From a Journal of Philosophy special issue on economic methodology, Varian and Gibbard claim that a good economic model fits in one of two categories: 1) As a caricature of reality, it points out clearly a feature of reality otherwise hidden in detail, and 2) It can be used to test hypothesis when a set of consequents sufficiently close to reality are generated as a result of a set of assumptions sufficiently close to reality and a mathematical model linking the two. This second claim is quite different from the famous claim in Friedman (1953), that economic models should not be judged on assumptions at all, but rather simply on their ability to predict things we care about. Varian and Gibbard find this incomplete, since economic models are often used to test new hypotheses, about which we could say nothing off the support. But, to me, this is simply a Hume-style argument about induction – the unrefutable assumption that a model’s assumptions are “close to true” for the purposes of new estimates is no different from the unrefutable assumption that we cannot predict off the support using a Freidman model.
“Economic Models,” Gibbard and Varian, 1978