A Dynamic Programming Approach to Educational Choice: Theory, Econometrics, and Practice

Moshe Buchinsky, University of California, Los Angeles
Douglas M. McKee, University of California, Los Angeles

Dynamic choice models, and in particular discrete choice dynamic programming (DDP) models, have become an increasingly popular way for social scientists to model individual behavior such as educational investment over the life cycle taking into account resource limitations, future consequences of current decisions and uncertainty about the future. But DDP models are difficult to estimate due to their substantial computational requirements and their seemingly extreme data requirements. In this paper we use simulated data as well as the National Longitudinal Survey of Youth (NLSY) to identify aspects of the DDP that can be relaxed in practice. We also examine how much information across time per individual is ``enough'' to estimate DDP parameters. Finally, we present an application of an extensive DDP model of educational investment using data from the NLSY and simulate several policy changes and their welfare implications.