Abstract
The optimal prediction or forecasting of time series values from the observations required many things such as checking the identification accuracy, model diagnosis, and data free from violations (outliers, for instance). Therefore, the researchers are always wondering if the used model or the supported method is sufficient to represent the data or there are more information that can be provided and probable increasing of precision as a consequence in the forecasting. This paper is an attempt to propose a new hybrid model building that can be denoted by AR-Holt (p+5). Also, suggest a new algorithm to estimate the parameters of this new hybrid model with its forecasting for inside and outside the series. Furthermore, the comparison has been done between this new hybrid model with AR(p) model which was identified as well as its parameters were estimated by many traditional methods which are Yule-Walker, Burg, robust RA, LS, Mcov and LMS methods for contaminated time series data. Simulation experiments have been conducted with different levels of contamination (p=0, 0.05, 0.15) to evaluate the superior of the performance of this new model according to different sample sizes (n=30, 70, 150). A real data application of the barley crops in Iraq is taken into consideration.