Market Value Interpretation Model Based On Machine Learning Factor

Overseas,quantitative investment has been accepted by more and more investors after decades of development.Quantitative investment has the background knowledge of mathematical statistics,and can avoid the interference of artificial emotions.Its investment performance is stable,and the scale and share of the market are also growing.With the continuous improvement of China's capital market and the continuous listing of innovative financial derivatives,quantitative investment has also been rapidly developed in the domestic capital market.Based on this background,this paper regards all the constituent stocks of CSE as the research object,establishes the market value interpretation model,constructs the quantitative stock selection factors of valuation errors,uses Python grammar programming to test the factors in group simulation transaction,and carries on the empirical analysis to the stock selection ability of the factors.Based on the theoretical background of value investment and the decomposition of market value by Rhodes-Kropf,Robinson and Viswanathan(2005),this paper constructs a market value model interpreted by financial indicators,and improves the model by machine learning.Through the market value interpretation model,this paper constructs the valuation error factors.Under the framework of the analysis of factor effectiveness in practice and industry,this paper makes an empirical analysis of the factors constructed.The research period chosen in this paper is from August 1,2011 to June 1,2019,which includes bear market,bull market,rebound and shock stage.It is more reasonable to test the validity of valuation error factors in different market conditions.In this paper,the data preprocessing of the constructed valuation error factors is carried out,including median de-extremum,standardization and industry market value neutralization.The purpose is to unify the dimension of the factors and facilitate comparison.Then,the stock return is simulated by grouping.The maximum excess return is 490.7% for the linear market value interpretation model,502.8% for the random forest market value interpretation model and 502.8% for the XGBoost market value interpretation model.The maximum excess return of the error factor is554.9%.The excess return obtained by the error factor constructed by the three methods is significantly higher than the benchmark return of the full index of the stock market,which proves the validity of the error factor.Although the performance of valuation error factors constructed by linear market value interpretation model is good,the relationship between financial indicators and market value is complex,which may include non-linear relationship,which is difficult to explain by linear market value interpretation model.Therefore,the residual information of the linear market value interpretation model is impure.Therefore,we adopt machine learning method which can solve the non-linear relationship to explain market value.The average excess return of the three models of linear market value interpretation model is 456.93%,and the average degree of market value interpretation is 0.945.The average excess earnings of the three models of the stochastic forest market value interpretation model are 495.33%,and the average degree of explanation for the market value is 0.974.The average excess earnings of the three models of XGBoost market value interpretation model are 527.97%,and the average degree of market value interpretation is 0.997.Compared with the linear model,the average excess return of machine learning model has been greatly improved,and the average explanation degree of market value has also been improved to a certain extent.It shows that machine learning market value interpretation can better explain the relationship between financial indicators and market value,and the evaluation error factor constructed is better.In addition,this paper also compares the traditional valuation class factor with the valuation error factor constructed in this paper,which further verifies the validity of the valuation error factor.In this paper,machine learning method is applied to market value interpretation model.The average interpretation degree of cross-sectional financial indicators to market value is more than 99%.An empirical analysis of the error factors of valuation is carried out to verify the validity of the factors.It is an extension of the application of machine learning to the financial field.In addition,this study has a certain reference for exploring factors in domestic investment and academic fields.