The RATS Software Forum

Dear all,
I have a question concerning the Unit Root Test ADF to test the stationarity of the series.
To be exact, I have to do several tests for different sub-periods: 1950:2011, 1960:2011, 1970:2011, 1980:2011, 1990:2011.
However, if I do only a test ADF for the overall period from 1950:2011 and if I see that the series is stationary in this overall period, can I say that this series is also stationary in the sub-periods mentioned above ?
I’m wondering if I’m obliged to do a test ADF for each sub-period ?

Many thanks,
Best regards

cecedi wrote:Dear all,
I have a question concerning the Unit Root Test ADF to test the stationarity of the series.
To be exact, I have to do several tests for different sub-periods: 1950:2011, 1960:2011, 1970:2011, 1980:2011, 1990:2011.
However, if I do only a test ADF for the overall period from 1950:2011 and if I see that the series is stationary in this overall period, can I say that this series is also stationary in the sub-periods mentioned above ?
I’m wondering if I’m obliged to do a test ADF for each sub-period ?

Many thanks,
Best regards

I assume that you are thinking of doing the subsamples because you suspect that the process might not be stable across the entire range. The literature on unit root tests with broken trends shows that that tends to create false acceptances of unit roots, not false rejections. So you're very unlikely to overturn your result in the subsamples, and if you did, it's not clear how you would interpret it---maybe some type of threshold effect, but the process would have to be inside the zone where the correction doesn't apply for a long time in order to give you an acceptance of a unit root.