Analysis Of Predictive Performance Using Machine Learning Methods:Evidence Of Delisted Firms In China During 1998-2016

Over the last decades,China as the second largest economy in the world has been the focus for academics and analysts in evaluating the internal and external financial shocks.Corporate financial distress has been an essential concern because,after crisis and fatal events,the companies are one of the agents in the economies that have been more affected and have to exit the markets.This study compares the performance of six machine learning models to solve the dichotomous classification problem of the listed and delisted companies on Shanghai and Shenzhen Stock Exchange Markets in China for the period 1998-2016.This research includes 2,863 companies that were split into training,validate,and test sample(50/25/25),to evaluate the effectiveness of the models.Accordingly,the study compares decision trees(DT),boosting,random forests(RF),support vector machines(SVM),logistic regression and artificial neural networks(ANN)models.A feature of variable selection modeling has been built by using a forward stepwise method on 44 financial ratios of the financial statements of CSMAR related to solvency,ratio structure,earning capacity analysis,industrial economic sector,operating capacity,cash flow,and growth rates.Finally,there are 24 financial indicators selected to fit the models in order to reduce the prediction error and get more reliable results.The empirical results reveal that random forests and boosting models give the best predictive performance;they present the smallest classification error and outperform in the evaluation measurements of AUC(ROC curve)and recall-precision plot.Additionally,we analyze the relative importance of the variables according to the random forests model,which is the method that outperforms the other models.This technique obtained that the three most important financial indicators for predicting early exit of Chinese companies are ROA(return on assets);EBITDA to total liabilities;and net profit margin on operation.