| With the increment of loan scale of commercial banks and the slowdown of China’s economy,especially the impact of COVID-19 in 2020,the scale of non-performing assets of commercial banks has been increasing.The commercial banks in China has been frustrate due to non-performing loans,so the reasonable disposal of non-performing loans has become a matter.So we need to do a good job in the pricing of non-performing assets.Using machine learning method to build non-performing assets pricing model has become a new idea of nonperforming assets disposal.With the development of machine learning in recent years,this paper takes the pricing demand of personal non-performing loans of Bank C,and deal the data of non-performing personal loans of Bank C in recent five years.From customer information,risk information,account information,credit information and contract information of non-performing personal loans,through extract and handle data,multi-dimensional characteristic data samples are constructed,and use XGBoost,LightGBM,and CatBoost model algorithms,build a nonperforming assets pricing model based on the recovery prediction.The model test results show that all model algorithms are good,the difference is small,the CatBoost model prediction ability is slightly better.At the same time,this paper puts forward the following suggestions for commercial banks to do better work for non-performing assets:first,build more useful the data pool of nonperforming assets and strengthen the basic information data of C bank;second,combine with the new sample data and recycling situation,continuously improve the pricing model,improve the work efficiency;third,deeply excavate the data to find the information of non-performing customers;The fourth is to expand the disposal channels of non-performing assets and enhance the value of assets;the fifth is to cooperate with the whole bank,move forward the risk and share information to prevent credit risk.This paper is an attempt to use machine learning method in the valuation and pricing of non-performing assets.I believed that with the in-depth study of non-performing asset pricing by more scholars in the future,there will be better characteristic data and algorithms applied to the non-performing asset pricing model. |
