Applied Time Series Forecasting Using Classical and Neural Models: Evidence From Saudi Basic Industries Corporation

The paper compares the predictive accuracy of the conventional time series model, Autoregressive Integrated Moving Average (ARIMA), Exponential Trend Smoothing (ETS), and neural augmented time series models: Hybrid Models, Multi-Layer Perception (MLP), and Neural Network Autoregression (NNAR) based on Saudi Basic Industries Corporation (SABIC) stock prices over different time horizons. Information used for this study was received through the Yahoo Finance website, reflecting years of one, three, and five duration, respectively, from 01/01/2019 onwards until 2024. The primary measure (besides other measures) used to gauge the forecasting precision of the models was Root Mean Squared Error (RMSE). The study was executed with RStudio version 4.3.2 while utilizing precise libraries for undertaking the required calculations. The results of the research indicated that NNAR outperformed all other models for all time horizons, reflecting its prowess in forecasting SABIC’s stock prices for the Saudi Arabian market. But the research highlights the need to take into account its limitations before using the correct forecasting method.