Re: [Gretl-devel] ARMA developments
In my posting of 27 April I said that our ARMA initializer should
"never be unusable" (that is, should never fail at the first
iteration).
As of that time, that was not strictly true: the MA polynomial would
never be "bad" but the AR part, if present, might be. That's now
fixed, I believe. So if anyone now sees the error message
"BFGS: initial value of objective function is not finite"
on AR(I)MA estimation in git/snapshots that's considered a bug, and
please let us know.
One other point. Since AS 197 is now the default algorithm (other
than for samples containing missing values) I've renamed the
(undocumented, testing) option to use AS 154 with the "arma"
command: instead of --as=154 it's now --as154 .
Allin