Should multivariate time series by normalized in any way before fitting a VAR model?

I have many time series on very different scales; some take on values in the millions, while some are restricted to $[-1,1]$. Should I preprocess them in any way, or will the found coefficients take account of all this on its own?