Application Research Of RF-PSO-SVM Model In Financial Distress Warning Of A-share Listed Companies

Since the reform and opening policy,the country’s real economy is thriving,and so is the capital market.As of August 31,2021,the total number of listed companies on the two exchanges reached 4,931,with a total market value of more than 87 trillion yuan.With the rapid development of the capital market,the country has been promoting relevant reforms,strengthening the supervision of listed companies,improving the quality of listed companies,improving,the socialist market economic system,and ultimately achieving the purpose of common prosperity.At present,in the face of COVID-19 and the impact of friction with the United States,it is of great significance for both government agencies and listed companies to improve the quality of listed companies,improve the exit mechanism of listed companies,accurately identify companies at financial risk,take precise measures and help them tide over difficulties.Data mining and machine learning are the hottest topics in recent ten years.Many scholars have introduced relevant algorithms into the field of financial distress prediction and achieved good prediction results.According to the operation situation of listed companies,select appropriate indicators and build a set of models that can accurately predict the financial difficulties of listed companies,which can effectively help companies identify risks and tide over difficulties.In this paper,197 a-share-listed companies that were specially treated for the first time from 2016 to 2021 are selected from the Oriental Fortune Choice data client,and the same number of normal companies are selected as the control group by one-to-one paired sampling method.Starting from seven dimensions of financial and nonfinancial perspectives,43 indicators are selected to reflect the operating situation of the company.After processing the missing values and outliers of the original data,and passing the normality test,significance test,and correlation analysis,the preliminaries are selected as the initial feature set of the input model.In addition,accuracy,recall,F1,Kappa,and AUC values are selected as evaluation indexes of different models,and the ten-fold cross-validation method is adopted to divide the training set and test set in the models.The feature set of the input model contains three types of indicators: raw indicators,indicators that have undergone dimension reduction through principal component analysis,and indicators that have been screened by random forest.Data need to be standardized before entering the model,and the principal components selected by principal component analysis need not be standardized.Support vector machines are divided into SM0-SVM and PSO-SVM according to different parameter optimization methods.The models constructed by different methods are compared and analyzed,and corresponding conclusions are finally obtained.The empirical results show that:(1)RF-PSO-SVM model can achieve good prediction effect,and the accuracy of prediction at T-1,T-2,and T-3 time points is87.5%,90%,and 92.5%,respectively;(2)Collinearity among indicators does not significantly affect the prediction performance of the classifier,and PCA dimension reduction does not significantly improve the classification prediction accuracy of the standard SVM model;(3)Listed companies in manufacturing,large and mediumsized,established more than 10 years ago,located in coastal areas and developed provinces are more likely to fall into financial difficulties;(4)The external performance of financial distress in different stages is different,which is reflected in the difference of characteristic indicators in different reporting periods.The closer the time of financial distress,the more characteristic indicators.