Reduced Form Estimation, Hedging against Possible Misspecification

Abstract

This paper substantiates the idea that a suitably weighted average of a restricted and an unrestricted reduced form estimator is generally superior to its constituent parts in case the simultaneous equation system is misspecified. The specification errors considered cover incompleteness of the sets of exogenous and endogenous variables and too sparsely parameterized structural form matrices. The weighted averages studied include pre-test and Stein-rule estimators. Copyright 1989 by Economics Department of the University of Pennsylvania and the Osaka University Institute of Social and Economic Research Association.