Three essays on nonlinear time series econometrics

Abstract

This dissertation is comprised of three chapters on time series econometrics. The first chapter investigates the asymptotic and finite sample properties of the lag selection procedure based on the final prediction error (FPE) when the additive structure is a priori known in the nonparametric autoregression. The simulation results show a significant improvement in the finite sample performance over the case of unrestricted estimation. The second chapter proposes a model-free cotrending rank selection procedure, in the presence of both stochastic and nonlinear deterministic trends. Simulation results suggest good finite sample properties of the new rank selection criteria. The proposed method is then illustrated through an application of Japanese money demand function allowing for the cotrending relationship among money, income and interest rates. The third chapter develops a smooth transition GARCH model with an asymmetric transition function, which allows an asymmetric response of volatility to the size and sign of shocks, and an asymmetric transition dynamics for positive shocks and negative shocks. The model is applied to the empirical financial data: the NASDAQ index and the individual stock IBM daily returns.