Inference with Many Weak Instruments

Abstract We develop a concept of weak identification in linear instrumental variable models which the number instruments can grow at same rate or slower than sample size. propose jackknifed version classical identification-robust Anderson–Rubin (AR) test statistic. Large-sample inference based on AR is valid under heteroscedasticity and identification. The feasible this statistic uses novel variance estimator. has uniformly correct size good power properties. also pre-test for that related to property Wald Jackknife Instrumental Variable Estimator. This new with many instruments.