| In this era of globalization,the economies of various countries are booming and market competition is becoming increasingly fierce,listed companies are facing a variety of risks and challenges.Due to the increasingness of business activities and financial status,as well as the factors that trigger the enterprise financial crisis,the probability of listed companies falling into financial distress is greater than ever.Generally speaking,it’s not in a sudden that listed companies get in financial distress,but in a dynamic and continuous process.Therefore,it is especially important to forecast financial crises in advance so that managers can detect crisis signals earlier in which situation enterprises,financial institutions and investors can take corresponding preventive measures.The establishment of an accurate and efficient early warning model for financial distress has become a significant direction in theoretical and practical research.This paper takes A-share listed companies in China’s Shanghai and Shenzhen markets as the research objects,and takes "ST" as the symbol for enterprises in financial distress.216 listed companies from 2015 to 2019 are selected as the research samples.To reflect companies’ profitability,growth capacity,operating capacity,solvency,cash flow and equity structure,33 indicators were selected based on the six aspects.And a financial warning indicator system was constructed for T-2,T-3 and T-4 years.The indicators with significant differences between the ST samples and the normal samples were initially screened by significance test based on the results of K-S normality test,in which the independent sample t-test was implemented for normal distributed indicators and the Mann-Whitney U-test was implemented for the indicators not conforming to the normal distribution.After that,29 indicators were retained in year T-2,25 indicators in year T-3,and 24 indicators in year T-4,showing the characteristic that the closer to the ST year,the more indicators with significant differences between ST companies and normal companies.Next,factor analysis was used to reduce the dimension of the indicators and eliminate multicollinearity.And logistic regression,decision tree,random forest,XGBoost and Stacking hierarchical models for each year were constructed to predict the financial status of year T,based on the public factor scores of each sample as the data base.Finally,use relevant evaluation indicators to make crossed comparison for the prediction effects of different models and different data periods.The results from the financial distress early warning models based on the final data of T-2,T-3 and T-4 years,show that the closer the year T is,the better the prediction effect of each model will be,and the integrated learning algorithm model for financial distress early warning effect is better than the basic classification algorithm model in each year.Among those models,the Stacking hierarchical model has the best prediction performance,and its prediction accuracy in year T-2 The model has an accuracy of0.886 and an AUC of 0.934 in year T-2,0.796 and an AUC of 0.866 in year T-3,and0.727 and an AUC of 0.765 in year T-4.The model also has a recall of 1 in year T-2,which means that it can fully identify companies in financial distress and is useful for financial early warning.This paper not only verifies that the financial situation of financially distressed companies has the characteristics of dynamic accumulation by empirical analysis,but also constructs a financial distress early warning model which gets a high prediction accuracy.So,it is highly feasible and practically meaningful to use machine learning algorithm to conduct financial distress early warning research for listed companies in China,so that companies can take correspondingly preventive measures according to the prediction effect of the model in different years to prevent the problems. |
PDF Full Text Request