[Gretl-users] Re: ERROR: sample script gmm
On Wed, 31 Jul 2019, Artur Tarassow wrote:
The weird thing is that if I execute only the 3rd example which fails
on my
1st machine by hand, it works fine and I don't get a "norm gradient" issue
returned. I can't really figure out why it fails...
It's failing because it's on the very edge of what works numerically,
so results may differ by system.
When you execute the third example by itself you have a different
initialization for alpha and delta -- the starting values of 0.5
and 0.5 as opposed to whatever has been estimated earlier.
Allin