Page 1 of 1
different results on different computers
Posted: Thu May 31, 2012 9:47 am
by schwert
My son and I both use Eviews 7.1 on window 7 64 bit machines, but we get different estimation results from regressions with ARMA errors (Non-linear least squares) When I estimate the models he has run I get the same results as when I estimate the models I have run, and he sees the same thing (which is not surprising, because we think we are estiamting the same model), but the results he gets on his computer are different than the results I get on two different machines that I use.
this is really puzzling .. .
I can send the eviews workfile if that would help
Re: different results on different computers
Posted: Thu May 31, 2012 10:29 am
by EViews Gareth
Check you are using the same version of EViews (really there's no reason for either of you to be using EViews 7.1!).
Also check you are using the same starting values.
Re: different results on different computers
Posted: Mon Jul 09, 2012 2:24 pm
by Rits
When you say "same starting values" are you refering to using the same first value of the sample being used?
Re: different results on different computers
Posted: Mon Jul 09, 2012 2:47 pm
by EViews Gareth
No. When performing non-linear estimation (such as ARMA), you need numerical starting values for the coefficients. Different starting values can yield different final estimates. In general EViews obtains starting values from the C vector in the workfile.
Re: different results on different computers
Posted: Mon Jul 09, 2012 2:49 pm
by Rits
Ok so it's the value of the C vector. I'll give it a shot. Thanks Gareth!