Econometric essays on identification and inference in the simultaneous equations model with weak instruments

The dissertation consists in the two main articles which investigate so-called 'weak identification problem' (equivalently, 'weak instrument problem'), a recently blooming research topic, in the simultaneous equation model. It is known that if instruments are weak, then usual IV estimators, for example, k-class estimators, are all incredible. Viewing the fact that instrument variable (IV) estimation is a central approach in the simultaneous equation model, the weak identification problem is an urgent issue. Phillips (2002) surveys the problem as one of six unsolved open questions in current econometrics with top priority.; "Optimal inference with weak instruments" suggests a novel estimation method to handle the weak instrument problem based on classical approach. Using robust covariance matrix estimator whose asymptotic bias is corrected, we suggest "tri-angular GLS" and studies bound identification suggested by Zellner (1972) and Learner (1981).; "Bayesian reduced rank regression in SEMs with weak identification" investigates a Bayesian simultaneous equation model. Gao and Lahiri (1999), using simulation, show indispensable handicaps of reduced rank regression adopted by Kleibergen and Van Dijk (1998). Our study analytically shows that reduced rank regression' rather aggravates the identification problem.