Study On Early Warning Of Financial Crisis Of Listed Companies Based On XGBoost Model

In recent years,as China's capital market economy continues to deepen the reform,Many companies began to enter the capital market in order to achieve better development,and the market competition increased sharply.Many enterprises are facing tremendous losses or even bankruptcy because of the financial crisis,the enterprise has a long financial crisis usually has a hidden period,If we can predict the financial crisis in advance,take early action,can greatly reduce the loss of investors and operators.How to predict the status of financial crisis in advance.What factors have a key impact on the financial position of the business.How to establish an early warning model with high efficiency and strong applicability is a hot topic that many scholars have studied in recent years.Therefore,the establishment of early warning model of financial crisis has great practical significance.Manufacturing listed companies in China's A-share market share of up to 60%,play a pillar role in the national economy.Therefore,this article takes the A-share manufacturing market of China from 2016 to 2017 as the research object of "financial abnormality*" by ST and 126 non-ST listed companies.27 financial indicators have been selected from the five perspectives of corporate solvency,profitability,development capability,operational capability and cash flow.Before modeling,the financial statements data of 189 listed companies(T-2)and(T-3)years(2013,2014 and 2015)were p-reprocessed by using SPSS 22.0 statistical software.First,the K-S test was used to sample data,and T-difference significant test was conducted on the variables subject to normal distribution.on the contrary,mann-Whitney U non parametric test was performed.Factor analysis will be conducted to make a significant distinction between ST and non-ST companies,finally,the index with higher information content and lower correlation is extracted as the initial variable of modeling.Then using the Logistic regression and the XGBoost model to establish the financial early-warning model based on(T-2),(T-3)annual financial data,at the same time for each early warning model select the "optimal segmentation point",and 135 training samples and 54 test samples were used to verify the model's accuracy and test.Research indicates:The five financial indicator variables:equity ratio,return on net assets,operating profit ratio,return on assets,and net profit growth can significantly distinguish the financial crisis situation of the company.They are long-term predictors;The asset cash recovery rate and operating cost rate are short-term discriminant indicators.Simultaneously,the two types of early warning model built in the company were ST in the first two years and the first three years have achieved good predictive results,And XGBoost warning model accuracy is much higher than the Logistic warning model,reached 97.04%,(T-2)year forecast accuracy is higher than(T-3)years.Thus,XGBoost early warning model achieved good forecasting results,has a certain real value.At last,This article points out the inadequacies of related research,to be improved in the follow-up study.