骗贷行为的特征、成因及影响——基于P2P网贷平台的实证分析

The growth of online peer-to-peer (P2P) lending provides a new financing channel for small enterprises and individuals. However, the widespread fraudulent loan requests pose serious threats to this market. Despite that some P2P lending platforms has developed some functions to tackle this problem, due to a lack of theoretical guidance, the effectiveness of such functions are not satisfactory. Such phenomenon is raising concerns among academia and it has become a new research direction of Fintech. Under this background, we aim to systematically study this problem. First, we use big data analytics to identify features of fraudulent loan requests from the massive, variety types of data, especially from the unstructured data, with which we develop the fraud detection model. Second, by examining the impacts of inter-person transmission factor and social network factor on the occurrence of loan fraudulence, we develop a model to explain why borrowers tend to deceive. Then, we study how loan fraudulence and anti-fraud functionality of lending platforms influence investors’ recognitions and decisions, with which we develop a consequence model. Finally, based on these outputs, we propose managerial implications. The fruits of this project may help extend ways to identify features of deception, enrich theories regarding the causes and consequences of deception, as well as enhance fraud detection and risk management capabilities. As similar fraud problems may also exist in other areas such as in electronic commerce and financial markets, the research outputs of this project may also be valuable for the mechanism design of these markets.

P2P网贷为小微企业及个人融资提供了新渠道，然而逐渐泛滥的欺诈现象却对该市场的健康发展构成了严重威胁。虽然部分网贷平台开始构建反欺诈机制，但是由于缺乏理论指引，效果并不理想。P2P欺诈问题日益引起学界关注，成为互联网金融学术研究最新国际趋势。在此背景下，本项目将系统研究P2P网贷市场中的骗贷行为。首先，利用大数据分析技术从海量数据尤其是非结构化数据中提炼欺诈特征，构建欺诈识别模型；然后研究人际传导和社会网络对骗贷行为的影响，厘清骗贷行为的形成机理；再次研究骗贷行为和机制设计对投资者认知和决策的影响，构建骗贷行为影响模型；最后在上述研究的基础上提出管理建议。相关研究成果将拓展欺诈识别的思路和方法，丰富欺诈成因和影响理论，提升网贷平台欺诈识别能力和风控水平。由于类似欺诈问题在其它电子商务活动和投融资市场中亦可能存在，研究结果对这些市场的机制设计也具有一定的参考价值。

本项目研究借款欺诈的特征及内在机理，包括欺诈检测识别、投资者决策以及欺诈行为模式三个方面。通过本项目研究，我们提出了一个基于欺诈三角理论的欺诈识别模型，该模型基于网贷平台的黑名单数据、基础交易数据以及从网络中提取的非结构化数据，综合利用传统机器学习算法以及深度学习算法，构建欺诈识别模型。该模型能够较准确地对欺诈借款进行识别，得到良好效果。在投资者决策方面，我们基于借款请求特征、借款人特征、市场信号等因素构建了投资决策模型，探讨上述因素对投资者欺诈认知的影响。在欺诈行为模式分析方面，我们分析了地域因素和社会网络因素与骗贷发生率之间的关系。上述研究结果一定程度上丰富了互联网交易反欺诈理论体系，对于完善交易机制、识别并遏制交易欺诈有实践指导意义。