The Evaluation of Robust Outlier Detection Procedure in Bilinear (1,0,1,1) Model
Abstract
The existence of outliers in bilinear time series model will causedistortion in parameter estimation, thus the existence must be detected before the next step is taken. In the outlier detection process, bootstrap method is commonly used to calculate the mean and variance magnitude of outlier effect. To improve the efficiency of detection process, this study proposes three robust estimators,namely, MOM to calculate the mean magnitude of outlier effect, while MADn and Tn to calculate the variance magnitude of outlier effect. Next, the effectiveness of the detection procedures was evaluated based on the probability of outlier detection obtained from simulation studies, focusing on two types of outlier which are often found in bilinear data i.e. additional outlier (AO) and innovational outlier (IO). The findings revealed that MOMTn with bootstrap procedure performs the best, followed by MOMMADn with bootstrap.