Survey on statistical inferences in weakly-identified instrumental variable models (150,00 руб.)

Anna Mikusheva APPLIED ECONOMETRICS ПРИКЛАДНАЯ ЭКОНОМЕТРИКА № 29 (1) 2013 Anna Mikusheva in weakly-identified instrumental variable models This paper provides a brief review of the current state of knowledge on the topic of weaklyidentified instrumental variable regression. <...> We describe the essence of the problem of weak identification, possible methods for detecting it in applied work as well as methods robust to weak identification. <...> Introduction that is, correlated with the error term in the structural equation. <...> This arises in many practically relevant situations when the correlation between X and Y does not correctly reflect the causation from X to Y, because, for example, some variables that influence both X and Y are omitted from the regression, or because there is reverse causality from Y to X. The idea behind IV regression is to use some exogenous variables Z (that is, variables not correlated with the error term) to disentangle some part of the variation in X that is exogenous and to estimate the causal effect of this part on Y using classical methods. <...> The typical requirements for the validity of the IV regression are twofold: the instruments Z are I required to be exogenous (not correlated with the error term) and relevant. <...> As we will see below the problem of weak identification manifests itself when an IV estimator is very biased and when classical IV inferences are unreliable. <...> To fix the ideas let us assume that one wants to estimate and make inferences about a k1-dimensional coefficient b in the regression YX Weii ii bg  +  + where Xi p1-dimensional regressors Wi =, is a k1 regressor potentially correlated with the error term ei itself. <...> We assume that are exogenous and that the coefficient g is not of interest by and ei , the OLS estimator of coefficient b is biased and asymptotically inconsistent, while all statistical inferences using it, such as OLS confidence sets and OLS tests based on t-statistics provide coverage (size) that is asymptotically wrong. <...> Theory and methodology Теория и методология 117 nstrumental variable (IV) regression is a very popular way of estimating the causal effect of a potentially endogenous regressor X on variable Y. <...>