Hybrid Statistical–Machine Learning Models for Forecasting the Saudi Stock Index (TASI): An Empirical Study

Model accuracy is a central concern in time series forecasting, particularly within applied statistical research where model performance directly influences the reliability of inference and decision-making. Classical statistical methods, including Autoregressive Integrated Moving Average (ARIMA) and Exponential Smoothing (ETS) models, have been widely used due to their clear structure and their proven capability to capture linear patterns in time-series data. However, these methods frequently exhibit limitations when confronted with non-linearity, structural breaks, and complex dynamic patterns often present in financial time series. In response to these constraints, hybrid modelling frameworks that integrate classical statistical methodologies with machine learning techniques specifically Artificial Neural Networks (ANNs) have gained increasing attention for their potential to enhance prediction accuracy. This study conducts a comparative analysis between traditional and hybrid forecasting models in predicting the Saudi Stock Index (TASI), using weekly data obtained from the Saudi Stock Exchange (Tadawul) for the period from January 28, 2024, to August 2, 2025. Empirical findings demonstrate that hybrid models substantially outperform individual approaches in terms of forecast precision and error reduction, highlighting their suitability for volatile market environments. The results further reinforce the argument that hybrid methods offer a more robust framework for time series forecasting within applied statistical contexts.