Application Of Survival Analysis In Personal Credit Assessment

Since the outbreak of COVID-19,downward pressure on the global economy has intensified,and international disputes have become prominent.Under the turbulent situation,the ability of financial market participants to fulfil their commitments will be significantly reduced,which puts forward higher requirements for predicting and preventing credit risks.How to use big data tools to mine the credit information of capital demanders,effectively prevent and control credit risks and regulate asset pricing is of great significance for maintaining financial stability and promoting the smooth operation of the economy.According to the characteristics of capital inflow,financial institutions not only face the risk of loss of principal and interest caused by default but also may suffer the loss of interest caused by prepayment.Estimating default and prepayment times gives financial institutions the flexibility to value loans and to allocate assets based on default levels over time.According to the characteristics of credit data,this paper innovatively constructs the actual survival time of default and prepayment and,through the theory of competitive risk in survival analysis,distinguishes two types of credit risks,default and prepayment,and studies the risk of default and prepayment respectively.As it is challenging to obtain domestic personal credit data,the research used the loan data with a loan term of 36 months in Lending Club for analysis.After cleaning the data,the continuous variables were sorted into chi-square boxes,and the WOE and IV boxes were calculated.Variables with IV value greater than 0.02 were selected,and redundant variables were eliminated.Finally,we obtained 403963 observations and 21 characteristic variables.In this paper,the Deep Recurrent Survival Analysis(DRSA)model is used for the first time to predict the risk event probability of personal credit.The credit risk is modelled in the time window of 0-12 months,13-24 months and 25-36 months.The results were compared with Logistic,Cox and Mixed Cure Model.The results of the model classification show that the prediction ability of the DRSA model is much higher than that of the other three models,indicating that the DRSA model is effective in credit risk assessment.Compared with prepayment,the accuracy of default prediction is higher,reaching 97.4%,98.8% and 99.8%respectively in the three-time windows.In order to solve the impact of class imbalance on the model’s classification accuracy,we used the cost-sensitive learning method to improve the model.The final results show that this method can significantly improve the classification ability of the model.