Mixed causality graphs for continuous-time stationary processes
share
In this paper, we introduce different concepts of Granger non-causality and contemporaneous uncorrelation for stationary continuous-time processes to model the different dependencies between the component series of multivariate time series models. Several equivalent characterisations for the different definitions are given, in particular by linear projections. We then define two mixed graphs based on different definitions of causality and contemporaneous uncorrelation, the (mixed) causality graph and the local (mixed) causality graph, to visualise and to analyse the different dependencies in stationary continuous-time processes. In these graphs, the components of the process are represented by vertices, directed edges between the vertices indicate causal influences and undirected edges indicate contemporaneous uncorrelation between the component processes. Further, we introduce various notions of causal Markov properties in analogy to Eichler (2012), which relate the different dependence structures of subprocesses, and we derive sufficient criteria for the (local) causality graph to satisfy them. Finally, as an example, the popular multivariate continuous-time AR (MCAR) processes satisfy our assumptions. For MCAR processes we show that the (local) causality graphs can be characterised explicitly by the model parameters.
READ FULL TEXT