I don't think that's quite true. You create a model not necessarily to predict, but to understand. A good model - which is by definition not intended to be ex

I don't think that's quite true. You create a model not necessarily to predict, but to understand. A good model - which is by definition not intended to be exactly right - helps one to come to insights that can in turn help to come up with predictions that can be tested.