I am trying to fit a 3 variable VAR model and I need to check if the qualitative features of impulse responses are approximately stable across subsamples. But I am not sure what impulse response stability is and how to check if they are stable. Can someone explain it to me please ?
You could graph the impulse responses for the different subsamples and see if they look similar, e.g. see if the response of one subsample fits into the 95% confidence interval of another subsample. That could be justified by the task formulation containing the expression qualitative features.
If you needed to assess that quantitatively, you would perhaps look directly at the models for each subsample and test whether they are very different (e.g. in terms of Kullback-Leibler distance or the like).