GMM estimation and inference
Endogeneity is a common phenomenon in applied econometrics and generally prevents a causal interpretation of ordinary least squares regressions. The availability of valid instruments can solve this problem. Instrumental variables are often used to estimate causal effects. While there are often lasting debates about the exogeneity of instruments, the relevance of the instruments is observable and thus the strength of the identification is unquestionable. The shortcomings of basic econometric techniques are also well understood. Many instruments and/or weak identification can affect the asymptotic properties of the usual 2SLS or two-step GMM estimator. Some identification robust techniques have been proposed in the recent years – among them the continuously updated estimator (CUE) and an appropriate variance estimator. In this project, I contribute to the literature in two fields. First, I show that the finite-sample properties of a recently proposed variance estimator for the CUE depend on the definition of the weight matrix. Second, I propose a modification of the CUE, which is consistent under usual and many weak moment asymptotics, and has a markedly smaller dispersion in Monte Carlo simulations. My application in political economy illustrates the importance of this issue in practice. Both contributions are more relevant in small samples, which make them particularly valuable for macroeconomic applications. Collaboration partner in this project is Prof. Frank Windmeijer, PhD (University of Bristol). This project is supported by the the Fritz Thyssen Stiftung.