Johansen Cointegration Test Procedure

[jomike]

Dear EViews-Team,

I'd like to pose you some small questions concerning the Johansen test procedure for finding cointegration. Here are my questions:

1. Concerning the choice of the lag length: in the literature one can find two ways of selecting the optimal lag length.
First, as Johansen suggested, one can set up the unresticted VAR and (under the restriction of no autocorrelation in the residuals) use the Lag Length Criteria of EViews to find the optimal number of lags (according to AIC,BIC, HQ,...).
Or, one could directly fit a VECM and search for the optimal number of lags here (also under the restriction of no autocorrelation in the residuals and according to AIC,BIC, HQ,..).

-> The second way seems more straightforward to me, as the results from AIC,BIC, HQ etc. obviously differ between the unrestricted VAR and the VECM. And since I will fit a VECM later on, it makes more sense to me to reassure, that for this model the adequate number of lags is chosen and that there is no autocorrelation in the residuals.
Or am I wrong?

2. For my work I have a quite large number of models to test for cointegration. I found that in most empirical applications, the optimal no. of lags is chosen (as described in question 1) before testing for the number of cointegrating relations. In my case it would be very much time saving if I could reverse the whole testing procedure, that is: I would first look for the number of cointegration relations of a model by varying the number of lags (say from 1-5 included lags) and then set up the model with the number of lags where I found significant cointegration.
-> In your eyes, do you think that one couls reverse the testing procedure as just described?

I'd be rally glad, if you found the time to answer my question!

Many greetings,

jomike