Research On Two-stage Personal Credit Loss Rate Forecasting Based On Heterogeneous Ensemble Learning

In recent years,the credit economy has developed vigorously in my country and has become an important driving force for economic growth.Consumer credit and credit economy have also become a new direction for China's economic transformation and upgrading.However,behind the rapid development of the consumer credit market,there are also credit risks that restrict the healthy development of the industry,such as high default rates.Credit risk is the most important risk faced by the credit industry.For banks and other credit institutions,there are frequent defaults due to insufficient credit risk management capabilities.It is urgent to build a scientific,efficient and accurate credit risk evaluation model.At present,academic research on credit risk assessment models is widely focused on the probability of default prediction model,which is a two-class model constructed to classify whether the borrower defaults or not.However,in the actual credit risk management problem,the default loss is also an equally important issue.Evaluation factors,this is because the borrower will cause different levels of loan losses to the lending institution due to the different credit levels in the event of default.Therefore,when evaluating the credit risk status of borrowers,it is necessary to comprehensively consider factors such as default probability and default loss.To make up for this lack of research,this paper constructs a two-stage credit loss rate prediction model based on heterogeneous ensemble algorithms,which can more completely characterize and evaluate the credit of borrowers,thereby reducing the risk cost of credit institutions and promoting the health of the credit industry development of.The main work of this paper is summarized as follows.First of all,in order to conduct a comprehensive assessment of customer credit risk,this article systematically sorts out the three key credit risk parameters of the "Basel Capital Accord III" : Probability of Default(PD),Exposure at Default(EAD),and Loss Given Default(LGD)On the basis of LR,from the perspective of credit loss research,a new concept of credit loss rate(LR)is derived and constructed,and the corresponding calculation formula is given;secondly,on the basis of the concept of LR,this paper is based on the heterogeneous ensemble algorithm construction A two-stage personal credit loss rate prediction model is developed.In the first stage,in order to distinguish the default status of credit customers,this paper constructs a heterogeneous ensemble classification model of default probability based on the stacking algorithm,so that all credit customers can be accurately classified into two types of default and non-default.In the second stage,for the customers who are predicted to be in default in the first stage,this paper builds a heterogeneous ensemble regression model based on the stacking algorithm to predict the EADR and LGD of the default customers.After obtaining the three key predictive values of PD,EADR,and LGD,the credit loss rate indicators are calculated according to the customer's default status.Finally,in order to verify the validity of the two-stage credit loss rate prediction model constructed in this paper,this paper uses a real credit sample of a domestic financial institution to conduct an empirical test.The empirical results show that the default judgment ability,cost control ability and loss prediction ability of the method proposed in this paper are significantly better than the traditional model,which proves that the two-stage credit loss rate prediction model based on the heterogeneous ensemble algorithm can be more comprehensive and accurate.Evaluating and predicting the credit risk of credit customers has strong practicability,innovation and scalability.In response to the empirical research results of this article,this article puts forward several countermeasures and suggestions to credit institutions,regulatory authorities and other relevant parties from the perspectives of credit risk assessment dimensions,credit risk prediction models,and credit risk supervision.First,when evaluating the credit risk of borrowers,credit institutions should add credit risk measurement indicators in the dimension of credit loss to comprehensively evaluate the credit risk status of borrowers;second,they should consider when constructing a full-element credit risk evaluation model for borrowers Adopt more accurate and efficient algorithm models such as heterogeneous integrated algorithms;third,in the current era of rapid environmental changes,regulators should improve their regulatory standards,and use modern financial technology-related methods to strengthen credit risk identification,early warning,and monitoring of banks and other credit institutions.Supervise and promote the healthy and sustainable development of the credit industry.