金融系统演化与危机量化预警研究——基于时滞稳定相关网络

Research on Quantitative Early Warning of Crisis Based on Time Delay Stability Correlation Network

金融体系中,某一个机构出现危机会在整个体系上扩散,造成大范围影响,导致整个体系的崩溃。这种体系结构带来的系统性风险,是主流金融理论难以描述的。“金融复杂系统”理论作为一种新的研究范式,越来越受到研究者的关注。已有关于金融复杂系统中金融危机状态的研究通常根据金融网络拓扑结构变化对应金融危机或主观选取金融指标进行预警,较少采用复杂网络演化特征提取进行金融危机量化预警。围绕上述问题,文章提出短序列相关性分析新方法:时滞稳定性相关,实证分析2003—2018年10个世界重要股指收盘价序列。将股指序列按指定窗口长度滑动,从每一个被窗口覆盖的序列片断得出关系网络,并构成网络序列。从网络序列中提取状态分布、状态转化、状态关联,用于金融复杂系统动态演化和危机早期信号预警,并且进行有效性分析。结果表明,在各个金融危机和重大事件(特别是2008年9月的全球金融危机和2015年6月的中国股灾)之前,发现相应的网络序列波峰,并作为量化预警信号。

The crisis of an institution will spread over the whole financial system in a wide range and lead to the collapse of the whole system.The systemic risk from structure cannot be described by the mainstream financial theory.More and more attention focus on a new research paradigm named the theory of financial complex system.The existing research on the theory of financial complex system usually uses the change of financial network topology to map the financial crisis or select subjective financial indicators for early warning and seldom uses the feature extraction of complex network evolution to carry out quantitative financial crisis early warning.To solve the above problems,a new method of short-sequence correlation analysis is proposed that is named"time delay stability correlation".By using 10 key stock indexes from 2004 to 2018,the stock index sequence is slide according to the specified window length and get the corresponding correlation networks.The consecutive financial networks are created into the network sequence.Extract the state distribution,state transformation and state correlation from the network sequence and study the dynamic evolution of the financial complex system.Based on the dynamic extracted regularities,trying to find the warning signal of the crisis and carry out the validity?analysis.Before the financial crisis and major events(especially the global financial crisis in September 2008 and China's crash in June 2015),corresponding peaks of curve are found and used as quantitative warning signals.