Dennis Kristensen, UCL

"CCP and the Estimation of Nonseparable Dynamic Models"

By Dennis Kristensen, University College London 


In this paper we generalize the so-called CCP estimator of Hotz and Miller (1993) to a broader class of dynamic discrete choice (DDC) models that allow period payoff functions to be non-separable in observable and unobservable (to the econometrician) variables. Such non-separabilities are common in applied microeconomic environments and our generalized CCP estimator allows for computationally simple estimation in this class of DDC models. We first establish invertibility results between conditional choice probabilities (CCPs) and value functions and use this to derive a policy iteration mapping in our more general framework. This is used to develop a pseudo-maximum likelihood estimator of model parameters that side-step the need for solving the model. To make the inversion operational, we use Mathematical Programming with Equilibrium Constraints (MPEC), following Judd and Su (2012) and demonstrate its applicability in a series of Monte Carlo experiments that replicate the model used in Keane and Wolpin (1997).

Contact person: Bertel Schjerning

CCE organizes a weekly seminar, usually on Tuesdays from 13:00 to 14:15. These seminars are open to everyone. 
The seminar covers topics from all research fields in structural econometrics and computational economics including theoretical, empirical, methodological issues.