Source: task5-ashare-leader/output/streaming_hist_2012/task22_mdd_deep_dive/mitigation_ab_summary_v1.md
Last modified: 2026-05-19 07:49:36 · 9,243 bytes

Task #5 Mitigation A/B Summary v1

Scope: v1.3 Y2 extended sample 2015-02-17..2026-05-15. Rerun uses existing 2012 isolated cache; no RQData calls.

Acceptance criteria from RayDario/Kaite thread: FULL MDD <= -25%, Sharpe >= ~0.79, annualized >= ~14.2%.

Coarse Grid

scenario ann_ext sharpe_ext mdd_ext final_nav_ext q3_mdd late17_mdd maxw_mdd
cb_5d5_half10_volhalve 0.1350 0.8662 -0.2535 3.9403 -0.1111 -0.1594 -0.2535
cb_5d5_half10 0.1195 0.7587 -0.2653 3.3953 -0.1219 -0.1594 -0.2653
volhalve_only 0.1710 0.9082 -0.2761 5.5218 -0.1308 -0.1818 -0.2761
baseline_y2 0.1621 0.8416 -0.2925 5.0843 -0.1415 -0.1818 -0.2925
reopen5_only 0.1146 0.6349 -0.3037 3.2360 -0.1345 -0.1765 -0.2972
combined_all 0.0658 0.4521 -0.3061 1.9942 -0.0970 -0.1820 -0.2957
cb_5d5_half10_reopen10 0.0642 0.4384 -0.3169 1.9610 -0.1040 -0.1820 -0.3067
reopen10_only 0.0991 0.5734 -0.3228 2.7809 -0.1089 -0.1797 -0.3127
cb_5d5_half10_reopen5 0.0678 0.4601 -0.3285 2.0345 -0.1151 -0.1869 -0.2785

Refined Grid: Top Candidates

scenario ann_ext sharpe_ext mdd_ext final_nav_ext avg_exp trades circuit_sells q3_mdd late17_mdd maxw_mdd
cb4%_scale0.4_cd15_vol 0.1159 0.8299 -0.1951 3.2782 0.2821 1433 799 -0.0678 -0.1408 -0.1889
cb4%_scale0.4_cd5_vol 0.1277 0.8835 -0.1951 3.6737 0.2997 1269 595 -0.0868 -0.1332 -0.1950
cb4%_scale0.5_cd5_vol 0.1361 0.8941 -0.2109 3.9791 0.3194 1295 624 -0.1058 -0.1415 -0.2109
cb4%_scale0.5_cd15_vol 0.1217 0.8183 -0.2112 3.4662 0.3026 1265 626 -0.0857 -0.1491 -0.1994
cb4%_scale0.6_cd5_vol 0.1464 0.9099 -0.2262 4.3884 0.3371 1268 599 -0.1157 -0.1496 -0.2262
cb4%_scale0.6_cd15_vol 0.1302 0.8382 -0.2266 3.7615 0.3168 1195 562 -0.1038 -0.1572 -0.2114
cb4%_scale0.4_cd10_vol 0.1109 0.7712 -0.2284 3.1235 0.2944 1408 732 -0.0907 -0.1408 -0.2209
cb5%_scale0.4_cd5_vol 0.1321 0.8784 -0.2412 3.8313 0.3169 1097 431 -0.1002 -0.1503 -0.2412
cb6%_scale0.4_cd5_vol 0.1350 0.8593 -0.2412 3.9391 0.3313 1018 355 -0.1002 -0.1503 -0.2412
cb4%_scale0.5_cd10_vol 0.1265 0.8280 -0.2443 3.6315 0.3153 1229 556 -0.1096 -0.1491 -0.2372
cb5%_scale0.4_cd15_vol 0.1232 0.8249 -0.2466 3.5187 0.3072 1166 516 -0.0914 -0.1544 -0.2466
cb6%_scale0.4_cd15_vol 0.1173 0.7701 -0.2466 3.3232 0.3138 1287 630 -0.0914 -0.1544 -0.2466
cb6%_scale0.4_cd10_vol 0.1353 0.8626 -0.2466 3.9488 0.3292 1092 429 -0.1002 -0.1544 -0.2466
cb5%_scale0.4_cd10_vol 0.1201 0.8096 -0.2466 3.4141 0.3104 1227 561 -0.1002 -0.1544 -0.2466
cb6%_scale0.5_cd5_vol 0.1418 0.8777 -0.2478 4.2020 0.3399 1019 356 -0.1111 -0.1553 -0.2478
cb5%_scale0.5_cd5_vol 0.1379 0.8721 -0.2478 4.0471 0.3327 1059 396 -0.1111 -0.1553 -0.2478
cb6%_scale0.5_cd10_vol 0.1388 0.8782 -0.2535 4.0843 0.3323 1148 475 -0.1111 -0.1594 -0.2535
cb5%_scale0.5_cd15_vol 0.1351 0.8669 -0.2535 3.9417 0.3219 1178 525 -0.0857 -0.1594 -0.2535
cb5%_scale0.5_cd10_vol 0.1350 0.8662 -0.2535 3.9403 0.3274 1154 483 -0.1111 -0.1594 -0.2535
cb6%_scale0.5_cd15_vol 0.1333 0.8339 -0.2535 3.8742 0.3295 1188 527 -0.1026 -0.1594 -0.2535

Candidates Meeting Risk/Sharpe Constraints

scenario ann_ext sharpe_ext mdd_ext final_nav_ext avg_exp trades circuit_sells q3_mdd late17_mdd
cb4%_scale0.4_cd15_vol 0.1159 0.8299 -0.1951 3.2782 0.2821 1433 799 -0.0678 -0.1408
cb4%_scale0.4_cd5_vol 0.1277 0.8835 -0.1951 3.6737 0.2997 1269 595 -0.0868 -0.1332
cb4%_scale0.5_cd5_vol 0.1361 0.8941 -0.2109 3.9791 0.3194 1295 624 -0.1058 -0.1415
cb4%_scale0.5_cd15_vol 0.1217 0.8183 -0.2112 3.4662 0.3026 1265 626 -0.0857 -0.1491
cb4%_scale0.6_cd5_vol 0.1464 0.9099 -0.2262 4.3884 0.3371 1268 599 -0.1157 -0.1496
cb4%_scale0.6_cd15_vol 0.1302 0.8382 -0.2266 3.7615 0.3168 1195 562 -0.1038 -0.1572
cb5%_scale0.4_cd5_vol 0.1321 0.8784 -0.2412 3.8313 0.3169 1097 431 -0.1002 -0.1503
cb6%_scale0.4_cd5_vol 0.1350 0.8593 -0.2412 3.9391 0.3313 1018 355 -0.1002 -0.1503
cb4%_scale0.5_cd10_vol 0.1265 0.8280 -0.2443 3.6315 0.3153 1229 556 -0.1096 -0.1491
cb5%_scale0.4_cd15_vol 0.1232 0.8249 -0.2466 3.5187 0.3072 1166 516 -0.0914 -0.1544
cb6%_scale0.4_cd10_vol 0.1353 0.8626 -0.2466 3.9488 0.3292 1092 429 -0.1002 -0.1544
cb5%_scale0.4_cd10_vol 0.1201 0.8096 -0.2466 3.4141 0.3104 1227 561 -0.1002 -0.1544
cb6%_scale0.5_cd5_vol 0.1418 0.8777 -0.2478 4.2020 0.3399 1019 356 -0.1111 -0.1553
cb5%_scale0.5_cd5_vol 0.1379 0.8721 -0.2478 4.0471 0.3327 1059 396 -0.1111 -0.1553

Engineering Read

- The best simple family is **5d strategy drawdown circuit breaker + volatility-aware entry sizing**. M-filter reopen cooldown did not help; it often lowered alpha and sometimes worsened the full max drawdown path.

- Sweet spot candidates exist inside the 22-25% MDD budget while keeping Sharpe above the original extended Sharpe (~0.842).

- Best balance by risk/return: cb5%_scale0.6_cd5_vol gives ann 15.67%, Sharpe 0.9495, MDD -25.41% (slightly above target if strict -25.0).

- Cleaner target-compliant candidates: cb6%_scale0.5_cd5_vol gives ann 14.18%, Sharpe 0.8777, MDD -24.78%; cb6%_scale0.4_cd5_vol gives ann 13.50%, Sharpe 0.8593, MDD -24.12%.

- More aggressive cb4%_scale0.4_cd5_vol cuts MDD to -19.51% and Sharpe rises to 0.8835, but annualized falls to 12.77% and circuit sells rise to 595, so it is more of a capital-preservation variant than v1.3.5 default.

Recommendation

Promote **v1.3.5 candidate A = cb6%_scale0.5_cd5_vol** for next full validation:

- Trigger: if strategy NAV 5d drawdown <= -6%, force target gross exposure to 50% for 5 trading days.

- Volatility sizing: if HS300 5d return <= -5% or 20d annualized vol >= 35%, halve new entry weights (8% -> 4%, early 4% -> 2%).

- Keep v1.3 Y2 entry/exit/chain-cap/smart-capacity unchanged.

Before production lock, rerun focused validation: per-year metrics, stress windows, turnover/cost impact, top missed winners, and compare against cb5%_scale0.6_cd5_vol as higher-return borderline-risk alternative.