> I don't suppose it's a case that users are
likely to bump up against
History:
to test auto_arima from lagreg I applied it
for every series in greene5_1.gdt
(non-percentage variables were transformed to logs)
One difference between libgretl and --x-12... is in
that for models for which nothing estimated
arima 0 d 0; 0 D 0; y --nc, where d+D>=0
--x-12... outputs lnl and ICs while libgretl does not
I inserted lnl and ICs manually for arima 0 d 0; y --nc (d>=1)
Immediately after that for log(pop) (pop=population)
best model became (0 2 0)
That was the first file with macro variables tested!
After that I started to write out lnl for arima 0 0 0; 0 1 0
and noticed the difference
Oleh
On Mon, 5 Nov 2018, oleg_komashko@ukr.net wrote:
> Dear all,
> script below illustrates the problem
Namely, that gretl's "arma" with --x-12-arima option was not working
right in the case of (a) seasonal integration, but (b) no AR, MA or
regression parameters (even a constant) estimated. (That is, only
the variance of the series in question is being estimated.)
That's now fixed in git. I don't suppose it's a case that users are
likely to bump up against, but anyway we now have it right.
Allin
_______________________________________________
Gretl-devel mailing list
Gretl-devel@lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-devel