Persistency in a VAR model

VAR, SVAR, BVAR, TVP-VAR, etc, I have a feeling that the persistency of the impulse responses will not change much irrespective of the model, even if impulse responses in the first period of the shock changes, for example, from negative to positive (or vise versa) depending on the model of choice. I mean persistency here is a feature of the data, not the model, right?