Abstract
The autocorrelation function (ACF) plays an important role in the context of ARMA modeling, especially for their identification and estimation. This study considers the robust estimation of the ACF of the AR(1) model if the white noise (WN) process is non- Gaussian. Three estimators including the ordinary moment estimator and two other (robust) estimators are considered. The impacts of the deviation from normality of the WN process on those estimators in terms of bias, MSE and distribution via Monte-Carlo simulation are examined. The empirical distribution of those estimators when the errors are normal, t, Cauchy and exponential are studied. Results show that the moment estimator is more affected by the change of the white noise distribution than other considered estimators.
DOI
10.22237/jmasm/1398917340
Included in
Applied Statistics Commons, Social and Behavioral Sciences Commons, Statistical Theory Commons
