Granger causality ("G-causality")

Basic intuition: "one series can be causal to the other if we can better predict the second time series by incorporating knowledge of the first one."

Trying out a Matlab package for Granger causality on simulated neural traces. I simulated three traces where neuron 1 drives neuron 2, and neuron 3 is isolated. The G-causality inference produced the correct result:

On the other hand, I don't know how well the traces can be modeled by the multivariate autoregressive (AR) model assumed by G-causality. (The toolbox gives a warning that the residuals after the AR model fit is not white.)