Multi-factor-Model-for-Stock-Selection/result_ml.txt at main · Parsnip77/Multi-factor-Model-for-Stock-Selection · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137

======================================================================
  Stage 3 · LightGBM Alpha Synthesis
======================================================================
1. Loading factors_clean.parquet and prices.parquet ...
   factors shape : (364200, 16)
   prices  shape : (361333, 9)

2. Computing 1-day forward returns ...

3. Merging factors and target labels ...
   merged shape  : (357978, 18)
   feature cols  : ['alpha001', 'alpha003', 'alpha006', 'alpha012', 'alpha038', 'alpha040', 'alpha041', 'alpha042', 'alpha072', 'alpha088', 'alpha094', 'alpha098', 'alpha101', 'alpha_5_day_reversal']
   training target : cs_rank_return  (cross-sectional pct-rank of forward_return, per trade_date)

======================================================================
  Walk-Forward Training
======================================================================
   Settings : train=16m  val=4m  test=4m  embargo=1d
   Estimated folds : 9

--- Fold 1 ---
  Train : 20190102 → 20200522  (n=99,349)
  Val   : 20200525 → 20200921  (n=24,609)
  Test  : 20200923 → 20210127  (n=24,768)
  Best iteration : 519

--- Fold 2 ---
  Train : 20190513 → 20200921  (n=99,263)
  Val   : 20200922 → 20210126  (n=24,768)
  Test  : 20210128 → 20210607  (n=24,719)
  Best iteration : 420

--- Fold 3 ---
  Train : 20190909 → 20210126  (n=99,030)
  Val   : 20210127 → 20210604  (n=24,716)
  Test  : 20210608 → 20211013  (n=24,648)
  Best iteration : 142

--- Fold 4 ---
  Train : 20200114 → 20210604  (n=98,769)
  Val   : 20210607 → 20211012  (n=24,648)
  Test  : 20211014 → 20220216  (n=24,847)
  Best iteration : 142

--- Fold 5 ---
  Train : 20200525 → 20211012  (n=98,741)
  Val   : 20211013 → 20220215  (n=24,846)
  Test  : 20220217 → 20220622  (n=24,757)
  Best iteration : 486

--- Fold 6 ---
  Train : 20200922 → 20220215  (n=98,978)
  Val   : 20220216 → 20220621  (n=24,761)
  Test  : 20220623 → 20221026  (n=24,890)
  Best iteration : 190

--- Fold 7 ---
  Train : 20210127 → 20220621  (n=98,971)
  Val   : 20220622 → 20221025  (n=24,888)
  Test  : 20221027 → 20230301  (n=24,854)
  Best iteration : 224

--- Fold 8 ---
  Train : 20210607 → 20221025  (n=99,143)
  Val   : 20221026 → 20230228  (n=24,854)
  Test  : 20230302 → 20230705  (n=24,831)
  Best iteration : 288

--- Fold 9 ---
  Train : 20211013 → 20230228  (n=99,349)
  Val   : 20230301 → 20230704  (n=24,830)
  Test  : 20230706 → 20231108  (n=24,771)
  Best iteration : 595

======================================================================
  Assembling Final ML Alpha
======================================================================
   final_alpha_df shape : (223085, 3)  (after 3-day smoothing)
   date range : 20200923 → 20231108

======================================================================
  IC Analysis — ML Synthetic Factor
======================================================================
   IC Mean : +0.0439
   IC Std  : 0.0982
   ICIR    : +0.4472

   IC chart saved: plots/ml_alpha_ic.png

======================================================================
  Layered Backtest (LayeredBacktester)
======================================================================

Layered Backtest Performance:

     Cum Return  Ann Return  Ann Vol  Sharpe  Max DD
G1      -0.4827     -0.1973   0.1607 -1.4144 -0.5220
G2      -0.1957     -0.0700   0.1667 -0.5998 -0.2693
G3      -0.0285     -0.0096   0.1655 -0.2392 -0.2371
G4       0.0623      0.0204   0.1688 -0.0572 -0.2455
G5       0.3644      0.1091   0.1761  0.4494 -0.2235
L-S      1.6169      0.3780   0.0995  3.4989 -0.1010

======================================================================
  Feature Importance (Average Across Folds)
======================================================================

Average Feature Importance (gain):

             feature  importance
            alpha038 5253.301237
            alpha042 4729.082828
            alpha101 4277.963360
alpha_5_day_reversal 3741.844163
            alpha041 3518.533852
            alpha001 3372.430376
            alpha094 3271.414493
            alpha012 3113.814430
            alpha040 2471.851905
            alpha088 2467.149983
            alpha006 2359.141435
            alpha072 2342.694817
            alpha003 2294.295588
            alpha098 2096.524049

   Saved: plots/feature_importance.png

======================================================================
  SHAP Analysis (Last Fold Test Sample)
======================================================================
   SHAP sample size : 300 rows
   Saved: plots/shap_beeswarm.png

======================================================================
  Report Written
======================================================================