Only one variable non-stationary?

[yanivd]

Hello,

I want to estimate a VAR model (with exogenous and seasonal variables) in which I have 4 endogenous variables (let's call them Y1, Y2,Y3 amd Y4).
Without going into specifics, I have data for 20 countries and it turns out that Y3 tests out to be stationary for only 8 of these countries (and the 1st diff. is stationary for the rest).
All other variables are always stationary - which seems odd to me because there is clear Granger causality between them. I would have differenced all variables but that really destroys many of the significant results (in the sense that the standard errors are way larger). Should I just difference this variable? (if do, it's a different model, essentially and, again, some of the results become worse). Is there any way to argue for me to use Y3 at level? (the residuals for all equations are always stationary...). Is there any paper that looks into that?
In other words, what do I do if only one (out of four) variable is nonstationary and only for some countries?

Open to any thought here...

Thanks!