cyb50-quant/research/dragon/v2/dragon_deep_oversold_selective_veto_experiments.md

Dragon Deep Oversold Selective Veto Experiments

• Goal: test whether weak deep-oversold alpha can be improved by vetoing only the most pathological local patterns, instead of blocking whole subtypes.
• Baseline objective is not preserved here; this is an alpha-first research pack.

Summary

• baseline: trades 107, avg_return 2.52%, profit_factor 3.31, real BUY 106, real SELL 105, deep weak trades 8
• selective_veto_positive_c1_lt_15_3: trades 106, avg_return 2.60%, profit_factor 3.48, real BUY 105, real SELL 104, deep weak trades 7
• selective_veto_shallow_jan_style: trades 105, avg_return 2.65%, profit_factor 3.55, real BUY 104, real SELL 103, deep weak trades 6
• selective_veto_positive_and_shallow: trades 104, avg_return 2.73%, profit_factor 3.75, real BUY 103, real SELL 102, deep weak trades 5
• selective_veto_plus_mixed_c1_lt_10_2_no_ql: trades 103, avg_return 2.81%, profit_factor 3.96, real BUY 102, real SELL 101, deep weak trades 4
• block_all_remaining_weak_subtypes: trades 102, avg_return 2.78%, profit_factor 3.76, real BUY 101, real SELL 100, deep weak trades 3

Delta Vs Baseline

• selective_veto_positive_c1_lt_15_3: delta_avg_return 0.08%, delta_profit_factor 0.17, delta_deep_weak_avg_return 0.48%, real BUY 105, real SELL 104
• selective_veto_shallow_jan_style: delta_avg_return 0.12%, delta_profit_factor 0.24, delta_deep_weak_avg_return 0.53%, real BUY 104, real SELL 103
• selective_veto_positive_and_shallow: delta_avg_return 0.21%, delta_profit_factor 0.44, delta_deep_weak_avg_return 1.32%, real BUY 103, real SELL 102
• selective_veto_plus_mixed_c1_lt_10_2_no_ql: delta_avg_return 0.28%, delta_profit_factor 0.65, delta_deep_weak_avg_return 2.40%, real BUY 102, real SELL 101
• block_all_remaining_weak_subtypes: delta_avg_return 0.26%, delta_profit_factor 0.45, delta_deep_weak_avg_return 0.61%, real BUY 101, real SELL 100

Quant Judgment

• Best branch in this pack: selective_veto_plus_mixed_c1_lt_10_2_no_ql with avg_return 2.81% and profit_factor 3.96.
• Compare this result to block_all_remaining_weak_subtypes to see whether narrow veto meaningfully preserves useful edge while still removing obvious losers.
• If a narrow veto matches most of the broad-block benefit with smaller date loss, it is the better alpha-first redesign candidate.