Research On P2P Credit Default Risk Prediction And Investment Decision

Peer-to-Peer(P2P)Lending refers to the method of debt financing that enables individuals to loan and lend money through an electronic trading platform.In March 2005,the world’s first online P2 P lending platform-Zopa-was established in London,England.Since then,P2 P lending has spread rapidly to other regions include the Americas and Asia.After over a decade of sustained development,the business model of P2 P lending has gradually stabilized globally and related services are gradually accepted by the public.The emerging of P2 P lending in China is relatively early in 2007.However,the development of P2 P lending in China crawls with twists and turns.After a period of barbaric growth,it has entered the adjustment period since 2014.In 2018,it has ushered in the concentrated exposure of risks,and various business indicators have undergone a considerable degree of correction.The problems encountered in the development of China’s P2 P lending business are mainly related to factors such as immature financial markets,relatively unregulated supervision,and lack of integrated financial services.However,judging from the actual situation of the development of current financial market in China,P2 P lending will play a vital role in China’s financial market.In order to further promote the healthy and stable development of China’s P2 P lending business,reduce information asymmetry,improve the accuracy of loan default prediction,and provide investors with investment decision-making assistance to help them effectively prevent risks,it has great practical significance.Therefore,the purpose of the thesis aims to provide individual investors with methods and tools to predict P2 P lending default accurately,help investors to prevent risks more effectively,and thereby improving the efficiency and profitability of their investment decisions.Since the emergence of P2 P lending business in 2005,related research has further developed rapidly.In particular,the United States scholars are doing leading-edge research in this field.Since 2012,with the rapid development of P2 P lending business,research on P2 P lending has also increased rapidly in China.At present,the topic of information utilization,loan characteristics and participant behavior analysis,information asymmetry and risk prevention become the mainstream in the field of P2 P lending research.In terms of information utilization,in addition to using general information(such as loan information,borrower’s credit and financial information),more studies focus on how to better explore and utilize social information such as borrower’s gender,race,friends and people,and even the information that hidden in texts and photos.As transaction data is relatively transparent in P2 P lending,quantitative research techniques and methods such as statistics,machine learning,mathematical and physical analysis have been extensively used.In particular,the research focus on topics such as loan characteristics,herding effect of investors’ behavior,credit evaluation of borrowers as well as loan default predict.Statistical research methods are mainly applied in the field of loan characteristics and investor behavior analysis,while machine learning research methods are widely used in the field of risk prevention.In the current research of P2 P lending,the research of loan default risk and investment decision support needs further explored.For instance,due to the failure to make use of investor-related information,the research conclusion is constrained by the information asymmetry between the borrowers and the lenders.Another limitation of previous literature is the absence of effective methods and tools to assist investors in their investment decision-making process.Furthermore,research methods and modeling require further improved as prior research mainly rely on a single model and a single method.Inspired by the theory of investment utility,modern portfolio theory and information entropy theory,the method of this thesis draws lessons from Term Frequency-Inverse Document Frequency(TF-IDF)algorithm in the field of information retrieval research.By constructing a loan default prediction model based on investor utility,and exploring the integration of different models and data sources,the thesis provides decision support to investors’ loan investment behavior in a form of portfolio investment.Firstly,a P2 P loan default prediction model based on investor utility is proposed.At present,most of the mainstream research on default prediction of P2 P loan relies on statistical or machine learning methods,training and learning of various kinds of information provided by borrowers.The prediction’s effect is constrained by the information asymmetry between borrowers and lenders.Recent studies have tried to make prediction through investor’s information,however,the differences among investors in risk preference,investment utility,investment ability and information they hold are not taken into account.The P2 P loan forecasting model based on investor utility draws lessons from the utility theory of economics,it constructs the investor’s investment income utility baseline,and uses the loan investor’s investment historical return rate,bid amount,interest rate of bid and other information to measure the loan default probability from the investor’s perspective.On this basis,aiming at accurately reflect the relationship between investors and borrowers,this paper incorporates factors such as the total amount of investors’ historical investment and the distribution of loans’ investment into the model,uses the TD-IDF algorithm in the field of information retrieval for reference,constructs the proportion factor of investment reverse investment for investors.Built on an empirical study of actual P2 P loan business data,it is found that the default prediction model based on investor utility performs well.Compare to the commonly used methods such as logistic regression and support vector machine,the accuracy of default prediction is improved by nearly 6% on average and it remains a stable performance across different scales of test data sets.Secondly,this paper designs a set of integrated learning training method(Training Algorithm Base on AUC and Q statistics,TABAQ).Taking AUC and Q statistics as the metrics for the accuracy and difference of the prediction model(learner),the proposed method systematically guides the whole process of construction,optimization and testing of integrated learning,so as to achieve integration of different learners,as well as the fusion of different sources of information on the borrower and the investor.Based on an empirical research,it can be found that TABAQ method can effectively guide the integrated learning process to get a better result learner.Integrating different learners and merging diverse information sources is helpful to reduce one kind of prediction errors.However,neither side of the model nor data source plays a significant role in further improving prediction accuracy.Only by integrating the two together,can the prediction accuracy be improved and type-I error be reduced simultaneously.Therefore,the method of ‘double integration’ between model and data is adopted.When the predicted AUC value is steadily improved,the number of forecast errors further reduced by 4.85%.The difference in data sources has a greater impact on the difference between learners than on models.Finally,an assistant decision-making method for P2 P loan investment is proposed.The loan expected return rate,loan yield variance,and loan default risk entropy are used as the measurement and evaluation indicators of return and risk of loan assets.This paper constructs loan portfolio by a single factor and double factor methods respectively,and provides investment strategy suggestions to investors as well.Empirical results show that the return variance and default risk entropy index can maintain the stability of investment returns,while the return of portfolio increases by about 4%-5% as default risk entropy index takes into account certain profitability at the same time.The expected rate of return indicator has also achieved the average rate of return has increased by nearly 10%.However,when the number of invested loans is relatively small,the yield of the portfolio has fluctuated.This research attempts to optimize the portfolio of single-factor metrics by the mean-variance method,but the results are not significant and should be limited by the assumption that there is no correlation between loans.By summarizing the results of above three parts,three conclusions can be attained.Firstly,information of investors in P2 P lending is indeed more credible than that of borrowers.Introducing more,newer and more effective information is conducive to breaking the restriction of information asymmetry and improving the accuracy of loan default prediction.Secondly,integrating different models through integrated learning and integrating diverse data is helpful to further improve the accuracy of loan default prediction.In this process,data plays a greater role than model.Thirdly,by means of building a portfolio of assets,the advantages of diversified investment and risk reduction can be fully utilized,and the balance between income and risk can be realized to a certain extent.From practical perspective,it brings significance and operability to guide investors to make more effective P2 P loan investment decisions.There are still many areas to be improved in the research of this paper.For instance,influence of investors’ continuous learning on investment ability is insufficient,the correlation between loans is not fully considered,and the lack of effective measurement methods for influences between various data and models.Further research can be carried out in these areas.