re:Re: Re: st: RE: dfuller: why do I get different results?

[Christopher Baum <kit.baum@bc.edu>]

<>

I still need the formal statistical test - for a research paper a
general plot will not be sufficient.

BTW: I just ran manually the first 10 panels. It is not a
representative sample, but as you can see below indeed, in most of
them the unit-root hypothesis was not rejected at the 1% significance
level:

Unit root tests have notoriously low power with < 100 observations -- that's why we have panel unit root tests.
Three rejections at 5% out of 10 samples suggests that "most" might be I(1), but then a df test with no constant nor trend is a queer bird indeed. The process you are modeling has to make sense under both null and alternative hypotheses for the test to be valid. A bit of algebra shows that a DF regression without a constant implies that the mean of Y is the mean of epsilon == 0 under the alternative hypothesis of stationarity. If your data in levels do not have a mean of zero, this model is incapable of reproducing the data with any choice of ]beta < 0, and so the test is flawed.

Kit

Kit Baum   |   Boston College Economics & DIW Berlin   |   http://ideas.repec.org/e/pba1.html
                             An Introduction to Stata Programming  |   http://www.stata-press.com/books/isp.html
  An Introduction to Modern Econometrics Using Stata  |   http://www.stata-press.com/books/imeus.html


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/