Strategies

ETH 1H Volume-Confirmed Momentum, Long-Short (ETHUSDT.BINANCE, 1H — faster horizon of the survivor, stronger surge gate to beat 1H fee drag)

Outcome: Abandoned

EthHourlyVolumeConfirmedMomentumLS

Outcome Summary

EthHourlyVolumeConfirmedMomentumLS was a faithful, well-engineered 1H port of a promoted 4H volume-confirmed momentum survivor, keeping the same edge but strengthening the surge gate, entry threshold, and trailing exits to fight higher 1H fee drag. Across 1192 trades over 6.7 years it made money gross (113.3% total return) but delivered only a 0.285 Sharpe with a confidence interval straddling zero, a 1.061 profit factor, and — decisively — a 0.1475% average per-trade return that fell below the 0.15% Binance fee floor. The backtest-review analyst abandoned it before optimization, judging the per-trade edge smaller than the round-trip taker cost on a sample too large to be noise. This was fee-fragility rather than a broken signal, and it matched the exact abandon condition the hypothesis had pre-registered for itself.

Strategy report

Hypothesis

Build on: ETH Volume-Confirmed Momentum, Long-Short (Single-Instrument ETHUSDT.BINANCE Perp, 4H Bars — Take Directional Momentum ONLY When a Volume Surge Confirms Genuine Participation, Flat Otherwise, ATR-Trailing, 3-Parameter) same edge on ETH 1h … (28,744 characters of source material given to the Research Lead)

Per-trade edge sits below the venue fee floor on a decisive 1192-trade / 6.7-year sample: avg_trade_return_pct 0.1475% < the 0.15% Binance USD-M floor, profit_factor 1.061 (in the fee_edge band [0.85,1.10]), and Sharpe 0.285 with CI [-0.54, 1.01] straddling zero. The 1H horizon port has too little headroom over the ~0.10% round-trip taker cost — commissions consume the gross edge (fee-fragility, not signal quality; matches L22 and L18). This is exactly the abandon condition the hypothesis pre-registered ('avg_trade_return_pct under 0.15% with a Sharpe CI straddling zero -> ABANDON, not tune'), and QA's edge concern is borne out. Optimization cannot lift an edge already below trading costs, so this is not worth 2 hours of compute. abandon_class=negative_expectancy.

Implementation

Long/short directional momentum on the ETHUSDT.BINANCE USD-M perpetual using 1-HOUR bars, taken ONLY when a concurrent volume surge confirms genuine participation, flat otherwise. Every bar it computes a continuous participation-weighted momentum signal: mom_atr * min(1, participation), where mom_atr is the 24-bar (1 day) price change in ATR(72) units and participation is the bar's volume over the median volume of the prior 120 bars (5 days) divided by vol_surge_mult. It goes long when that signal crosses up through +1.5 ATR with the surge actually met (participation >= 1), short on the mirror condition; the cross requirement stops a multi-bar high-volume trend from re-triggering entries for pure fees. Exits are structural: momentum reversing to 0.40 ATR against the position (with hysteresis so intra-leg pauses are held through), or a 4-ATR trailing stop from the best close since entry. Sizing risks 2% of equity to the trailing-stop distance, capped at 1x equity notional (leverage 1.0), floored at ETH's 0.001 step and $20 notional. Pure OHLCV, single instrument, no supplementary feeds; three tunable parameters (mom_lookback, vol_surge_mult, trail_atr_mult) with declared bounds matching the code clamps.

The hypothesis asks for the surviving 4H ETH volume-confirmed momentum strategy run at a faster horizon on 1H bars, with a stronger surge gate to beat 1H fee drag, so I ported EthVolumeConfirmedMomentumLS verbatim in mechanism and changed only the calibration that the horizon change forces. (1) Surge multiple 2.0 -> 2.5: the 1H volume distribution has a much fatter right tail than the 4H one, so 2.0x median is a common, weakly-informative event at 1H. (2) Entry threshold 1.0 -> 1.5 ATR: momentum is measured over mom_lookback bars in SINGLE-BAR ATR units and an n-bar move's dispersion scales ~sqrt(n); with a 24-bar 1H lookback vs the parent's 12-bar 4H one, keeping 1.0 would have selected a strictly weaker move in sigma terms, so 1.5 restores the parent's selectivity per unit of noise. (3) Exit hysteresis 0.25 -> 0.40 ATR and trail 3.0 -> 4.0 ATR: 1H has far more pauses and noise excursions per unit of realised trend, and widening the exit band is how turnover (and therefore fee drag, the stated risk of this reframe) is held down without touching the entry mechanism. All windows are set in calendar terms (1-day momentum, 3-day ATR, 5-day volume baseline). Frequency sanity: the surge gate fires on roughly 2-4% of 1H bars and the ATR threshold is met on a majority of bars, so hundreds of entry opportunities exist in the sandbox window -- no zero-trade risk. Cost: per-bar work is bounded by ~190 element reads plus one 120-element sort independent of history (index addressing, memoised prior-bar eligibility, O(1) trailing extreme), which matters because 1H is 4x the parent's bar count under the same 300s wall clock. Futures venue (BINANCE USD-M) is required because the strategy shorts; leverage stays 1.0 and the notional cap means no sizing path reads it. One honest caveat: at 1H the confirmed leg is ~4x smaller in price terms than at 4H while the round-trip fee is unchanged, so this port has materially less headroom over costs than the parent -- if avg_trade_return_pct comes back under 0.15% with a Sharpe CI straddling zero, the right call is to abandon the horizon change rather than tighten the gate further.

Verification Results

Confirm the base template consumes the overridden min_bars_required for warmup gating; if it uses its own value the extra warmup is harmless.

min_bars_required(self) overrides a framework method (base_shadow static warning). This appears deliberate and correctly scales warmup to the locked windows (vol_window+2, atr_period+2, mom_lookback+2); no functional problem observed in synthetic/sandbox runs.

Faithful horizon port, but the hypothesis's own honest caveat is borne out in the sandbox: avg_trade_return_pct 0.231% clears the 0.15% futures floor only modestly, and the Sharpe CI is wide and straddles zero (-0.83 to 1.79). Regime attribution shows the stressed vol tercile is negative (-3.5%, Sharpe 0.07) while gains concentrate in the normal tercile. The 1H reframe has materially less headroom over fees than the 4H parent; the full backtest + walk-forward should confirm the per-trade edge is stable and not overfit before promotion. Advisory only.

Backtest Review

Sharpe: 0.29
Total return: 113.32%
Max drawdown: 44.00%
Trades: 1192
Win rate: 35.9%
Profit factor: 1.06

Faithful, well-engineered horizon port of the promoted 4H survivor; trades match the hypothesis (587 long / 605 short volume-confirmed momentum, balanced, 1192 trades over 6.7 years).

Reliable sample (metrics_reliable=true, 1192 trades) so the result is decisive, not noise.

avg_trade_return_pct 0.1475% is BELOW the 0.15% Binance USD-M fee floor — the per-trade edge is smaller than the round-trip cost, so the strategy is not numerically viable and optimization cannot lift an edge below costs.

profit_factor 1.061 sits in the fee_edge danger band [0.85,1.10] with 1192 trades: the gross edge is real but consumed by commissions (fee-fragile).

Sharpe 0.285 with CI [-0.54, 1.01] straddling zero — no statistically distinguishable risk-adjusted edge.

The 1H reframe has materially less headroom over fees than the 4H parent, exactly the honest caveat the hypothesis itself pre-registered as an abandon condition (both triggers met: sub-0.15% per-trade AND Sharpe CI straddling zero).

max_drawdown 44%, calm vol regime deeply negative (-45%, Sharpe -0.04), tariff_shock_2025 stress window Sharpe -3.0.

Outcome Summary

Porting a viable 4H edge to a 1H horizon quadruples turnover and proportionally magnifies the ~0.10% round-trip fee bite; tightening confirmation gates cannot rescue a gross edge that commissions consume, so honor a pre-registered fee-floor abandon condition rather than tuning.

The backtest-review analyst issued an abandon verdict before optimization: the per-trade edge sat below the round-trip taker cost on a decisive sample, Sharpe CI straddled zero, and profit factor was in the fee-fragile band — exactly the abandon condition the hypothesis had pre-registered. Optimization, analyst promotion review, and risk review were never reached.

A faster-horizon port of a promoted 4H ETH strategy: take long/short directional momentum on ETHUSDT.BINANCE 1H perps only when a concurrent volume surge (2.5x median) confirms genuine participation, with a stronger entry threshold and wider ATR trailing exits to hold down 1H fee drag.

Over a 6.7-year, 1192-trade sample it returned 113.3% total (CAGR 9.94%) but with Sharpe just 0.285 (CI [-0.54, 1.01]), profit factor 1.061, 35.9% win rate, and 44% max drawdown. Critically, avg_trade_return_pct was 0.1475%, below the 0.15% Binance USD-M fee floor.

Abandon Reason

Per-trade edge sits below the venue fee floor on a decisive 1192-trade / 6.7-year sample: avg_trade_return_pct 0.1475% < the 0.15% Binance USD-M floor, profit_factor 1.061 (in the fee_edge band [0.85,1.10]), and Sharpe 0.285 with CI [-0.54, 1.01] straddling zero. The 1H horizon port has too little headroom over the ~0.10% round-trip taker cost — commissions consume the gross edge (fee-fragility, not signal quality; matches and). This is exactly the abandon condition the hypothesis pre-registered ('avg_trade_return_pct under 0.15% with a Sharpe CI straddling zero -> ABANDON, not tune'), and QA's edge concern is borne out. Optimization cannot lift an edge already below trading costs, so this is not worth 2 hours of compute. abandon_class=negative_expectancy.

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