Research On Credit Risk Of Online Loans In The Account Management Cycle

In recent years,with the vigorous development of financial technology and digital finance,people’s demands for financing and raising cash through the Internet have also increased.As an important financing tool,online loans play an important role in the Internet financial business and lending market.Compared with commercial bank loans,online loans have a series of advantages such as high efficiency,low transaction costs,fewer borrowing barriers,and more accessible.However,online loans are usually unsecured credit loans,which lead to a higher credit risk.Once default occurs,investors have limited ways to recover defaulted loans.Therefore,early warning and precautionary measures are main ways for online loans to control credit risk,collection and recovery actions are just taken when loans are defaulted.It is required that investors screen high-risk borrowers in advance during the application processing and reject their loan applications.In the account management,investors can predict the status of the account using the borrower’s repayment behavior and pay attention to warning signals of the potential risk.In the collection and recovery process,investors need to accurately estimate the loss as well as formulate and adopt effective collection and recovery strategies.In order to facilitate the steady development of online loans and provide a comprehensive and systematic theoretical reference for the prediction,control,and management of credit risk,this thesis takes the fixed-term online loan as an example and then systematically studies the credit risk at stages from the loan application to the account closure.It provides models and methods for evaluating the credit risk of online loans in the account management cycle.This thesis can provide investors with a wealth of credit risk models,and improve their expertise in risk management as well as reduce their investment losses.This thesis can also enrich the scoring models of online lending companies,credit bureaus and thirdparty scoring agencies,and improve the quality of their credit risk assessment services.Financial regulators may attain insights and obtain reference for making online lending supervision policies from this thesis.This thesis is divided into seven chapters.Chapter One is the introduction.This chapter puts forward research questions of this thesis on the basis of summarizing the advantages and disadvantages of the online loan business,and then introduces the research significance,research methods,research content,research ideas and contributions of this thesis.Chapter Two is a literature review.In this chapter,we first review the concept and content of credit risk in the literature to provide a basis for defining the credit risk of online loans.Then we summarize the credit risk model used in consumer loans at each stage in the account management cycle.In fact,the online loan in this study is also a kind of consumer loans,and the credit risk modeling experience from other types of consumer loans can be used for reference.Finally,this chapter summarizes the research of online loan’s credit risk,clarifies the current research progress,and identifies the research gap.Chapter Three is regarding credit risk assessment in the account management cycle.This chapter analyzes the credit risks of this type of online loan at each stage in the account management cycle.Finally,it introduces widely used credit risk assessment methods and tools and elaborates the principles of credit scoring.Chapters Four,Five and Six are the core parts of this thesis,which study the credit risk in the process of loan application,account management as well as collections and recoveries respectively.Chapter Four is focused on the application processing.In this chapter,we use a competing risks proportional hazards model to develop an application score and a profit score,which are used to evaluate and predict the default and prepayment risks of a single online loan,as well as the return and profit of a loan portfolio.This chapter also provides a method for investors who lack professional knowledge and skills to determine the interest rate of online loans based on credit risk.Chapter Five is the credit risk model at the stage of account management.This chapter takes Markov chain to model the borrower’s repayment behavior,and analyzes the influencial factors of the account status,and further predicts the future status of individual loan and the credit risk of the loan portfolio.Chapter Six is about the credit risk model in the collections and recoveries.This chapter firstly studies methods to improve the predictive accuracy of loss given default(LGD)from the dimensions of variables and models.Based on the LGD prediction,the expected loss of the defaulted loan is then estimated.Finally,this chapter uses Markov decision process to study the formulation of optimal collection and recovery strategy.Chapter Seven presents the research conclusions and suggestions for the Chinese online loan market.The conclusions and implications can be useful to stakeholders such as investors,online lending companies,credit bureaus,third-party scoring agencies,and regulators.And suggestions and prospects for the future research on online loans are also provided.The main conclusions of this thesis are associated with application processing,account management as well as collections and recoveries.At the stage of application processing,the default and prepayment hazard rates are significantly different for online loans with different terms,so it is necessary to customize and develop scoring models for loans with different terms.The competing risks proportional hazards model shows that four types of information-loan characteristics,borrower characteristics,credit history and macroeconomic conditions-will have a significant impact on default risk and prepayment risk.Compared with analyzing the influencing factors of risks,this thesis pays more attention to the model’s predictive ability for risks.The "2+2" dimension evaluation system shows that the competing risks proportional hazards model is comparable to common classification methods-logistic regression,support vector machine and random forest-in predicting online loan’s default and prepayment risks.It significantly outperforms the traditional proportional hazards model and artificial neural networks,moreover,has a great advantage in development efficiency and saving costs.It also finds that the ability of competing risks proportional hazards model in predicting the risk of default is stronger than predicting the risk of early repayment,which may attribute to the lower correlation between the collected information and the risk of early repayment.The competing risks proportional hazards model can further be used to predict the profit of individual loan and loan portfolos,which cannot be performed by other common classification methods.So it is another advantage of the model.In order to meet the needs of institutional investors to maximize profits,this thesis further estimates and predicts the profits of a single loan and loan portfolio based on a competing risks proportional hazards model.In terms of predictive accuracy,the profit scoring model developed in this thesis performs well,and its performance on highprofit groups exceeds the widely used linear profit scoring model.After considering the risk of early repayment,the prediction of profitability is more accurate,stable and consistent,which also shows that the risk of early repayment should be considered seriously.Finally,this thesis provides an interest rate estimation model to assist investors who lack experience and expertise to determine the risk adjusted interest rates.By making a comparison,it is found that the interest rate provided in this thesis is similar to the interest rate from the online lending company in terms of the distribution,but this thesis has lowered the interest rate of some loans.At the account management stage,this thesis uses the Markov chain model to model,analyze and predict the state of the account and the borrower’s repayment behavior.Results show that macroeconomic conditions,months on book and outstanding balances have an important influence on the transition probability matrix.Regarding the macroeconomic conditions,loans in lower score bands are more likely to be affected by the economic recession and further become defaulted,while the transition probability of loans in high score bands is less affected by the economic recession.Concerning month on book and outstanding balance,although the two have a confounding effect on the transition probability matrix,we still find that the month on book has a positive effect on the probability of moving from each state into default until it reaches 15 th to 22 nd month on book,after that this effect becomes negative.The higher the ratio of the remaining outstanding balance to the total payables,the greater the transition probability of moving into default.When we predict the account status and default risk of loans and loan portfolios,cumulative logistic regression performs best in predicting the transition probability,while first-order and second-order Markov chains are better in predicting the total number of accounts in each state.At the stage of collections and recoveries,this thesis studies ways from the dimensions of variables and models to improve the predictive accuracy of the LGD.With regard to variables,the study finds that application scores and behavior scores used at the application processing stage and the account management stage respectively can improve the LGD predictive accuracy,and the improvement is still robust when we use different models and different metrics.This shows that we should consider the LGD prediction in the account management cycle,as well as try to discover new and useful information from other stages.Regarding models,the predictive performance of the LGD model is affected by the data set,performance metrics,and time of prediction.Another benefit of using application scores and behavior scores to predict LGD is that we can further predict expected losses at any time before default occurs.This thesis establishes a loss prediction model based on process of predicting LGD with scores,and selects the important time when the loan is overdue for the first time to evaluate predictive performance of the model.Results show that the model constructed in this thesis performs better than the strategy of using the historical average as the future predicted value.The main contributions of this thesis are as follows.First,this thesis systematically studies the credit risk of online loans,and provides models and methods for evaluating,predicting,and managing credit risks at the stages of application processing,account management as well as collections and recoveries.Compared with the research that only focuses on the credit risk of online loans at a certain stage,the research content and methods of this thesis are more comprehensive.Secondly,this thesis uses the competing risks proportional hazards model to study and predict the default risk and prepayment risk simultaneously at the application processing stage.The prepayment risk of online loans has not attracted enough attention yet.In addition,this thesis not only establishes an application scoring model,but also developed a profit scoring model which accounts for default and early repayment risks at this stage to meet investors’ diverse needs of controlling the credit risk and maximizing profits.Next,when studying the credit risk at the account management stage,this thesis analyzes the dynamic factors that affect the borrower’s repayment behavior and account status,and further predicts the credit risk of a loan portfolio.Finally,when studying the credit risk in the collections and recoveries,the contribution of this thesis is that we use the application score and the behavior score to improve the LGD predictive accuracy and further construct the loss prediction model.