BtcDailyAtrCompressionRangeExpansionBreakoutRegimeLong

BTC Daily ATR-Compression Range Expansion Breakout Long — Volatility-Regime-Conditional Entry

A LONG-ONLY breakout strategy on BTCUSDT.BINANCE daily bars that enters only when two conditions stack: (1) the market is in a confirmed volatility-compression regime (ATR(14) sits in the bottom 15th percentile of its trailing 90-day distribution — Tony Crabel's NR7/range-contraction concept), AND (2) a directional range-expansion bar fires (today's high breaks above the prior 5-day range high by at least 0.5 × ATR(14)). This is structurally distinct from every mechanism falsified in this session: it is NOT Tier 1 trend-following (no MA crossover, no Supertrend, no Parabolic SAR — the only trend filter is the breakout itself), NOT a Tier 2 fixed-lookback Donchian channel breakout (entry requires the volatility regime to be in compression, otherwise the same 5-day breakout is skipped — most days are filtered out), NOT a Tier 3 oscillator (no RSI/Stoch/MACD), NOT a volume breakout (the pipeline-resident BTC/BNB/SOL Volume Breakouts use volume as the signal — this strategy ignores volume entirely), NOT cross-asset, NOT cross-sectional, NOT calendar, NOT sub-daily, NOT accumulation, NOT supp-data dependent, NOT cross-venue. The edge mechanism is the empirical clustering of large directional moves immediately AFTER low-volatility consolidation periods (documented in Crabel 1990, Raschke-Connors 1996 'Street Smarts', and confirmed in crypto by Vivar-Vaytsekhovsky 2022). Trade frequency is naturally low (~15-25 entries per year on BTC daily bars) so fees do not dominate, and the holding period is short (2-7 days) so each trade resolves before mean reversion sets in. Daily bars on BTCUSDT.BINANCE have well-audited multi-year history — no MATIC-style deprecation risk, no Hyperliquid coverage limit, no BINANCE_SPOT 4H ingestion gap.

Iteration 2 addresses the sole Layer-3 blocker: trade density (33 trades / ~6yr, ~5.5/yr, below the ~62 walk-forward floor and the strategy's own 15-25/yr projection; sharpe_ci_low -0.96 straddling 0). Per the analyst's explicit prescription, I raise density by loosening the two TUNABLE knobs while keeping the compression-THEN-expansion mechanism fully intact: (1) compression_pct 15 -> 30 selects the bottom THIRD of the trailing ATR distribution — still a genuine lower-vol regime that filters ~70% of days, but fires ~2x more often; (2) breakout_atr_mult 0.5 -> 0.25 lets more compressed-regime expansions qualify while still requiring a new-range expansion beyond the prior high, so it does NOT degenerate into a plain Donchian breakout (the regime gate remains the edge). breakout_lookback stays 5 and all exit/risk/sizing logic is unchanged. Combined, these should push trade_count comfortably above 60 (roughly doubling from the compression loosening, plus additional qualifying expansions from the lower multiple) so the 225-trial optimization has adequate sample for the deflated-Sharpe gate, without gutting the per-trade edge (PF/Sharpe driven by the compression clustering, which is preserved). Signal computation, look-ahead safety (regime/breakout measured off the prior bar's ATR), and all Layer 1-2 structure are untouched.

Failed the multiple-testing and forward gates decisively on attempt 1, confirming the pre-optimization significance risk. Deflated Sharpe is 0.0176 with the optimized Sharpe (1.021) FAR below the 225-trial expected-max luck bar (2.8901), is_significant=FALSE, sharpe_ci_low -0.6953 (CI deeply through 0), PBO 0.5807. The walk-forward is is_overfitted=TRUE with avg IS 2.464 collapsing to avg OOS -1.283 and OOS windows [0.0, 0.0, -3.849] — two zero-trade windows plus one deeply negative, because the compression gate filters ~70% of days and structurally starves the forward windows. Sensitivity FAILED with 4 cliffs on the core defining parameters (atr_period, pct_window, compression_pct, breakout_lookback), so the edge exists only at fragile cliff-edge values with no robust plateau. The holdout 'pass' is spurious (ratio 0 computed against the negative WF-OOS; holdout_sharpe 5.124 is a tiny-window artifact), and the optimized equity is near-flat (CAGR 1.52%) so its Sharpe is a low-volatility artifact, not a real edge. Not iterate: with 4 cliffs on the core knobs, a best-of-225 Sharpe far below the luck bar, and a compression gate that produces zero-trade OOS windows, there is no robust region to tune toward — a second sweep re-finds the same fragile cliff config with negative forward average. Not revise_hypothesis: the compression-breakout mechanism is not a proven sibling stranded on a dead target (BTC daily is a fine instrument); the edge is simply too weak and too fragile to survive deflation. FAILURE PATTERN: a differentiated, low-drawdown, academically-grounded regime-gated breakout with a marginal base (Sharpe 0.57, CI through 0, PSR 0.786) does not survive optimization — best-of-225 lifts the point Sharpe to 1.02 but cannot clear the 2.89 luck bar (DSR 0.018), the regime filter that makes it distinctive also starves the walk-forward windows into zero-trade/negative OOS, and its defining parameters are all cliffs. A clean, low-DD, plausible mechanism is not enough when the underlying edge is too thin and too regime-fragile to be statistically distinguishable from best-of-N noise.

Long-only volatility-regime breakout on BTCUSDT.BINANCE daily bars. Enters when ATR(14) as of the prior bar sits in the bottom 30th percentile of its trailing 90-day distribution (compression regime) AND today's high breaks the prior 5-day range high by at least 0.25xATR (directional range expansion). Exits on a 1.5xATR stop, 3.0xATR target, or a 7-bar time cap. ATR-risk-based sizing (1% equity per trade), pure OHLCV, single instrument, leverage 1.0.

Functioning, differentiated edge: profit_factor 1.271 (>1.2), positive expectancy, win_rate 49%, not a PF<1 loser

Iteration-2 density fix worked: 65 trades (above the ~62 walk-forward floor), trade pattern matches hypothesis (long-only compression-gated breakouts, short holds)

Well-behaved, non-artifact returns: kurtosis 19.3 (moderate), positive skew 1.57, tail_ratio 1.17 — edge is NOT carried by 2-3 outlier days

Strong diversification profile: max_drawdown 10.4%, exposure 18.9%, beta 0.025, low benchmark correlation 0.35 — a genuine low-vol regime diversifier

Academically grounded, distinct mechanism (Crabel/Raschke-Connors compression-then-expansion), differentiated from the pipeline's Donchian/volume breakouts

Thin base Sharpe 0.570 with sharpe_ci_low -0.924 (CI straddles 0) and PSR 0.786 — NOT yet significant; real risk of failing the 225-trial deflated-Sharpe gate after optimization

Tiny absolute return (CAGR 1.46%, total 11.4%) — an artifact of conservative sizing (risk_pct 0.01, 23.5% avg position), but it means the strategy only matters if the Sharpe is real and can be scaled

Mixed annual returns (2021 -1.4%, 2022 -7.9%, 2025 -0.03%) — the edge is inconsistent across regimes

Single-asset daily breakouts on majors have repeatedly failed deflation in this session

Differentiated, academically-grounded mechanism (Crabel compression-then-expansion), low drawdown (opt 7.0%), low beta (0.015) — a clean diversification profile in principle

Not an implausible artifact in the base (kurtosis moderate, tail_ratio ~1.6-1.97); trade count adequate (65) after the iter-2 density fix

Failed deflated Sharpe: DSR 0.0176, optimized Sharpe 1.021 FAR below the 225-trial expected-max luck bar 2.8901, is_significant=FALSE

sharpe_ci_low -0.6953 (CI deeply through 0); PBO 0.5807 (>0.5)

Walk-forward is_overfitted=TRUE: avg IS 2.464 -> avg OOS -1.283, with OOS windows [0.0, 0.0, -3.849] (two zero-trade windows + one deeply negative)

Sensitivity FAILED: 4 cliffs on the core defining parameters (atr_period, pct_window, compression_pct, breakout_lookback) — no robust plateau

Holdout 'pass' is spurious: ratio 0 computed against a negative WF-OOS; holdout_sharpe 5.124 is a tiny-window artifact

Optimized equity is near-flat (CAGR 1.52%, most annual returns <2.5%) — high Sharpe is a low-volatility artifact, not a real edge; compression gate structurally starves OOS windows

Strong per-trade quality (Sharpe 1.20, PF 1.63, 54.5% win, max_dd 5.6%, positive 5/7 years) but the sample is too thin to optimize: only 33 trades over ~6 years (~5.5/yr), below the ~62 walk-forward floor and far below the hypothesis's own 15-25/yr projection. A 225-trial optimization on 33 trades fits noise and will very likely fail the deflated-Sharpe gate (the 37-trade DOGE breakout came back DSR 0.0; sharpe_ci_low here is already -0.96, straddling 0). FIX — increase density to >60 trades WITHOUT gutting the compression edge (it's tunable, unlike a fixed-period crossover): (1) loosen the compression gate — raise compression_pct from 15 toward ~25-35 (the bottom-third of the ATR distribution still captures 'lower-vol regime' while firing far more often); (2) optionally lower breakout_atr_mult (0.5 -> 0.25) or shorten breakout_lookback so more compressed-regime expansions qualify. CRITICAL: keep the compression-THEN-expansion mechanism intact (do not loosen so far it becomes a plain Donchian breakout — the regime filter IS the edge). Re-backtest and confirm trade_count > 60 AND the per-trade edge holds (PF stays >~1.3, Sharpe doesn't collapse) AND a positive sharpe_ci_low before resubmitting for optimization. If a denser config can't exceed ~60 trades while preserving the edge, the compression mechanism is too rare on BTC daily to validate standalone and should be abandoned (or extended to a multi-instrument basket of majors to pool the signal).

This strategy applied Crabel/Raschke-Connors compression-then-expansion logic to BTC daily bars, entering long only when low-volatility consolidation was followed by a range-expansion breakout, and its iteration-2 fix loosened two density knobs to raise trade count above the walk-forward floor. It passed the pre-optimization review (verdict: optimize) as a differentiated, low-drawdown, low-beta diversifier, but with a flagged-marginal base Sharpe of 0.570 whose confidence interval straddled zero. Optimization confirmed the risk: the deflated Sharpe was 0.0176, the optimized Sharpe of 1.021 fell far short of the 2.89 luck bar, the walk-forward was overfit with a negative average OOS, and sensitivity failed with four cliffs on the core parameters. The analyst concluded the compression gate that made it distinctive also starved its forward windows into zero-trade/negative territory, judged the edge too weak and fragile to survive deflation, and abandoned it after exhausting both attempts.

A regime filter that makes a breakout distinctive can also structurally starve the walk-forward windows — filtering ~70% of days produced zero-trade and negative OOS periods — so a clean, low-drawdown, academically-grounded mechanism still fails when its edge is too thin and too regime-fragile to be distinguished from best-of-N noise, and its defining parameters are all cliffs rather than a robust plateau.

The analyst abandoned it at the post-optimization ANALYZING stage (iteration 2 of 2): it failed the deflated Sharpe test (DSR 0.0176 vs 0.95) with the optimized Sharpe 1.021 far below the 225-trial expected-max luck bar of 2.8901, an overfitted walk-forward (avg IS 2.464 collapsing to avg OOS -1.283 with windows [0.0, 0.0, -3.849]), PBO 0.5807, and a failed sensitivity surface showing 4 cliffs on the core defining parameters.

A long-only BTCUSDT daily breakout that entered only when ATR(14) volatility sat in a compressed regime (bottom-third percentile of its trailing 90-day distribution after the iteration-2 density fix) AND a directional range-expansion bar broke the prior 5-day range high, exploiting the tendency of large moves to cluster after low-volatility consolidation.

The base backtest showed a thin edge — Sharpe 0.570, total return 11.4% (CAGR 1.46%), 65 trades, profit factor 1.271, max drawdown 10.4%, and sharpe_ci_low -0.924; optimization lifted the point Sharpe to 1.021 with a 7.0% drawdown but a near-flat 1.52% CAGR and deflated Sharpe of just 0.0176.