系统性风险实时预警与监控研究：基于信息流的动态多级网络建模

In the latest years, there become intensive campaigns for the financial regulators in various countries to guard and defuse the financial systemic risk (SR). However, due to the unavailability of full data stemming from unaffordable regulation cost and complexity of risk channels resulting from financial innovation, it still confronts big challenge to practice realtime warning and monitoring toward SR. The reconstruction of financial network based on specifying information flows is a novel candidate to break the bottleneck, as it could be a feasible way to overcome the stalemate of data unavailability meanwhile permit one to establish the risk dependency network for even non-bank institutions. Therefore, this project takes such methodology as a breakthrough point. In the beginning, two different kind of reconstruction methods will be explored and extended. One of them is to reconstruction the potential connections between institutions based on information flow measures, which helps to dissect the realtime warning for SR. Whereas the other one is to reconstruction the real transaction links between institutions based on information flows configurations, which serves the monitoring design for SR. Then, by means of those methods, this project advances to diagnose SR from the perspective of dynamic multilayered reconstructed network. On the one hand, we will attempt to develop approaches to detect the decoupling states within multilayered network in terms of information inflow, that is especially devised for warning emerging susceptibility of near-collapsed financial markets. On the other hand, we will explore the monitoring procedure for three vital contributors to SR, i.e., level of leverage, degree of investment overlaps and extent of obligation diversity those, with simultaneously coordinating the major risk contagion channels including default cascades, fire-sale spillover effects, etc., in different network layers and considering most financial sub-systems such as banks, funds, real estates and even shadow banks. Finally, a real-time warning and monitoring system being both data driven and mechanism examined for SR will be established through integration and optimization. This project is of positive significance for reexploring characteristics of SR and supplementing the method system of warning and monitoring for SR.

当前，防范和化解系统性风险(SR)已成为各国金融监管的焦点工作。然而，囿于高额监管成本，完整的关系数据很难获得，加之金融创新增加了风险渠道的多样性，使得SR的实时预警和监控依然充满挑战。基于信息流建模的金融网络重构法提供了一种新思路，有望突破数据可得性的限制，并允许还原非银行机构间风险依存网络。本项目将首先研究基于信息流测度的潜在网络重构和基于信息流约束的真实网络重构两种方法。进而在动态多级网络联动的视角下展开SR的预警监控研究。一方面开发针对多级网络间解耦状态的信息流探测方法，对系统濒临瓦解时涌现的不稳定性进行预警；另一方面，通过联立银行、基金、地产、影子银行多个子系统，联动违约级联、抛售外溢等多个网络层次的风险传染渠道，探索对杠杆水平、投资重叠、债务同质三大指标的监控。最后通过集成，建立数据驱动和机理挖掘并举的SR分析系统。本研究对于SR的特征挖掘和预警监控方法体系的补充具有积极意义。

基于信息流建模的金融网络重构法能够突破数据可得性限制，允许还原非银行机构间的风险依存关系。本项目围绕信息流测度和信息流约束下的金融市场主体间风险依存关系的重构网络，研究了系统性风险的预警与监控方法以及相关指标。具体来说，在信息流测度方面，项目基于限制性随机重组方法开发了“投机影响网络（SIN）”中的实质性影响关系识别方法，更好的服务于对行业和机构系统性风险险位的监控；同时，按照价格不稳定性的不同来源，研究了若干适用于信息流建模的资产风险特征，包括市场自验性同步预期反转效应、集体性偏差均值回复、市场内禀交易分歧，仔细研究了这些风险特征的建模方法、参数估计、统计检验，并探索了对应于系统性风险的若干监控指标和预警指标BBAR。在信息流约束建模方面，基于BiCM网络构型算法开发了单模投射下的“有效投资组合重叠网络（PON）”构建方法，并开辟了基于PON的关于抛售外溢、投资重叠、策略相似、流动性传染等原因的风险扩散效应分析方法，通过拓扑演化特征探索了一些列的风险监控指标，基于有效PON的动力学建模，推导并实证出一个理想的系统性风险预警拓扑指标：有效抛售临界感染率。 更近一步，PON的基础上叠加基于信息约束建模的基金持股二部重构网络进行冲击仿真的压力测试，在初步捕捉由于重叠资产抛售（贬值）造成的流动性风险传染效应的基础上，更深入的追踪风险扩散的动态自激（价格贬值-赎回冲击）正反馈效应，并挖掘出相关的监控指标，开发出新颖的预警指标 SRW。