Panel unit root - interpreting the statistical significance

[John Coleman]

Hi everyone,

To cut to the chase, I have performed 'xtunitroot ips test' (Im et al 2003, unit root test) on my panel data.

More specifically the code I entered:

xtunitroot ips naturalloggdp, lags (1)

The result that Stata returned to me was:
Ho: All panels contain unit roots Number of panels = 21
Ha: Some panels are stationary Number of periods = 31

............Statistic...... p-value
W-t-bar -12.4188 ......0.0000


I am aware that this indicates that we can reject null hypothesis. However at what level of significance can I reject the null? How is the level of significance determined.

I have familiarised myself with the Im et al article and also this stata xtunitroot manual (http://www.stata.com/manuals13/xtxtunitroot.pdf) but I am alittle bit lost.

Many thanks for reading,

John

[Jesse Wursten]

I'm not entirely sure what you're asking? The p-value is 0. Hence, you can reject the H0 at any significance level, just like you would in a regression. Note that this doesn't mean there's no unit roots in your panel. All it tells you is that not all your panels have a unitroot.

[John Coleman]

I'm not entirely sure what you're asking? The p-value is 0. Hence, you can reject the H0 at any significance level, just like you would in a regression. Note that this doesn't mean there's no unit roots in your panel. All it tells you is that not all your panels have a unitroot.

Hi Jesse,

Thank you so much for your time and for your response.

I made a mess of my original question. I want to know where I can find the critical value tables of the IPS, LLC and Breitung test?

[Guest]

Maybe I am completely wrong ... but I think the xtunitroot manual states critical values for ips for example on page 14 (Example4) and describes the background calculations on the same page. For the other tests it might either be in there as well. It could also be a good idea to look at the original papers for a reference to the test statistic.
(see: http://www.stata.com/manuals13/xtxtunitroot.pdf)

[John Coleman]

Hi Justus.

Thank you for your response. So for the IPS test the critical value does appear. However when I run the IPS test with lags and trend the critical value disappears. You are left with the output of the W-T-statistic and a p-value.

I have looked at the papers also so maybe I am misunderstanding...

Thanks for your help

[Guest]

Hi John

the sentence I forgot to cite was: " This statistic is appropriate when you assume
that both N and T fixed; exact critical values reported in IPS (2003) are reported immediately to
its right."
Sorry for being inaccurate. I suggest again you look at the original paper (such as: http://www.personal.psu.edu/hxb11/BREITUNG.PDF) but it appears to me that the critical values have to be obtained by simulations. More for my interest why are you concerned about the exact test statistic?
You might also look at this: http://cemood.people.wm.edu/panelur.pdf

On Page 4 he refers to critical value tables. Even though the last sentence on the same page is not entirely true to my knowledge...

[John Coleman]

Hi John

the sentence I forgot to cite was: " This statistic is appropriate when you assume
that both N and T fixed; exact critical values reported in IPS (2003) are reported immediately to
its right."
Sorry for being inaccurate. I suggest again you look at the original paper (such as: http://www.personal.psu.edu/hxb11/BREITUNG.PDF) but it appears to me that the critical values have to be obtained by simulations. More for my interest why are you concerned about the exact test statistic?
You might also look at this: http://cemood.people.wm.edu/panelur.pdf

On Page 4 he refers to critical value tables. Even though the last sentence on the same page is not entirely true to my knowledge...

Hi Justus,

That is disappointing to me if the critical values have to obtained through simulations and not by the 'xtunitroot' command itself. And it is beyond my skillset! The reason for being concerned with the test statistic is so that I can reject null hypothesis at certain levels of significance and put that in my paper.

Thanks,

J