Re: [Gretl-devel] svars

On Thu, 7 Jan 2010, Sven Schreiber wrote: [...] My take on SVARs: if you want to do Structural VARs with free software, the possible alternatives that come to my mind are: 1) use R + urca + vars 2) use jMulti 3) use my collection of scripts I am deliberately excluding A. Warne's program since it isn't free software (which is a shame, since it's very very good). My preference goes to alternative 3 (astonishing, isn't it?). The only 2 people who have seen my scripts so far are Allin and Sven, since I don't consider them ready for prime time yet. However, since I've wanted to work on these for a long time, but never did, I'm putting them on a location they can be downloaded from: http://www.econ.univpm.it/lucchetti/gretlSVAR.tgz. Hopefully someone else will find them useful and improve them. As they say, "release early, release often". When, in the future, we find that the algorithms are sufficiently robust and reliable, we may think about translating them to C and incorporating SVARs into gretl proper. Why not jMulti? Well, because despite being a very good package when it comes to the algorithms, I'm sorry to say it's pretty much dead. A Linux version hasn't been released for ages, and all the progress I've seen for a while was some occasional bugfix. Plus, I'm not exactly enthusiastic about the combination of gauss (or R) for number crunching, java for the GUI and postscript for the output (too many things that can go wrong). Last but not least, it's not scriptable. Why not R, then? To begin with, I think our own implementation of VARs + cointegration-related stuff is better. Next, there's the usual gripe I have with R: either you work with it all the time, or its syntax is, let's say, unpleasant. I like to think that gretl's now excellent interface to R allows us to use R as a fallback tool. But in the long-term, I'd rather use gretl as much as possible. For the brave souls who will try to get some work done with my functions: I've tried to test pretty much everything to the best of my skills, except for bootstrapping, which is still untested and probably contains hordes of bugs (as opposed to the odd bug here and there in the rest of the package). However, the tarball also contains a pdf file which will guide you through the basic syntax and is the nearest thing to proper documentation that I can provide. PS: Sven, with my functions you CAN do restricted SVECMs. The \beta matrix is one of the inputs, and the function by itself doesn't care about the way it was estimated (assistance by superconsistency is gratefully acknowledged). Riccardo (Jack) Lucchetti Dipartimento di Economia Università Politecnica delle Marche r.lucchetti(a)univpm.it http://www.econ.univpm.it/lucchetti