Am 02.09.20 um 15:02 schrieb Olasehinde Timmy:
I already know how to use the syntax in estimating the SVAR/SVEC model
in the Gretl. However, I couldn't found how to get the table for the
forecast error variance decomposition.
you can use the FEVD() function in the addon to get the matrix with
those numbers. See the function documentation in the appendix of the
help for the SVAR addon.
Moreover, can the set identification be used in the SVEC model too?
not possible. The cointegration restrictions represent
infinitely small sets (points), so the standard algorithm would "never"
find acceptable draws. We do have longer-term plans to allow mixing of
traditional short-run zero restrictions and set restrictions, but that's
of course different from the cointegration restrictions. Do you know of
any example in the literature where a SVEC(M) is combined with set
I also humbly suggest that in the future that the VAR/VEC model is
designed to support other forms of IRF & FEVD such as the generalized
Last time I checked, the generalized IRFs are only the collection of
Cholesky-identified shocks where the recursive ordering is re-shuffled
for every shock. So it is already possible to get them, probably also
already without the SVAR addon using only the basic built-in tools in
native gretl. Just re-run the model N times with the different orderings
and save the IRFs for the respective first shock each time (or was it
the last, I don't remember). This may sound like a lot, but because each
model is recursive, this is really not a problem. It is debatable
whether the structural model as a whole is really identified with this
approach, and that's part of the reason we don't offer an extra
interface for them. (I hope I'm not misrepresenting the opinions or
feelings of SVAR's co-author Jack here.)