Hi Esben,
I am not aware of an easy way to compute the confidence intervals of the impulse-responses using vgx functionalities. Try one of the following:
i) use the estimated innovations process returned by the last output argument of vgxvarx, and conduct the residual bootstrap to simulate multiple paths of impulse-responses.
ii) use the estimated standard errors of estimated parameters (the second output argument of vgxvarx) to build multiple VAR spec, and then use the parametric bootstrap for the confidence intervals.
iii) use the (somewhat cumbersome) analytic formulae of the estimated impulse-response covariance matrix to compute the asymptotic confidence intervals.
I apologize for the inconvenience.
