Introduction to multi-factor screening analysis
Overview
Multi-factor stock selection model in the model building, often involves a lot of stock price impact factors, and may derive a large number of candidate models. Therefore, the evaluation and screening of the multi-factor stock selection model is particularly critical.
Process
There are many fundamental data factors (features), and it is particularly important to find the factors that correspond to high stock returns.
The process of mining factors
First analyze some of the factors that are effective for stock returns from hundreds of factors:
- Filter in each category of factors, and select effective N factors from each category of factors, including factors such as quality, valuation, growth, etc.
- Strict: e.g. 20 effective factors
- Not strict: e.g. 50 effective factors
Perform correlation analysis among the single factors screened, and merge the factors with strong correlation
- In the end, effective, weakly correlated factors, generally around 10
- Audition -> N factors -> Featured -> n factors
Effectiveness analysis
Factor IC analysis:
- IC (Information Coefficient): Information coefficient, which represents the correlation strength between the factor and stock returns
Factor return analysis:
- Determine the stock direction of the factor
Factor direction
| Factor direction | Factor description |
|---|---|
| Factor ascending | The smaller the factor value, the better. Such as market value, valuation (price-earnings ratio, price-to-book ratio, price-to-sales ratio, etc.) |
| Factor descending | The larger the factor value, the better, such as ROE, profit, profit growth rate factors |
| Factor neutral | The direction of the factors is uncertain, such as turnover rate, asset-liability ratio and other factors |