Collaborative Research: Flexible and Robust Data-driven Inference in Nonparametric and Semiparametric Econometrics

University Of California-Berkeley, Berkeley CA

Investigators

Abstract

This award funds research that will develop new statistical methods for use in analyzing economic data and testing economic theory. The goal is to develop new procedures that are more robust (i.e. less sensitive) to the specifics of their implementation. The project advances science because the new methods will give economists and other social scientists better methods for statistical testing of economic models. Drawing valid inferences from available data is important as well in a variety of policy settings. The ideal method would be flexible, simple, and robust in the sense that the results would not be overly dependent on the specific way the method is applied in any given context. While previous work in econometrics has developed new nonparametric and semiparametric methods to meet the first two goals, these methods are not very robust. Their use in practice requires choices whose effect in finite samples is not known. As a result, many empirical social scientists employ parametric methods. By developing new nonparametric and semiparametric methods that are robust, the PIs hope to develop methods that avoid misspecification bias while also being attractive for practical use. They plan to employ an alternative asymptotic theory to derive novel distributional approximations for non-and semi-parametric statistics. They plan four lines of research: (i) alternative asymptotic results and bootstrapping validity for non-linear semiparametrics; (ii) alternative asymptotic results and consistent standard-errors for linear semiparametrics, (iii) higher-order expansions for the alternative asymptotic results, and (iv) new robust nonparametric and semiparametric methods employing local-polynominal techniques.