Page 1 of 1

Different results for running state space models

Posted: Thu Jan 10, 2019 2:02 pm
by adamsmith212
Hi,

I am running a state space model (time varying parameter) and trying to plot the smoothed states. I am able to do this. However, when I change the specification, estimate, and then change it back to the EXACT original specification - my estimates are completely different. A concrete example:

Here is what I do:

I load up a workfile with the following specification:

@ename e1
@ename e2
@ename e3

@evar var(e1) = exp(C(1))
@evar var(e2) = exp(C(2))
@evar var(e3) = exp(C(3))


@signal spread_1d = mp1*beta1 + path*beta2 + e1
@state beta1 = beta1(-1) + e2
@state beta2 = beta2(-1) + e3

I then edit the spec and run this:
@ename e1
@ename e2
@ename e3

@evar var(e1) = exp(C(1))
@evar var(e2) = exp(C(2))
@evar var(e3) = exp(C(3))


@signal spread_1d = mp1*beta1 + e1
@state beta1 = beta1(-1) + e2

I then edit the spec and run this:
@ename e1
@ename e2
@ename e3

@evar var(e1) = exp(C(1))
@evar var(e2) = exp(C(2))
@evar var(e3) = exp(C(3))


@signal spread_1d = path*beta1 + e1
@state beta1 = beta1(-1) + e2

and after running the original spec, I cannot get the original estimation results.

Do you have any insight or idea into this? In other words, when I change the specification and re-estimate, I cannot get back the original results. My smoothed estimates are just a flat horizontal line.

Thank you

Re: Different results for running state space models

Posted: Thu Jan 10, 2019 3:55 pm
by startz
Starting values (in the C vector) matter.

Re: Different results for running state space models

Posted: Fri Jan 11, 2019 9:16 am
by adamsmith212
Thank you. How would you suggest that I estimate my code so that each time I do so, I get the same estimation values? In other words, get the same results each time I run the code.

Re: Different results for running state space models

Posted: Fri Jan 11, 2019 9:29 am
by EViews Gareth