A Prediction Research On Borrowers' Default Activity In Lending Through Internet Platform Based On Ordinal Logistic Model

With the development of modern information technology,mobile payment,social networking and cloud computing which are based on the Internet will have a profound influence on traditional financial pattern.P2P platform which is based on Internet has transformed traditional financial mode especially loan mode.According to the statistics of an internet called home of internet loan,by the end of March 2017,the number of net lending platforms has reached 5888.The historical cumulative volume has reached 4.105269 trillion RMB.In March,transaction volume has reached 250.843 billion RMB.There are 4.2 million investors and 2.4 million issuers being involved in nearly 4600 online net lending platforms.With the expansion of Internet financial business,credit default risk of P2P platform is also gradually exposed.A total of 3611 online platforms has experienced run away,shut down,withdrawal difficulties,and economic intervention.This caused huge losses to property investors.Therefore,effective identification of the default risk is a solid guarantee to increase profit of investors and maintain the benign development of the Internet financial loan business.This thesis is based on the perspective of maximizing investors' benefit in net platform loan.It focuses the differences on customer's payment behavior.Based on dynamic performance of customers' monthly payment,this thesis reversely analyzes the state transition matrix of markov chain to construct credit state transition equation.This equation can lead to the default probability of classified customers.It then classifies its customers into "default" and "non-default" groups,and it further classifies its customers into "early settlement","currently normal" "suspicious" and"loss" categories.It uses ordinal category logistic model and ROC inspection to explore the relationship between the customer characteristics and the dynamic repayment behavior.The empirical results can come to the following conclusions.(1)Age,financial condition,education level,product type,position and borrowing type are important indicators to predict default.These indicators have a larger degree of differentiation.Age and other variances have the sudden change of default risk.(2)Social relation variables have the strong binding force.Social relationship can reflect the authenticity of customers' borrowing and has a major influence on defining customer credit risks.(3)With the increase of level,the default risk would not decrease for some variables which have clear hierarchy and class characteristics.When the education degree or the member' s hierarchy reach certain level,the default risk does not have a clear differentiation.(4)With reference of the state transition matrix,it is much easier to reversely conclude the possibility of the default risk of four types of customers.This can avoid defining the critical point of default probability and make the prediction result more carefully.Ordinal logistic model has higher AUC value and lower rate of false rate.