Research On P2P Investment Decision Considering Investor's Risk Preference

With the rapid development of Internet technology,the online lending mode of P2P network lending has sprung up rapidly.For borrowers,P2P network lending enlarges their financing channels,provides access to funds from investors around the world,and the loan rate is reasonable.For investors,P2P network lending platform loans,is a new type of financial products,the hands of surplus funds can be invested in the platform loans,gainning investment returns.This new Internet-based business model has great potential to meet the needs of many customers who are unable to obtain services from banks because they do not meet bank standards.However,in recent years,the default risk and low investor satisfaction of P2P network lending have endangered the development of the platform.Previous research on P2P investment decision-making either only focused on the loan itself,without considering the investor's risk preference,or considered the investor's risk preference,but the models used are not practical enough,and the final decision-making results often can not satisfy the decision-makers.In recent years,more and more studies show that portfolio theory has a strong practicability in P2P investment decision-making.The whole process of using portfolio theory to make P2P investment decision is to calculate the default probability of the loan first,then predict the return and risk of the loan on this basis,and bring it into the P2P portfolio model to find the optimal portfolio selection.On this basis,this paper uses machine learning technology to improve the default probability calculation part of the process,and considers the risk preference of investors when constructing the P2P portfolio model,so as to enhance investors'satisfaction with the P2P network lending investment.The research content includes three parts:(1)Establish a P2P network loan default probability prediction model based on RFLR(Random Forest model-Logistic Regression model).This paper analyzes the default factors that affect the loan,explains the RF selection characteristics and the principle of LR computing the default probability.Get data from P2P platform Lending Club official website and preprocess the data.The optimal feature subset is selected by RF model and combined with the target variable to form a new data set.The optimal LR model is trained by grid search and cross validation.The default probability of the loan is calculated by this model.The experiment found that there were 24 important characteristics,among which loan interest rate,income-liability ratio,the earliest time of credit record,monthly repayment and annual income were the five most important characteristics.These characteristics are the most effective against future default of loans.Investors and P2P network lending platform should pay attention to these indicators when borrowers apply for loans.(2)Build an instance based P2P network lending return and risk prediction model.The idea and process of the instance model is introduced,and the principle of determining the weight parameter by KR(Kernel Regression)model is introduced.On the basis of the previous part of the study,the loan samples are divided into historical loan samples with known return and new loans with predicted return and risks.The optimal width of K/R model is obtained by using a cross validation and minimization error method.The KR model with the input optimal width is used to calculate the relative weight between the historical loan and the new loan.On this basis,the weighted average value of the historical loan return combined with the relative weight is taken as the forecast value of the new loan return,and the variance of the historical loan return is taken as the risk of the new loan return.(3)Construct a portfolio selection model with investors' risk preference.This paper introduces the traditional portfolio model,the concept of risk preference and its impact on investors' investment decisions.Starting from the reality,a portfolio model considering investors' risk preference is constructed as the P2P investment decision model in this paper,and numerical experiments are carried out to verify the model.The objective of the model is to subtract the greatest risk from the return,and to adjust the risk preference to reflect the trade-off between the two.The decision variables of the model are the proportion of investment and the 0-1 variable reflecting whether or not to invest.The constraints of the model include:the return of all loans invested can not be lower than the minimum expected return of investors;the investment amount of an investor once investing in a loan is greater than the minimum investment amount of each loan required by the P2P platform;the investment amount of an investor once investing in a loan is smaller than the amount of the loan itself;and the investment amount of the investor once investing in a loan is smaller than the amount of the loan itself;he sum of capital ratios is 1.After the completion of the model,numerical experiments are carried out.Because this is a mixed integer programming model,this paper uses Cvx package in Matlab to call Gurobi to solve it.In the course of the experiment,the new loan with known return and risk is taken as the product of the investor's investment.Under different risk preferences,the new loan is brought into the constructed model to solve the optimal portfolio.The practicability and feasibility of the model are verified by comparing the results of investment decision and the characteristics of the corresponding risk preference investors.It is found that different investors with different risk preferences have different investment decisions when facing the same investment products.By analyzing the investment portfolio,we find that the distribution of loans matches the degree of risk preference of the corresponding investors,which proves the rationality and scientificity of the P2P network lending investment decision-making model considering the risk preference of the investors.