Do Convolutional Neural Networks outperform Linear Models? Evidence from the Saudi Stock Market

Purpose: The study investigates the effectiveness of artificial convolutional neural networks (CNNs) in predicting stock returns in the Saudi stock market and evaluates their predictive performance relative to traditional linear forecasting models commonly used in the financial literature. Methodology: The research employs daily data from the Saudi Stock Exchange covering the period from January 2009 to August 2022 and includes a sample of 116 listed stocks. Both univariate and multivariate convolutional neural network models are developed and compared with benchmark linear models. Model performance is assessed using three standard forecasting accuracy measures: mean absolute error (MAE), mean squared error (MSE), and root mean squared error (RMSE). Findings: The results indicate that convolutional neural network models significantly outperform linear models across all accuracy metrics. The superior performance of CNNs is observed consistently across the sampled firms, highlighting their robustness in capturing complex and nonlinear patterns in stock price dynamics. Unique Contribution to Theory, Practice and Policy: The study contributes to the growing literature on machine learning applications in financial forecasting by providing empirical evidence on the superiority of deep learning models in an emerging market context. From a practical perspective, the findings support the adoption of advanced artificial intelligence techniques by investors and financial institutions. From a policy standpoint, the results underscore the potential of data-driven forecasting tools to enhance market efficiency in developing economies such as Saudi Arabia.