Research on Enterprise Financial Crisis Warning Based on Multiple Eigenvalue Screening Methods

Abstract

There are many factors that affect listed companies being ST. Previous research mainly relies on experience, intuition, historical research results and indicators that have seriously deteriorated, and uses a large number of statistical testing methods to screen out the more important indicators (characteristic indicators) in the model. At present, there are many feature index screening methods in machine learning, different screening methods have different effects, each method has its own advantages and disadvantages. This paper takes China's A-share listed companies as the research background, 48 indicators have been preliminarily determined in 6 aspects of capacity, operating capacity, cash capacity, profitability, development capacity and social responsibility indicators. After statistical analysis of the initial indicators, we have obtained the results that can effectively distinguish the sample enterprises of financial crisis and the sample enterprises of financial health 37 indicators, and then use RF, REF, MIC and Lasso to screen the indicator eigenvalues, obtain the eigenvalues and pictures under the four methods, and then obtain 18 eigenvalues according to the application principle of the comprehensive screening method. Finally, the eigenvalues screened by the five indicators are used as the input variables of the financial crisis early warning model, and are brought into the model for empirical analysis. The study found that the characteristic indicators obtained by the comprehensive screening method have the highest accuracy in predicting financial crisis.