DWstat for Panel Least Squares

[PROFI_USER]

Dear experts,

I have found out that Eviews 7 shows different DW stat if I include some NA values to the end of the sample. This modification does not change estimation results except DW stat.
Screen shots show data and estimations results.

Thanks!

Capture.PNG (24.47 KiB) Viewed 5974 times

Capture2.PNG (16.62 KiB) Viewed 5974 times

[startz]

You've also included NA values in the middle of the sample. That changes the pattern of serial correlation.

[PROFI_USER]

Thank you for your answer!

Could you please clarify another one aspect of DWstat for panel data. I am confused with terms which contain resids from different cross-sections. Namely

term.PNG (1.21 KiB) Viewed 5937 times

How can we interpret such term? I was not able to find buckground for this.

In particular Alok Bhargava [1982] gives formula which looks quite consiquent.

WikiFormula.PNG (13.02 KiB) Viewed 5937 times

How do you feel about it?
Thanks!
Bhargava, Alok; Franzini, L.; Narendranathan, W. (1982). "Serial Correlation and the Fixed Effects Model". Review of Economic Studies 49 (4): 533–549.

[startz]

In the Bhargava formula you give, there are never residuals from different cross sections in a term. They all have the subscript i.

[EViews Glenn]

The DW for the original example looks odd to me as the NAs are added at the end of the cross-section sample and thus shouldn't have an effect (that's my first instinct). I'll take a look when I get into the office this morning. That said, personally, I find the DW in a panel context to be less than illuminating.

[edit] Illuminating or not, we've got a bug. There are some cases when we compute the DW where we lose the panel workfile context. So in this case, the examples not only don't match, but neither is correct. I'll let you know when I have a fix. Sorry for the inconvenience.

[EViews Glenn]

I've posted a fix in the main EViews code-base and it should show up in the next patch. Again, our apologies.

[PROFI_USER]

Thank you for your assistance!