A1252
Title: Bootstrap-based inference for weak instruments in IV regression in finance
Authors: Yongdeng Xu - Cardiff University (United Kingdom) [presenting]
Abstract: Econometric work on weak instruments typically analyzes a worst-case endogeneity setting (rho =1) with symmetric two-sided tests, under which 2SLS t-tests over-reject. In empirical finance, endogeneity is usually small to moderate; standard t-tests are instead conservative and, crucially, their sampling distribution is skewed, with rejections concentrated in one tail. Motivated by this gap, restricted-bootstrap t-tests that impose the null during resampling are developed. In extensive Monte Carlo experiments, a restricted wild efficient bootstrap attains nominal size and correct tail behavior across a wide range of instrument strengths and endogeneity levels, addressing both conservativeness and asymmetry. The approach also extends the VtF idea of a prior study beyond the just-identified case by accommodating overidentified models with multiple instruments and delivering valid one- and two-sided inference. For practice, unified critical-value tables indexed only by the first-stage F and the endogeneity level (rho) are provided, together with MATLAB/Stata code. These tools are drop-in replacements for standard t-tests, enabling reliable 2SLS inference.
