Generalised linear models for aggregate claims; to Tweedie or not?