EthFourHourVolumeConfirmedBreakoutLongDailyRegime

ETH 4H Volume-Confirmed Breakout Long with Daily Regime Filter

A long-only single-instrument breakout strategy on ETHUSDT perpetual futures using 4-hour bars and OHLCV-only data, with a daily-bar regime filter. Architecturally MIRRORS the proven BNB 4H Volume-Confirmed Breakout (Sharpe 2.03, paper_stage), applied to ETH as the FIFTH validation point of this construction class (BNB ✓ proven; SOL, LINK, BTC ⏳ in pipeline; ETH proposed here). ETH is selected on a strong empirical basis: ETH Daily Golden Cross Momentum produced Sharpe 4.04 (the highest in this pipeline) — definitive evidence that ETH has clean trend persistence at daily timeframe. The 4H volume breakout is the intraday momentum-continuation extension of the same trend dynamic, so the architecture-asset fit is empirically supported (NOT a thesis-only argument like BNB Golden Cross had). Importantly, this proposal explicitly addresses the recently-documented failures: (a) NO bait-and-switch — strictly ETHUSDT.BINANCE perp, well-validated infrastructure (ETH Golden Cross is already running cleanly at paper_stage), (b) NO architecture-clone-without-evidence — ETH has explicit trend-persistence proof, (c) NO multi-condition AND-gate, (d) sensitivity_passed=true REQUIRED before optimization. Existing ETH strategy in pipeline (EthDailyGoldenCrossMomentumLong) uses daily bars + EMA-crossover events — this proposal uses 4H bars + intraday volume-confirmed breakouts: orthogonal in BOTH timeframe AND mechanism, providing genuine portfolio diversification with minimal signal correlation to existing ETH exposure. Single-dominant-filter design (200-day SMA daily regime + 20-bar 4H Donchian high + volume confirmation), explicitly NOT a multi-condition AND-gate. Calibrated for ~30-50 entries/year — well above the sparsity-failure threshold.

Implements the hypothesis's single-dominant-filter breakout: the 20-bar 4H Donchian close-breakout is the trigger, with volume and a 200-day-equivalent regime SMA as confirmations, all derived from the single ETH 4H feed (no fragile extra-bar daily dependency). Iteration 3 fix for the reported Layer-3 'No trades produced': I loaded the real ETHUSDT.BINANCE 4H catalog (14,214 bars) and replayed the EXACT strategy class through the verification proxy — the prior code already produced 239 entry signals, so the zero-trades feedback was stale relative to this 4H-derived-regime version. To honor the diagnostic's explicit 'loosen the entry logic / thresholds' instruction and add safety margin against parameter perturbation, I lowered vol_mult from 1.5 to 1.2 (the ONLY change). Verified against real ETH 4H data this yields 319 entry signals (up from 239) while retaining genuine volume confirmation (20% above the trailing average), and the synthetic Layer-2 scenarios still pass with a continuous, non-frozen breakout-margin signal. All other logic — breakout, regime, exits, capital-relative sizing — is byte-identical to the prior iteration, so no earlier passing layer regresses.

Failed deflated Sharpe and the forward walk-forward on attempt 1. DSR=0.0067 (vs 0.95 bar) with the optimized Sharpe 1.047 BELOW the 225-trial expected-max luck bar of 2.297 (is_significant=false, sharpe_ci_low -0.23 through 0, PBO 0.5344) — the selected config is statistically indistinguishable from best-of-N noise, and the PSR=0.981 vs DSR=0.0067 split is the classic multiple-testing inflation tell. The walk-forward is is_overfitted=TRUE with a NEGATIVE average OOS: avg IS 1.463 -> avg OOS -0.098, windows [0.0, 0.995, -1.290] (one deeply negative). The holdout 'pass' is spurious — ratio 0 computed against the negative WF-OOS, with holdout_sharpe 4.338 a single-window artifact. The result is also implausible and over-leveraged: the optimizer breached the intended 0.95 sizing by pushing position_pct to 1.415 (>1x), driving avg_position_pct to 168% of equity and a +904% headline that is 152% un-exited open MTM (end_unrealized_pct), with max_drawdown ~50%, annualized vol ~145%, kurtosis ~88 and tail_ratio 0.39 — the edge rests on a handful of parabolic single days (2024-03-14 +97%, 2025-08-01 +68%). Not iterate (attempt 1 of 2): the binding failure is an overfit, negative-average-OOS walk-forward with an over-levered outlier-fit config, not an under-explored parameter region (sensitivity is already a clean 0-cliff plateau), so a second best-of-225 sweep re-selects the same sub-significant, leverage-inflated fit. Not revise_hypothesis: the 4H volume-confirmed breakout family is already validated/promoted on the BNB sibling (Sharpe 2.03) and ETH is not a dead target, so this is ETH-arm asset-selectivity failing deflation, not a proven mechanism stranded on a dead instrument. FAILURE PATTERN: the FourHourVolumeBreakout family does NOT transfer to the ETH arm — a clean 0-cliff sensitivity grid and a strong-looking holdout ratio mask a modest, outlier-carried edge whose best-of-225 selected Sharpe (1.05) sits below the 2.30 luck bar (DSR 0.007) with a negative average OOS (-0.098) and an over-levered (168% position, 152% unrealized) headline. This mirrors the abandoned SUI arm of the same family: a validated family and a favorable single holdout window do not rescue an arm that fails deflated Sharpe with an overfit, negative-OOS walk-forward; the promoted BNB instance remains the configuration where this mechanism clears the bar.

Long-only 4H Donchian breakout on ETHUSDT.BINANCE USD-M perpetual futures. Enters long when the 4H close breaks above the prior 20-bar high (Donchian breakout), volume exceeds 1.2x the prior-20-bar average (volume confirmation), and the close is above its 1200-bar (~200-day) SMA (uptrend regime, derived from the 4H series itself). Exits on an 8% hard stop, a lower-Donchian (10-bar) structure break, or regime turning off. Pure OHLCV, no supplementary data, leverage 1.0.

Verification failed (Layer 3 — sandbox backtest): No trades produced Bar type used: ETHUSDT.BINANCE-4-HOUR-LAST-EXTERNAL, Bars processed: 13980 Diagnostics: should_enter() returned a side 0 times over 13959 evaluated bars -> your ENTRY CONDITION never triggered. Loosen the entry logic / thresholds. Ensure your strategy produces trades with the given data and parameters.

Trades match the hypothesized mechanism exactly: 99 long-only breakout entries, 0 shorts, no dropped/zero-qty entries

Regime filter demonstrably works — kept the strategy flat through 2022 bear (absent from annual returns), avoiding the worst drawdown year

Positive edge signature: Sharpe 0.72, profit_factor 1.27 (>1.2), sortino 2.48, calmar 2.75, positive alpha (+5%) with low beta (0.24)

99 trades over ~6.5 years is a sufficient sample for meaningful parameter optimization without fitting pure noise

Heavily outlier-carried: return_kurtosis 78.8, skew 7.87, tail_ratio 0.39 — a few huge single days (+63%, +60%, +55%, +52%) drive the headline; the post-opt deflated-Sharpe/holdout gates must confirm the edge survives without them

Base config not yet statistically significant: probabilistic_sharpe 0.906 (<0.95), sharpe_ci_low -0.65 (CI straddles 0), max_drawdown 47.6%

end_unrealized_pct 64.3% — most of the +215% total return is un-exited open-position MTM, not realized

Trade frequency (~15/yr) is below the hypothesized 30-50/yr

Clean sensitivity grid: 0 cliffs, smooth monotonic surface across all 8 params (base Sharpe 0.72 robust to +-20% perturbation).

Differentiated design: 4H volume-confirmed Donchian breakout with a 200-day-equiv regime filter, orthogonal in timeframe and mechanism to the ETH daily Golden Cross sibling.

Adequate trade density (99 base / 66 optimized entries) and low commission drag (2.3-2.8% of gross).

Fails the multiple-testing hard gate decisively: deflated_sharpe=0.0067 with the optimized Sharpe 1.047 BELOW the 225-trial expected-max luck bar of 2.297 (is_significant=false, sharpe_ci_low=-0.23 through 0, PBO=0.5344). The probabilistic_sharpe 0.981 vs DSR 0.0067 split is the textbook best-of-N inflation tell.

Walk-forward is_overfitted=TRUE with a NEGATIVE average OOS: avg IS 1.463 -> avg OOS -0.098, windows [0.0, 0.995, -1.290] (one zero, one deeply negative) — the edge does not survive out of sample.

The holdout 'pass' is spurious: ratio 0 computed against a NEGATIVE WF-OOS (-0.098); the holdout_sharpe 4.338 is a single-window artifact, not a real forward read.

Optimizer breached the intended sizing: position_pct pushed to 1.415 (>1x), driving avg_position_pct to 168% of equity — it chased bull-year return with leverage, not edge. Base already ran avg_position 121%.

Outlier-carried and dangerous: max_drawdown ~48-50%, annualized_volatility 100-145%, return_kurtosis 79-88, skew ~8, tail_ratio 0.39 — the result rests on a handful of parabolic single days (e.g. 2024-03-14 +97%, 2025-08-01 +68%, 2020/2021 +52-79% days).

Optimized +904% headline is mostly un-exited open MTM: end_unrealized_pct=152% — realized track record is far smaller, not a validated edge.

Do NOT optimize — the backtest is corrupted by reckless sizing and likely artifacts, so it isn't a fair test of the breakout premise. Fix before re-running: 1. SIZING (critical): position_pct=0.95 (near all-in) compounds to avg_position_pct 159.6% with 126% annualized vol and 44% drawdown. Replace with risk-based sizing: size so the 8% stop risks ~1% of equity, i.e. qty = (equity * risk_frac) / (price * stop_loss_pct/100) with risk_frac ~0.01, OR cap notional at a modest fraction (e.g. 0.20-0.30) of equity. The BNB sibling's Sharpe 2.03 was almost certainly NOT achieved with 0.95 all-in sizing — match its sizing approach. 2. INVESTIGATE IMPLAUSIBLE DAYS (critical): daily_returns shows +100.7% (2025-07-22), +75.6% (2024-03-05), +51% (2021-01-11), +50% (2020-08-07). A 1x-leverage ETH position cannot move +100% in a day. Determine the cause: (a) is the compounded position reaching far more than 1.6x effective leverage at those moments? (b) is there a bad/Spike bar in the ETHUSDT 4H data at those timestamps (print the OHLCV around each)? These 4 days drive the entire result (skew 7.99, kurtosis 85) — without them the strategy is negative (2021 -21%, 2023 -17%). The result cannot be trusted until this is explained. 3. END-OF-BACKTEST EXIT (important): end_unrealized_pct is 99.47 — ~59% of the +169.9% headline is unrealized open-position paper gain on the final bar. Ensure the strategy's exits (stop / structure break / regime-off) are the only drivers and judge on REALIZED return; a horizon-end MTM headline that's mostly unrealized is not a validated edge. Resubmit only after a properly-sized run with the artifact explained. If, with sane sizing and clean data, the realized Sharpe and profit_factor remain marginal (PF <1.2), the ETH 4H breakout has no robust edge and should be abandoned.

EthFourHourVolumeConfirmedBreakoutLongDailyRegime ported the promoted BNB 4H volume-confirmed breakout architecture (Sharpe 2.03) onto ETH as the family's fifth validation point. Its base backtest looked promising — Sharpe 0.72, profit factor 1.27, positive alpha, and a rare fully clean 0-cliff sensitivity grid — so it advanced to optimization despite a 48% drawdown and a return driven by a handful of parabolic days. Optimization inflated the headline to Sharpe 1.047 and +904%, but every forward check rejected it: deflated Sharpe 0.0067 against a 2.297 luck bar, PBO 0.53, a walk-forward whose OOS averaged −0.098, and a holdout 'pass' that was really a single-window artifact, all propped up by the optimizer pushing position size to 168% of equity. After three iterations it was abandoned as an ETH-arm asset-selectivity failure, with the promoted BNB instance remaining the only configuration where this mechanism clears the bar.

A validated, promoted architecture does not automatically transfer to a new asset arm; a clean sensitivity plateau and a favorable single holdout window can mask an outlier-carried edge that fails deflated Sharpe, goes negative out-of-sample, and only produces a big headline via >1x leverage and un-exited open MTM.

The analyst abandoned it because it failed the multiple-testing gate: deflated Sharpe 0.0067 with the optimized 1.047 below the 225-trial expected-max luck bar of 2.297 (is_significant false, sharpe_ci_low −0.23, PBO 0.5344), and an is_overfitted walk-forward with negative average OOS (IS 1.463 → OOS −0.098, windows [0, 0.995, −1.290]); the holdout 'pass' was a spurious single-window artifact and the optimizer had breached intended sizing with leverage.

A long-only ETHUSDT 4-hour Donchian(20) breakout strategy requiring volume confirmation and a ~200-day-equivalent (1200-bar) SMA uptrend regime filter, mirroring the pipeline's promoted BNB 4H volume-confirmed breakout (Sharpe 2.03) onto ETH as a fifth validation point of that construction class.

The base backtest posted Sharpe 0.72, total return 215.69, profit factor 1.27 and positive alpha (+5%) over 99 trades with a clean 0-cliff sensitivity grid, but a 47.6% max drawdown and a heavily outlier-carried profile (kurtosis 78.8). Optimization lifted the headline to Sharpe 1.047 and +904% return (66 trades, PF 1.61) largely by pushing position_pct above 1x (avg position 168% of equity) with a ~50% drawdown.