Identification of Dynamic Discrete Choice Models off Short Panels

Abstract

This paper considers nonparametric identification of nonstationary dynamic discrete choice models when the agent's time horizon extends beyond the length of the data. We show conditions under which ow payoffs are identified subject to standard normalizations and, when the payoff function does not depend directly on time, show identification even when the time horizon extends beyond the length of the data. We further establish identification for a class of nonstationary dynamic discrete choice games and show how the nonstationarity of the problem can be helpful in unbundling the agent's state-specific payoffs from the expected payoffs where the expectation is taken over the actions of the agent's competitors.