Sufficient Statistics for Markovian Feedback Processes and Unobserved Heterogeneity in Dynamic Panel Logit Models

Abstract: In this paper, we examine identification in a dynamic panel logit model with state dependence, first-order Markov feedback processes, and individual unobserved heterogeneity by introducing sufficient statistics for the feedback process and unobserved heterogeneity. If a sequentially exogenous discrete covariate follows a first-order Markov process, identification of the coefficient on the covariate via conditional likelihood is infeasible, whereas identification of the coefficient on the lagged dependent variable is feasible when there are at least three periods after the initial-condition period. If the feedback depends only on the lagged dependent variable, the coefficient on the covariate is identified with at least two periods, and the coefficient on the lagged dependent variable is identified with at least three periods.