On Fri, 2 May 2014, Tim Nall wrote:

> In an earlier post (with a different subject line) I mentioned that
> the SVAR graphs could be more informative by adding "IRF by shock" and
> IRF by var". That same principle might apply to adding the model
> option to the graph title, e.g., "Plain model" or "KPSW Model."

Yes, that could be done. I'll give it a good look next week. However, I'm 
a bit reluctant to give "official" status to the terms "plain" and "KPSW", 
as they're non-standard in the literature, and I just used them in the 
context of my SVAR package as working expressions. And bear in mind that 
graphs are editable, so if you like those phrases (which I don't very 
much), you're welcome to add them in your own work.

Moreover, there's a more general thing I'd like to point out. The SVAR 
package is meant to be used primarily via scripts. The GUI interface must 
be considered a convenient shortcut for some of its basic features, but 
the specification, estimation and analysis of a SVAR model are too complex 
to be handled through a GUI to a satisfactory degree.

I don't plan to spend too much time working on the GUI interface to SVAR: 
I'm simply not interested in that. I'd much prefer devote my time and 
energy in debugging and extending its number-crunching part; for example, 
I'd like to incorporate the algorithm for assessing identification, 
sign-identified models à la Uhlig, heteroskedasticity-identified models à 
la Rigobon with Bacchiocchi's generalisation... there's so much more that 
could be done.

If you fell like spending time perfecting the GUI to it. you're very 
welcome, and I'll be happy to offer all the help I can.

> And at a higher level of menu items: if I understand the SVAR
> documentation correctly (and the possibility that I do not is quite
> high), the KPSW Model of SVAR is actually a SVEC? If so, then to a
> naive user at least (e.g., me) moving that option off of the SVAR and
> into its own sub-menu-item of Time Series would make sense.. call it
> SVEC (KPSW) perhaps.. Thanks TMN

Yes, up to a point. A VECM is, from a statistical viewpoint, a particolar 
subspecies of a VAR in which we assume that we know how many unit roots 
are in the system so that a very clever system of constraints can be put 
on the VAR representation. This has no implication on the "structural" 
part _per se_. You can use a triangular (or Cholesky, if you prefer, or 
recursive, if you prefer) identification scheme on a VECM and it will 
become a SVECM. What *I* call a KPSW model is a model in which you taker 
the extra step of using your assumed knowledge of the cointegration space 
to classify structural shocks into permanent and transitory. So a KPSW 
model is a SVECM, but the converse doesn't necessarily hold.

-------------------------------------------------------
   Riccardo (Jack) Lucchetti
   Dipartimento di Scienze Economiche e Sociali (DiSES)

   Università Politecnica delle Marche
   (formerly known as Università di Ancona)

   r.lucchetti@univpm.it
   http://www2.econ.univpm.it/servizi/hpp/lucchetti
-------------------------------------------------------