| Fundamental analysis plays a very important role in the field of securities investment,and the development of quantitative investment also provides many new perspectives for the research of securities investment.Using fundamental data to construct factors for quantitative research is a research method that combines the two.However,many fundamental factors underperform in empirical test.This paper makes an in-depth study on this phenomenon.This paper holds that the main reason for the poor empirical test results of most fundamental factors lies in the insufficient use of information in the process of factor construction.Existing fundamental factors usually only use the latest fundamental data in the construction process and ignore the value of their historical data.At the same time,the existing research methods focus on the test of a single fundamental variable,and there may be synergy between different fundamental variables.From the above two aspects,this paper improves the construction method of fundamental factors and makes an empirical test in the A-share market.First,this paper selects four fundamental factors:return on equity,operating profit to net income ratio,earnings before interest,taxes,depreciation and amortization per share,and current assets turnover.Next,the cross section regression method is used to improve the four original fundamental factors.The validity test results show that the improved fundamental factors have a significant improvement in effectiveness compared with the original fundamental factors,which can better explain the cross-sectional differences of stock returns.This result shows that in addition to the latest fundamental data,the historical data of fundamental variables are still important for the construction of fundamental factors.Finally,the paper constructs the compound fundamental factors by using different models to comprehensively consider the four fundamental variables.The empirical results show that the effectiveness of the compound fundamental factors obtained by using the appropriate model exceeds that of all single fundamental factors.In this paper,different complexity models are used to discuss the complexity of the interaction between fundamental variables.The models used in this paper include the multiple linear regression model,the Elastic Net model,and the XGBoost model that performs well in machine learning tasks.By comparing the test results of different models,we found that the XGBoost model performed best,with an average monthly return of 1.6%for its long-short portfolio.Therefore,this paper holds that the interaction between different fundamental variables is complex,and the construction of composite fundamental factors based on their interaction relationship can further improve the ability of fundamental information to explain the cross-sectional difference of stock returns. |
