>A criticism VARs face is that they are atheoretical. They are not built on some economic theory that imposes a theoretical structure to the equations. Every variable is assumed to influence every other variable in the system, which makes direct interpretation of the estimated coefficients very difficult.
This is maybe me needing a refresher, but in what way does this make interpretation of the coefficients difficult?
I'd much rather side with Sims on this:
Sims advocated VAR models as providing a theory-free method to estimate economic relationships, thus being an alternative to the "incredible identification restrictions" in structural models. [1]
I was also expecting a bit more material on Granger causality, as I remember using that as the final nail in testing predictive validity.
[1] https://en.wikipedia.org/wiki/Vector_autoregression#Applicat...
