Question regarding Johansen, please advise
Posted: Fri Jun 14, 2013 6:21 am
by jacoub.sleibi
I'm running the Johansen cointegration test. I have the following questions:
Can I run the test on lets say 4 variables if one of them is stationary i.e one of them does'nt have a unit root?
What if the result shows None* to be significant? what does None* means ??
thank you
Re: Question regarding Johansen, please advise
Posted: Fri Jun 14, 2013 6:58 am
by startz
No. Series can't be cointegrated without being integrated.
Re: Question regarding Johansen, please advise
Posted: Thu Sep 19, 2013 3:29 pm
by Pandelis
a) As long as the number of integrated variables is more than 1 then it is possible that they are co-integrated. Since you have three then it is OK to check co-integration with them only (the stationary variable can be included as an exogenous variable, but that jumps to a different topic).
b) If "none" is the result of the co-integration test (be careful, it has nothing to do with significance, it is not a t-est or an F-test), that means that none co-integrating vectors have been found. In other words, you have NO co-integration. Your VAR model is not stationary, thus your estimation results would be spurious, unreliable for any kind of conclusions.
What you should do? Use all 3 integrated variables in first differences instead of levels. In that manner your VAR model will be stationary and you will have statistically meaningful results.