Cross Validated is a question and answer site for people interested in statistics, machine learning, data analysis, data mining, and data visualization. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

I have two variables. They're both $I(1)$ even when I fit constant and trend terms into the ADF test. The $p$-values for the stationarity tests are around 0.5 so it's not a marginal case.

However, when I execute the Johansen procedure with a constant in the error correction term the $\Pi$ matrix is full rank by the trace and maximum eigenvalue tests ($p < 0.01$). I've read a textbook and a paper on the consequences of $\Pi$ being full rank and they both state that this implies that both variables in the system are $I(0)$. It's important to note that if I exclude a constant from the ECT or if I add a deterministic time trend, then the rank of $\Pi$ is estimated as 1 (instead of full rank).

Is this most likely a type 1 error or could there be more going on here (such as partial integration)?

share|improve this question

You have answered your own question. If you get cointegration when adding trend, or removing constant, this means that these terms are important, especially when without these terms you get conflicting results. To test the intuition I suggest doing Monte-Carlo simulation with simple unit roots. Also if you have two variables you can also test cointegration using Engle-Granger approach.

share|improve this answer

If two variables are I(1) but are not co-integrated, it simply means that there is no long run-relationship between the two variables. However, there may be short run relationship. So you need to run the VAR in first difference. If two variables are I(0),we can always use OLS. There is no justification to use co-integration.

Johansen’s co-integration model describes five different cases regarding the intercept and trend in the co-integrating equation. For details refer to the paper by Ahking (2002) in Journal of Macroeconomics. The results will differ but which to choose depends upon the data and the theory.

share|improve this answer
This doesn't answer my question. I asked about the scenario of $\Pi$ being full rank while at the same time both variables being $I(1)$ (hence apparent contradiction). – Jase Jan 10 '13 at 17:29
My answer did refer to the paper which suggests that you couldn't just add /remove trend just to get the answer. That should be based on the data and theory. – Metrics Jan 10 '13 at 20:40
But I didn't ask how to specify a VECM. I asked what causes a full rank $\Pi$ when both variables are non-stationary. – Jase Jan 11 '13 at 2:57
I can't be specific, but since this is not true for all cases, I think this has to do with data per se. – Metrics Jan 11 '13 at 3:55
@Metrics, it sounds like you know a lot about cointegration. It would be helpful if you answered the poster's actual question, instead of answering a different question that he/she is not asking. – user25064 Apr 30 '13 at 15:29

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.