Page 1 of 1

State Space issue

Posted: Fri Mar 02, 2018 2:21 am
by pgrign
Hi, I'm trying to estimate a SS model, and it seems to me that, when I try to estimate variance, the values vary significantly according to the coefficient I use. In other words, if in one model I estimate the variance as [var=exp(c(2))], and in the other one as [var=exp(c(20))] the results are completely different, even for the space estimation. I attach a workfile in which I put 3 state space models that have the very same structure and use the very same data, but different completely as the coefficients are different.
Could somebody explain me why that happens?
Many thanks

Re: State Space issue

Posted: Fri Mar 02, 2018 7:09 am
by startz
One possibility is different starting values in c(2) versus in c(20).

Re: State Space issue

Posted: Fri Mar 02, 2018 8:15 am
by pgrign
Thanks for the answer... I've also thought of that, but even if I try to impose priors I get the same results. I don't understand how the number put in the brackets of c() would change the estimations.

Re: State Space issue

Posted: Fri Mar 02, 2018 8:25 am
by EViews Gareth
Can you provide the workfile?

Re: State Space issue

Posted: Fri Mar 02, 2018 8:32 am
by pgrign
Yes, you may find it in the first post. I attach it again just to be sure :)

Re: State Space issue

Posted: Fri Mar 02, 2018 10:04 am
by EViews Gareth
Sorry, early morning - missed it in the first post.

Anyway, if I open up ss01 and ss02, and estimate both (setting C=1 before estimating each), I obtain identical results.

As Startz said, it is just a starting value issue.

Re: State Space issue

Posted: Mon Mar 05, 2018 7:10 am
by pgrign
Oh thanks, I've completely misunderstood what startz was saying... now it works fine
Thank you!