Judging Instrument Relevance in Instrumental Variables Estimation

Abstract

Recent research has emphasized the poor finite-sample performance of the instrumental variables (IV) estimator when the instruments are weakly correlated with the regressors. We show how the canonical correlations between regressors and instruments can provide a measure of instrument relevance in the general multiple-instrument-multiple-regressor case. However, our simulation results indicate that any such relevance measure probably has little practical merit, as its use may actually exacerbate the poor finite-sample properties of the IV estimator.