BOCPD Entry-Filter Backtest — Trend Strategies
PromotedBOCPD Entry-Filter Backtest — Trend Strategies
2026-05-19 · status: promoted · 100.9s
Hypothesis: Blocking entries when BOCPD p_short > threshold improves trend-strategy risk-adjusted PnL on BTC. Test at thresholds 0.3 / 0.5 / 0.7 against the no-filter baseline across EMA_PURE, EMA_MACD, HOLY_GRAIL over a 2-year+ window.
Verdict: SHIP — BOCPD filter promotes 3/6 strategies. Add
bocpd_filterto those configs in production.
Key metrics
| metric | value |
|---|---|
| n_strategies | 6 |
| n_promote | 3 |
| n_marginal | 2 |
| n_hurt | 1 |
| window_from | 2023-01-01 |
| window_to | 2026-04-01 |
| EMA_PURE_best_threshold | +0.3000 |
| EMA_PURE_pnl_lift_pct | +0.4776 |
| EMA_PURE_dd_delta_pp | +0.0007 |
| EMA_MACD_best_threshold | +0.3000 |
| EMA_MACD_pnl_lift_pct | +0.6891 |
| EMA_MACD_dd_delta_pp | +0.0013 |
| HOLY_GRAIL_best_threshold | +1.0000 |
| HOLY_GRAIL_pnl_lift_pct | +0.0000 |
| HOLY_GRAIL_dd_delta_pp | +0.0000 |
| BB_EXTREME_3_best_threshold | +0.3000 |
| BB_EXTREME_3_pnl_lift_pct | +6.8836 |
| BB_EXTREME_3_dd_delta_pp | +0.0083 |
| BB_EXTREME_1_best_threshold | +0.3000 |
| BB_EXTREME_1_pnl_lift_pct | +12.6457 |
| BB_EXTREME_1_dd_delta_pp | +0.0011 |
| BB_EXTREME_2_best_threshold | +0.3000 |
| BB_EXTREME_2_pnl_lift_pct | +17.0414 |
| BB_EXTREME_2_dd_delta_pp | +0.0121 |
Approach
Run three trend strategies (EMA_PURE, EMA_MACD, HOLY_GRAIL) over 2023-01-01 → 2026-04-01 with the new bocpd_filter at four thresholds: 1.0 (off), 0.7, 0.5, 0.3. For each (strategy × threshold) cell we report total PnL, # trades, win rate, trade-Sharpe, and max drawdown. The baseline (1.0) shows what each strategy does without the filter; lower thresholds block more entries.
Counterfactual analysis: for the baseline runs we record the BOCPD p_short value at each trade's entry. Trades with p_short > θ would have been blocked at threshold θ. Their net contribution to PnL tells us why the filter would help or hurt.
Total PnL ($) — rows=strategy, cols=threshold
| strategy | 0.3 | 0.5 | 0.7 | 1.0 |
|---|---|---|---|---|
| BB_EXTREME_1 | 17.23 | 16.76 | 16.2 | 15.3 |
| BB_EXTREME_2 | 20.67 | 19.76 | 18.9 | 17.66 |
| BB_EXTREME_3 | 12.21 | 12.12 | 12.12 | 11.42 |
| EMA_MACD | -18.84 | -19.37 | -19.03 | -18.97 |
| EMA_PURE | -17.58 | -18.46 | -18.08 | -17.67 |
| HOLY_GRAIL | -0.83 | -0.83 | -0.83 | -0.83 |
# Trades
| strategy | 0.3 | 0.5 | 0.7 | 1.0 |
|---|---|---|---|---|
| BB_EXTREME_1 | 146 | 149 | 156 | 165 |
| BB_EXTREME_2 | 115 | 120 | 128 | 139 |
| BB_EXTREME_3 | 95 | 97 | 110 | 124 |
| EMA_MACD | 1225 | 1231 | 1233 | 1234 |
| EMA_PURE | 1006 | 1020 | 1027 | 1039 |
| HOLY_GRAIL | 27 | 27 | 27 | 27 |
Win rate
| strategy | 0.3 | 0.5 | 0.7 | 1.0 |
|---|---|---|---|---|
| BB_EXTREME_1 | 0.1233 | 0.1208 | 0.1218 | 0.1152 |
| BB_EXTREME_2 | 0.1652 | 0.1583 | 0.1484 | 0.1367 |
| BB_EXTREME_3 | 0.0842 | 0.0825 | 0.0727 | 0.0645 |
| EMA_MACD | 0.3282 | 0.3274 | 0.3293 | 0.3298 |
| EMA_PURE | 0.326 | 0.3255 | 0.3281 | 0.3282 |
| HOLY_GRAIL | 0.5556 | 0.5556 | 0.5556 | 0.5556 |
Max DD (%)
| strategy | 0.3 | 0.5 | 0.7 | 1.0 |
|---|---|---|---|---|
| BB_EXTREME_1 | -0.03 | -0.03 | -0.03 | -0.03 |
| BB_EXTREME_2 | -0.03 | -0.03 | -0.04 | -0.04 |
| BB_EXTREME_3 | -0.02 | -0.02 | -0.02 | -0.02 |
| EMA_MACD | -0.19 | -0.19 | -0.19 | -0.19 |
| EMA_PURE | -0.18 | -0.19 | -0.18 | -0.18 |
| HOLY_GRAIL | -0.01 | -0.01 | -0.01 | -0.01 |
Sharpe (trade-level)
| strategy | 0.3 | 0.5 | 0.7 | 1.0 |
|---|---|---|---|---|
| BB_EXTREME_1 | 2.148 | 2.066 | 1.953 | 1.735 |
| BB_EXTREME_2 | 2.914 | 2.724 | 2.521 | 2.182 |
| BB_EXTREME_3 | 3.08 | 3.023 | 2.669 | 2.362 |
| EMA_MACD | -2.19 | -2.242 | -2.2 | -2.185 |
| EMA_PURE | -2.515 | -2.61 | -2.542 | -2.43 |
| HOLY_GRAIL | -2.895 | -2.895 | -2.895 | -2.895 |
Counterfactual: PnL of trades that would be blocked
| strategy | threshold | n_blocked | blocked_pnl | blocked_mean_pnl | kept_mean_pnl | n_kept | kept_pnl | total_pnl_baseline |
|---|---|---|---|---|---|---|---|---|
| EMA_PURE | 1 | 0 | 0 | — | -0.02 | 1039 | -17.67 | -17.67 |
| EMA_PURE | 0.7 | 25 | 0.06 | +0.00 | -0.02 | 1014 | -17.73 | -17.67 |
| EMA_PURE | 0.5 | 33 | 0.09 | +0.00 | -0.02 | 1006 | -17.76 | -17.67 |
| EMA_PURE | 0.3 | 55 | 0.22 | +0.00 | -0.02 | 984 | -17.89 | -17.67 |
| EMA_MACD | 1 | 0 | 0 | — | -0.02 | 1234 | -18.97 | -18.97 |
| EMA_MACD | 0.7 | 18 | -0.42 | -0.02 | -0.02 | 1216 | -18.55 | -18.97 |
| EMA_MACD | 0.5 | 24 | -0.88 | -0.04 | -0.01 | 1210 | -18.09 | -18.97 |
| EMA_MACD | 0.3 | 37 | -0.31 | -0.01 | -0.02 | 1197 | -18.66 | -18.97 |
| HOLY_GRAIL | 1 | 0 | 0 | — | -0.03 | 27 | -0.83 | -0.83 |
| HOLY_GRAIL | 0.7 | 1 | -0.15 | -0.15 | -0.03 | 26 | -0.68 | -0.83 |
| HOLY_GRAIL | 0.5 | 1 | -0.15 | -0.15 | -0.03 | 26 | -0.68 | -0.83 |
| HOLY_GRAIL | 0.3 | 2 | -0.43 | -0.22 | -0.02 | 25 | -0.4 | -0.83 |
| BB_EXTREME_3 | 1 | 0 | 0 | — | 0.09 | 124 | 11.42 | 11.42 |
| BB_EXTREME_3 | 0.7 | 5 | -0.22 | -0.04 | 0.1 | 119 | 11.65 | 11.42 |
| BB_EXTREME_3 | 0.5 | 8 | -0.36 | -0.04 | 0.1 | 116 | 11.78 | 11.42 |
| BB_EXTREME_3 | 0.3 | 12 | -0.55 | -0.05 | 0.11 | 112 | 11.98 | 11.42 |
| BB_EXTREME_1 | 1 | 0 | 0 | — | 0.09 | 165 | 15.3 | 15.3 |
| BB_EXTREME_1 | 0.7 | 8 | 5.45 | +0.68 | 0.06 | 157 | 9.85 | 15.3 |
| BB_EXTREME_1 | 0.5 | 11 | 5.2 | +0.47 | 0.07 | 154 | 10.1 | 15.3 |
| BB_EXTREME_1 | 0.3 | 14 | 4.78 | +0.34 | 0.07 | 151 | 10.52 | 15.3 |
| BB_EXTREME_2 | 1 | 0 | 0 | — | 0.13 | 139 | 17.66 | 17.66 |
| BB_EXTREME_2 | 0.7 | 8 | 5.51 | +0.69 | 0.09 | 131 | 12.15 | 17.66 |
| BB_EXTREME_2 | 0.5 | 12 | 5.12 | +0.43 | 0.1 | 127 | 12.53 | 17.66 |
| BB_EXTREME_2 | 0.3 | 16 | 4.63 | +0.29 | 0.11 | 123 | 13.02 | 17.66 |
Reading the table: blocked_pnl is the net PnL of trades whose entry BOCPD score exceeded the threshold. If blocked_pnl is negative, the filter would have removed losing trades → net win. If positive, the filter would have removed profitable trades → net loss.
Per-strategy verdicts
-
EMA_PURE: MARGINAL_UNPROFITABLE — θ=0.3: PnL -17.58 (+0.5%), DD -0.1789.
-
EMA_MACD: MARGINAL_UNPROFITABLE — θ=0.3: PnL -18.84 (+0.7%), DD -0.1886.
-
HOLY_GRAIL: FILTER HURTS — baseline (off) is the best variant.
-
BB_EXTREME_3: PROMOTE — θ=0.3: PnL +12.21 (+6.9%), DD -0.0157 (baseline -0.0240, Δ +0.83 bps).
-
BB_EXTREME_1: PROMOTE — θ=0.3: PnL +17.23 (+12.6%), DD -0.0319 (baseline -0.0330, Δ +0.11 bps).
-
BB_EXTREME_2: PROMOTE — θ=0.3: PnL +20.67 (+17.0%), DD -0.0327 (baseline -0.0448, Δ +1.21 bps).
Overall verdict
SHIP — BOCPD filter promotes 3/6 strategies. Add bocpd_filter to those configs in production.