Diagnostic test for time series regressions.

[rozita saadan]

Dear Statalist,

I have unbalanced daily panel data consist of 2000 firms and 20 years. I would like to run time series regressions for each firm and get the R-squared value for each regression.
My question here is, do I need to perform diagnostic test such as unit-root test, normality, and other tests before do the time series regression?
Previously, I had run the fisher-type unit root test, however, I was confused because I want to run the time-series regression, not to pool the result of R-square of all firms.

Can anyone help me to answer this question?
Thank you.

Regards,
Rozita

[Phil Bromiley]

It is unusual to run separate regressions for each firm. Obviously, you can use various procedures including writing a little program and using rangestat, or just looping over all the firms.

With 2000 firms, what would you do with any of these tests? Suppose you find 100 firms that fail the unit-root test, then what?

How many of these tests one does varies a bit with academic area - some areas prefer more tests than others. By the way, the normality test should be normality of the error term, not normality of the variables themselves.