Research On Financial Crisis Early Warning Of Listed Companies Based On Statistical Learning

With the integration of the world and advancement of economic globalization,China's economic development is rapid and the market is changing quickly.China's companies have more opportunities to expand,but also face greater challenges.The risk is growing in business,and a mistake of strategy in business may lead the enterprise into a crisis.It makes the enterprise lower the speed of business development and miss the opportunities of development,which is worse,face the delisting and declare bankruptcy.In the background,the financial early warning system plays an important role increasingly.An effective financial early warning system can not only conduct early warning of corporation crisis,but also effectively help enterprise to find the warning source and eliminate crisis as soon as possible to ensure that development of enterprise is proceeding smoothly.Based on reviewing and summarizing a large number of literatures on financial crisis early warning model,and analyzing the advantages and disadvantages of domestic and foreign researches,this thesis determines the financial early warning index according to the specific conditions of listed companies in China,and builds financial warning models based on statistical learning methods.The support vector machine,Logistic regression,BP neural network,the correlation analysis and principal component analysis were applied to construct the financial early warning model in this thesis.The financial early warning model based on the statistical learning method built in this thesis can effectively overcome the shortcomings of the recent Z-score method and the Fisher discriminant method which need the sample to meet limited conditions.What is more,as new training samples appear,the models based on statistical learning method in this thesis,can be used to train the model immediately,adjust the model parameters,and improve the ability of model forecasting.This thesis selects 62 ST companies and 62 non-ST companies under the A-shares of Shanghai and Shenzhen stock market as the research objects,uses the Scikit-learn machine learning package in Python to build model,and also uses the cross validation and grid search technique for model training and parameters tuning.The empirical result shows that three kinds of financial early warning models constructed in this thesis can reach over 80%predicting accuracy on the test set,which demonstrates that the effectiveness of the three kinds models.At the same time,this thesis notices that the early warning model based on support vector machine has better stability in the condition of high accuracy,which is better than the early warning model based on Logistic regression and BP neural network.