Monte Carlo Simulation, Schueller & Spanos (eds), 2001 Balkema, Rotterdam, ISBN 90 5809 188 0
Testing Separate Dynamic Nonlinear Econometric Models
Statistics Department, Instituto Technologico Autonomo de Mexico (ITAM), Mexico City, Mexico
A. Ronald Gallant
Department of Economics, University of North Carolina, Chapel Hill NC, USA
ABSTRACT: We propose a very general procedure for contrasting two nested or non-nested dynamic nonlinear econometric models. It is based on the idea of testing the specification of the null model by using the alternative model to generate moment conditions for a method of moments estimation procedure. Particular emphasis is given to the Efficient Method of Moments (EMM), which is a simulation-based inference method that is appropriate when the null model is difficult to fit using the usual estimation procedures yet for which it is relatively easy to compute expectations by Monte Carlo simulation, by quadrature, or by analytic expressions. The procedures is illustrated when the null model corresponds to a system of three differential equations with random parameters and observable data are available for the third equation only. Data comes from a highly seasonal time series example (monthly chickenpox reports in New York City). The alternative specification is a semi-nonparametric (SNP) model. A qualitative discussion of the large sample properties of the EMM estimator, the score function, and the associated specification test under model misspecification is also included.