Comparative Study On The Predictive Ability And Index Explanation Strength Of Corporate Default Models Before And After The Epidemic

The outbreak of the COVID-19 epidemic in early 2020 has brought a huge impact on the global economy.The import and export and trade exchanges(except the medical industry)of various countries have decreased.At the same time,domestic economic activities have also been forced to reduce due to the impact of the epidemic.Many companies have suffered losses,and then cause these companies to default on their debts.At present,corporate default models are mainly divided into two categories: traditional models and machine learning models.After the outbreak of the COVID-19 epidemic,the predictive ability of the above two models for corporate defaults has not been well tested.Therefore,in the new market environment brought about by the current new crown epidemic,the prediction ability and index interpretation strength of the company’s default model are analyzed to screen out the company’s default model and index system that are more suitable for the current new market environment,so as to better It is particularly important to manage the default risk of the company.Based on Logistic regression and random forest model,this thesis compares and analyzes the predictive ability and index interpretation strength of corporate default models before and after the epidemic,and then provides basic guidance on model selection and index selection for the actual work of financial risk management.The research in this thesis can help users of the corporate default model to better deal with the losses caused by corporate defaults,and conduct more scientific management of credit risks related to bond issuers,so as to ensure the healthy and orderly development of my country’s financial industry.Based on the above research purposes,this thesis selects the relevant financial quarterly data of companies that publicly issue credit bonds in my country from December 31,2017 to September 30,2021.First of all,this thesis takes the outbreak of the COVID-19 epidemic in my country on December 31,2019 as the time point before and after the epidemic,and divides the collected sample data into pre-epidemic samples and post-epidemic samples.In order to construct and verify the company default model,this thesis divides the samples before and after the epidemic into the training set and the validation set data according to the ratio of 4:1.The training set data is used for model construction,and the validation set data is used for model prediction.Ability and explanatory power analysis of indicators.In terms of the selection of financial indicators,this thesis selects a total of 21 indicators that reflect the company’s solvency,operating ability,profitability and growth ability as the input variables of the model.In terms of model prediction ability analysis,this thesis uses four indicators,such as recall rate,precision rate,F1 value and AUC value of ROC curve,as the evaluation standard of model prediction ability.In terms of index explanation strength analysis,this thesis uses the SHAP tool to visualize the contribution of each model input index to the model output and the ranking of their relative importance.This paper draws four main conclusions through comparative research:(1)In terms of the predictive ability of the logistic regression model,the predictive ability of the logistic regression model after the epidemic is better than that of the logistic regression model before the epidemic.The logistic regression model has good robustness.Under the new impact of the COVID-19 epidemic on the bond market,it can still better predict corporate defaults and adapt to the new bond market environment.(2)In terms of the prediction ability of the random forest model,the prediction ability of the random forest model after the epidemic is slightly better than that of the random forest model before the epidemic.The overall robustness of the random forest model is good,it can better capture the new characteristics of corporate defaults in the new market environment brought about by the COVID-19 epidemic,and the ability to predict corporate defaults is relatively stable.(3)In terms of the explanatory power of the Logistic regression model,the top 5 indicators that the Logistic regression model before the epidemic has the strongest interpretation of the model output results are the gross profit margin of sales,the turnover rate of current assets,the turnover rate of total assets,and the total assets.Year-on-year growth rate and current ratio.Among the four major categories of financial indicators selected in this paper,the overall operational capability indicators are the most prominent in explaining the model output results.In the logistic regression model after the epidemic,the top five indicators that contributed the most to the output of the model were the total asset turnover ratio,cash ratio,gross profit margin,asset-liability ratio,and year-on-year growth rate of total assets.Among the four major categories of financial indicators selected in this paper,the solvency index as a whole has the most prominent explanation for the model output results.(4)In terms of the index interpretation strength of the random forest model,the top five indicators that the random forest model before the epidemic has the strongest explanation for the model output results are earnings per share,return on equity,return on total assets,and interest guarantee.Multiples and Cash Ratios.Among the four categories of financial indicators selected in this paper,profitability indicators have the greatest impact on the model output.In the post-pandemic random forest model,the top five indicators with the strongest explanatory power are earnings per share,cash ratio,return on equity,year-on-year growth rate of total assets,and current ratio.Among the four categories of financial indicators selected in this paper,profitability indicators and solvency indicators have a greater impact on the model output.