Should We Go One Step Further? ? Accurate Comparison of One-step and Two-step Procedures in a GMM Framework

University Of California-San Diego, La Jolla CA

Investigators

Abstract

This award funds a project that develops new knowledge about the statistical properties of methods that are widely used to test hypotheses developed from economic theory. The overall goal is to give researchers who use these methods new tools for understanding which method to choose for their particular application. Employing these tools could lead to more precise and reliable conclusions. The project advances science by improving the methods used for data analysis. Questions about the relative efficiency of estimators and tests are fundamental in econometrics. In the popular Generalized Method of Moments (GMM) setting, it is standard practice to use a two-step procedure to improve efficiency. The two-step procedure requires the estimation of a weighting matrix. Standard asymptotic theory predicts that the estimator error in the weighting matrix is asymptotically negligible, and therefore the two step procedure outperforms the one step procedure in large samples. However, this asymptotic result, while elegant and convenient, ignores possible estimation uncertainty in the weighting matrix. This uncertainty can be very high. The PI plans to develop a new asymptotic framework that fully accounts for estimation uncertainty. He will compare the performance of the one step and two step GMM procedures, as well as other procedures, under this framework. The PI plans to consider both time series and spatial settings as well as parametric, nonparametric, and semiparametric models. The results should help applied researchers choose the most efficient statistical procedures for their specific problems.