[Gretl-devel] Exo vars ordering in system (bug and feature request)

[if this post comes through twice, have used a wrong email address before, again...] Hi, first of all, sorry for the long post... Consider the following system where for illustration only a subset of all exogenous variables (here: seasonal dummies) are used in each equation: <hansl> open denmark genr dummy system method=ols equation LRY dq2 const equation LRM dq1 const end system matrix B = $sysB print B </hansl> just like the order in which they appear in the system spec. Ok; now there are two problems with that: First what I think is a bug -- in the estimation printout 'const' always comes first, and it has been said before that the ordering in the coefficient accessors should be like in the printout. And BTW, which ordering is the one used by $xtxinv? Secondly I have a more subtle problem: In my application I do not specify the equations manually but as a result of some model reduction procedure. Thus I do not know in advance which exogenous variables "survive" in which of the equations, and therefore I do not know in advance in which order they will appear in the final system (and by implication, in $sysB). But I want to use the final system for forecasting, and so I need my list (or matrix) of exogenous system regressors in the corresponding order matching the $sysB structure. Currently it seems almost impossible (or very difficult) to achieve that. Of course it is impossible for gretl to guess what the wanted ordering is. One possibility would be to introduce a new 'sysB_order' keyword which could be made an optional part of the system spec. For example the above system would become: <hansl> system method=ols equation LRY dq2 const equation LRM dq1 const sysB_order const dq1 dq2 end system </hansl> (This would also affect $xtxinv I guess.) Another (and maybe even better) option would be to make the system-wide list of exogenous variables accessible, matching the $sysB ordering. Why not use $xlist for this, which currently isn't used in the system case? thanks, sven