Unit root tests - boarder case
Posted: Mon Dec 26, 2011 2:47 am
Hi,
I have a small question. I have a time series (n=500) and want to check for stationarity. Therefore I performed some unit root tests, ADF , PP and KPSS.
When including just a constant, all tests say that the process is not stationary.
Using a constant and a trend, then the ADF and PP reject the null hypothesis at the 10% level. Also the KPSS tests rejects the null at the 10% level.
Can I accept that the process is stationary ? Since ADF type of tests have a really bad power when including a constant and a trend and all tests just reject at the 10% level, whereas KPPS has the reversed null hypothesis.
Eventually both stationarity and non-stationarity can be argued, but what is more reasonable.
Thank you
Bye Bye
I have a small question. I have a time series (n=500) and want to check for stationarity. Therefore I performed some unit root tests, ADF , PP and KPSS.
When including just a constant, all tests say that the process is not stationary.
Using a constant and a trend, then the ADF and PP reject the null hypothesis at the 10% level. Also the KPSS tests rejects the null at the 10% level.
Can I accept that the process is stationary ? Since ADF type of tests have a really bad power when including a constant and a trend and all tests just reject at the 10% level, whereas KPPS has the reversed null hypothesis.
Eventually both stationarity and non-stationarity can be argued, but what is more reasonable.
Thank you
Bye Bye