EthIntradayLargeImpulseBarContinuationLS1H

ETH Intraday Large-Impulse-Bar Continuation (Single-Instrument USD-M, Long-Short, 1H, Ride Directional Impulse Bars, Trailing-Stop Winners, Tight Initial Stop)

A long-SHORT, SINGLE-INSTRUMENT, pure-OHLCV intraday TREND-CONTINUATION strategy on ETHUSDT.BINANCE USD-M perpetual futures using 1-HOUR bars. It is built directly on the single fact this session has TRIPLE-CONFIRMED from independent failures: intraday crypto moves TREND/CONTINUE, they do NOT revert. Three fades died anti-predictively this session -- VWAP-excursion mean-reversion (BTC -90%, SOL -80%), volume-spike capitulation-flush BOUNCE (-45%), OI-divergence fade -- each with the same signature (avg_loss ~2x avg_win because the move kept going). The analyst's explicit prescription: stop fading intraday moves; use an OHLCV-derivable large-range-bar / volume-spike CONTINUATION. This strategy does exactly that: it detects a LARGE-IMPULSE BAR (an abnormally large directional 1H move -- the OHLCV signature of momentum / forced flow) and enters WITH the impulse direction, the literal inverse of the abandoned flush-bounce. EVERY design choice encodes a specific kill: (1) DIRECTION = continuation, NOT fade -- the repeatedly-refuted anti-predictive family is closed; continuation is the validated direction. (2) PAYOFF SHAPE = let-winners-run / cut-losers-fast -- the flush-bounce ALSO died on a losing-by-construction 1:3.3 reward:risk (TP 1.5% / SL 5%, needs 77% win rate, got 50%); this uses a TRAILING stop to let the continuation run and a TIGHT initial stop, producing the avg_win > avg_loss asymmetry that fades and the mis-built bounce both lacked. (3) PURE OHLCV, SINGLE INSTRUMENT -- immune to the SEVEN supplementary-stream data-coverage kills (OI/taker/L-S/liquidations) and the basket data-collection wall (thin alts abort runs). (4) ETH -- a confirmed-full-history major, balancing the BTC-heavy book and distinct from the pending BTC VWAP-continuation by both trigger (single-bar impulse vs VWAP-anchor deviation) and asset. (5) NOT a funding-confirmed-trend clone (ETH/BTC-only and uses no funding), NOT VWAP-MR, NOT a fade/bounce, NOT cross-asset ratio, NOT supplementary-primary, NOT engine-incompatible continuous-spot-rebalance. Intentionally minimal: 3 tunables (impulse_z, vol_lookback, trail_atr_mult).

Iteration 2 addresses the backtest-review verdict (over-trading + redundancy) at its root. (1) OVER-TRADING: v1 fired 1210 trades because a close-to-close return z-score at impulse_z 2.5 is loose. The trigger is now a RANGE-EXPANSION impulse that requires BOTH a large-range bar (TR/ATR >= 1.75) AND a decisive directional close, which is intrinsically selective — it lifts per-trade edge above the ~35%-of-gross impact cost by firing only on high-conviction impulses, rather than merely tightening a threshold on the identical signal. (2) MEANINGFUL DIFFERENTIATION: the analyst explicitly required a DIFFERENT TRIGGER CONSTRUCTION (not a looser impulse_z on the same mechanism) versus the passing sibling EthImpulseBarContinuationTrendLS1H, which uses the return-z-score construction. This version uses bar-SHAPE / range-expansion (TR/ATR x intrabar close location) — a genuinely distinct OHLCV momentum signature: a gap-and-fade bar (big return, weak close) that fires the sibling is REJECTED here, and a wide-range bar that slams shut at its extreme fires here even when the net return is moderate. Same validated continuation FAMILY and payoff shape, different trigger. Everything that already passed verification (ATR, trailing-stop machinery, sizing, exits, imports, continuous-signal property) is unchanged; only the impulse computation and its two edge tunables changed, so Layers 1-3 remain green. Whether the book wants a second ETH member of this family remains a Research-Lead call; this delivers the required distinct, selective version if kept.

Not worth the optimization budget: a cost-dominated, toy-capacity intraday churner with only a marginal net edge, and redundant with a superior promoted sibling. 1179 trades at turnover 585 with impact_cost_pct 27.1% + commission_pct_of_gross 8.85% leave profit_factor just 1.13 and Sharpe 0.373 (CI through 0, sharpe_ci_low -0.40); capacity_usd is only $1.36M, so the edge exists only at trivial scale — a structural cost/impact wall that parameter tuning cannot lift to a deployable, promotable level. This is iteration 2 of an over-trading problem and the selectivity fix barely moved it (v1 Sharpe 0.23/impact ~33% -> 0.37/impact 27%), while the mechanism only pays in high-vol years (2020 +56%, 2022 +53%) and has decayed recently (2025 -7.5%). A meaningfully SUPERIOR promoted sibling already exists (ETH-1H return-z impulse continuation, Sharpe ~1.40), so this 'differentiated' range-expansion reformulation is both weaker and redundant, and the 4H-cadence sibling of the same family was already abandoned post-optimization for failing deflated Sharpe with a negative holdout and heavy impact — the same outcome this run is very likely to reach. Not iterate: the developer already spent an iteration cutting trade count with no material gain, and the binding constraints (per-trade edge below impact cost, $1.36M capacity) are structural to this high-turnover ETH-1H variant, not a tunable region.

ETH USD-M perp (ETHUSDT.BINANCE) 1H long/short intraday trend-continuation. Enters WITH a large directional impulse bar detected via a RANGE-EXPANSION construction: impulse = (2*close_location - 1) * (true_range / ATR). A qualifying bar must be genuinely large (TR/ATR >= min_range_ratio) AND close decisively in one direction (|impulse| >= impulse_z). Rides the continuation with an ATR trailing stop (let winners run) behind a tight initial ATR stop (cut losers fast), plus a max-hold time cap. Pure OHLCV, single instrument, leverage 1.0. Three edge tunables: impulse_z, min_range_ratio, trail_atr_mult.

Intended payoff asymmetry works (avg_win $4357 ~2x avg_loss $2216); mechanism pays strongly in high-vol regimes (2020 +56%, 2022 +53%)

Large sample (1179 trades), clean sizing (avg_position_pct 57.6%), trades match the hypothesis (long-short range-expansion continuation, 612 long / 567 short)

Cost-dominated: impact_cost_pct 27.1% + commission_pct_of_gross 8.85% on turnover 585; profit_factor only 1.13 — the gross edge barely survives trading costs

Toy capacity: capacity_usd $1.36M — edge exists only at trivial scale; a standalone promotion blocker

Marginal risk-adjusted return: Sharpe 0.373 with CI through 0 (sharpe_ci_low -0.40), PSR 0.818 — weak before any deflation penalty

Iteration-2 selectivity fix barely helped (v1 Sharpe 0.23/impact ~33% -> now 0.37/impact 27%); recent decay (2025 -7.5%); a SUPERIOR promoted sibling (ETH-1H return-z impulse continuation, Sharpe ~1.40) already covers this space, making this differentiated variant redundant and weaker

Do not optimize this as-is — two issues. (1) OVER-TRADING: impulse_z 2.5 is too loose, firing 1210 trades (turnover 489), which drives impact_cost_pct to 34.9% (impact ~1/3 of gross), caps capacity at only $819k, and leaves Sharpe at 0.23 / PF 1.08 (sharpe_ci_low -0.58, CI straddles 0). Raise the default impulse_z toward the more selective ~3.0 region used by the validated ETH sibling so entries fire only on genuine large-conviction impulses — this should lift the per-trade edge above costs and improve capacity. (2) REDUNDANCY (defer to Research Lead): the companion SOL hypothesis states EthImpulseBarContinuationTrendLS1H already PASSED optimization at Sharpe 1.40 and is in risk_review. This strategy (EthIntradayLargeImpulseBarContinuationLS1H) is a second ETH instance of the same 1H impulse-bar-continuation mechanism — the ETH slot for this family appears already filled by a much stronger sibling. Confirm whether a second ETH impulse-continuation is actually wanted: if not, abandon as book-redundant rather than optimizing a weaker near-clone; if yes, it must be MEANINGFULLY differentiated (different trigger construction or timeframe), not just a looser impulse_z on the same asset/mechanism. The new-asset members of this family (BTC, and any non-ETH major) are the higher-value targets, not a duplicate ETH.

EthIntradayLargeImpulseBarContinuationLS1H was built on the session's triple-confirmed finding that intraday crypto moves continue rather than revert, entering with large range-expansion impulse bars and using a trailing stop to let winners run. In its second iteration it swapped the v1 return-z trigger for a differentiated bar-shape (TR/ATR + decisive close) trigger to attack over-trading, and it did deliver the intended payoff asymmetry (avg win ≈ 2× avg loss) and paid well in high-vol years (2020 +56%, 2022 +53%). But across 1,179 trades the gross edge barely survived costs — impact 27.1% plus 8.85% commission left profit factor 1.13 and Sharpe 0.373 with a CI through zero, and capacity was only $1.36M. The backtest-review gate abandoned it before spending any optimization budget, judging the cost/impact wall and trivial capacity structural rather than tunable, and the variant redundant with and weaker than the already-promoted ETH-1H return-z impulse-continuation sibling.

A high-turnover intraday ETH-1H continuation edge can be real and correctly-shaped (avg_win > avg_loss) yet still be un-promotable because per-trade edge sits below impact cost and capacity is trivial ($1.36M) — a structural wall that tightening selectivity or parameter tuning cannot lift, especially when a superior sibling already occupies the space.

The backtest-review gate abandoned it before optimization: it failed on capacity ($1.36M vs $50M), Sharpe (0.373 vs 0.5), profit factor (1.13 vs 1.2), sharpe_ci_low (-0.40) and impact cost (27.1% vs 25%) — a structural cost/impact wall, and this iteration-2 selectivity fix barely improved the over-trading (v1 Sharpe 0.23/impact ~33% → 0.37/impact 27%) while a superior promoted sibling (ETH-1H return-z impulse continuation, Sharpe ~1.40) already covers the space.

A long-short, single-instrument, pure-OHLCV intraday trend-continuation strategy on ETHUSDT.BINANCE USD-M perp (1H bars) that detected a large range-expansion impulse bar closing decisively in one direction and entered WITH the impulse, riding winners with an ATR trailing stop behind a tight initial stop.

Over 1,179 trades (612 long / 567 short) it returned +227.56% with the intended payoff asymmetry (avg win $4,357 ≈ 2× avg loss $2,216), but Sharpe was only 0.373 (CI low -0.40), profit factor just 1.13, and it was cost-dominated (impact 27.1%, commission 8.85% of gross on turnover 585) with a toy capacity of $1.36M.