Strategies

BTC 4H Wide-Range Bullish Bar Volatility-Expansion Momentum Long with Daily Bull Regime Filter

Outcome: Abandoned

BtcFourHourWideRangeBullishBarVolatilityExpansionMomentumLong

Outcome Summary

This strategy isolated a genuinely distinct signal — a single 4H bar's total range relative to recent volatility, with a bullish upper close — on BTC, and earned an 'optimize' verdict on a thin but positive, high-density base edge (356 trades, Sharpe 0.39), though flagged for heavy costs, tiny capacity, and recent decay. Optimization produced an encouraging clean walk-forward (Sharpe 1.83, non-overfit, all OOS windows positive), but the decisive gates rejected it: deflated Sharpe 0.0 against a 5.50 luck bar, a negative-Sharpe holdout (-2.82), and six core-parameter cliffs, with optimized annual returns fading to -4.2% in 2026. The analyst abandoned it at the analyzing stage on its second iteration as a knife's-edge, sub-luck-bar selection artifact with no robust plateau to tune toward — the same signature as the abandoned ADA volume-breakout sibling — so it stopped before risk review.

Strategy report

Hypothesis

A long-only single-instrument RANGE-EXPANSION MOMENTUM strategy on BTCUSDT perpetual futures using 4-hour bars and OHLCV-only data, with a daily-bar bull regime filter. This proposal is designed to AVOID every documented failure pattern this session: (1) it uses OHLCV-only data (avoiding the supplementary-data infrastructure failures that killed funding-rate, liquidation, and OI strategies), (2) it operates on BTC 4H (the most-validated data path), (3) it uses a fundamentally different signal type than every existing BTC strategy, (4) it's a high-density signal (30-50 trades/year, well above the analyst's 30-trade walk-forward threshold), (5) it's NOT a mean-reversion or momentum-continuation transfer from any failed strategy, (6) it's NOT a symmetric inversion. The trigger is the classical Wide Range Bar (WRB) concept documented by Toby Crabel (1990) and Linda Raschke ('Street Smarts' 1996): a single 4H bar whose range (high - low) materially exceeds the recent volatility baseline, combined with bullish close direction. Crucially, WRB captures a DIFFERENT signal-type than all existing BTC strategies: (a) BtcFourHourVolumeBreakoutLong uses Donchian-high break + volume (positional + volume); (b) BtcFourHourBollingerUpperBandMomentumContinuationLong uses std-dev distance from mean (statistical extension); (c) BtcFourHourAggressorBuyImbalanceMomentumLong uses order-flow direction (microstructure). NONE measures the bar's TOTAL RANGE relative to recent volatility — that's a separate single-bar geometric property. Economic mechanism: when a BTC 4H bar produces 1.5×+ normal range AND closes bullishly in the upper portion of that range, this signals (a) a capital-flow event (institutional ETF rebalancing, macro catalyst, news-driven momentum) that brought meaningful directional capital, (b) buyers maintained control through the wide range to the close (not a wick-then-reverse pattern), (c) systematic momentum traders observe the range-burst and enter, providing 4-12 hour follow-through. This is distinct from BB-Upper (which captures multi-bar extension) and from Vol Breakout (which requires Donchian high — a positional condition independent of bar range). Two bars with identical Vol Breakout signals can have very different range-expansion characteristics. Only 4 explicit parameters (range multiplier, range lookback, regime SMA, stop pct).

Iteration-2 minimal fix for the Layer-3 'No trades produced' failure. Root cause: the entry required self._regime_ok, which read from self._daily_close populated only by on_extra_bar from a separate DAILY extra-bar feed; that feed never populated in the sandbox, so _regime_on() was permanently False and the WRB trigger (range>=1.5x baseline + bullish-upper close, which fires often on BTC 4H) never produced an entry over all 14,019 bars. I changed ONLY the regime computation: _regime_on() now computes the identical 50-day bull filter from the existing 4H close buffer (regime_sma_period*bars_per_day = 300-bar SMA), and the daily extra feed is removed from config (single feed). min_bars_required and the buffer cap were raised to guarantee the 300-bar regime window is available before trading. Imports, the WRB range-expansion signal math, entry/exit/sizing logic, and the BINANCE futures routing (leverage 1.0) are unchanged, so Layers 1 and 2 stay green while the strategy now produces long entries whenever BTC prints a wide bullish range bar within a confirmed 4H-derived uptrend.

Encouraging walk-forward that fails the decisive forward and multiple-testing gates on a thin, knife's-edge edge. The optimizer produced a clean-looking in-sample/walk-forward profile (Sharpe 1.83, PF 1.45, max_DD 8%, is_overfitted=FALSE, all 3 OOS windows strongly positive avg 3.45, PBO 0.4524), but the gates that test generalization reject it: deflated_sharpe 0.0 (is_significant FALSE) with the optimized Sharpe 1.83 FAR below the expected-max luck bar 5.50 over 225 trials — after honest multiple-testing correction the edge is indistinguishable from best-of-225 selection noise (sharpe_ci_low just 0.096); the HOLDOUT FAILED with a NEGATIVE Sharpe -2.82 (ratio -0.818) — the recent one-shot forward window loses money; and sensitivity FAILED with SIX cliffs (range_mult, bars_per_day, close_pos_min, max_hold_bars, range_lookback, regime_sma_period) meaning nearly every core parameter sits on a knife's edge with no robust plateau. The recent decay flagged at pre-optimization materialized (optimized 2024 +1.3% -> 2025 -1.9% -> 2026 -4.2%, rolling Sharpe negative through 2026). Not promote: DSR 0.0 below the luck bar + negative holdout + 6 cliffs are each disqualifying; the positive walk-forward is in-sample best-of-225 selection. Not iterate (attempt 1 of 2): the failure is sub-luck-bar deflation + a negative holdout + pervasive cliffs across the core params + recent decay, not a robust region the sweep under-explored — with 6 core-param cliffs there is no plateau to tune toward, so a second sweep re-finds another knife's-edge, sub-luck-bar config that fails the recent holdout. Not revise_hypothesis: WRB momentum on BTC is a thin single-asset edge with no proven sibling stranded on a dead target — it simply does not deflate. FAILURE PATTERN: long-only Wide-Range-Bar volatility-expansion momentum on BTC 4H optimizes to a flashy Sharpe 1.83 with a clean NON-overfit positive walk-forward (avg OOS 3.45, PBO 0.45), yet fails deflated Sharpe (0.0, far below the 5.50 luck bar), a deeply negative holdout (-2.82), and sensitivity with 6 core-param cliffs — a thin, fee-fragile single-bar momentum edge that is front-loaded and decaying (2025/2026 negative). A positive walk-forward with a negative holdout, DSR 0.0, and pervasive cliffs is best-of-225 selection on a knife's edge, the same signature as the abandoned ADA volume-breakout sibling.

Implementation

Long-only Wide Range Bar volatility-expansion momentum on BTCUSDT.BINANCE USD-M futures, 4H bars, pure OHLCV single feed. Enters LONG when a 4H bar's range is >= 1.5x the prior 20-bar mean range AND it closes bullishly in the upper 60% of its range AND the 50-day bull regime is on (now computed from the same 4H buffer as a 300-bar SMA instead of a separate daily feed). Exits on a 6% take-profit, a 4% stop, a regime flip, or a 6-bar time stop. The signal is a continuous range-expansion score signed by close position; the discrete WRB gate drives entry. leverage 1.0.

Verification Results

Verification failed (Layer 3 — sandbox backtest): No trades produced Bar type used: BTCUSDT.BINANCE-4-HOUR-LAST-EXTERNAL, Bars processed: 14040 Diagnostics: should_enter() returned a side 0 times over 14019 evaluated bars -> your ENTRY CONDITION never triggered. Loosen the entry logic / thresholds. Ensure your strategy produces trades with the given data and parameters.

Backtest Review

Sharpe: 0.39
Total return: 33.39%
Max drawdown: 19.84%
Trades: 356
Win rate: 48.6%
Profit factor: 1.09

Highest trade density of the batch: 356 signals -> 356 submitted, 0 dropped (~55/yr) — ample sample for sensitivity/walk-forward; ITER-2 regime fix works

Implemented correctly and trades as specified: all long, exposure 14.6%, avg hold ~22h, WRB range-expansion + bullish-upper-close gate firing as designed

Net-positive edge (unlike today's no-edge abandons): PF 1.092, Sharpe 0.386, total +33.4%, positive in 4 of 7 years

Quality-filter design gives the optimizer a concrete lever (raise range_mult/close_pos_min) to lift avg-trade-return and cut fee drag

Thin per-trade edge ~0.11% (expectancy $59.6 on ~$55k notional) — at/below the 0.15% futures viability target; PF 1.092 barely above 1

Heavy cost drag: commission_pct_of_gross 8.77%; impact_cost_pct 21.9% — fees/impact consume a large share of gross

Tiny capacity (~$2.08M): edge real only at toy scale — a promotion blocker the optimizer must address by trading more selectively

Regime-mixed returns (2021 -10.5%, 2024 -4.0%) and rolling Sharpe swings deeply negative in stretches (2024 ~-7) — generalization risk for the holdout

Analysis

Walk-forward is_overfitted=FALSE with all 3 OOS windows strongly positive (avg OOS 3.45), PBO 0.4524 (<0.5), PSR 0.9781

Optimized config looks clean in isolation: Sharpe 1.83, PF 1.45, Sortino 3.65, max_drawdown 8.1%, capacity improved to ~$19M by trading more selectively

Implemented correctly with high density (356 base / 157 optimized trades); ITER-2 regime fix works

Failed deflated Sharpe: DSR=0.0 (bar 0.95) with the optimized Sharpe 1.83 FAR below the expected-max luck bar 5.50 over 225 trials — indistinguishable from best-of-N selection

Holdout FAILED with a NEGATIVE Sharpe -2.82 (ratio -0.818): the recent one-shot forward window loses money

Sensitivity FAILED with 6 cliffs — range_mult, bars_per_day, close_pos_min, max_hold_bars, range_lookback, regime_sma_period all cliffs; no robust plateau, the config is a knife's-edge point pick

is_significant FALSE; sharpe_ci_low 0.096 (barely above 0)

Recent decay (flagged at pre-opt): optimized 2024 +1.3% -> 2025 -1.9% -> 2026 -4.2%, rolling Sharpe negative through 2026

Heavy fee drag (commission 8.75% of gross) — fee-fragile thin edge

Outcome Summary

A clean, non-overfit positive walk-forward is not enough — when deflated Sharpe is 0.0 below the luck bar, the holdout is negative, and nearly every core parameter sits on a sensitivity cliff, the result is best-of-N selection on a knife's edge, not a robust edge; a thin, fee-fragile single-bar momentum signal that is front-loaded and decaying does not deflate.

It cleared the backtest-review gate with an 'optimize' verdict and posted an encouraging non-overfit walk-forward, but the post-optimization analyst abandoned it: deflated Sharpe 0.0 (<0.95) with the optimized Sharpe 1.83 far below the 5.50 expected-max over 225 trials, the holdout FAILED with a negative Sharpe -2.82, and sensitivity failed with 6 cliffs across nearly every core parameter — a knife's-edge, fee-fragile edge that had decayed to negative by 2025-2026.

A long-only Wide-Range-Bar volatility-expansion momentum strategy on BTCUSDT 4H Binance futures (1.0x leverage, 0.50 position) — entering long when a 4H bar's range exceeded 1.5x its 20-bar baseline AND it closed bullishly in the upper 60% of that range, inside a 50-day bull regime — a single-bar geometric range property distinct from the portfolio's Donchian/Bollinger/order-flow BTC strategies, with TP/stop/regime/time exits.

The base backtest had a thin positive edge — 356 trades (~55/yr), +33.4% total, Sharpe 0.39, profit factor 1.092, expectancy +$59.6/trade (~0.11%, below the 0.15% viability target) — burdened by heavy costs (8.8% commission, 21.9% impact) and tiny ~$2.08M capacity; the optimized config looked clean (Sharpe 1.83, PF 1.45, 8.1% drawdown) and even passed walk-forward (not overfit, all 3 OOS windows strongly positive).

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