The RATS Software Forum

Tom
it's ok to do generalised IR functions in rats. But would you also have ready a few lines for the generalised FEVD?
Thanks a lot for support
Fabio

Does a generalized FEVD even make sense? Without an orthogonalization, you don't have a "decomposition" of the variance. Now, if you do a @FORCEDFACTOR, you can fill out an orthogonalization that will let you specify one or two columns of the factor and get their share of any factorization.

I know Tom but i was trying to replicate the results in a recent set of papers by Diebold and Yilmaz about spillovers. They define spillovers as selected ratios of fevd at given horizons. While they start with the case of cholesky they then adopt the GFEVD as in Pesaran et al., as if you want to measure spillovers then you want to 'keep' the contemporaneous relations among variables. So, you dont want to use the results in a traditional sense.
Basically you suggest to use something like
forcedfactor sigma ||1,1,1,1,1,1|| f
in a 6 variables var, so that the unique component loads on all the series in the var. Am i right on this?

The Diebold-Yilmaz papers were posted (later) at https://estima.com/forum/viewtopic.php?f=8&t=986. The remainder of this thread has been deleted as it was influenced by some typos in the DY paper.