Two-stage-least-squares (2SLS) estimates are biased towards the probability
limit of OLS estimates. This bias glows with the degree of over-identifica
tion and can generate highly misleading results. In this paper we propose t
wo simple alternatives to 2SLS and limited-information-maximum-liketihood (
LIML) estimators for models with more instruments than endogenous regressor
s. These estimators can be interpreted as instrumental variables procedures
using an instrument that is independent of disturbances even in finite sam
ples. Independence is achieved by using a 'leave-one-out' jackknife-type fi
tted value in place of the usual first stage equation. The new estimators a
re first order equivalent to 2SLS but with finite-sample properties superio
r, in terms of bias and coverage rate of confidence intervals, compared to
those of 2SLS and similar to those of LIML, when there are many instruments
. Copyright (C) 1999 John Wiley & Sons, Ltd.