A Study On Non-performing Loans' LGD

So far,China's online lending platform has developed rapidly.After nearly fifteen years of development,there are more than 2000 network loan platform.The most obvious risk to these online lending platforms is credit risk.Basel II adopts a new credit risk management approach,the internal rating method,which relies on four key parameters: exposure at default,probability of default,loss given default and expiration date.Therefore,it is of great significance to study the loss given default of network lending to control credit risk and manage profit of the network loan platform.This paper chooses Lending Club,a large mature network loan company in the United States,established earlier,mature in development,rich in business experience and more standardized business model,as the research object,which is of reference to the development of the domestic network lending platform.Combined with existing theories and research,this paper will study the loss given default based on previous scholars' research on the online loans and other types of loan defaults.Although the resulting non-performing loans have been considered bad debts,but in reality the recovery of non-performing loans has not stopped,instead there are still borrowers after default to continue to repay,this repayment is also a considerable income for creditors.Therefore,it is of practical and economic significance to study the loss given default of non-performing loans.Since censored data occupies the collection process,we choose survival analysis models,which match well the property of data to be researched.After the characteristic engineering and data mining of the application information,the observation period of survival analysis is derived,and the macro-economic variables such as unemployment rate,government bond interest rate,stock market index are added in,and key business variables like debt settlement or not,whether to join the subsidy plan are also considered to model recovery rate.After the modeling process,we find that the survival analysis models perform better than the liner regression model,and the stratified Cox model performs better than the single Cox model.Meanwhile,when the two types of control variables(debt settlememt and hardship)are added in,the concordance index grows and the standard error decreases,in relation to the single Cox model.Through this study,we surprisedly find that creditors can enhance their ability to predict loss given default through releasing limits in ways to repay and actively collecting debtors' repayment data.In this way,it gets easier for financial institutions or debt companies to predict loss given default and better control credit risk.