AANIV: Stata module to compute unbiased IV regression

The conventional instrumental variable (IV) or two-stage least squares (2SLS) estimator may be badly biased in overidentified models with weak instruments. While the 2SLS estimator performs better in the exactly identified case, in the sense that its median rapidly approaches the true value as instruments become strong, it has no first moment. That is, the estimator has no mean, and no finite higher moments, either. For papers on the finite-sample properties of IV estimators, see e.g. Phillips (1980), Phillips (1983), and Hillier (2006), and references therein. The estimator implemented in aaniv is an unbiased IV estimator for a special case of an exactly identified model with one endogenous variable and one instrument, from Andrews and Armstrong (2017), which relies on a sign restriction in the first stage. That is, if we know the sign of the effect of of the instrument on the endogenous treatment variable, we can construct an unbiased estimate of the effect of treatment on the treated.