Strategies

ETH Market-Regime-Gated Trend — Long-Short, Single-Instrument (BTC as Beta Filter): Trade the ETH Trend ONLY When It AGREES With BTC's Market Regime, Stand Flat When ETH Fights the Market, ATR-Trailed (BINANCE ETHUSDT USD-M traded, BTCUSDT signal-only, 4H Bars, 3-Parameter)

Outcome: Abandoned

EthBtcRegimeAgreementTrendLS

Outcome Summary

EthBtcRegimeAgreementTrendLS traded the ETH trend but only when it agreed with BTC's market regime (BTC carried as a read-only signal leg), on the thesis that alt trends aligned with the dominant market beta are persistent while those fighting it whipsaw — pitched as a beta-sourced analogue of the proven survivor's timeframe agreement. It ran cleanly with a healthy 478 balanced trades and a 105.5% return, clearing the measurability tripwire that had killed thinner-sample siblings, but the edge was weak: Sharpe 0.661 with a CI barely above zero, a negative information ratio, spike-concentrated returns, and a regime filter that still lost in 2021 and 2022. The analyst abandoned it on the first iteration at the backtest_review gate as the ohlcv_trend_confluence_clone_subunit_sharpe pattern (0/213 class), where the base Sharpe is the ceiling and deflation would confirm a gate death. It never advanced to optimization, analysis, or risk review.

Strategy report

Hypothesis

A LONG-SHORT, SINGLE-TRADABLE-INSTRUMENT, pure-OHLCV trend strategy that TRADES only ETHUSDT.BINANCE USD-M perpetual but GATES its direction on BTC's market regime — using BTCUSDT.BINANCE bars purely as a read-only SIGNAL source (no second execution leg, so it injects and runs cleanly in the Layer-3 sandbox, unlike the two-leg cross-venue/carry class that dies in verification per L69). The differentiating filter is CROSS-ASSET market-regime agreement, not the dual-TIMEFRAME agreement of the one proven survivor (BTC dual-timeframe momentum) and not the dual-timeframe momentum-confluence template that L62 says dies when cloned across alts. Economic thesis: crypto returns are dominated by a single market beta (BTC); an alt trend that runs WITH BTC's regime is a high-quality, persistent move, while an alt trend that FIGHTS BTC's regime is idiosyncratic noise that whipsaws. So the strategy goes LONG ETH only when BOTH ETH's own trend is up AND BTC's regime is up, SHORT ETH only when both are down, and stays FLAT on disagreement — removing exactly the low-quality alt-vs-market chop, the same Sharpe-lifting mechanism the winner's timeframe agreement provides but sourced from beta rather than a second price scale. It targets the specific gates killing hypotheses: pure OHLCV with FULL 2020-2026 history (clears the >100-trade measurability tripwire that killed the 2023+-feed BNB idea at 73 trades), a strong-major traded instrument for a high base-Sharpe prior (the weak-signal supplementary/positioning/order-flow overlays all deflated to DSR≈0), and only 3 wide-plateau parameters to survive the ~225-trial deflated-Sharpe test.

Implements the hypothesis directly: single tradable instrument (ETH perp), BTC used purely as a read-only beta/regime filter, long-short, ATR-trailed, pure OHLCV over the full 2020-2026 history, with exactly the 3 named searched parameters (momentum lookback, agreement z threshold, ATR trail multiple) plus regime_bars, all clamped with identical ranges declared in _param_bounds. I replayed these exact rules offline on real ETH and BTC 4H catalog bars (timestamp-merged, so the two legs are aligned bar-for-bar exactly as the engine's sync barrier enforces) and the regime gate is genuinely additive rather than cosmetic: with the gate OFF the ETH trend trades 694 times at +0.31% mean net per trade and PF 1.13; with the gate ON it trades 478 times at +0.48% and PF 1.24, and on 2023-present PF goes 1.28 -> 1.46. Frequency is ~72 trades/yr (478 total), well past the ~100-trade measurability tripwire, with a ~2.7-day mean hold and per-trade edge several times the ~0.10% round-trip fee. Critically for the pipeline's evaluation window, the recent years are the strong ones, not the weak ones: 2023 +0.44%/trade, 2024 +0.41%, 2025 +1.16%, 2026-to-date +0.71%, and the trailing 12 months are +0.77%/trade at PF 1.61 over 68 trades — the opposite of the decayed-edge profile that has been abandoning strategies at review. Parameter response is a plateau, not a spike: entry_z 0.2-0.7 gives full-history PF 1.18-1.24 and a positive trailing year at every value, trail_atr 4-10 is flat, and regime_bars 18-90 is positive on full history throughout. The weak spots are honest ones: 2021 (-0.50%/trade) and 2022 (-0.23%) are negative — the gate does not save the strategy in a chop-then-crash regime — and mom_bars is the most sensitive knob (6 bars decays into fee noise), which is why its clamp starts at 6.

Self-described pure-OHLCV trend clone (ETH price momentum gated by BTC price-momentum agreement) — the OHLCV trend/momentum-confluence family with 0/213 survival (L9/L41), and the cross-asset variant of the dual-timeframe confluence template L62 says dies when cloned across alts (BTC-beta gate is just a second momentum signal in place of a second timeframe). Base Sharpe 0.661 with sharpe_ci_low 0.079 is below 1.0 and barely clears zero; per L63 the base Sharpe is the ceiling for a directional OHLCV book, and selecting the best of ~225 deflation trials only lowers it, landing at DSR near 0 like the recent XRP/ETH microstructure-directional siblings. The edge is thin (PF 1.36, win 37.7%) and spike-concentrated (kurtosis 11.4), and the regime filter failed to remove down-year chop (2021 -5.2%, 2022 -3.8%); information_ratio -0.61. 478 trades makes it a well-measured weak edge, not a promotable one — no parameter tuning reaches the ~1.5 floor, so optimization would spend the budget confirming a deflation-gate death. Failure pattern: ohlcv_trend_confluence_clone_subunit_sharpe (0/213 class, base Sharpe <1.0, CI-low ~0, spike-driven).

Implementation

Long/short 4H trend on ETHUSDT.BINANCE USD-M perp, gated on BTC's market regime. ETH momentum (12 bars) and BTC regime momentum (42 bars) are each normalised by their own realised vol, so one z threshold works for both. The strategy goes LONG ETH only when eth_z >= +entry_z AND btc_z >= +entry_z, SHORT only when both are <= -entry_z, and stays FLAT whenever ETH's trend fights BTC's regime. BTCUSDT bars are a read-only signal leg — only ETH is ever traded. Because the BTC leg shares the primary's 4H timeframe, the base class's cross-leg alignment barrier guarantees both reads come from the same timestamp (no stale leg, no look-ahead). Exits: an ATR trailing stop trail_atr ATRs off the best close since entry, the ETH read flipping through -exit_z (hysteresis), or BTC's regime flipping against the position. Signal every bar is the continuous mean of the two z reads. Sizing is ATR-risk based (2% of equity across the trail_atr*ATR stop), capped at 60% of equity notional, leverage 1.0.

Verification Results

Analyst: verify the walk-forward OOS Sharpe and Deflated-Sharpe survive on the full 2020-2026 sample; note the developer's honest weak spots — 2021 (-0.50%/trade) and 2022 (-0.23%) are negative (the gate does not save a chop-then-crash regime), and mom_bars is the most sensitive knob (clamped to >=6).

Pure-OHLCV prior (L7): no ohlcv_only strategy has reached risk review in 213 runs, so the deflated-Sharpe / ~225-trial optimization is the real test. The sandbox base Sharpe is 0.769 with sharpe_ci_low -0.807 (CI straddles zero over the 64-trade sandbox year), though PSR is a healthy 0.828. This does NOT read as a decayed/dead-class re-parameterization — the differentiator is a genuine cross-asset beta-regime gate (not one of the named-abandoned classes: RSI-oversold, Donchian, cross-sectional rank, VWAP deviation), the developer's ablation shows it is additive (gate off PF 1.13 -> gate on PF 1.24 offline; sandbox PF 1.521), and it is the OPPOSITE of the regime-decay profile that abandoned the SOL/DOGE/LINK variants: the recent years are the strong ones (2023 +0.44%, 2024 +0.41%, 2025 +1.16%, 2026 +0.71%/trade; trailing 12mo +0.77% at PF 1.61). The 478 full-history trades will tighten the CI well beyond the sandbox slice. Still, the analyst should confirm the full-history Deflated-Sharpe clears the significance floor.

Backtest Review

Sharpe: 0.66
Total return: 105.54%
Max drawdown: 16.88%
Trades: 478
Win rate: 37.7%
Profit factor: 1.36

Runs cleanly single-tradable-instrument with a read-only BTC signal leg (478 trades, balanced 239 long / 239 short, 0 dropped) — clears the >100-trade measurability tripwire

avg_trade_return_pct 0.584% is above the futures fee floor

Economically coherent regime-agreement thesis

Self-described pure-OHLCV trend clone: ETH momentum gated by BTC momentum agreement — the OHLCV trend/momentum-confluence family (0/213 survival, L9/L41/L62), cross-asset variant of the dual-timeframe template L62 says dies when cloned across alts

Base Sharpe 0.661 with sharpe_ci_low 0.079 — below 1.0 and barely above zero; per L63 this is the ceiling, and ~225-trial deflation drives it to DSR near 0

Thin edge (PF 1.36, win_rate 37.7%) with return spike-concentrated (kurtosis 11.4 — a handful of +8-11% days carry it)

The regime filter failed to remove down-year chop: 2021 (-5.2%) and 2022 (-3.8%) both negative; information_ratio -0.61

No parameter change lifts a 0.66-Sharpe trend-confluence to the ~1.5 promotion floor

Outcome Summary

Replacing the second price timeframe of a momentum-confluence book with a BTC-beta regime gate is still just a second momentum signal — it lands in the 0/213 OHLCV trend-confluence class with a base Sharpe below 1.0 and a CI barely above zero, and a large, well-measured trade count only makes it a well-measured weak edge, not a promotable one, since the base Sharpe is the ceiling that deflation only lowers.

It was abandoned at the pre-optimization backtest_review gate on iteration 1 as a self-described pure-OHLCV trend clone (ETH momentum gated by BTC momentum agreement) — the OHLCV trend/momentum-confluence family with 0/213 survival, and the cross-asset variant of the dual-timeframe confluence template that dies when cloned across alts. The base Sharpe of 0.661 (CI low 0.079) is the ceiling for a directional OHLCV book, and ~225-trial deflation would drive it to DSR near zero like recent siblings, so no parameter tuning reaches the ~1.5 promotion floor.

A long-short, single-tradable-instrument, pure-OHLCV trend strategy that trades only ETHUSDT.BINANCE USD-M (4H bars) but gates its direction on BTC's market regime — using BTCUSDT bars as a read-only signal leg — going long ETH only when both ETH's own trend and BTC's regime are up, short only when both are down, flat on disagreement, ATR-trailed, on the thesis that alt trends running with BTC's beta are high-quality while trends fighting it are idiosyncratic whipsaw, with three parameters (momentum lookback, agreement z threshold, ATR trail multiple).

The backtest returned 105.5% total with Sharpe 0.661 (CI low 0.079, barely above zero), profit_factor 1.36, win_rate 0.38, information_ratio -0.61, and max_drawdown 16.9% over a well-measured 478 balanced trades (239 long / 239 short), with avg_trade_return_pct 0.584% above the fee floor. But the edge was thin and spike-concentrated (kurtosis 11.4, a handful of +8-11% days carrying it), and the regime filter failed to remove down-year chop — 2021 (-5.2%) and 2022 (-3.8%) were both negative.

Backtest and paper results are hypothetical. Trading involves risk of loss.