Semiparametric Transition Models

Résumé

A new semiparametric time series model- the semiparametric transition model is introduced which generalizes the threshold and smooth transition models by assuming the transition function to be unknown. We propose an estimation strategy based on the nonlinear least squares, where the transition function can be estimated separately by a general nonparametric estimator of a particular varying-coeffi cient model. The consistency and asymptotic normality for the slope estimator of the semiparametric transition model are derived and shown to be fi rst-order asymptotically independent of the nonparametric estimator of the varying-coe fficient model. Moreover, Monte Carlo simulations show that the estimation of the semiparametric transition model is more robust than the parametric estimations of the threshold and smooth transition models to the type of transition between models. (avec Pavel Cizek)