ABSTRACT

The Generalized Method of Moments (GMM) is a popular estimation
technique among econometricians which is used to fit models in
which the number of parameters is exceeded by the number of
moment conditions identifying them. Since the large sample
properties of the GMM estimators depend on whether the model is
correctly specified or not, we are interested to study the
properties of various bootstrap tests under misspecification.
Specifically, we study the properties of the standard bootstrap,
centered-bootstrap, and empirical-likelihood bootstrap. We show
that although some bootstrap estimators of the null distributions
of the Wald, likelihood ratio-type, score-type, and the J-test of
over-identifying restrictions are consistent when the model is
correctly specified, they are inconsistent under misspecification
and vice-versa. Further higher order expansions of the size
distortions of the bootstrap tests enable us to develop a general
bootstrap methodology which is highly accurate under correct
model specification and consistent under misspecification. In an
empirical study, we explore the finite sample behavior of the
bootstrap tests for correctly specified and for misspecified
versions of an over-identified dynamic panel data model.

This work is done in collaboration with Dr Brett Presnell.