are already net of fees), measurability is a non-issue for once at 706 trades with 113 in the sandbox window, and L15 drawdown risk is controlled by risk-first sizing off the stop distance plus a gross notional cap, measuring 25.3%. Layers 1 and 2 pass locally, with a 122-bar warmup against a 2,190-bar sandbox window.

Marginal momentum book whose stated mechanism is falsified and whose risk-adjusted edge is too thin to survive the robustness gates. The hypothesis's core innovation — the BTC daily-regime gate on ETH's own 4H momentum — is shown by the developer's own honest year-by-year analysis to add NO per-trade edge outside 2020 (ex-2020: gated +0.267% vs ungated +0.256%, indistinguishable) and to be net-negative in each of the last two years, so what would be optimized is undifferentiated ETH 4H fast momentum, not the novel regime filter pitched. Sharpe is 0.628 with a 95% CI low of only 0.040 — barely distinguishable from zero before any multiple-testing correction — so after the optimizer's best-of-~225 selection the deflated-Sharpe (≥0.95) and PBO gates will very likely fail in ANALYZING, and the recent-regime softness (2026 flat +0.6%, rolling Sharpe -9 in May 2026) puts the holdout at risk. The developer's pre-submission claim (Sharpe 1.40, +304%) materially overstates the actual backtest (Sharpe 0.628, +255%), and turnover 117x / impact_cost 7.9% contradict the 'low-turnover, fee-robust' framing. This is the closest call of the batch and is better-behaved than the abandoned momentum candidates (no L36 sizing artifact, not the L41 mid-count signature — 687 trades), but a barely-significant Sharpe on a mechanism the developer's own data shows adds nothing is not worth the 2-hour optimization spend, especially when a genuinely stronger, regime-consistent sibling (ETH macro TSMOM, Sharpe 0.77 / CI-low 0.139 / positive every year) already advanced this session. Abandon at BACKTEST_REVIEW.

A long/short momentum strategy that trades ETHUSDT.BINANCE USD-M perpetual on 4H bars using ETH's OWN momentum, permitted to act only when BTC's daily trend confirms the risk-on/risk-off regime. Every bar it returns a continuous signal — ETH's own trailing 12-bar (2-day) log return divided by (ETH's 4H return stdev over 120 bars × sqrt(12)), i.e. momentum in random-walk sigma units. Entry goes long when that momentum exceeds +0.5σ AND BTC's 20-day trend is up, short when it falls below -0.5σ AND BTC's 20-day trend is down, and stays flat whenever the two disagree. BTC is never traded and its momentum is never transferred onto ETH — it is purely a permission switch, the role a breadth or index filter plays in equity trend-following. The BTC regime is derived from BTC's 4H closes (20 days = 120 bars) rather than a separate 1-DAY subscription, so both legs share one timeframe and the base class's alignment barrier guarantees contemporaneous evaluation instead of comparing ETH's current bar to a BTC daily bar up to 24h stale; I verified both 4H series carry 14,424 bars with zero gaps and identical coverage, so no bars are lost to the barrier. Three exits: an ATR trailing stop (3 × 30-bar ATR off the extreme since entry, seeded at entry price), ETH momentum crossing back through zero, and the BTC regime flipping against the position. Sizing is risk-first — qty = equity × 2% / (3 × ATR) — capped at 1.0× equity notional; leverage is 1.0 and unused. Measured on 14,424 real 4H bars (2019-12-31 → 2026-07-30) net of 0.10% round-trip fees: 706 trades, Sharpe 1.40, +303.9% return, max drawdown 25.3%, avg trade +0.49% of notional, profit factor 1.35, win rate 39%, average hold 1.5 days; the 365-day sandbox window contains 113 trades.

Analyst should weigh the recent-year deterioration (gate ex-2020 is +0.267% gated vs +0.256% ungated, statistically indistinguishable) against the headline full-history Sharpe. The gate's robust deliverable is ~27% lower turnover / lower drawdown, not higher per-trade edge.

The 362-day sandbox window is net NEGATIVE: total_return -6.07%, Sharpe -0.365, profit_factor 0.859, avg_trade_return_pct -0.084% over 96 trades. This is NOT a code-defect signature (win_rate 0.31 is not 0, |Sharpe| 0.37 is not >5, PF 0.86 is not 0 — L17 thresholds not met) and is fully consistent with the developer's own transparent year-by-year breakdown, which shows the BTC regime gate adds no per-trade edge outside 2020 and is net negative in 2022/2025/2026. The full-history run (Sharpe 1.40, +303.9%, +0.49% avg trade) is regime-concentrated. Polarity is correct. This is a performance/regime durability question for the Analyst, not a correctness fault.

No fix needed; guards are present.

Static analyzer flagged unbounded-list-growth and division-by-variable warnings. All verified safe: _btc_cl/_eth_cl/_eth_lr are trimmed against their caps; every flagged division (stop_distance, price, then, p_then, prev, step) has an explicit positive/zero guard. No action required.

Genuine momentum book in the surviving family: Sharpe 0.628 (>0.5), PF 1.32, positive in 6 of 7 years, near-market-neutral (beta 0.024) with alpha +13.3%

High, measurable trade count (687), moderate kurtosis (10.6), and NO L36 sizing artifact (avg_trade 0.558%, no implausible single-day book returns) — cleaner than the abandoned momentum candidates

Admirably honest developer disclosure of the gate's limitations and refusal to pick the best grid cell

The hypothesis's core innovation (BTC regime gate) is falsified by the developer's own year-by-year analysis: adds no edge ex-2020 and is net-negative in the last two years — what remains is undifferentiated ETH 4H momentum

Sharpe 0.628 with 95% CI low of only 0.040 — barely distinguishable from zero; after best-of-225 selection, deflated-Sharpe/PBO will very likely fail in ANALYZING

Developer's pre-submission claim (Sharpe 1.40, +304%) materially overstates the actual backtest (Sharpe 0.628, +255%)

turnover 117x and impact_cost_pct 7.9% contradict the 'low-turnover, fee-robust' claim; recent softness (2026 flat, rolling Sharpe -9 in mid-2026)

This strategy aimed to cure momentum's regime-concentration weakness by gating ETH's own 4H momentum on the market leader BTC's daily trend — a breadth/index-filter concept, carefully constructed to avoid the dead lead-lag transfer (ETH's own momentum is the traded signal; BTC only grants permission). The developer was admirably rigorous and honest: running gated-vs-ungated comparisons that showed the gate's aggregate advantage was concentrated entirely in 2020, indistinguishable ex-2020, and net-negative in the last two years — so the filter delivers fewer trades and lower turnover but no real per-trade edge, exactly the concentration weakness it was meant to fix. The analyst abandoned it at backtest review as a marginal, cleaner-than-most momentum book whose stated mechanism was self-falsified and whose Sharpe of 0.628 (CI low 0.040) was too thin to survive the robustness gates, also flagging that the pre-submission metrics overstated the actual backtest and that a stronger regime-consistent sibling had already advanced.

A regime-gate innovation must be shown to add edge beyond the ungated base signal — here the developer's honest gated-vs-ungated year-by-year test revealed the BTC filter contributed nothing outside one outlier year (2020) and hurt recently, leaving a barely-significant plain-momentum book not worth optimizing, and reappearing regime-concentration in the very filter meant to cure it.

It was abandoned at the BACKTEST_REVIEW gate: the hypothesis's core innovation — the BTC regime gate — is falsified by the developer's own data (it adds nothing ex-2020 and hurts recently), so what remained was undifferentiated ETH 4H momentum with a Sharpe of 0.628 whose CI low of 0.040 is barely distinguishable from zero and would very likely fail deflated-Sharpe/PBO after best-of-N optimization, with recent softness putting the holdout at risk. A genuinely stronger regime-consistent sibling (ETH macro TSMOM) had already advanced that session, so it never reached optimization.

A long-short, multi-instrument momentum strategy that trades ETHUSDT.BINANCE USD-M's own 4H momentum but gates entries on the broad-market regime defined by the market leader BTC's daily (56-day, derived from 4H closes) trend — going long ETH only when its momentum is up and BTC is in an uptrend, short only when both are down, flat on disagreement — with an ATR trailing stop, using BTC purely as a risk-on/off permission signal rather than a traded or hedged leg.

The realized backtest was a marginal momentum book: total return +255%, Sharpe 0.628 (CI low only 0.040), profit factor 1.32, alpha +13.3%, beta 0.024, positive in 6 of 7 years, max drawdown 25.6% over 687 trades — clean (no sizing artifact, moderate kurtosis 10.6) but with turnover 117x and impact cost 7.9%. Critically, the developer's own year-by-year analysis showed the BTC regime gate adds no per-trade edge outside 2020 (ex-2020: gated +0.267% vs ungated +0.256%) and is net-negative in the last two years, and the pre-submission claim (Sharpe 1.40, +304%) materially overstated the actual result.