State space number of iterations

[elnoch]

Hello,

I'm running Eviews 9.5, enterprise edition. I've been trying to estimate a state space model. The program asked me for an update and I downloaded it and installed it. However, now when I run my model, I cannot make my model to be estimated with more than 500 iterations. I tried both the name_model.ml(maxit=1000) command and using the estimate button, but both of them stop at 500 iterations. Can you check if this is a bug please?

I'm sorry to post this here but I didn't find any other option.

[EViews Glenn]

Posting here is just fine. You can also email support@eviews.com.

I'm having trouble replicating this. Can you tell me your build date?

The option for maxit= seems to be accepting input properly as does the dialog. I've tested with a smaller number of iterations, but not with 1000 as I don't have a sspace handy that doesn't converge before that. If you want to post your workfile with the sspace object or send it to support that would help.

[elnoch]

Hi,

Here's the workfile and the program. The odd thing is that if I change a small thing in the program (for example, using annual growth instead of quarterly), then the program runs more than the 500 iterations.
The build is Aug 11 2016.

Thanks.

[EViews Glenn]

I'm still not seeing the behavior that you report.

If I open your workfile and then run the program after changing the maximum number of iterations to 5000, EViews performs 1954 iterations and then stops with a report of a failure to improve. At a glance, it looks as though for this specification and those starting values that the estimate of the variance of the INFLA equation is being pushed to zero, which is why you're getting a failure to improve.

[elnoch]

Hi,

I still have the problem but I wasn't getting good results so I decided to move to another model, specifically the one presented in this paper: https://www.imf.org/external/pubs/ft/wp ... p15144.pdf, which is similar to the one I was trying to estimate. I have one question though: some of the state equations include contemporaneous values of other state variables (see for example equation 2 and 3, page 7) and expectations (equations 16 and 17, page 9) so this would imply including leading values. Is it possible to estimate this model in Eviews? Is there a way around these issues?

Thank you.

[elnoch]

No answer?

[EViews Glenn]

I'm sorry but the request is a little beyond the kind of support that we can provide. From your brief description however, you are going to have problems unless you can rewrite the specification in terms of lags as the state equations must be specified in AR(1) form.