State space results differ

[AmateurEviewsUser]

Hey all,

I have been trying to estimate a state space model using Eviews 7. After figuring out how to provide initial values for state variables and parameters, I managed to converge the model and the results look fine.

However, problem arises when I try to plot the smoothed estimates of the state variable. When I do view -> state views -> Graphs state series- > smoothed state estimates and compare it with the series I get using proc -> make state series -> smoothed state estimates, I get totally different results.

Shouldn't I be getting the same result using both the options? I cant figure out where I am going wrong.

Would appreciate some help

Thanks

[EViews Gareth]

Could you post your workfile?

[EViews Glenn]

And what's the date on your EViews 7 (Help/About EViews).

[AmateurEviewsUser]

Hey Glenn,

Eviews v7.1 has a build date of Feb 10, 2011 on it.

Hey Gareth,

Please find attached the graphs generated using the two options. Note: In the first graph the final value (SV1) is below 14 whereas in the second graph it goes close to 16. I don't understand the discrepancy. They look similar in their behavior but the magnitude differs.

[EViews Glenn]

Before we go on, can you update to the latest version?

[akkerman]

I am having the same problem. Using Eviews 9.

[EViews Glenn]

I'll have to see a workfile with the estimated sspace object as I"m not seeing this behavior in my examples.

[akkerman]

I have attached the workfile.

[EViews Glenn]

Thanks. I'll let you know when I figure out what's going on.

[EViews Glenn]

It looks as though our code for producing the state space graphs gets confused when you have an irregular dated workfile with gaps in the dated workfile structure. I'm working on a fix.

For now, you can produce the graphs by saving the results using the Procs and then constructing the graphs manually. I know that this is a bit of a pain and apologize for the inconvenience.

We'll get a fix out as soon as we can. Apologies again.

[akkerman]

Thank you Glenn.

[EViews Glenn]

A fix will appear in the next patch for both 9.5 and 10. Thanks for helping us locate the problem and sorry again for the inconvenience.