Hello all,

I have a multi-level unbalanced panel (i.e., banks nested in countries) which is micro (N=800; T=72, quarterly frequency). For my regression model, I apply several specifications, such as Hierarchical Linear Modeling; also, because my panels are cross-correlated, I apply a PCSE model. To control for endogeneity, GMM/IV models are not suitable in the presence of cross-sectional dependence, and from my readings, Correlated Random Effects could control to some extend for this problem.
In a second stange, I want a apply a panel VAR for my two main variables (bank stability - the dependent at the bank level and business cycle, the main covariate - country level). Doing the panel unit-root tests incorporated in Stata, I find that some of the panels are stationary, that is, I reject the null that all panels contain unit roots. Further, I test for cointegration and find that my two variables are cointegrated (all tests within Stata which support unblanced panels reject the null of no cointegration). Here are my questions:

1) Does make sense to test for cointegration if some panels are stationary? (Because my data is unbalanced, one cannot say that all panels are stationary; I made my panel being balanced loosing 600 IDs(banks) and it says that all panels are stationary. But this is for the sake of testing only.)

2) If cointegration exists, VECM models instead of VARs are suitable. But there is no panel VECM in Stata or user-written program, it's just Panel VAR. But there are others user-written programs such as -xtpmg- or -xtdcce2- that estimate error correction models. As far as I understand, these models are usually applied to macro data (e.g., Penn World Tables). Given the structure of my data, can I apply this models in my case?

Thank you in advance for all explanations.