Comparison between adaptive filtering methods and Box-Jenkins methods for mixed models in time series using simulation

Abstract

This study focuses on forecasting by comparing adaptive filtering and Box-Jenkins methods to predict future values of a lower-order mixed model (ARMA), utilizing simulation techniques for three sample sizes: small, medium, and large. Using the R statistical programming language, the most important conclusions were reached, including that adaptive filtering (MW) is superior to Box-Jenkins (BJ) methods, as it includes the former's ability to handle non-stationary time series, as the essential steps of this method are neither dependent on nor affected by simple changes in the data. Furthermore, Box-Jenkins (BJ) methods are superior to (MW) methods for forecasting before using forecast control. Among the most important recommendations reached by the researcher was the generalization of adaptive filtering to include seasonal and multivariate time series models