> On Mon, 20 Aug 2007, Summers, Peter wrote: > >> Do any of you have any experience using gretl for structural VAR >> work? > > Structural VARs are not yet supported as such, although I think > Jack Lucchetti wrote a gretl script that headed in that direction. > Hopefully, Jack will be back on the list before long, and he can > surely offer more help than I.
The current situation is that I have a pretty comprehensive set of ox functions to deal with structural VARs. The three classes (K, C and AB models) of models presented in Amisano-Giannini (1997) are all covered, with short- and long-run identification constraints. Cointegraion is also handled, so for example you can replicate King, Plosser, Stock & Watson (1992). My ox class also performs identification analysis as per my own ET paper (2006). My original plan was to incorporate the whole thing in gretl, once the VECM dust has settled, the "proper" way: that is, recode the whole thing in C. This, however, will take quite a long time. Alternative routes we may take: a) recode everything as a set of gretl functions b) build a gretl function that invokes my ox class (like Sven did with python) (a) may be good for gretl, as it would also provide an excellent testbed for user-defined functions and matrix primitives. However, the amount of work is quite big, and my gut feeling is that this would only end up as a poor duplicate of a native implementation. (b) may perhaps be more promising; yet, it would require you to install ox on your machine and is quite complicated anyway, because parameter passing is definitely not trivial. In the meantime, I guess your best bet is to use JMulTi, which is pretty good at these things. The fact that gretl provides the facility to export your data in JMulTi format should make this relatively painless. HTH, Riccardo (Jack) Lucchetti Dipartimento di Economia Università Politecnica delle Marche r.lucchetti(a)univpm.it http://www.econ.univpm.it/lucchetti
