Enhancing Stock Price Forecasting with Deep Learning: Insights from the Saudi Stock Market

Investments are the financial market’s backbone, allocating resources and assets to generate future profits. One notable aspect of investing is trading stocks rep- resenting ownership in a company. Investors buy these shares hoping their value will rise, but forecasting stock prices is challenging due to the complex market dynamics and data patterns that are non-linear and have noise. Technical anal- ysis and Deep Learning (DL) have been investigated to solve the problem of stock forecasting, but rarely in the Saudi stock market. This study explores the utilization of DL models to forecast stock prices. It identifies effective technical indicators tailored to the Saudi stock market. Also, it evaluates DL architectures, specifically Recurrent Neural Networks (RNN) like Long short-term memory (LSTM), Gated recurrent units (GRU), and Bidirectional recurrent neural net- work (BiRNN) using Tadawul dataset. The findings of the study suggest that technical indicators of the Saudi stock market play a significant role. GRU and LSTM, consistently outperformed BiRNN and benchmark models. Additionally, the study shows that data distribution can specify which RNN model should be used. GRU demonstrated versatility in handling various data distributions, whereas LSTM showed superiority in managing data with slight variations, and BiRNN showed promising results with highly varied data. Finally, the proposed Ensemble Averaging (EA) model demonstrates superiority in forecasting, achiev- ing the greatest win rate and the lowest loss rate. Therefore, EA is the best performing technique in forecasting stock prices.