Judging Instrument Relevance in Instrumental Variables Estimation

Recent research has emphasized the poor finite-sampleperformance of the instrumental variables (IV) estimator when the instruments are weakly correlated with the regressors. To address this problem, a variety of statistics intended to judge instrument relevance have been proposed in the literature. We show how the canonical correlations between the regressors and instruments can provide a measure of instrument relevance in the general multiple-instrument-multiple-regressor case. However, our simulation results show that the use of any such relevance measure in choosing among alternative instrument sets may actually exacerbate the poor finite-sample properties of the IV estimator