结构向量自回归模型因果图的信息论辨识方法

Identification Structural Vector Autoregressive Causal Graphical Models by Information Theory Criteria

由观测数据确定变量间的因果关系是时间序列分析的重要内容.本文利用图模型方法研究结构向量自回归模型变量间的因果关系,通过时间序列因果图的建立将问题转化为时间序列因果图结构的辨识.基于信息论方法提出了因果性定向的三步准则,利用关联积分估计互信息和条件互信息.模拟结果显示本方法能更有效地辨识结构向量自回归模型因果图的因果结构.

Detecting the causal relationships among variables is an important content of time series analysis.In this paper,the causal relationships among variables of structural vector autoregressive model are studied using graphical models,time series causal graph is presented and the structural identification problem of the causal graph is investigated.A three-step procedure is developed to orient the causal direction based on the information theory criteria.The mutual informations and the conditional mutual informations are estimated by the correlation integral.Numerical results demonstrate that the proposed method is able to identify the causal structure of causal graph of structural vector autoregressive model very effectively.