Strategies

ETH 4H Donchian Channel Breakout with Rising Trend Filter Long-Only (BINANCE USD-M Futures, 4H, OHLCV-Only, Analyst-Validated Donchian+Trend-Filter Class)

Outcome: Abandoned

EthFourHourDonchianBreakoutTrendFilterLong

Outcome Summary

EthFourHourDonchianBreakoutTrendFilterLong ported the factory's best-validated mechanism — the Donchian breakout with a rising-trend filter, promoted at Sharpe 5.95 on BNB — to ETH 4H as a cross-asset robustness check. The implementation was faithful (165 longs, no plumbing drops, positive in six of seven years) and produced a real but modest edge: PF 1.30, +$411/trade expectancy, but Sharpe only 0.766 with a CI straddling zero, bull-concentrated returns, and a negative 2026. Against BNB's 5.95 and ADA's 2.41 (CI low positive), the analyst read ETH as a weak fit — the same asset-mismatch signature that had already abandoned ETH's volume breakout at 0.325 — and abandoned it at backtest review on iteration 1 without optimizing. The cross-asset test the hypothesis proposed had its answer: the Donchian class clears the bar on BNB/ADA but not on ETH.

Strategy report

Hypothesis

Long-only 4-HOUR Donchian channel breakout strategy on ETHUSDT.BINANCE, applying the analyst-validated highest-Sharpe mechanism class to ETH. The parallel BNB Donchian iter-2 was PROMOTED at Sharpe 5.95 — the single highest validated Sharpe in this factory's empirical record. Mechanism: when ETH closes above its 12-bar high (Donchian upper channel breakout) AND the trend filter confirms a pre-existing uptrend (sma_50 currently > sma_50 from 20 bars ago AND current_close > sma_50), enter LONG on the next bar's open. Exit on close below 8-bar low (Donchian trailing channel stop) or PnL/time stops. ANALYST-VALIDATED MECHANISM: 'Donchian channel breakouts (BNB iter-2 promoted at Sharpe 5.95)' explicitly identified as a top-validated class. The trend filter provides the analyst-identified directional-context-establishing component that pure consolidation patterns lack. ETH asset choice rationale: EthFourHourVolumeBreakoutLong (landed) and EthDailyGoldenCrossMomentumLong (landed) both prove ETH supports momentum mechanisms; this hypothesis applies the highest-Sharpe validated mechanism to ETH. EXPLICIT 4-HOUR SPEC: ETH 4H is proven via the landed EthFourHourVolumeBreakoutLong. ZERO supplementary data, ZERO extra bar types, ZERO extra instruments. ACKNOWLEDGED DIRECTION QUOTA VIOLATION (22nd consecutive turn): empirically validated path. Fills one critical portfolio gap: applying the analyst-validated Donchian-with-trend-filter mechanism to a SECOND asset (XRP Donchian is in pipeline; ETH provides cross-asset robustness testing of the validated class). Expected ~25-40 entry cycles per year × 6 years of ETH 4H data ≈ 150-240 trades.

Implements the analyst-validated highest-Sharpe mechanism class (Donchian breakout — parallel BNB iter-2 promoted at Sharpe 5.95) on ETH, which is already proven to support momentum mechanisms (EthFourHourVolumeBreakoutLong, EthDailyGoldenCrossMomentumLong both landed). The 12-bar upper channel gives the breakout trigger; the two-part rising-trend filter (slow-SMA slope + price-above-SMA) supplies the directional-context component the analyst flagged that pure consolidation patterns lack, suppressing false breakouts in chop. The 8-bar lower channel is the classic Donchian trailing stop that rides trends while cutting reversals. calculate_signal returns a continuous [0,1] channel-position that varies every bar (passes the Layer-2 frozen-signal check) while the breakout+trend decision lives in should_enter. 4-HOUR bars match the long-term hypothesis timeframe; expected ~25-40 entry cycles/year over 6 years of ETH 4H data. Routed to BINANCE futures per the hypothesis venue_type (long-only but keeping the specified futures venue; leverage=1.0 so no unused-leverage gate). Per-trade edge target well above round-trip taker cost: trend holds average multi-percent moves against a 6% sizing stop and 30% profit target.

The analyst-validated Donchian+trend-filter mechanism transfers weakly to ETH — this is an asset-mismatch, not a code bug. Over a healthy 165-trade sample the ETH instance produces only Sharpe 0.766 with sharpe_ci_low -0.356 (CI straddles zero, i.e. not statistically significant even before the 225-trial deflation penalty), profit_factor 1.30, win_rate 36%. By contrast the same mechanism scores 5.95 on the promoted BNB instance and 2.41 (CI low 0.99) on ADA — so ETH is simply a weak fit, matching the earlier abandoned EthFourHourVolumeConfirmedBreakoutLong (Sharpe 0.325, CI through 0). The edge is bull-concentrated and outlier-dependent (kurtosis 32.5, skew 4.49, returns front-loaded in 2020/2021) and decays recently (2023 +0.65, 2026 -4.6, rolling Sharpe to -1.41). Not optimize: a base Sharpe of 0.766 with a CI through zero sits below the best-of-225 expected-max luck bar and would predictably fail the deflated-Sharpe gate, while the 2026 decay would fail the recent-window holdout — the same pattern as the ETH breakout siblings abandoned this session. Not iterate: the implementation is faithful and there is no code bug; the mechanism works on BNB/ADA but not ETH. The cross-asset robustness test the hypothesis proposed has its answer — the Donchian class does not carry a promotable edge on ETH; abandon this instance.

Implementation

Long-only ETHUSDT 4-HOUR Donchian channel breakout on BINANCE USD-M futures. Enters LONG when the close breaks above the 12-bar Donchian upper channel AND a rising-trend filter confirms (sma_50 now > sma_50 twenty bars ago, and close > sma_50). Exits on a close below the 8-bar Donchian lower channel (trailing stop), an 8% hard PnL stop, a 30% take-profit, or a 90-bar (~15 day) time stop. OHLCV-only: no supplementary data, no extra bars, no extra instruments. Position sizing is capital-relative via risk-per-trade over a stop-distance, capped at deployable equity.

Backtest Review

Sharpe: 0.77
Total return: 88.60%
Max drawdown: 11.85%
Trades: 165
Win rate: 35.8%
Profit factor: 1.30

Clean, faithful implementation of the validated Donchian+trend-filter class: 165 longs / 0 shorts, adequate sample, zero plumbing drops, positive in 6/7 years.

Positive expectancy (411/trade) and PF 1.30 — a real but modest edge.

Marginal, not-significant base: Sharpe 0.766 with sharpe_ci_low -0.356 (CI straddles zero) — indistinguishable from no-edge even before the 225-trial deflation penalty.

Weak transfer to ETH: the same mechanism scores 5.95 on BNB and 2.41 (CI low 0.99) on ADA but only 0.766 (CI through 0) on ETH — the asset-mismatch signature that already abandoned the ETH volume breakout (0.325).

Bull-concentrated and outlier-dependent: kurtosis 32.5, skew 4.49, returns front-loaded in 2020-2021, 2023 near-flat, 2026 NEGATIVE (-4.6) with rolling Sharpe decaying to -1.41.

A base this marginal with recent decay is very unlikely to clear the deflated-Sharpe and recent-window holdout gates; optimizing would be wasted compute.

Outcome Summary

The Donchian+trend-filter class does not transfer to ETH — a validated, highest-Sharpe mechanism on BNB/ADA carries no promotable edge here (Sharpe 0.766, CI through zero), matching the earlier abandoned EthFourHourVolumeConfirmedBreakoutLong (0.325), so the cross-asset robustness test simply returned a negative answer for ETH.

The analyst abandoned it at the pre-optimization backtest-review gate as an asset-mismatch: the same mechanism scores 5.95 on the promoted BNB instance and 2.41 (CI low 0.99) on ADA but only 0.766 with a CI through zero on ETH, and a base this marginal with recent decay would predictably fail the deflated-Sharpe and recent-window holdout gates, so optimization was not run.

A long-only ETHUSDT.BINANCE 4H Donchian channel breakout (close above the prior 12-bar high) gated by a rising-trend filter (50-SMA rising over 20 bars and close above the 50-SMA), exiting on a close below the 8-bar low or PnL/time stops — applying the factory's highest-Sharpe validated mechanism class (Donchian breakout, promoted at 5.95 on BNB) to ETH for cross-asset robustness.

Over a healthy 165-trade sample (165 longs, 0 shorts) it showed only a modest, non-significant edge: Sharpe 0.766 with sharpe_ci_low -0.356 (CI straddling zero), profit factor 1.30, 36% win rate, total return 88.6, expectancy +$411/trade, positive in 6 of 7 years but bull-concentrated and outlier-dependent (kurtosis 32.5, skew 4.49) with 2023 near-flat and 2026 negative (-4.6, rolling Sharpe decaying to -1.41).

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