That's ok,
But I have noticed a very bad innovation for
the last twenty-four hours:
while I proposed to increase gradient norm maximum
for --nc models when the roots bumps at the
unit circle, I have noticed that maximum
for gradient norm was increased for models
with constant that leads to many false bumbing
at the unit circle when trivial {mean(y), 0.0001,...
leads to quick and normal convergence
1) theory when n->Inf const->mean(y),
so it's a theoretically valid start values
2) I have estimated more than million arimas
to check empirical peerformance
Now they are out-dated by last changes
Now I must invent how to catch false
bumping at the unit circle
Oleh
On Thu, 15 Nov 2018, oleg_komashko@ukr.net wrote:
> Dear Allin,
> The file mentioned was attached in one of the recent messages
> gretl version 2018d-git
> Current session: 2018-11-15 22:29
>
> ? set verbose off
> Read datafile /home/oleh/gretl/very_bad_data.gdt
>
> 'arima 2 1; var1 --x-12-arima' results:
[...]
I'm very busy with classes right now, but I'll get back to this point
when I can.
Allin