Research On Credit Evaluation Of Listed Companies Based On Random Forest Model

Since the Central Economic Work Conference in December 2018 first mentioned the construction of new infrastructure(hereinafter referred to as "new infrastructure"),the concept of new infrastructure has become a hit in society.The Chinese economy,which is already under the dual pressure of economic structural transformation and trade wars,has been hit by the epidemic of COVID-19.New infrastructure is expected to counter economic downturn risks and promote the transformation and upgrading of the entire society.However,in recent years,the number of credit defaults of new infrastructure listed companies has been increasing year by year,and they are facing a series of credit risks caused by problems such as excessive investment,limited technology,and inadequate supervision.The establishment of a scientific credit evaluation model to evaluate the credit risk of new infrastructure listed companies can not only strengthen the company's own early warning and optimize the investor's asset portfolio,but also encourage the government to promote the orderly and stable development of new infrastructure listed companies.Based on this,the research conducts following research: how to scientifically evaluate the credit risk level of listed companies of new infrastructure.This article starts with the credit default problem of listed companies of new infrastructure in A-share market,and uses i Find database and CSMAR database as the data sources for the research.This article deliberates the credit evaluation of listed companies of new infrastructure mainly from three parts: theoretical analysis,credit risk characteristic analysis,and empirical research.The theoretical part is mainly expounded from the perspective of corporate credit risk,describing the cause and transmission mechanism of corporate credit risk.This article analyzes the credit risk characteristics of listed companies of new infrastructure and summarizes the five major credit risks based on former literature and theory.In the empirical research,this article takes the listed company of new infrastructure as example,comprehensively considers the impact of multiple factors on the credit of the listed companies of new infrastructure,and extracts 29 variables that may be related to the credit forecast of the listed companies of new infrastructure and compares the current mainstream machine learning algorithm,adopting random forest algorithm to evaluate the credit risk of new infrastructure listed companies.The empirical part can be divided into three steps.First,this article uses the OOB error rate and Gini coefficient method in the random forest model to select 18 indicators with the highest importance among a total of 29 candidate indicators.Then this article uses SMOTE algorithm and grid search method to optimize data and the random forest model,separately and rank the feature variables in the optimized random forest model based on their importance.Finally,this article compares the SMOTE-RF model with SVM model,logistic regression model and CART model.The research finds that:(1)Listed companies of new infrastructure face five major risks: excessive investment,single financing channels,insufficient working capital,improper corporate governance structure and limited technical level.(2)Factors that have a greater impact on the credit risk of listed companies of new infrastructure include18 indicators such as the number of patents,earnings per share,and total asset turnover.Among them,the number of patents,EPS and turnover rate are the top three important factors.The three most trivial indicators are the fixed asset turnover rate,the net cash flow of operating income,and the cash operation index.(3)From the comparison results before and after model optimization,this article finds that the prediction accuracy rate of optimized SMOTE-RF is as high as 90.91%,which is better than the SMOTE-RF model before optimization,and the prediction accuracy of the SMOTE-RF model before optimization is higher than that of the RF model.It shows that SOMTE algorithm and grid method are effective methods to improve the accuracy of prediction.(4)Comparing different models,the prediction accuracy of the optimized SMOTE-RF model outperforms the SVM model,logistic regression model and CART model.The performance(AUC value)of the model is as follows: SMOTE-RF model(0.91),SVM(0.82),CART(0.79),logistic regression(0.74).The main contribution of this article lies in the fact that when the existing literature conducts credit evaluation research on listed companies,more consideration is to select financial indicators and use traditional credit evaluation models.Based on the analysis of the credit risk characteristics of listed companies of new infrastructure,this paper innovatively combines financial and non-financial indicators,using optimized random forest model to evaluate the credit risks.This method not only enhances the comprehensiveness of the credit evaluation index system,but promotes the accuracy of credit risk evaluation,which can provide ideas for the credit evaluation research of listed companies of other industries.