Improving Models Accuracy Using Kalman Filter and Holt-Winters Approaches Based on ARFIMA Models

Abstract-The analysis, modeling, and forecast of oil prices are among the most important studies related to global and local economic trends. Such studies are necessary to increase investments and reduce risks because oil prices exert a significant impact on supply and demand in global markets. In the current work, four models are proposed, namely, autoregressive fractionally integrated moving average (ARFIMA), ARFIMA with additive Holt-Winters (ARFIMA-AHW), ARFIMA with multiplicative Holt-Winters (aRFIMA-MHw), and ARFIMA with Kalman filter (ARFIMA-KF), for modeling monthly Brent crude oil prices. Accordingly, this study aims to extend the researchers' previous work by comparing the performance of the proposed statistical methods to provide an accurate individual or hybrid model for the efficient and reliable modeling of these prices. In addition, the characteristics of the optimal and most accurate method are identified to refinement the prediction outcomes of the ARFIMA model by using the Kalman filter and Holt-Winters methods in hybridization. The capabilities of these proposed models are evaluated in view of the root-meansquare error and by conducting the autoregressive conditional heteroscedasticity with Lagrange multiplier and Ljung-Box tests. This study shows that the ARFIMA (2,0.3589648,2)-KF model outperforms the other proposed models on the basis of the test results.