Research On Credit Rating And Influencing Factors Of Listed Companies Based On Multi-scale Convolutional Neural Network

In recent years,economic globalization has encountered setbacks,and the international economic cycle has undergone in-depth adjustments.More importantly,the trend of anti-globalization has also been exacerbated by the new crown epidemic.With economic development entering a new normal and the impact of the new crown epidemic,China’s economy is currently facing some challenges,such as the triple pressure of demand contraction,supply shock,and weakening expectations.With multiple shocks,the Chinese economy is facing increasing downward pressure.In the process of the development of the real economy,the important role of the bond market can be seen.It broadens the financing channels of enterprises,promotes the development of the real economy,and diversifies financial risks.Therefore,with the development and expansion of the bond market,the primary direct financing method for enterprises is the issuance of bonds.The rapid development of corporate credit bonds,the proportion of credit bonds in the bond market has reached the level of three pillars with national debt and financial bonds,it can be seen that credit bonds are of great significance to corporate financing.In order to solve various problems such as corporate financing,the first thing to be solved is how to conduct corporate credit rating.According to the influencing factors of corporate credit rating,combined with the existing research of scholars in this field at home and abroad,this paper selects a multiscale convolutional neural network to establish a corporate credit rating model.The data are mainly divided into two categories:corporate financial indicators and corporate governance capability indicators.Added financial distress and financing constraints technical indicators.Then perform data cleaning on the sample data,perform multicollinearity test on the indicators,use Random Forest and XGBoost to screen the importance of indicators,obtain the top 20 most important features of each model,and select the same 12 most important indicators for discussion.After removing unimportant indicators,two sampling methods of undersampling and SMOTENC are used to deal with the imbalanced data set of the training set,and the influence of multicollinearity on each model is discussed.Next,we select financial distress and financing constraints technical indicators to discuss whether these technical indicators can provide additional explanatory power on Multi-scale Convolutional Neural Networks.Finally,technical indicators are selected to train random forest,XGBoost,Neural Network and Multi-scale Convolutional Neural Network,select the best results of each model for comparison,compare and analyze multiple multi-class evaluation indicators,and the optimal results of each model are selected for comparison.The best model is the Multi-scale Convolutional Neural Network model.Compared with the evaluation indicators of other enterprise credit rating models,the Multi-scale Convolutional Neural Network model improves the accuracy by 4.37%,the Macro F1 score by 4.82%,the Macro precision by 4.56%,the Macro recall rate by 4.77%,and the Macro AUC value by 1.87%.