Partially adaptive estimation of nonlinear models via a normal mixture

Abstract

This paper extends the partially adaptive method Phillips (1994) provided for linear models to nonlinear models. Asymptotic results are established under conditions general enough they cover both cross-sectional and time series applications. The sampling efficiency of the new estimator is illustrated in a small Monte Carlo study in which the parameters of an autoregressive moving average are estimated. The study indicates that, for non-normal distributions, the new estimator improves on the nonlinear least squares estimator in terms of efficiency.

Keywords:

• ARMA process
• nonlinear regression model
• quasi maximum likelihood
• JEL Classifications:C13,C20