On Wiener-Granger causality, information and canonical correlation

A simple procedure is suggested for testing Wiener-Granger causality in stationary multivariate time series, which is based on Geweke's measure for Wiener-Granger causality and canonical correlation coefficients. The procedure is direct in the sense that no auto-regressive (AR), or mixed autoregressive moving average (ARMA), models have to be estimated first, in this way avoiding potential estimation problems, such as multicollinearity, which may occur.