P2P網絡借貸平臺風險預測模型研究

Risk Forecast Model for Online P2P Lending Platform

Student thesis: Doctoral Thesis

View graph of relations

Author(s)

  • Daoshun ZHANG

Related Research Unit(s)

Detail(s)

Awarding Institution
Supervisors/Advisors
Award date21 Jul 2016

Abstract

近年來,P2P網絡借貸發展極爲迅速,這使得對借款人的風險評估和預測變得尤爲重要。儘管以往的研究從“硬信息”(即款項信息、借款人自身特點等)和“軟信息”(社會關係、社會網絡資本等)等多方面利用計量經濟學方法來評估P2P網絡借貸借款人的風險,但專門針對學生貸款平臺,挖掘影響學生風險因素的研究相對缺乏,研究逾期行爲的研究更少。本研究針對學生P2P借款平臺,利用實際二手數據,從款項層面和單筆還款層面分別考察逾期和違約兩個階段的風險影響因素,基于决策樹分析的方法提出風險預測模型,幷用另一批實際數據對模型進行檢驗。同時,我們利用梯度提升决策分析樹和計量模型對預測效果進行進一步驗證。最後,基于模型結果我們分別計算了本文所提出的風險預測模型所帶來的平臺資本損益。我們發現許多區別于傳統文獻研究的變量在風險預測方面發揮著重要作用。例如,學生院校和家鄉城市的距離、學生院校城市和家鄉城市的經濟水平差距、學生同父母的關係、學生的學歷、住房條件、性別等因素都會直接影響到學生的逾期和壞賬風險。利用我們所提出的風險預測模型能有效提升P2P借貸平臺的風險評估水平,幷帶來很大的正向的資本收益。本研究在研究對象(“拍來貸”學生P2P網絡借貸平臺)和結合學生背景的影響因素、逾期和違約兩階段分析以及利用數據挖掘的方法論等方面具有較强的理論貢獻。此外,本研究在用戶風險預測和信用評估、P2P借貸平臺損益分析和運營策略等方面具有很强的實踐意義。
In recent years, online P2P lending grows remarkably, which highlights the significance of risk evaluation and risk forecast of borrowers. Although past studies on risk evaluation in online P2P lending platform were from many aspects of hard information (i.e. loan information, borrower characteristics etc.) and soft information (i.e. social relationship, social network capital etc.) with econometric method, few research has paid attention to mining influential factors of risk regarding to student lending platform, and study considering delinquency behavior is rare. Focusing on student P2P lending platform and using actual second-hand data, this study investigated the influential factors of risk on borrowers’ delinquency and default behaviors on the level of loan and level of monthly payment respectively. We proposed the risk forecast model with decision tree analysis, and adopted another real dataset to test the forecast model. Meanwhile, we used gradient boost decision tree (GBDT) and econometric model to further validate the forecast model. Finally, based on the proposed forecast model, we conducted capital gain and loss analysis for online P2P lending platform. We found that many factors that are different from past studies were important on student borrower risk prediction. For example, the distance between school (university or college) and hometown, the economic gap between school (university or college) and hometownt, the relationship between student borrowers and their parents, housing condition, education level, gender will have direct impact on student risk (delinquency or default behaviors). The proposed forecast model will effectively improve the risk evaluation of online P2P lending platform, and bring large positive capital revenue. This study has several theoretical contributions on research object (i.e. student online P2P lending) and student-related risk factors, two-stage risk forecast on borrowers' delinquency and default behaviors, as well as adoption of data mining methodology. Moreover, the study has strong implications for online P2P lending platform on borrower risk prediction and credit assessment, P2P lending cost-benefit analysis and platform operation.