Some Effects of Incorrect Specification on the Small-Sample Properties of Several Simultaneous-Equation Estimators

This is the authors' abstract. We don't add key points for this paper.

KNOWLEDGE ABOUT the small-sample properties of estimators of simultaneous equations has been confined mainly to models which are correctly specified.2 This paper provides some sampling experiment results exploring the effects of certain misspecifications on several simultaneous equation estimators. The main type of misspecification investigated is the incorrect exclusion of variables from certain equations or from the model itself. Such misspecifications could arise in practice from use of partial or incomplete economic theory which overlooks some aspects of the generation of economic data. The need to impose a prior restrictions on the structural equations to obtain identification might also lead to the incorrect exclusion of variables from equations where theory suggests that they do not play important roles. Section 2 of this paper reports experiments where important non-zero coefficients were specified to be zero in the structure estimated. Section 3 investigates the specification of small coefficients to be zero to obtain identification. In Section 4 the omission of an exogenous variable from the estimated structure is considered. Section 5 takes up the related aspect of omitting an equation. Finally, Sections 6 and 7 investigate the effects of failing to specify as zero all the structural coefficients for which such a specification would be valid. Although this is not really a misspecification, it is of interest here since the danger of incorrectly specifying coefficients to be zero might lead investigators to be over-cautious in their specifications. The results of these sections indicate that such failures of specification may