Economics Webinar - Nonparametric Estimation of Finite-Mixture Models of Dynamic Discrete Choices

This paper provides new identification results for mixtures of stationary Markov processes. Our results complete and sharpen those of Kasahara and Shimotsu (2009). Our approach is constructive and, contrary to Hu and Shum (2012), does not require monotonicity conditions on component distributions. Identification is achieved from knowledge of the cross-sectional distribution of as little as three effective time-series observations. Nonparametric maximum likelihood is considered for the purpose of estimation and inference. Implementation via the EM algorithm is straightforward and evaluated in a Monte Carlo exercise.