IFactor = AI + Multifacor A multifactor model combinded with machine learning algorithm on China A-shares.
Support q_factor(Hou.[2015}]) and FF5(Fama.[2015]) to be continued.
AI model can't run without GPU...
Requirement: Python 3.8.10 sklearn dateutil Alphalens-reloaded Tushare ArcticDB
Result:
- Q_factor worked on A-share from 20100101 to 20240424, annual return(without commision) of Long/Short portfolio is about 2.5%, cumulative return is 145%, IC mean is 0.018.
Information Analysis
1D 10D 20D
IC Mean 0.007 0.014 0.018
IC Std. 0.050 0.059 0.061
Risk-Adjusted IC 0.135 0.243 0.291
t-stat(IC) 7.921 14.260 17.092
p-value(IC) 0.000 0.000 0.000
IC Skew -0.061 -0.054 0.067
IC Kurtosis 0.553 1.080 0.594
- Frankly speaking, the result of q_factor is under expectation, main reason is ROE factor which does not work on A-share, maybe it's suit for US market but not same with A-share, many reasons..
So, I simply replace ROE with PB, result is much better: annual return(without commision) of Long/Short portfolio is about 6.1%, cumulative return is 220%, IC mean is 0.043.
Information Analysis
1D 10D 20D
IC Mean 0.021 0.034 0.043
IC Std. 0.114 0.135 0.142
Risk-Adjusted IC 0.181 0.250 0.302
t-stat(IC) 10.660 14.669 17.768
p-value(IC) 0.000 0.000 0.000
IC Skew 0.112 0.168 0.162
IC Kurtosis 0.166 -0.462 -0.483
Simulated trading(keep updating)
Portpolio name: IFactor
Open position on April 22th, 2024
Trade details:
Performance from April 22th, 2024 to May 23th, 2024
CSI500 fell 0.91% on May 24th, but my fortofolio flatted.
To do:
- Find more factors for A-share.
- How to run deep learning model with my poor computer?