Strange formula in pooled model

[kiber_master]

Hello!

I tried to calculate pooled model (without effects) with AR and SAR, and found out that fitted in "Actual, Fitted, Residual Table" differs from static forecast on the same period. I then discovered that the formula for fitted series for pooled model in EViews differs from the same formula for linear regression (also with AR and SAR) by one summand.
I've prepared an example for you in workfile and calculation in Excel, so you can see the difference. Could you tell me, is there is a mistake or formula for fitted in pooled model should really be different from the same formula in linear regression?

[EViews Gareth]

Might be a bug, I'll investigate. Note that the fitted value from proc->forecast, and the residual series from proc->make resid, are correct.

[kiber_master]

the residual series from proc->make resid are the same as in "Actual, Fitted, Residual Table". So, if we named them resid01, and the fitted value from proc->forecast - fit01, their sum didn't give the actual series.
Please, check it.

[EViews Gareth]

Yeah, I noticed afterwards that the residuals are not right, but forecast is.

[EViews Gareth]

This has been fixed in the latest patch.

[kiber_master]

I've got it, thank you.

[kiber_master]

Sorry that I'm so late. I've check newest update of E7, and found that estimates of coefficients differs from previous version. Why is it so? Did you change something in the optimization method?
You can check it if open file poole_sar.wf1, push stats, freese the table and reestimate the model.

[EViews Gareth]

There was a rare bug in panel SAR estimation that your workfile helped us spot. We fixed it.

[kiber_master]

So, previous coefficients was not correct?

[EViews Gareth]

Right.

[kiber_master]

Thank you