Forecast SE for pooled model

[kiber_master]

Hello!

As I understand, to estimate model with fixed effect, we just need to estimate simple linear model with dummies for effects. I made such model, it gave the same results for regressor coefficient, the same fitted, forecast, but forecast SE was different. Why it is so? Technicaly I understand that in pooled model we use vector c and x? and covariance matrix is 2x2, and in simple linear model matrix is 4x4 and regressors are different, but it is the same model. So why should SE be different?

Please, help me to resolve this question.

I've made an example in E7 and in Excel

[EViews Gareth]

I don't follow. Could you explain exactly what you've done in a bit more detail?

Looking at your workfile, none of the equations seem to match in any way?

[kiber_master]

Sorry, workfile was not prepared correctly. Check this one.
In xlsx-file please check column R for both pages. There are formulas for forecast standard error in pooled model with fixed effect. At the first page covariance matrix 2x2 is used. At the second page it is 4x4. So results differ, although it is the same model, as at the first page.

By the way, I think, there is a bug: SE differs when I do not include constant in pooled model with fixed effects, although in this model constant is estimated in any case.

[EViews Gareth]

The panel forecast does not compute coefficient uncertainty for the fixed effects.

[kiber_master]

No, it does. You can check the formula in Excel file, it is for coefficient uncertainty, because depend on x_t. And near there is SE series from EViews, the same as I've calculated.

[kiber_master]

So, what is your answer?

[EViews Gareth]

I cannot follow your Excel file, so cannot comment on it.

However the answer to the question, why do forecast standard errors differ between the panel equation and the manual equation you estimated on a different page, is that the panel equation does not compute coefficient uncertainty for the fixed effects.

[kiber_master]

You even didn't check your statement! Compute forecast SE for model with fixed effects with and without coefficient uncertainty, and get different results.
How can you tell me the answer, if you didn't know, how things going on in the program?

Tell, what is wrong with my excell file? I'll change it, if it's needed.

[kiber_master]

Sorry, if I was too harsh. It's seems to me that you don't want to answer, just kick me away. :(

[EViews Gareth]

It does calculate coef uncertainty for the other regressors, just not the fixed effects, so you'll still get different SEs.