Semiparametric transition models

A new semiparametric time series model is introduced – the transition (SETR) that generalizes threshold and smooth models by letting function to be of an unknown form. Estimation based on a combination (local) least squares estimations regression parameters. The asymptotic behavior for coefficient estimator SETR established, including its oracle property. Monte Carlo simulations demonstrate proposed more robust form than parametric methods precise varying estimators.