IPS Unit Root Test Options

[Julia Susann]

Dear Statalisters,

I am using the IPS unit root test to test for unit roots in my macro panel (N=22) (T=20).

I want to do the test with constant and with constant and trend and let the Aikaike information criteria choose the optimal lags. From my understanding, this translates into the following STATA commands:

• for constant only: xtunitroot ips variable, lags(aic)
• for constant and trend: xtunitroot ips variable, trend lags(aic)

Did I understand that correctly?

The Stata help file also suggests the option demean. I am not sure when exactly it should be used and if - in case it is used- it can simply be added to the above mentioned commands. Could anyone explain?

Sorry if this is a very basic question, but I am not a STATA or econometrics expert.

Thnak you!

[Jesse Wursten]

I think you also need to specify a maximum number of lags, e.g. xtunitroot ..., lags(aic 5).

As for demean

demean requests that xtunitroot first subtract the cross-sectional averages from the series. When specified, for each time period xtunitroot computes the mean of the series across panels and subtracts this mean from the series. Levin, Lin, and Chu suggest this procedure to mitigate the impact of cross-sectional dependence.

The problem in macro panels is that your panels are rarely independent of one another. E.g. affairs in Belgium affect what happens in the Netherlands and Luxemburg. While often ignored, this can lead to some issues, especially in panel unit root testing. Demeaning the panels per time unit reduces these issues.

[Julia Susann]

You are right, I should define a maximum number of lags. The demean option makes sense now. I will add it as well. Thank you for your help!

[Micah Alvarez]

Hello, I read in the Stata manual (https://www.stata.com/manuals/xtxtunitroot.pdf) that "Once we control for cross-sectional correlation by removing cross-sectional means (when we use the demean option), we can no longer reject the null hypothesis of a unit root at the 1% significance level, though we can reject at the 5% level." Somehow, we need a p-value something like 0.0187 in order to reject the null hypothesis of a unit root, but we can't reject it when it is smaller than 0.01. Someone correct me if I'm wrong, thanks.