According to Enders (2014), p. 303 as long as there are identical regressors in each equation in the VAR, OLS estimates are consistent and asymptotically efficient. However, I imposed restrictions in my VAR such that not all variables are included on the right side, in which case he recommends seemingly unrelated regressions (SUR) since they provide efficient estimates.
Now, I am using the R package vars in which I can easily make forecasting and impulse responses, but it estimates the coefficients using OLS. Now, my question is whether the coefficients are still consistent? And if yes, if I can do the forecasting and impulse response using OLS, but for inference I should use SUR?