大数据驱动的中小微企业全息风险评估与介观调控机制研究

针对中小微企业借贷中因传统市场化风险定价障碍产生的“融资难”问题，本项目利用大数据支持的复杂网络方法和信息融合技术，在统计力学框架下建立企业成长的结构性耦合随机系统模型。通过对企业动态风险基因的全方位解析，深入研究介观尺度下中小微企业全息风险评估方法和成长性能调控机制，并以此指导金融机构的贷款决策，这是大数据金融创新在支持普惠金融服务方面的重要尝试。. 理论上，该研究有效克服了传统方法中由于“动态结构缺位”对企业宏观定向成长机制和非线性风险响应特征缺乏解读能力和直观描述的弊端，又避免了从微观层面表征个体随机动力学行为带来的方程求解和理论分析困难，尤其在处理企业复杂环境（非线性关联、时滞反馈、市场记忆、非周期驱动等）方面具有天然优势。同时，以HIGGS企业大数据平台为支撑的参数估计、样本统计、模型校验和实证分析，能有效解决传统研究中的对象局限性问题。

This project is proposed in view of the financing difficulty caused by the traditional market-oriented risk pricing of loans for small and medium-sized enterprises (SMEs). Based on complex network method and information fusion technology supported by big data, the coupled stochastic dynamical system for SME's growth is established by regarding the mesoscopic structure as a whole under the framework of statistical mechanics, and then the enterprise risk genes can be fully analyzed. Furthermore, the holographic risk assessment method and mesoscopic performance control mechanism are studied in depth. This method can effectively guide financial institutions in the decision of SME loans, and it is also an important attempt of financial innovation in supporting inclusive financial services.. In theory, based on the studies of this project, it is avoided that the classical statistical methods can not effectively explain or intuitively describe the generation mechanism of some macro growth behaviors because of not considering the dynamical structure. Based on complex system theory, the individual behavior can be described from micro level, but it brings the new problems of equations solution and theoretical analysis. However, the method proposed in this project performs better in these aspects, especially for the enterprises in complex environments, such as nonlinear relationship, time delayed feedback, market memory, aperiodic driving effect, etc. Furthermore, we employs the enterprise big data platform HIGGS, based on which we can effectively complete the parameter estimation, system simulation and large sample analysis to avoid the traditional problems, including limited samples, model validity test, and empirical analysis.