Study On Identification And Forecast Of Default Risk Factors Of Listed Companies

Default risk is also regarded as credit risk,which put into use the likelihood of a corporation or individual being subjected to losses due to certain factors.Affected by the new crown pneumonia epidemic,firms are coming the stress of being impossible to go on work and manufacture and insolvency,which has even led to some companies on the verge of bankruptcy and some companies face the risk of default.However,the occurrence of corporate default risk does not happen overnight.It is a product of a stage.It was initially reflected in a certain degree of abnormality in corporate assets,liabilities,and owner's equity,and then fell into financial distress for a period of time and was accompanied by loan default risks.From time to time until it deteriorated to bankruptcy.At this time,it is particularly important to find out the characteristics of corporate default risk objectively and timely,and availably avert the wrong end of the risk of company default risk.Hence,so as to hold the regular progress and good competitiveness of enterprises.The default risk element recognition system must have the capacity to distinguish default risk,and the default risk prediction model must have universal applicability.Based on this,the research goal of this article is how to improve the identification system of default risk factors;the second is how to complete the default risk indicators.The system is to screen out the combination of indicators with the maximum redundancy and minimum redundancy in identifying the default status of listed companies;the third is how to build a default risk prediction model to improve the model's ability to classify default risk prediction and have a certain interpretability.The first research method of this paper is to use the five feature selection methods of variance selection method,F-Classif,SVM-RFE,Lasso and XGBoost were used for the raw data set and the reasons for each feature selection was ranked by feature importance,and obtain five feature subtypes set.The second is to vote on these five feature subsets using the mode voting method based on integrated feature selection,based on the principle of default risk factors containing the largest amount of information and the highest number of votes,and the purpose is to filter the ability to identify default risks from the original data set the strongest feature system.The third is to use clustering and under-sampling means to settle a matter of lopsided data sets,integrated with the Light GBM model with mighty classification capabilities,to resolve the problem of unbalance of analogy and promote the classification precision of the model arithmetic.The experimental verdicts of the default risk prediction of quoted companies manifest that the 35 default risk factors selected from the 212 original index systems using the integrated feature selection method not only have the ability to identify default status,but also are a scientific and reasonable factor identification system.Secondly,among the 35 indicators obtained through the integrated feature selection method,the internal accounting ratio of the firm "return on assets","days in inventory" and other indices,the "mean age of top managers" in the company governance structure,and the "firm price" in market factors"The indicator has the highest number of votes and is a key indicator that affects the risk of corporate default.Finally,the integrated CUS-LGB model(Cluster-based Under Sampling with Light GBM,be called for short CUS-LGB)referred in this thesis is compared with the financial forecast models used in other 7 comparisons,and it is found that the accuracy of the CUS-LGB model after identification of default risk factors coming up to 95.67%,the AUC value is 92.86%,and the type II error rate is also the smallest at 21.32%.It shows that the CUS-LGB model not only has high classification accuracy,but also can balance the unbalanced data set.