A0608
Title: Asymptotic efficiency for two-stage conditional M-estimators
Authors: ELysee Aristide Houndetoungan - Cy Cergy Paris Universite (France) [presenting]
Kassimou Abdoul Haki Maoude - University of Montreal (Canada)
Abstract: Two-step estimation strategies are popular for dealing with many issues in applied econometrics, such as endogeneity, missing data, and latent variables. These approaches consist in estimating a control, which is used in a second estimation strategy. The asymptotic properties of the second stage estimator depend on the uncertainty in the first stage and can be challenging to establish. We study these properties for a general class of first-stage estimation procedures, where the second-stage estimator is an M-estimator. For example, our approach allows a Bayesian estimator or nonparametric estimator at the first stage. We develop a straightforward approach to consistently estimate the variance of the estimator at the second stage. Our method is computationally more attractive than the Bootstrap method as we do not perform several estimations.
