Research On Financial Crisis Warning Via Penalized COX Model

With the increasingly fierce competition,listed companies are facing more uncertainties.Listed companies can not ignore the financial risks while pursuing their interests.Therefore,how to select important indicators from a wide range of financial indicators and establish an effective financial crisis warning model is particularly important.This paper adopts the COX model which has the advantages of being free from the distribution of dependent variables,working with censored data,computing the probability of a company's financial crisis at a specific point in time,and dynamically display the company's financial status.We consider the group structure based on economic significance of financial indicators,use CMCP method to select financial indicators,and establish CMCP-COX financial crisis early warning model.In this paper,we first reviews the literature on financial crisis early warning and variable selection,and summarizes the financial crisis early warning model and indicator screening method commonly used in current research.Secondly,we introduce the basic theory of survival analysis and COX proportional risk model,the CMCP penalty variable selection method and the theory and solution method of CMCP-COX model.In the simulation,the data are set according to the characteristics of the real data and verify the effectivity of the CMCP method selection financial indicators under different correlation scenarios of the group indicators.The results show that under all three correlation scenarios,compared with four alternative models(including Lasso-COX,stepwise COX,COX,and stepwise Logistic),the CMCP-COX always presents better variable selection and prediction performance in terms of TPR,FPR,and MSE,especially in high correlation.In the empirical analysis,we select the data of 425 listed companies and construct the CMCP-COX financial crisis early warning model based on the financial indicator group structure.The results of parameter comparison analysis show that the model forms of CMCP-COX model and Lasso-COX model are more simple effective and easier to explain than the stepwise COX model,COX model and stepwise logistic model.In the long-term forecast,the prediction accuracy of the test set of the CMCP-COX model is higher than 85%,especially in the“crisis”category,the prediction accuracy is higher than 90%.Its prediction accuracy and AUC value both are superior to traditional stepwise logistic model and other COX models,and the prediction results are better.The time-point prediction effect of the CMCP-COX model has better performance than other COX models.The prediction results of the CMCP-COX model are stable in each period and reflect the changes in the financial status of the enterprise in a timely manner.In summary,the CMCP-COX model has better performance and more practical significance for the financial crisis early warning.Finally,we propose corresponding policy recommendations based on the research findings and some ways to extend this work.