Econometrics provides the statistical tools for estimating economic relationships from real world data. These estimates are used to test economic theories and for policy analysis. But econometric practice also tends to impose strong assumptions as a prelude to estimation. The objective of this project is to improve current practice by developing econometric methods that require less restrictive assumptions and, for problems where restrictive assumptions are inescapable, measuring the trade-offs between estimability and the strength of the assumptions. The specific econometric methods studied by this project involve analysis of dynamic choice. The usual approach to dynamic decision making is backwards recursion, i.e., the agent should begin by solving each of the terminal problems that he might face and then work backwards to determine the optimal action today. But direct analysis of dynamic choice by backward recursion poses formidable modelling and computational problems. This project develops an exciting new approach to dynamic choice that does not require restrictive assumptions about the future. The project shows that inference can be based on a relatively, simple, partly-reduced form static representation of backwards recursion. This approach is applied to versions of the life- cycle consumption problem, repeated games and sequential sampling. The methods are used in a major empirical study of schooling and career choice.
