Dynamic forecast issue

[mandix]

I know that there are a gazillion posts and questions, all very patiently responded by (mostly) Gareth. I read many of them, and they were very helpful.
BUT... I still have a question that is killing me.
I am running a very simple ARMA(1,1), with monthly data, 1960m1 to 1996m3 (it is out of Pindyck and Rubinfeld, the interest rate of 3-month T-bills).
When I ask for a *dynamic forecast*, all forecasted values from 1961m1 onward *are the same*. That is, I go into the forecasted values created by Eviews,
and I get all the same numbers.
Did anybody come across this? Does anybody know why?

[EViews Glenn]

Could you post your workfile with the saved estimated equation...

[mandix]

Yes, I would and want to; but I do not see how to attach a file...
There is no "paper clip" (like in Outlook) that I can see, and it seems that I cannot "copy and paste" the file inside this dialog box (right-clicking the mouse does not enable me the Paste function).
Which is the reason I did not attach in the first time around...sorry.

[EViews Gareth]

To upload an attachment to your post, click on the "Upload attachment" tab underneath the dialog box.

[mandix]

Sorry, I think I am blind. My apologies.
Attached is the file.
I am running a simple ARMA(1,1) on the *differenced* series (so, it is more of an ARIMA (1,1,1)).
The forecasted series is from a dynamic forecast. You will see that after a few months in the beginning, all values are the same.

Thanks again, Manfred

[startz]

The forecast values are all just the intercept. The MA term essentially only matters for one period. The AR term declines geometrically. .3^10 is close to zero.

I expect that's what you're seeing.

[mandix]

Ahh, it was that obvious. Sorry. Thanks so much Startz.

[mandix]

No - wait - I do not think it is that obvious.
Startz, correct me if I am completely off, but what you are saying is true of out of sample forecasts, but not *in sample*.
The MA for in sample *does not disappear*, it disappears with the last observation (more or less).
But what I am saying is that in my example the dynamic forecast is the same number for most of the *in sample* period.

Ok - where I am getting this wrong?

[startz]

Dynamic forecasts in EViews are mult-step forecasts, so the in-sample is no different from the out-of-sample...I think. I always find this one confusing.

[mandix]

Startz, ok, then this makes me feel a little bit better...
Because I also do find it somewhat confusing.
Not sure if Gareth or Glenn could chip in to clear this up (probably for the n-th time).
To repeat my problem: I ran a very simple ARMA(1,1) on the differenced time series of 3-month T-Bills.
The sample is 1960m1 to 1996m3 (this follows an example from Pindyck and Rubinfeld, ch 18).
Then, I ask for a *dynamic* forecast.
And lo and behold, almost ALL of *in-sample* forecast (i.e., estimates) are the same number.
Why?

[EViews Gareth]

For the reason that Startz gave you - your AR term dies off quickly. In sample or out of sample doesn't matter - you're doing a multi-step dynamic forecast.

[mandix]

Ok, I will chew on this, I do stand corrected then.
Thank you all for your kind patience.

[EViews Glenn]

To put it another way, you are doing a dynamic forecast from the start of your forecast sample (which appears to be the start of the estimation sample). The actual data are not used for the forecast period. That's why your MA vanishes after 1 period since there are no additional lagged residuals since the forecast is dynamic from the first period. The rest of the variation is from the AR dying down (again since there are no additional lagged residuals).

[mandix]

Glenn, thanks again. Manfred