VARs-how to deal with inverted AR roots outside unit circle

[huugh]

Hi,
any tips on how to deal with inverted AR roots of VAR lying outside the unit circle in the multivariate VAR (specifically, it is the application of Toda-Yamamoto procedure, i.e. the VAR is estimated with I(1) variables in levels)?
I tried to vary the number of lags (both add and remove) but to no avail. I was hoping I could resolve the problem with higher number of lags, however I have run out of available degrees of freedom in the model before the AR roots problem was solved.

It does not seem there is a structural break in the series.
Any tips?