On Fri, 21 Oct 2011, Allin Cottrell wrote: > On Fri, 21 Oct 2011, Riccardo (Jack) Lucchetti wrote: > >> On Fri, 21 Oct 2011, Anutechia Asongu wrote: >> >>> Could anyone gist me on why Grelt crashes each time I'm estimation via a >>> TSLS approach; especially when the endogenous explaining >>> variables exceed 3 and the instruments >4. >> >> Does it? Here, the following script runs just fine [...] > > Here too; and I also tried estimating the same model via the GUI.
Anutechia sent me his dataset and model (thanks!) and we were able to determine (a) that the crash in the original tsls model does not occur with current gretl, but (b) a modified version of the model produced a crash that I could reproduce. This is now fixed in my source tree and the snapshots for Windows and OS X, but unfortunately not yet in CVS since sourceforge CVS is down (they're apparently working on it). The problem arose in this case: you have an endogenous regressor which is not all-zero over the entire current sample range, but is all-zero in the subset of observations that are actually used for tsls, when within-sample missing values are skipped. The fix was to detect and remove such variables before they cause trouble. Allin Cottrell
