BtcFourHourVolatilityBreakoutRegimeLong

BTC Volatility Breakout Long-Only (Single-Asset BINANCE 4H, Vol Regime Transition Low-to-High, Bars-Only)

Single-asset long-only strategy implementing the analyst's EXPLICIT recommendation following the BtcPostVolSpikeRecovery failure: 'Volatility EXPANSION strategies: enter long when vol REGIME SHIFTS from low to high (volatility breakout, opposite sign convention). Sinclair 2010 also documents this — and it may work better in crypto where vol expansions tend to coincide with bull rallies.' The post-vol-spike RECOVERY strategy (buy the calm after the storm) has been conclusively shown NOT to work in crypto (PF 0.96 over 150 trades) because crypto vol spikes often precede continued downside, not recoveries. This proposal implements the OPPOSITE-direction mechanism: enter long DURING vol expansion (volatility regime breakout) when price action is bullish. The thesis: in crypto, sudden vol expansions COMBINED with upward price action signal the early phase of breakout rallies (2020-Q4 BTC rally, 2024-Q1 ETF rally, 2024-Q4 Trump rally all started with vol expansions accompanying price breakouts). The combination filter — vol expansion AND price-above-MA — discriminates between bullish vol expansions (rallies) and bearish vol expansions (capitulations). This addresses EVERY documented failure mode: (a) bars-only — NO supp_data dependency (analyst has now flagged 7+ failures from supp_data fallbacks: premium_index, L/S ratio, taker_ratio, HL funding, COIN-M all unavailable in production), (b) single-asset BTC, (c) single-venue Binance USD-M (sane backtest engine), (d) long-only (no short-leg blowup), (e) MID-horizon (4H bars with 1-7 day holds — NOT macro horizon which the analyst confirmed is structurally incompatible with factory walk-forward), (f) genuinely novel mechanism class — vol BREAKOUT differs from vol REGIME-SWITCHING (BtcPostVolSpikeRecovery), vol COMPRESSION (BtcSpotVolatilityCompression), and KER/Aroon/ADX (which are trend-quality not vol-regime). Compliant with EVERY documented factory infrastructure constraint. NO fallback path: if Parkinson vol can't be computed (warm-up insufficient), strategy produces zero trades — honest data behavior per analyst guidance.

Iteration 3 is a config-routing fix only, per the feedback: the prior backtest was misrouted to BTCUSD.HYPERLIQUID-1-MINUTE over 5 days with empty parameters {}, while the hypothesis and code target BTCUSDT.BINANCE 4H over full history. The strategy code is unchanged (it already passed static + synthetic layers). The corrected FactoryStrategyConfig sets instrument_id='BTCUSDT.BINANCE' and bar_type='BTCUSDT.BINANCE-4-HOUR-LAST-EXTERNAL' so the run spans 2020-2026 (metrics_reliable=true, non-empty walk-forward/holdout windows), and populates the densified default parameters that came through empty. vol_mult stays OUT of config.parameters so it remains pinned at the code default 1.40 and excluded from the optimizer sweep (the iteration-2 over-selection fix); only the stable, non-cliff params (ma_period, stop_loss_pct, take_profit_pct, baseline_window) are exposed for optimization. This reproduces the base Sharpe-2.36 run so its true edge (PF>1.2, non-zero-straddling CI required) can be judged.

A genuinely better-than-usual result that still fails two decisive gates on the final optimization attempt (2 of 2). The walk-forward is honestly non-overfit (is_overfitted=false, all three OOS windows positive [1.39, 1.81, 1.19], avg OOS 1.46), sensitivity is a clean 0-cliff surface, and sharpe_ci_low=0.535 is above zero — so there IS a real in-sample/walk-forward edge. But: (1) it fails deflated Sharpe — DSR=0.5536 (<0.95), is_significant=false, PBO=0.6945, with the optimized Sharpe 1.84 only marginally above the 225-trial expected-max luck bar of 1.75, so after multiple-testing correction the selected config is not statistically significant (PSR 0.9962 vs DSR 0.55 is the classic trap); and (2) the untouched time-ordered HOLDOUT FAILED and is NEGATIVE (holdout_sharpe -0.358, ratio -0.245 vs WF-OOS 1.463), the realization of the 2025-2026 decay flagged at pre-optimization review (rolling Sharpe ~6 in 2024 collapsing to negative through 2026; optimized annual 2026 -2.49). Not iterate: this is attempt 2 of 2 (budget exhausted), and iteration 2 was ALREADY a significance-targeted fix (vol_mult pinned off its monotonic cliff and clamped ≤1.7, entry densified to lift trade count) that raised DSR only to 0.55 — short of 0.95 — while the forward holdout turned negative; parameter tuning cannot manufacture significance for an edge that barely clears the luck bar, nor reverse a decaying recent regime. Not revise_hypothesis: BTC 4H vol-breakout long is not a proven mechanism stranded on a dead instrument — it is a modest single-asset edge that fails best-of-225 deflation and a negative recent holdout, so there is nothing validated to re-target. FAILURE PATTERN: a clean, non-overfit walk-forward with positive OOS across all windows and a positive Sharpe CI is NOT sufficient for promotion when the deflated Sharpe stays below 0.95 (optimized Sharpe only marginally over the expected-max luck bar) and the untouched forward holdout is negative from genuine 2025-2026 decay; a significance-targeted prior iteration that reaches only DSR 0.55 confirms the edge is sub-significant, not mis-tuned.

Single-asset, long-only BTCUSDT.BINANCE USD-M perp on 4H bars. Enters long when the recent-window Parkinson volatility breaks out to >= vol_mult (1.40, pinned) times the long-window Parkinson baseline AND price is above its SMA(ma_period) — a bullish volatility-regime breakout. Exits on a percentage stop, take-profit, a trend break (close back below the MA), or a max-hold time stop. Risk-based sizing capped at 1x equity notional. Bars-only, no supplementary-data dependency; zero trades if warm-up is insufficient.

Trades implement the hypothesis faithfully: long-only (120 longs, 0 shorts), vol-breakout + bullish-MA entries fire every year 2020-2026, clean plumbing with zero drops.

Healthy sample (120 trades) — the iteration-2/3 densification worked, giving optimization real data to fit rather than a thin single-regime sample.

Positive, non-degenerate edge: Sharpe 1.50 with sharpe_ci_low 0.148 (>0), PF 1.65, max_drawdown 5.75%, Sortino 3.19, positive alpha; bars-only single-venue design avoids the supp-data/cross-venue failure modes.

Clear recent decay: 2025 -0.8%, 2026 -3.6%, rolling Sharpe falling from ~6 (2024) to -1.57 (2026-05) — the recent-window holdout and deflated-Sharpe gate are the likely failure point at final analysis.

Long win rate only 44% and max_consecutive_losses 14 — edge relies on winners being larger than losers (PF 1.65); fee drag is 4.16% of gross across 120 trades.

Best-of-N optimization over 225 trials will need the genuine edge to clear the deflation luck bar despite the weakening recent regime.

Genuinely non-overfit walk-forward: is_overfitted=false, avg IS 2.46 → avg OOS 1.46 with all three OOS windows positive [1.39, 1.81, 1.19].

Clean sensitivity surface (0 cliffs) and a positive Sharpe CI lower bound (sharpe_ci_low 0.535), on a healthy sample (120 base / 134 optimized trades).

Faithful long-only implementation with low fee/impact drag; the iteration-2 significance fix did improve robustness (walk-forward no longer overfit, CI low positive).

Fails deflated Sharpe: DSR=0.5536 (<0.95 bar), is_significant=false, PBO=0.6945 (>0.5) — the optimized Sharpe 1.84 sits only marginally above the 225-trial expected-max luck bar of 1.75, so after multiple-testing correction the selection is not significant (PSR 0.9962 vs DSR 0.55 is the trap).

Holdout FAILED and is NEGATIVE: holdout_sharpe -0.358, ratio -0.245 vs WF-OOS 1.463 — the untouched recent window loses money.

Genuine recent-regime decay (flagged at pre-opt review): rolling Sharpe fell from ~6 in 2024 to negative through 2026; optimized annual 2025 +1.3, 2026 -2.49.

Optimization attempt 2 of 2 — budget exhausted; iteration 2 was already a significance-targeted fix that reached only DSR 0.55 and could not clear the bar or arrest the decay.

only marginally above

The backtest was run on the WRONG instrument and timeframe: config shows BTCUSD.HYPERLIQUID-1-MINUTE-LAST-EXTERNAL over 5 days with empty parameters {}, but the hypothesis and strategy code target BTCUSDT.BINANCE on 4H bars over full history (2020-2026). This is the same misroute seen on the prior strategy. Re-run with the correct FactoryStrategyConfig: instrument_id='BTCUSDT.BINANCE', bar_type='BTCUSDT.BINANCE-4-HOUR-LAST-EXTERNAL' (the daily/4H bar type your catalog uses), and pass the densified default parameters (they came through empty {}). Verify the run spans multiple years so metrics_reliable=true and walk-forward/holdout windows are non-empty. The iteration-2 note references a prior valid run (base Sharpe 2.36, entries every year 2020-2026) — reproduce THAT run, then judge. Per the pre-committed caution for this vol-breakout family, the corrected run must show a clear positive edge (PF>1.2, non-zero-straddling CI) to proceed to optimization; otherwise abandon.

BtcFourHourVolatilityBreakoutRegimeLong was the batch's strongest near-miss: a bars-only, single-asset, single-venue long-only design built to satisfy every documented factory constraint, trading bullish volatility-regime breakouts as the analyst-recommended inverse of the failed post-vol-spike recovery. It produced a genuine edge — 120 trades with Sharpe 1.50, PF 1.65, positive alpha, and after optimization an honestly non-overfit walk-forward (all three OOS windows positive, avg OOS 1.46) on a clean zero-cliff sensitivity surface. But it failed the two gates that matter most at final analysis: deflated Sharpe 0.55 (vs the 0.95 bar) with the optimized Sharpe 1.84 only marginally above the 1.75 luck bar, and a negative forward holdout (-0.358) that realized the flagged 2025–2026 decay. On attempt 2 of 2 with budget exhausted and a prior significance-fix already reaching only DSR 0.55, the analyst judged it a modest, sub-significant, decaying edge and abandoned it rather than iterate or re-target.

A clean, non-overfit walk-forward with positive OOS across all windows and a positive Sharpe CI is not sufficient for promotion — when the deflated Sharpe stays below 0.95 (optimized Sharpe barely over the expected-max luck bar) and the untouched forward holdout is negative from genuine recent decay, the edge is sub-significant rather than mis-tuned, and a significance-targeted iteration reaching only DSR 0.55 confirms it.

Despite the honest walk-forward, the analyst abandoned it on the final optimization attempt (2 of 2) for failing two decisive gates: deflated Sharpe 0.5536 (below the 0.95 bar) with is_significant=false, PBO 0.6945, and optimized Sharpe 1.84 only marginally above the 225-trial luck bar of 1.75; and a negative untouched holdout (holdout_sharpe -0.358, ratio -0.245) realizing the 2025–2026 regime decay (rolling Sharpe ~6 in 2024 collapsing to negative).

A long-only BTCUSDT.BINANCE 4H strategy (bars-only) that enters long on a low-to-high volatility regime shift — short-window Parkinson vol breaking above its long-window baseline — filtered to bullish expansions by requiring close above a 50-period SMA, implementing Sinclair's vol-expansion thesis as the deliberate opposite of the failed post-vol-spike recovery mechanism.

The base backtest took 120 long trades (0 shorts) with entries every year 2020–2026: Sharpe 1.50 (sharpe_ci_low 0.148), profit factor 1.65, total return 32.9, max drawdown 5.75%, 44% win rate, positive alpha; the optimized refit reached Sharpe 1.84, PF 1.70, and total return 42.9 with a non-overfit walk-forward (avg IS 2.46 → avg OOS 1.46, all three OOS windows positive) and a clean 0-cliff sensitivity surface.