How large is the class of generalized extreme value random utility models?

Abstract

The Generalized Extreme Value Model (GEV) was developed by McFadden (cf. McFadden, 1981) with the purpose of extending the Luce Model to account for interdependent utilities. While the Luce model satisfies the IIA property it has not been clear whether or not the GEV class implies theoretical restrictions on the choice probabilities other than those that follow from the random utility framework. The present paper extends the GEV class to the intertemporal situation and proves that the choice probabilities generated from random utility processes can be approximated arbitrarily closely by choice probabilities from an intertemporal GEV model.