Directed Interaction Tests for Time-Series Analysis Based on VAR Model

Abstract

Exploring directed influence relationships at different temporal and spatial scales is an important issue in time-series research. This paper develops a method for testing the directed interactions of multivariable time-series with a vector autoregressive (VAR) model. The calculation of Granger causality between the reference time-series and the other time-series is not rely on a priori specification of a model for pre-selected time-series, but aims at testing or contrasting specific hypotheses about time-series interactions. The measurement error interference on parameter estimates were evaluated by using VAR modeling, and then Granger causality relationships of time-series were detected in computational simulations. The simulation results demonstrate that the proposed method has a satisfactory performance on analyze directed interactions, when its applicability and usefulness are tested using multiple units of time-series.