Am 26.08.2022 um 15:03 schrieb Arturo José Galindo:
Thanks a lot Artur and Sven. This was very helpful. Another question,
I saw in the documentation that the possibility of combining sign
restrictions and zero restrictions on the C matrix is still in the
works. Is there a way to work around this while a new version of the
SVAR addon comes out?
In short, not really. One workaround that might come to mind is to
approximate an exact zero restriction with a small set around zero, but
this would have a huge computational cost, because you would be drawing
gazillions of rotations that will be rejected. Also it would be
difficult to justify the workaround to the audience.
Another theoretical option might in principle be applicable if, say, the
first shock is exactly identified. For example there are quite many
applications where the first shock is identical to the first
reduced-form error. Then it might be possible to consider only the
orthogonal subspace, projecting off that first shock, and somehow
continue from there. But the SVAR addon isn't set up for that, and so
it's not clear to me (even though I'm a co-author of that addon) how the
existing functions could be recycled to achieve something like that.
In any case, the more demand is for the feature, the more likely it is
that eventually the work will proceed.