Title: An investigation of a generalized least squares estimator for non-linear time series models
Authors: Xiaoling Dou - Waseda University (Japan) [presenting]
Abstract: Ochi's estimator for the autoregressive coefficient of the first-order autoregressive model (AR(1)) uses two constants for the end points of the process. Classical estimators for AR(1) , such as the least squares estimator, Burg's estimator, and Yule-Walker estimator are obtained as special cases by choice of the constants in Ochi's estimator. By writing the first-order autoregressive conditional heteroskedastic model, ARCH(1), in a form similar to that of AR(1), we extend Ochi's estimator to ARCH(1) models. This allows the introduction of analogues of the least squares estimator, Burg's estimator and Yule-Walker estimator. We compare the relations of these with Ochi's estimator for ARCH(1) models. We then provide a simulation for AR(1) models and examine the performance of Ochi's estimator. Also, we simulate Ochi's estimator for ARCH(1) with different parameter values and sample sizes.