BnbFourHourVolumeConfirmedBreakoutLong

BNB 4H Volume-Confirmed Breakout Long

A long-only breakout strategy on BNBUSDT perpetual futures using 4-hour bars. The strategy enters long when price closes above the highest high of the prior N bars (Donchian-style breakout) AND the breakout bar's volume exceeds 1.5x the rolling average volume — a simple volume-confirmation filter to avoid false breakouts on thin liquidity. BNB exhibits distinctive volume-driven momentum cycles tied to Binance ecosystem catalysts (token burns, IEO/Launchpool launches, BNB Chain narrative pumps), and breakouts on elevated volume historically have higher follow-through than headline-only moves. Holding period is days, not minutes, so per-trade returns easily clear the 0.10% round-trip fee on USD-M futures. Single instrument, single timeframe, OHLCV-only — no supplementary data dependencies that have caused repeated verification timeouts in prior hypotheses.

Implements the hypothesis exactly: single instrument (BNBUSDT), single timeframe (4H), long-only, and entry gated on precisely the two named conditions — Donchian N-bar breakout (close > prior-N highest high) and volume confirmation (volume >= 1.5x rolling average). Iteration 4 removes the extraneous SMA(50) uptrend filter that the semantic auditor flagged as an undocumented third entry gate: the `uptrend` term is dropped from the entry decision and the trend filter is disabled, so no breakout+volume bar is silently rejected. Signal, exit, and sizing logic that already passed earlier verification layers are unchanged; imports remain clean (stdlib + nautilus_trader only).

Failed deflated Sharpe and the forward holdout decisively on optimization attempt 2 of 2. DSR=0.3959 (vs 0.95 bar) with the optimized Sharpe 1.927 BELOW the 225-trial expected-max luck bar of 2.0801 (is_significant=false, PBO=0.6997 >0.5) — after multiple-testing correction the selected best-of-225 config is statistically indistinguishable from best-of-N noise; probabilistic_sharpe 0.9996 vs DSR 0.3959 is the textbook PSR-vs-DSR selection-inflation trap. Decisively, the untouched time-ordered HOLDOUT FAILED (ratio 0.184 < 0.70; holdout_sharpe 0.2598 vs WF-OOS 1.4157) — the recent regime collapses, realizing genuine decay (annual 2025 -1.6%, 2026 -2.9%; rolling Sharpe negative into 2026). The strategy also has a NEGATIVE information ratio (-0.74 base and optimized) with benchmark_meaningful=true, so it loses to its own BNB buy-hold risk-adjusted, and its full-sample edge is bull-market-concentrated (2021+2024 carry it; kurtosis ~27, skew ~2.9). Not iterate (final attempt, and unwarranted regardless): the sensitivity surface is clean (0 cliffs) but the ENTIRE grid sits at ~1.2-1.7 Sharpe, uniformly below the 2.08 luck bar, so there is no robust region above significance to tune toward — a further best-of-225 sweep re-selects the same sub-significant, decaying config. Not revise_hypothesis: this is a multiple-testing/forward-generalization failure on a perfectly valid instrument, not a proven mechanism stranded on a dead target — a promoted BNB Donchian breakout sibling already exists, and adding a volume-confirmation filter did not create a distinct promotable edge. FAILURE PATTERN: a volume-confirmed single-asset BNB 4H Donchian breakout, even with a clean 0-cliff sensitivity surface and a non-overfit-flagged walk-forward (IS 2.72 -> OOS 1.42), still fails promotion because its best-of-225 optimized Sharpe (1.93) cannot clear the 2.08 expected-max luck bar (DSR 0.40, PBO 0.70) and its untouched forward holdout collapses (ratio 0.184) from genuine 2025-2026 decay, with a negative information ratio vs its own buy-hold. Clean sensitivity and a positive walk-forward OOS measure consistency, not significance or forward-generalization, and do not rescue a bull-market-concentrated single-name breakout that fails deflated Sharpe and the forward holdout simultaneously.

Long-only volume-confirmed Donchian breakout on BNBUSDT 4H USD-M futures. Enters long when the bar closes above the highest high of the prior breakout_period bars AND bar volume >= volume_mult x the average volume of the prior volume_window bars. Exits on take-profit (12%), stop-loss (6%), or a 36-bar time stop. Capital-relative sizing (18% notional capped by 1.5% risk-per-trade), leverage 1.0.

179 long trades over ~6 years (~30/yr, avg hold 4d 9h) — ample sample for walk-forward optimization and matches the hypothesis's days-holding, long-only breakout premise

Clean entry diagnostics: 179 signaled → 179 submitted, zero dropped (size-zero/min-notional/cash-short all 0) — strategy is functioning, not frozen or broken

Genuine gross edge that clears fees: profit_factor 1.51, win_rate 55%, Sharpe 1.43, Sortino 2.35, commission only 2.43% of gross

Trades implement the stated mechanism (all long, breakout+volume, no shorts) with a moderate 13.4% max drawdown

Recent-regime decay: annual returns negative in 2025 (-1.6%) and 2026 (-2.9%), rolling Sharpe sliding negative into 2026 — a holdout-gate risk to watch post-optimization

information_ratio -0.74 vs buy-hold (directional long riding BNB beta 0.07/corr 0.74); real alpha is modest (alpha 0.016)

return_kurtosis 26.9 / skew 2.9 — some outlier dependence (Feb 2021 month alone +28.8%) inflating full-sample metrics

Clean sensitivity surface: 0 cliff parameters across all three tuned params (breakout_period, volume_mult, volume_window); the Sharpe heatmap is a smooth plateau ~1.2-1.7.

Walk-forward is NOT overfit-flagged (is_overfitted=false): avg IS 2.72 -> avg OOS 1.42, all three OOS windows positive [1.74, 1.21, 1.30].

Code<->hypothesis alignment is correct this iteration (breakout AND volume, no hidden trend gate); trades are long-only breakout entries matching the stated mechanism.

Optimization modestly improved the full refit (Sharpe 1.43 -> 1.93, max_dd 13.4% -> 8.3%, PF 1.51 -> 1.78, sortino 3.39) and edge clears fees (avg_win $1966 vs avg_loss $1560, PF 1.78, commission only 2.35% of gross).

FAILS deflated Sharpe: DSR=0.3959 (vs 0.95 bar), is_significant=false, and the optimized Sharpe 1.927 is BELOW the 225-trial expected-max luck bar of 2.0801 — after multiple-testing correction the selected config is indistinguishable from best-of-N noise.

PBO=0.6997 (>0.5): the parameter selection is more likely than not overfitting the search.

Forward HOLDOUT FAILED decisively: ratio 0.184 (<0.70 bar), holdout_sharpe 0.2598 vs WF-OOS 1.4157 — the untouched recent window barely holds up, realizing genuine decay (annual 2025 -1.6%, 2026 -2.9%; rolling Sharpe negative into 2026).

probabilistic_sharpe 0.9996 vs DSR 0.3959 is the textbook PSR-vs-DSR selection-inflation trap (PSR ignores the 225-trial count DSR corrects for).

NEGATIVE information ratio (-0.74 base, -0.74 optimized) with benchmark_meaningful=true — the strategy loses to its own BNB buy-hold on a risk-adjusted basis.

Edge is bull-market-concentrated (2021 +25.6%/+30.9%, 2024 +13.8%/+14.9%; return_kurtosis ~25-27, skew ~2.9-3.0; Feb-2021 alone drives ~29-31% of the year), decaying to negative in the most recent two years.

Code↔hypothesis misalignment found by the semantic auditor — the code does NOT implement the hypothesis. Re-code the strategy to implement the hypothesis EXACTLY (instrument, timeframe, direction, the named edge/mechanic, sizing). Concrete issues: The hypothesis specifies entry as exactly two conditions — Donchian breakout (close > prior-N-bar highest high) AND volume >= 1.5x rolling average — but the code adds a mandatory third entry gate (`use_trend_filter=true`, `sma_trend_period=50`): entry requires close > SMA(50) via the `uptrend` term ANDed into `_entry_ok`. Bars satisfying breakout+volume below the SMA are silently rejected. This trend filter appears nowhere in the hypothesis text and materially narrows the stated entry rule.

The BNB 4H Volume-Confirmed Breakout Long passed backtest review with an 'optimize' verdict on a genuine fee-clearing edge (Sharpe 1.43, PF 1.51, 179 trades, 13.4% drawdown), and by iteration 4 its code correctly matched the hypothesis's exact two-condition entry (breakout AND volume, no hidden trend gate). Optimization tuned three parameters into a smooth, cliff-free plateau and a non-overfit walk-forward (IS 2.72 → OOS 1.42), modestly lifting the refit to Sharpe 1.93. But the selected best-of-225 config could not clear the 2.08 expected-max luck bar (DSR 0.40, PBO 0.70, not significant), and the time-ordered holdout collapsed (ratio 0.184, holdout Sharpe 0.26) as the strategy realized genuine 2025–2026 decay while losing to its own buy-hold on a risk-adjusted basis. On the second and final optimization attempt the analyst abandoned it as a multiple-testing and forward-generalization failure rather than a fixable target, noting a promoted BNB Donchian breakout sibling already exists and the volume filter added no distinct promotable edge.

A clean sensitivity surface (0 cliffs) and a non-overfit-flagged walk-forward measure consistency, not statistical significance or forward-generalization — a bull-market-concentrated single-name breakout can still fail deflated Sharpe and a recent-regime holdout simultaneously.

The analyst abandoned it at the post-optimization ANALYZING stage (attempt 2 of 2): it failed the deflated Sharpe test (DSR 0.3959 vs 0.95 bar) with optimized Sharpe 1.93 below the 225-trial expected-max luck bar of 2.08, PBO 0.70, and decisively failed the untouched forward holdout (ratio 0.184 < 0.70), plus a negative information ratio (-0.74) vs its own BNB buy-hold.

A long-only BNBUSDT 4H strategy that bought N-bar Donchian breakouts only when the breakout bar's volume exceeded 1.5x its rolling average, aiming to capture BNB's volume-driven momentum cycles while clearing round-trip futures fees on multi-day holds.

The initial backtest looked strong over ~693 days and 179 trades: total return 67.2%, Sharpe 1.43, win rate 55%, profit factor 1.51, and max drawdown 13.4%. Optimization improved the full refit (Sharpe 1.93, max drawdown 8.3%, profit factor 1.78, 140 trades) but the forward holdout Sharpe collapsed to 0.26.