GMM with Many Weak Moment Conditions and Nuisance Parameters: General Theory and Applications to Causal Inference

Weak identification is a common issue for many statistical problems -- for example, when instrumental variables are weakly correlated with treatment, or when proxy variables are weakly correlated with unmeasured confounders. Under weak identification, standard estimation methods, such as the generalized method of moments (GMM), can have sizeable bias in finite samples or even asymptotically. In addition, many practical settings involve a growing number of nuisance parameters, adding further complexity to the problem. In this paper, we study estimation and inference under a general nonlinear moment model with many weak moment conditions and many nuisance parameters. To obtain debiased inference for finite-dimensional target parameters, we demonstrate that Neyman orthogonality plays a stronger role than in conventional settings with strong identification. We study a general two-step debiasing estimator that allows for possibly nonparametric first-step estimation of nuisance parameters, and we establish its consistency and asymptotic normality under a many weak moment asymptotic regime. Our theory accommodates both high-dimensional moment conditions and function-valued nuisance parameters. We provide high-level assumptions for a general setting and discuss specific applications to the problems of estimation and inference with weak instruments and weak proxies.