[Gretl-users] restricted SUR estimation

This is a follow-up to my posting of January 9, in response to Sung Jin Lim http://lists.wfu.edu/pipermail/gretl-users/2012-January/007212.html The script I posted showed essentially the same results from gretl and R (systemfit) for an SUR system, estimated (a) without restrictions and (b) with restrictions and iterated. However, I found that if I estimated the restricted system _without_ iteration, the results from gretl and R differed quite substantially. This turned out to be due to a difference in the initial OLS estimation: gretl initialized using unrestricted OLS, while systemfit initializes with restricted OLS by default (though it has an option to use unrestricted OLS). I looked a bit further into this and decided that gretl's use of unrestricted OLS in this context was not really consistent (in the ordinary, non-econometric, sense of the word), since when we do restricted 3SLS we initialize with restricted 2SLS. I've therefore modified gretl's "system" command to use restricted OLS when initializing for restricted SUR, but have added an option to the "estimate" command, namely --unrestrict-init, which gives you the old behavior. This is in CVS and snapshots. Allin Cottrell