EViews.com

Hello guys, I'm new to this forum and I need some help with stationary data.

I'm trying to make a simple paired Granger procedure using eViews, but first of all I must make the data stationary.
I am using two methods to make sure it is stationary. 1st I check correlograms and then I run a Unit root test. I have made my Y data stationary so far by using second differences (both correlograms and unit root test are ok), but I have problems doing the same with X data. When I check unit root test, i can reject the null hypothesis that my data has a unit root but when I look at correlograms, they aren't within the limitation brackets and shows that data is not stationary. So my question is what should I count on - unit root test or correlograms? Or maybe there is some other way to ensure stationarity of the data? For better understanding, I will add the pics of what I get.

Here is the graph of the data i have -> this is the raw data.



Here is the graph of logged and 2 times differenced data:



And here is the correlograms of the logged and two times differenced data:



Unit root test of the same data:

If the ADF results are all positive in second differences. etc ( no unit root ) and data is declared stationary.
While the correlgrams implies something else, why don't you just take a individual stance at it?
This is not a uncommon problem when doing multiple analysis and tests of the same antecedent.

If I were you I would look up comparative research around this.

I've seen a similar issue with trace and maximum eigen value test in johansen cointegration test.
These are two tests with the same mission withtin the same procedure, but they can sometimes produce different results..( one finds cointegration equation and other not )

Helmut Lütkepohl did a research on this
http://ideas.repec.org/a/ect/emjrnl/v4y2001i2p8.html