Comparing How Python and R Estimate Granger-Causality in the Frequency Domain

This paper deals with the estimation of unconditional and conditional Granger-causality spectrum in the frequency domain. We describe two Python routines that parallel the existing R routines in computing these two quantities via package grangers. We present a simulation study showing that under zero-causality processes Python routines tend to perform slightly better than R, while under low-causality processes R routines perform quite better, because Python is less sensitive than R to small causality parameters. This difference can be attributed to the intrinsic VAR order selection procedure of the two packages.