Financial Distress Prediction Of Listed Companies Based On Lasso-logistic And XGBoost

Under the background of economic globalization,the relationships among all the participators in the economic activities are closer and closer.Listed companies will have a greater probability of financial distress in this environment.Generally speaking,it will not suddenly happen that the listed companies are in financial difficulties.They must have a process from normal to worse.Only when the deterioration state continues and cannot be reversed will they fall into financial difficulties.Therefore,to a certain extent,it is foreseeable whether the listed companies will meet the financial distress.To the listed companies,accurately predicting the issue of financial distress can help managers find problems and take effective measures and financial policies to deal with them in advance.To investors and creditors,it helps to make rational decisions and adjust investment or credit strategies.To the regulatory authorities,they could recognize the bad condition early when something wrong with the listed company's operating,therefore,they could timely guide and play the role of supervision.So,establishing an accurate prediction model financial distress is very meaningful for stakeholders in all departments.This dissertation firstly reviews the literatures of previous research results on financial distress prediction.Then elaborates the theoretical basis of Lasso-logistic and XGBoost methods.It is studied on the basis of the definition on financial distress with 96 samples of ST and 2682 non-ST listed companies in China's 2018 A-share market.Also,this research combines the status of the listed companies in China,from 36 variables in 7 perspective of financial and non-financial factors which is,solvency,growth ability,cash flow,profitability,operational capability,shareholding structure and governance mechanism.Before the establishment of prediction model,the three-year index data of 2,778 listed companies in T-2,T-3 and T-4(i.e.2016,2015 and 2014)were pre-processed.Then,Lasso-logistic is used to build the financial distress prediction model based on the data of T-2,T-3 and T-4 years respectively.By adding the penalty term for the model parameters,screen variables and estimate parameters in the meantime,and compare the prediction effect against the full variable Logistic.XGBoost is used to establish the financial distress prediction models of T-2,T-3 and T-4 respectively to analyze the importance of variables and sort them.In addition,the random forest and GBDT algorithm were selected to predict the financial distress to compare with the above two methods.The results showed that in the variable importance ranking analysis,the most contributing variables are earnings per share,net cash flow per share from operating activities,and equity concentration,which are all included in the financial distress prediction models of T-2,T-3,and T-4 years.In the XGBoost financial distress prediction model,they are long-term predictors;while the working capital turnover rate is only significant in the T-3 model,which is a short-term discriminant.A comparative analysis of all the results reveals that the closer the time is to financial distress,the higher the accuracy of financial distress prediction.On the other hand,whatever in T-2,T-3 or T-4,Lasso-logistic is better than the full-variable Logistic.The prediction effect of XGBoost is obviously better than that of random forest,GBDT and Lasso-logistic.The success rate of XGBoost is 0.96 in T-2.After comprehensive comparison,XGBoost has great advantages in the identification of the minority of companies which meet the problem of financial distress,and it can also sort the variables by importance,which has certain practical significance.