VAR instead of VECM?

[Vviews]

Dear all,

I have four variables in log form (sa,m,g,y) which are non-stationary but cointegrated.
When estimating a VECM (sa m g y), the model shows up strong signs of instability according
to Eviews and other packages. The questions is why (!???) but I couldn't figure it out.
As I am mainly interested in the Impulse Response Functions etc., I have estimated
a VAR (not in differences as variables are jointly stationary).
The model appears to be stable according to the standard diagnostics tests and the results are "satisfying".

According to almost all literature sources, one should estimate a VECM if variables are cointegrated.

Now, is there any issue with my approach?

Best,
V.

PS: I have attached my workfile.