A Comparison of the 2SLS and Modified LIML Estimators for a Nonlinear Econometric Model When the Degree of Overidentifiability is large

Abstract

This paper studies a comparison of the estimation results by 2SLS and modified LIML estimators and the simulation performances of a nonlinear simulataneous equations model when the degree of overidentifiability (L) is large. This model consists of 16 behavioral equations, 13 identities, a 20 sample sizes and 28 predetermined variables. The MF-LIML estimator appears to perform better than the F-LIML estimator in terms of the Hauseman test, and better than the 2SLS estimator in terms of the root mean squared errors ((RMSEs) and Theil's inequality coefficients (TICs) during both sample and postsample periods. We concluded that the modified LIML estimator is practically better than 2SLS estimator using a nonlinear simultaneous equations model when the value of L is large.