Identification and Estimation of Nonstationary Dynamic Discrete Choice Models

Under common assumptions for dynamic discrete choice models with general forms of nonstationarity, we prove a novel Markovian property that allows us to bypass the state transition distribution in the identification and estimation of flow utility parameters. We show that under mild additional assumptions, the identification of the flow utility parameters amounts to the unique solution of a linear system of equations, which depends on a testable rank condition. The theoretical and empirical implications of the rank condition in our model are thoroughly discussed. We propose a three-step conditional-choice-probability-based semiparametric estimator that circumvents estimation of and simulating from the state transition distribution. Simulation experiments show that the rank condition is easy to fulfill, and our estimator gives comparable finite sample performance as the Hotz-Miller estimators but is computationally much less demanding. The asymptotic distribution of the estimator is provided, and the sensitivity of the estimator to the key additional assumption is also examined.