B0321
Title: Semiparametric efficiency in deep instrumental variable models
Authors: Zuofeng Shang - New Jersey Institute of Technology (United States) [presenting]
Abstract: Endogeneity is fundamentally important as many empirical applications may suffer from the omission of explanatory variables, measurement error, or simultaneous causality. Recently, researchers propose a Deep Instrumental Variable (IV) framework based on deep neural networks to address endogeneity, demonstrating superior performances to existing approaches. We aim to understand the empirical success of the Deep IV theoretically. Specifically, we consider a two-stage estimator using deep neural networks in the linear instrumental variables model. By imposing a latent structural assumption on the reduced form equation between endogenous variables and instrumental variables, the first-stage estimator can automatically capture this latent structure and converge to the optimal instruments at the minimax optimal rate, which is free of the dimension of instrumental variables. Given that the network architectures are well chosen, we further show that the second-stage estimator achieves the semiparametric efficiency bound. In comparison with classical methods, the deep IV method does not require an explicit functional form and has a faster convergence rate, which is more computationally tractable in modeling the interactions between IVs. Simulation experiments on synthetic data and a real-world application will be provided.