In VAR what matters is that you have enough lags with respect to your integrations. If you have I(d) variables then estimating VAR(p+d) is fine which is the case here. Regarding the IRFs, please note that these are not structural responses. Hence they don't have any economic meaning.
↧