基于动态因果结构推断的SVAR模型识别:算法和仿真

给定任一满足递归结构假设的SVAR模型,存在一个与其是相同数据生成过程的线性动态因果结构模型,且SVAR模型系数矩阵与动态因果结构之间存在特定的对应关系,故同期变量间因果结构推断可以为SVAR模型提供正确的识别条件.同时还证明同期变量与滞后变量间动态因果结构推断能为同期变量间因果结构的正确推断提供有用信息.据此,在PC算法基础上,构建了动态因果结构推断的具体算法,将基于同期变量因果结构推断的SVAR模型识别拓展到基于动态因果结构推断,从而使SVAR模型得以完全识别的情形得到有效拓展,并给出了该方法下SVAR模型识别的充要条件,这些结论均得到了Monte Carlo仿真结果的有力支持.

Identification of SVAR model based on the inference of dynamic causal structure: Algorithms and Monte Carlo simulation

For a SVAR model in which contemporaneous variables have recursive causal structure, we proved that, there exsists a linear dynamic causal model which is the same data generating process with the SVAR model. Therefore, causal inference of contemporaneous variables can be used to establish the identification conditions of SVAR model correctly. Meanwhile, the dynamic causal structure contain the information of causal structure of contemporaneous variables. Accordingly, the paper proposed an algorithm of dynamic causal inference, and then we extended the approach that identifying the SVAR model based on causal inference of contemporaneous variables to dynamic causal inference, which generalized the case that SVAR model can be identified completely. The paper has also given a necessary and sufficient condition for identifying SVAR model with the method.