Dynamic Stock Industrial Classification
Use graph-based analysis to re-classify stocks and experiment different re-classification methodologies to improve Markowitz portfolio optimization performance in the low-frequency quantitative trading context.
Note that for strategy confidentiality, many files are hidden.
This project contains the following five modules:
- factor generation: compute and store factors alpha factors and risk factors for low-frequency trading;
- backtest: low-frequency backtest framework;
- factor combination: combine factors using ML models;
- portfolio optimization: Markowitz portfolio optimization, with turnover, industrial exposure, style exposure, and various other constraints.
- graph clustering: experiment different graph-based clustering on stocks.
China A-Share stocks, the corresponding major index data (sz50, hs300, zz500, zz1000), and the member stock weights from 20150101 to 20211231, provided by Shanghai Probability Quantitative Investment.
It’s very easy to use this platform!
- run each module at a time;
- change config for corresponding module in respective files (file location indicated inside run.py).
To run each module, in current directory:
- factor generation:
python run.py gen
python run.py backtest
- factor combination:
python run.py comb
- portfolio optimization:
python run.py opt
- graph clustering:
python run.py cluster
Special thanks to coworkers and my best friends at Shanghai Probability Quantitative Investment: Beilei Xu, Zhongyuan Wang, Zhenghang Xie, Cong Chen, Yihao Zhou, Weilin Chen, Yuhan Tao, Wan Zheng, and many others. This project would be impossible without their data, insights, and experiences.