Faithin Omaha eBook Collection

Download or read book Essays on Identification in Linear IV Models written by Chen Zhang and published by . This book was released on 2021 with total page pages. Available in PDF, EPUB and Kindle. Book excerpt: This dissertation includes two chapters on identifications in linear IV models. Chapter 1: the exogeneity assumption in instrumental variable (IV) regressions is too strong in some empirical applications. A small deviation from the assumption would lead many classical tests to have distorted asymptotic sizes. Thus, the inferences derived from the exogeneity assumption can be subject to critique. For their reason, this paper introduces a new inference method for the structural parameter in linear IV regressions. The method is robust to local deviations of the exogeneity assumption and as powerful as the Wald test when exogeneity holds. To do so, the paper introduces a partial identification approach that only assumes that the covariance between the instruments and the unobservables is in a prespecified set. Based on this assumption, the paper proposes a cone-based (CB) test and shows that (i) the test has correct asymptotic size, and (ii) the test is asymptotically equivalent to the Wald test when the identified set shrinks to a singleton at a rate faster than root n. The paper then examines the linear IV regression model in Conely, Hansen, and Rossi (2012) and shows that the confidence interval constructed by the CB test is asymptotically smaller than the one in that paper. Finally, the paper demonstrates the performance of the CB test through Monte Carlo studies and two empirical applications. Chapter 2: weak IV is often a great concern in empirical research. While there are many weak IV robust inference methods for testing hypothesis about the structural parameters in the linear IV models, there is no clear power ranking among these methods. This chapter introduces a new conditional likelihood ratio (CLR) type test in linear IV regression models. In the chapter, we show that the proposed test has correct asymptotic size in the parameter space allowing for Kronecker Product structure covariance matrices; the test is asymptotically similar and rotationally invariant; the test is nearly uniformly most powerful among a class of invariant similar tests in the parameter space that allows for Kronecker product covariance matrices. In Monte Carlo studies, we show that the test: (i) performs very close to Moreira's CLR test under homoskedasticity; (ii) the test has correct null rejection probability in a larger parameter space that allows for Kronecker product covariance matrix while the original Moreira's CLR test overrejects. (iii) The test performs very close to the heteroskedasticity-- robust AR test under weak IV, but it outperforms the heteroskedasticity-- robust AR test when the model is overidentified and identification is strong.