State Space -Forcast

[Hank]

I am reading the Eviews 8 manual on State Space models (Chapter 19 in User Guide II). I have the following questions I hope you can help me. I thank you in adavnce.

1. In the State-Space forecasting part, on page 623, it says, when they are multiple equations for a signal varaible, Eviews can not create a forecast series using wildcard"*". My question is, is there a case that there are multiple equations for a signal variable? Could you give me an example? Will Eviews be able estimate a model like that? I just can not imagine such a case.

2. On the same page in the manual, the second last paragraph, it says that "For the one-step ahead and smoothed methods, this means that at the earliest, the forecast period will begin n-1 observations into the estimation sample". Why n-1? not n?

3. The last sentence in the same paragraph, "for other initialiaztion methods, forecast sample endpoint adjustment is not required", what does it mean? Do we ever need to adjust the endpoint?

4. On top of the page 613, Is the recursive coefficent in the model the same as constant? Since you can say sv2=sv2(-1)=sv2(-2)=..., Does it become the same as a constant? Is that right?

Thank you very much.

Hank

[EViews Glenn]

Off the top of my head, I can answer 1. You could specify that model and we'll try to estimate it, but as with other state space models, just because you can write it out that doesn't mean that it makes since. I can't think of one with this structure, but that doesn't mean that one doesn't exist.

On the others, I'll have to look at the docs since I don't have them in front of me. It may take a bit to get to this as I'm deep in working on the EV9 beta, so if someone wants to jump they're most welcome to do so

[Hank]

Thanks, Glenn.

What I mean is Eviews actually allows two equations share the same signal variable? Normally that does not happen.

I hope you can help me with other questions.

When will Eviews 9 come out?

Thanks.

Hank

[EViews Glenn]

It's been quite a while since I thought about this I haven't looked carefully, but I think that is that you get the initial condition and then n-2 periods run from the beginning of the estimation period to get the n-1 periods of initialization. The endpoints for when we start the forecast need adjusting since we use the estimation sample to obtain these initial conditions. For cases where we have user-specfied forecast initialization or Algebraic Riccatti, we don't need to use data to determine the initialization, we just initialize the period prior to the start of the forecast using the specified method.

[Hank]

Thanks for your answer, Glen.

Your answer confirmed what I was thinking, although it was not directly to my question. That paragraph is a bit confusing. It talked about n-step forecast. If the initial condition started at period 0, the beginning of the forecast period should be N instead of N-1. The last setentance of the paragraph is is unclear.

Do you have any thoughts about a recursive coefficient? what is the diference between a recursive coefficient and an unknown constant?

Thank you very much.

Hank

[EViews Glenn]

Reclusive coefficients are modeled as unknown states with zero variances. Dependent on what you mean by a "constant" I guess you can think of them in those terms.

[hychen]

I have a question about Auto-specification of State Space model. what is the difference between regressors with fixed or recursive coefficients? Why do we need to separate them into two categories? Given that recursive coefficient does not have a error term, it should be the same as fixed coefficient? why does the auto-specification separate regressors with fixed coefficients from regressors with recursive coefficients?

I wonder if someone can help me with this question.

Thanks,

Hank