BOCPD Filter — Universal Scan Across All Viable Entries
DoneBOCPD Filter — Universal Scan Across All Viable Entries
2026-05-19 · status: done · 60.9s
Hypothesis: bocpd_filter@0.3 helps multiple entry types beyond BB-Extreme. Strategies where the underlying baseline is profitable AND lift > +5% are candidates for megasweep COMPATIBLE_FILTERS inclusion (pending shuffle-test verification).
Verdict: BB-EXTREME ONLY — only BB-Extreme reproduces the lift; no other profitable strategy benefits. BOCPD's edge is mean-reversion-specific. Don't expand the megasweep filter pool.
Key metrics
| metric | value |
|---|---|
| n_strategies_tested | 8 |
| n_filter_helps | 1 |
| n_marginal | 1 |
| n_filter_hurts | 1 |
| n_baseline_unprofitable | 5 |
| strategies_filter_helps | ['BB_EXTREME_2'] |
Approach
Run each strategy with/without bocpd_filter@0.3 over 2023-01-01 → 2026-04-01. Goal: identify entries beyond BB-Extreme where BOCPD filter lifts PnL meaningfully on a profitable baseline. Strategies with unprofitable baselines are control — apparent 'lift' there is trade-count noise (per C7-D meta-lesson).
Results per strategy
| strategy | off_pnl | on_pnl | off_trades | on_trades | off_sharpe | on_sharpe | off_dd_pct | on_dd_pct | lift_pct | is_profitable | category |
|---|---|---|---|---|---|---|---|---|---|---|---|
| BB_EXTREME_2 | 17.66 | 20.67 | 139 | 115 | +2.182 | +2.914 | -0.04 | -0.03 | +17.0% | True | FILTER HELPS (profitable + lift) |
| HOLY_GRAIL | -0.83 | -0.83 | 27 | 27 | -2.895 | -2.895 | -0.01 | -0.01 | +0.0% | False | BASELINE UNPROFITABLE |
| DONCHIAN_20 | 0 | 0 | 0 | 0 | — | — | — | — | — | False | BASELINE UNPROFITABLE |
| DONCHIAN_50 | 5.52 | 4.7 | 120 | 120 | +1.208 | +1.027 | -0.03 | -0.03 | -14.9% | True | FILTER HURTS (profitable, drag) |
| SFP_20 | -4.72 | -5.29 | 375 | 360 | -0.909 | -1.068 | -0.06 | -0.07 | -12.2% | False | BASELINE UNPROFITABLE |
| RSI_MR_LONG_ONLY | 0.01 | 0.01 | 10 | 10 | +0.026 | +0.026 | -0.01 | -0.01 | +0.0% | True | MARGINAL (profitable + small lift) |
| BURJ_KHALIFA | -4.46 | -4.07 | 147 | 147 | -1.861 | -1.704 | -0.07 | -0.07 | +8.7% | False | BASELINE UNPROFITABLE |
| FIBDIV | -1 | -1.13 | 338 | 336 | -0.236 | -0.268 | -0.03 | -0.03 | -13.2% | False | BASELINE UNPROFITABLE |
Verdict per category
Filter HELPS (profitable + lift ≥ 5%)
- BB_EXTREME_2: baseline=$+17.66 → $+20.67 (+17.0%), Sharpe 2.18 → 2.91, trades 139 → 115
Marginal (profitable + 0-5% lift)
- RSI_MR_LONG_ONLY: baseline=$+0.01 → $+0.01 (+0.0%), Sharpe 0.03 → 0.03, trades 10 → 10
Filter HURTS (profitable + drag)
- DONCHIAN_50: baseline=$+5.52 → $+4.70 (-14.9%), Sharpe 1.21 → 1.03, trades 120 → 120
Baseline unprofitable (filter signal cannot be assessed)
-
HOLY_GRAIL: baseline=$-0.83 → $-0.83 (+0.0%), Sharpe -2.90 → -2.90, trades 27 → 27
-
DONCHIAN_20: baseline=$+0.00 → $+0.00 (+nan%), Sharpe nan → nan, trades 0 → 0
-
SFP_20: baseline=$-4.72 → $-5.29 (-12.2%), Sharpe -0.91 → -1.07, trades 375 → 360
-
BURJ_KHALIFA: baseline=$-4.46 → $-4.07 (+8.7%), Sharpe -1.86 → -1.70, trades 147 → 147
-
FIBDIV: baseline=$-1.00 → $-1.13 (-13.2%), Sharpe -0.24 → -0.27, trades 338 → 336
Overall verdict
BB-EXTREME ONLY — only BB-Extreme reproduces the lift; no other profitable strategy benefits. BOCPD's edge is mean-reversion-specific. Don't expand the megasweep filter pool.