Sensitivity analysis using approximate moment condition models

We consider inference in models defined by approximate moment conditions. show that near?optimal confidence intervals (CIs) can be formed taking a generalized method of moments (GMM) estimator, and adding subtracting the standard error times critical value takes into account potential bias from misspecification In order to optimize performance under misspecification, weighting matrix for this GMM estimator and, therefore, differs one is optimal correct specification. To formally near?optimality these CIs, we develop asymptotic efficiency bounds locally misspecified setting. These may independent interest, due their implications possibility using selection procedures when conducting condition models. apply our methods an empirical application automobile demand, adjusting shrink CIs factor 3 or more.