Application Of Ensemble Method In Financial Early Warning Research With Data Based On MD&A And Financial Indicators

The healthy and vigorous development of listed companies is the backbone of supporting the high-quality development of China's economy.As a leading company in its region and industry,listed companies have injected impetus,vitality,and leadership into the economic,technological,and innovative development of high-quality economic.However,if a financial crisis occurs during the development of a listed company,it will influence the development of its managers,investors,governments,and the entire economic society.Therefore,it is necessary to conduct financial early warning of listed companies' potential financial crisis issues,so that listed companies stakeholders take timely preventive and resolution measures to avoid or reduce the losses of financial crisis.This paper selects from 2010 to 2018 A-share listed companies in Shenzhen Stock Exchange and Shanghai Stock Exchange as the research object of financial early warning.Considering that the financial status of listed companies is affected by macro and micro factors,This paper not only uses the company's financial indicator data in research data selection,but also Management Discussion & Analysis(Hereinafter referred to as MD&A)of listed companies' annual reports that can reflect economic and financial environment information is taken into consideration,that is combining financial indicator data and MD&A data into mixed data for financial early warning analysis,to discuss whether MD&A as incremental information of financial indicator data is conducive to the improvement of financial early warning capabilities.In terms of research model selection,this paper first build single classifier models based on the ideas of single classifier Decision Tree(DT),Logistic Regression(LR)and Support Vector Machine(SVM)for financial early warning research.In addition,considering the limitations of the single classifier and the advantages of the ensemble method,the ideas of the Bagging,Boosting,and Stacking in the ensemble methods are used to build an integrated model based on the single classifiers DT,LR,and SVM.The financial early warning performance of different ensemble methods,integrated models and single classifier models are compared and analyzed,and the factors affecting the performance of integrated models are discussed.The best-performing model in the empirical results is the Bagging-DT(RF)model based on the Bagging ensemble method and DT as the base classifier which used mixed data.Its accuracy(Accuracy)reaches 86.1%,and AUC(Area Under the Curve)is 0.912,which realizes more accurate forecasting and judgment on the issue of financial early warning.The research in this paper shows that MD&A as the incremental information of financial indicator data can improve the accuracy of financial early warning.At the same time,the application of ensemble methods can better learn the characteristics of the data.The integrated model based on the ensemble method enhanced the financial crisis early warning capability.The research provides a reference for future financial early warning research in data selection and method selection,and provides useful information for listed company stakeholders to make decisions.It has certain theoretical and practical significance.