BnbDailyRangeExpansionBreakoutLongShort

BNB BINANCE Futures Daily Range-Expansion Breakout Long-Short (Linda Raschke Style, 3-Parameter, OHLCV-Only)

Long+short directional range-expansion breakout strategy on BNBUSDT.BINANCE (USD-M perpetual) using daily bars. This is a mechanism class explicitly distinct from every existing portfolio entry: it requires not just a directional breakout (today's close beyond yesterday's high or low) but ALSO that today's range itself is meaningfully wide compared to recent history (today's true range > 1.5× the 10-day median true range). This dual requirement filters out 'narrow-range breakouts' — situations where price grinds above yesterday's high without conviction — which are the dominant noise mode that breaks generic Donchian and TSM strategies. The 'range expansion' filter is Linda Raschke's (1992) classical contribution to systematic trading, well-documented across futures and equities. The mechanism is mechanically orthogonal to all existing portfolio mechanisms: not Donchian (uses N-bar high, not range comparison); not TSM (uses signed return, not volatility expansion); not Bollinger squeeze (uses width compression, not range expansion at breakout); not Dow theory (uses pattern peaks, not single-bar expansion); not pullback (buys dips, not breakout days); not ORB (no session anchor). This is a single-bar high-conviction signal that's mechanically uncorrelated with the existing book. BNB BINANCE futures is the chosen venue because: (a) BNB is the 4th-largest crypto with ~$500M-$1B daily futures volume (massively above the ETC liquidity floor that killed cross-instrument generalization), (b) BNB has 5+ years of continuous Binance perp data (long enough for robust 3-window walk-forward), (c) BNB's distinct fundamental drivers — exchange token, quarterly burns, BNB Chain ecosystem launches — make it less correlated with BTC-driven momentum signals than ETH/SOL, providing genuine diversification when added alongside BTC/ETH/SOL TSM strategies already in portfolio, (d) Binance futures venue allows BOTH directions natively (SPOT cannot), filling the long+short quota gap (currently 12.0% vs ~45% target — the largest under-represented DIRECTION bucket), (e) the 0.10% RT fee floor on Binance futures is appropriate for daily-frequency strategies (clearable by typical 4-10% daily moves). Pure OHLCV — NO supplementary data dependency (avoids the 14+ session abandons from infra gaps). 3 parameters total: range_lookback=10, range_multiplier=1.5, atr_stop_mult=1.5. NOT a cross-venue port (the HL SOL TSM abandon proved cross-venue claims are unreliable). NOT a cross-instrument port (this is a freshly-motivated test of a mechanism not yet tried on this asset). Hard stop included from start.

Single-instrument long+short daily strategy implemented with the standard FactoryStrategy 4-hook pattern, so the base _process_primary drives entry/exit/sizing on BNBUSDT directly (no centralized multi-leg management needed). The dual Raschke requirement is computed in calculate_signal: a directional breakout (close beyond the PRIOR bar's high/low, cached as _breakout_dir) gated by a range-expansion test (today's true range > range_multiplier x the MEDIAN true range over range_lookback, cached as _range_expanded). Median (not mean) makes the filter robust to a single prior outlier range day. calculate_signal returns a continuous, every-bar diagnostic ((close-midpoint)/ATR x range_ratio) so Layer-2 frozen-signal detection passes, while the binary entry decision lives in should_enter reading the cached booleans. should_exit applies an ATR hard stop (atr_stop_mult x entry ATR) and a reverse-breakout exit, with defensive reconstruction of entry context from the open position if state is lost. position_size is capital-relative: risk_pct of equity divided by the ATR stop distance, capped at max_notional_frac of equity, quantized down to BNB's 2-dp step and checked against the $5 min notional. Futures venue is required because the strategy goes both long and short (filling the under-represented long+short direction bucket) and BNB perps have deep liquidity and 5+ years of history for robust walk-forward. Three tunable parameters (range_lookback, range_multiplier, atr_stop_mult) keep the optimization surface minimal; leverage stays 1.0 and is referenced in sizing without amplifying beyond equity, so no unused-leverage gate applies.

Failed deflated Sharpe: DSR=0.6738 (<0.95), is_significant=FALSE, sharpe_ci_low=-0.211 (CI straddles 0); the optimized Sharpe 0.66 is only marginally above the expected-max luck bar 0.556 over 225 trials. PBO 0.5767 (>0.5) and walk-forward is_overfitted=TRUE (avg IS 0.97 -> avg OOS -0.35, with OOS windows [0.30, 1.30, -2.65]) confirm overfitting. Root cause is the outlier-dependence flagged at pre-optimization: optimization drove return_kurtosis from 542 to 793 and tail_ratio to 0.19 while thinning to 82 trades, so the entire edge rests on a handful of explosive single days (2021-04-18 +123%, 2025-09-22 +29.5%); deflated Sharpe's skew/kurtosis adjustment correctly discounts this to 0.67, and the PSR 0.9975 / DSR 0.6738 split is the diagnostic tell. The passing holdout (1.72) and 0-cliff sensitivity are single-window/in-sample artifacts that cannot override a CI straddling zero, PBO 0.58, an overfit walk-forward with a -2.65 OOS window, and a sub-significant DSR. Not iterate (attempt 1 of 2): the sensitivity surface is flat (~0.5 everywhere with no high-Sharpe plateau to tune toward), the edge is intrinsically fat-tail-carried (kurtosis 793 on 82 trades), and one OOS window is catastrophically negative — there is no robust region a second sweep would find; it would re-select another outlier-dependent, sub-significant config. Not revise_hypothesis: this is the fat-tail arm of the single-asset directional range-expansion breakout family, not a proven mechanism stranded on a dead target. FAILURE PATTERN: single-asset long-short range-expansion/breakout on a crypto major (BNB) produces a headline carried by a few explosive expansion days (kurtosis 540-790, tail_ratio <0.2); after best-of-225 selection it fails deflated Sharpe (0.67, barely above the 0.56 luck bar) with a CI straddling zero, PBO 0.58, and is_overfitted=TRUE with a deeply negative OOS window — the same outlier-luck signature as the abandoned SOL Keltner (kurtosis 124, DSR 0.65) and XRP Chandelier (DSR 0.01) trend siblings. A passing single-window holdout and a clean sensitivity grid do not rescue an edge resting on a handful of fat-tail days.

Long+short daily range-expansion breakout on BNBUSDT.BINANCE USD-M perpetual (Linda Raschke 1992 style). Enters long when the daily close breaks above the prior bar's high AND today's true range exceeds 1.5x the 10-day median true range; enters short on the mirror condition (close below prior low + wide-range day). The range-expansion filter rejects narrow-range grind-through breakouts. Exits on an ATR(14) x1.5 hard stop from entry or a reverse range-expansion breakout. Risk-based sizing (2% equity risk over the ATR stop distance, capped at 95% notional). Pure OHLCV, no supplementary data, leverage 1.0, three tunable parameters.

Genuinely distinct, liquid, long+short mechanism: 99 trades (51 long / 48 short), capacity $478M with impact 1.45%, PF 1.91, max_DD 22.8% — fills the under-represented direction bucket

Edge is distributed across recent years (2024 +32%, 2025 +20%, 2026 +9%; positive in 5 of 7 years), unlike the bull-concentrated breakout-family siblings that decayed to negative

No correctness bug; clean risk-based sizing (avg_position_pct 40%, no liquidation)

Extreme outlier-dependence: return_kurtosis 542, skew 21, tail_ratio 0.37; the +77.8% day on 2021-03-22 dominates 2021 — deflated Sharpe (which penalizes skew/kurtosis) will be the binding optimization gate, as it was for the SOL Keltner / XRP Chandelier fat-tail siblings

Marginal raw Sharpe 0.65 with sharpe_ci_low -0.25 (CI straddles 0)

Trade frequency ~16/yr split across 3 walk-forward windows is adequate but not abundant

Clean robustness surface in places: sensitivity PASSED with 0 cliffs, holdout PASSED (1.72), and recent annual returns positive (2024 +32%, 2025 +20%, 2026 +26%)

Liquid asset, low fee drag (commission 1.27% of gross), genuinely distinct long+short mechanism

Failed deflated Sharpe: DSR 0.6738 (<0.95), is_significant=FALSE, sharpe_ci_low -0.211 (CI straddles 0); optimized Sharpe 0.66 only marginally above the expected-max luck bar 0.556

PBO 0.5767 (>0.5); walk-forward is_overfitted=TRUE with avg IS 0.97 collapsing to avg OOS -0.35 and one OOS window -2.647

Outlier-dependence worsened under optimization: return_kurtosis 793 (base 542), tail_ratio 0.19 — the edge rests on a few explosive days (2021-04-18 +123%, 2025-09-22 +29.5%)

PSR 0.9975 vs DSR 0.6738 split confirms the headline is fat-tail/best-of-225 inflation, not a generalizing edge

BnbDailyRangeExpansionBreakoutLongShort introduced a genuinely distinct, liquid long+short mechanism — Raschke-style range-expansion breakouts on BNB daily — to fill the portfolio's most under-represented direction bucket, and its base backtest looked promising (301% return, PF 1.91, edge spread across recent years) enough to earn an 'optimize' verdict despite extreme kurtosis. Optimization improved the profit factor but deepened the outlier-dependence (kurtosis 793 on 82 trades), and the forward gates split: the holdout passed (1.72) and sensitivity was clean, but the deflated Sharpe was only 0.67 against a 0.556 luck bar, the CI straddled zero, PBO was 0.58, and the walk-forward was overfit with a -2.65 OOS window. The PSR 0.9975 / DSR 0.6738 split confirmed the headline rested on a few explosive days. On its first iteration the analyst abandoned it, citing the same outlier-luck signature as the abandoned SOL Keltner and XRP Chandelier fat-tail siblings.

A passing single-window holdout and a clean sensitivity grid cannot rescue an edge resting on a handful of explosive expansion days — deflated Sharpe's skew/kurtosis adjustment correctly discounts a fat-tail-carried headline (kurtosis 793, tail ratio 0.19), and the PSR-high / DSR-low split is the diagnostic tell of outlier luck rather than a generalizing edge.

Despite a passing holdout (Sharpe 1.72), clean sensitivity (0 cliffs), and positive recent annual returns, the analyst abandoned it post-optimization because the deflated Sharpe was 0.674 (below the 0.95 bar, only marginally above the 0.556 luck bar) with the CI straddling zero, PBO was 0.58, and walk-forward was_overfitted=TRUE (avg IS 0.97 collapsing to avg OOS -0.35 with one window at -2.65).

A single-instrument, OHLCV-only long/short Linda Raschke-style range-expansion breakout on BNBUSDT daily futures — entering long when the close exceeds the prior bar's high (short on the prior low) but only on wide-range days where today's true range exceeds 1.5x the 10-day median true range, filtering out low-conviction narrow-range breakouts, with an ATR hard stop and reverse-signal exit and just 3 tunable parameters.

The base backtest produced a 301.1% return (CAGR 23.1%) over 99 trades (51 long, 48 short) with profit factor 1.91, max drawdown 22.8%, and edge distributed across recent years, but Sharpe was only 0.65 (CI low -0.25) and the distribution was extremely fat-tailed (kurtosis 542, skew 21). Optimization raised PF to 2.69 (Sharpe 0.66, 82 trades, max DD 16.4%) but worsened the outlier-dependence (kurtosis 793).