基于多维度大数据的住房抵押贷款风险管理决策支持研究

The total outstanding volume of housing mortgage loan balance in mainland China has kept increasing rapidly during the past decade, while the sector of mortgage-backed securities (MBS) also started to surge dramatically recently. It is against such background that the research project proposed, with the cooperation with several leading data companies and financial institutes, focuses on the application of big data analysis to improving financial institutes’ management on housing mortgages’ default risk in China. Multiple big data sources will be introduced and analyzed in this project, such as application and repayment of housing mortgages, listing and transaction of housing units, cell phones usage records, information from media and we-media, and remote sensing data. Based on both such big data analysis and the theoretical findings from previous real estate economics and finance research, several “big data analysis modules” will be designed in order to achieve in a quantitative estimate on the current and future risk conditions of both the housing market where the collateral housing unit locates and the borrower household. With these estimates as key inputs, a mortgage default risk model will then be developed, which can be expected to quantitatively evaluate and forecast the default risk of any specific housing mortgage at a higher accuracy level and a higher frequency, but with a lower cost. As the most important deliverable outcome, several packages will finally be developed based on this model to support financial institutes’ risk management practices, including the censoring and approval of mortgage applications, the continuous monitoring of outstanding mortgages, and the design and pricing of MBS.

本项目在当前我国住房抵押贷款规模持续高速扩张和住房抵押贷款支持证券（MBS）快速兴起的双重背景下，研究大数据分析对金融机构在住房抵押贷款一级和二级市场中的违约风险管理决策的支持作用。项目拟通过科研机构、数据企业和金融机构的协作，将住房抵押贷款申请与还款、住房挂牌与交易、移动终端使用、媒体与自媒体资讯、空间遥感等多维度大数据资源和房地产经济学、金融学等领域理论研究成果相结合，研制若干大数据分析支持模块，重点在“住房市场波动状况”和“住房抵押贷款借款人行为”两个方向上实现对当前风险水平和未来变化趋势的定量分析判断；进而以此为输入项，研究设计具备更高分析精度、更高分析频率和更低应用成本的住房抵押贷款违约风险评价与预测模型。项目研究成果最终将固化为在贷前审核、贷后跟踪监测、MBS产品设计与定价等环节直接支持金融机构管理决策的若干工具包，推动金融机构住房抵押贷款违约风险管理的决策流程优化和水平提升。

本课题在当前我国住房抵押贷款规模持续高速扩张和住房市场转入“新常态”的背景下，研究大数据分析对金融机构住房抵押贷款违约风险管理决策的支持作用。课题将多维度大数据资源和房地产经济学、金融学等领域理论研究成果相结合，重点在“住房市场风险评价与预测”和“借款人特征识别与行为监测”两个方向上研发了一系列大数据支持模块。住房市场风险评价与预测方向上，针对中国住房市场当前存在的数据缺口，研发了住房挂牌价格指数、住房价格自动评估系统、微博舆情指数、开发企业风险评估系统等大数据模块，有条件在更丰富的维度、更大的范围和更高的频率上，更准确的反映住房市场风险状况的变化。借款人特征识别与行为监测方向上，一方面基于住房抵押贷款海量历史还款数据，改进现有的借款人逾期/违约概率“预测”模型，尤其是重点研究单位性质、借款人职务级别等具有中国特点的借款人特征的影响，另一方面深度挖掘住房公积金缴存记录、信用卡消费记录、股票账户投资记录等大数据资源在动态“监测”借款人异常状况方面的作用。综合利用上述两类大数据模块，课题以某大型国有商业银行总行和两个重点城市住房公积金中心作为典型应用场景，针对贷前审核和贷后跟踪监测两个核心的风险管理环节，设计形成住房抵押贷款风险管理工具包，推动了金融机构住房抵押贷款违约风险管理的决策流程优化和水平提升。本课题已累计发表论文20篇，包括在China Economic Review、Regional Science and Urban Economics等SSCI期刊发表论文8篇，另有十余篇论文正在审稿过程中；获得省部级科技进步三等奖1项和多项学术会议奖项；课题负责人于2017年入选教育部“长江学者奖励计划”青年学者。课题研究成果已经在多家重要金融机构投入实际应用，并对中央政府住房市场监测分析和调控决策发挥了直接的支持作用。