correlogram-differences

[Pitchforkp]

Hi,

I have a quation concering a correlogram in unstructured vs panel data. When i estimate an equation using a panel structured dataset using just period fixed effects, the correlogram looks very different than when I use an unstrucutred dataset but use year dummies using the @expand function to control for year effects. My question is why they look so different (panel gives positive autocorrelation while unstructured gives no autocorrelation). Don't the two measure the same thing? the coefficients are the same, histogram as wel. If i don't need to control for cross-section effects does it mean that i cant trust the unstructructed dataset results of no autocorrelation?

regards,

Paul

[EViews Gareth]

There's a couple of things going on here...

First of all, it isn't immediately clear to me how you performed a correlogram with fixed effects. As far as I know there isn't a way to do this in EViews...

Secondly when you deal with an unstructured workfile, there is no knowledge of where each cross-section begins, so lags can cross from one cross-section to another. For example a one period lag on Firm Y for the first observation will actually return the last value for Firm X. In a panel, EViews knows where each cross-section begins and ends so won't let lags cross over.

[Pitchforkp]

Hey gareth, thanks for swift reply

the correlogram of the residuals in the fixed effects was just obtained by lookin at the correlogram of the saved residuals from the fixed effects panel regression. while the correlogram of the unstructured equation's residuals are obviously given in the residual option. thanks for the explanation though on the matter,

i have one question then remaining. if i want to estimate this equation though, of which cross-section effects are not needed and i just need year dummies, can i use an unstructured workfile as wel as a panel dataset or do i need the panel dataset for reliability of standard errors?

the dataset is an unbalanced cross-section time series. i guess i need to structure it in a panel dataset and use those regressions right? also since i suspect serial correlation i can correct for this by using robust coefficient variance estimator i suppose. sounds to me like running a regression in an unstructured dataset while controlling for year effects doesn't mean the serial correlation tests are of any use since they neglect cross-sectional differences.

regards,

paul