Assessing the strength of many instruments with the first-stage F and Cragg-Donald statistics
This paper investigates the behavior of Stock and Yogo (2005)'s first-stage F statistic and the Cragg-Donald statistic (Cragg and Donald, 1993) when the number of instruments and the sample size go to infinity in a comparable magnitude. Our theory shows that the first-stage F test is oversized for detecting many weak instruments. We next propose an asymptotically valid correction of the F statistic for testing weakness of instruments. The theory is also used to construct confidence intervals for the strength of instruments. As for the Cragg-Donald statistic, we obtain an asymptotically valid correction in the case of two endogenous variables. Monte Carlo experiments demonstrate the satisfactory performance of the proposed methods in both situations of a single and multiple endogenous variables. The usefulness of the proposed tests is illustrated by an analysis of the returns to education data in Angrist and Keueger (1991).
PDF Abstract