A comparative study of deep learning algorithms in univariate and multivariate forecasting of the Malaysian stock market

• Main Authors: Mohd. Ridzuan Ab. Khalil, Azuraliza Abu Bakar
• Format: Article
• Language: English
• Published: Universiti Kebangsaan Malaysia 2023
• Online Access: http://journalarticle.ukm.my/21934/1/ST%2022.pdf; http://journalarticle.ukm.my/21934/; http://www.ukm.my/jsm/index.html
• Institution: Universiti Kebangsaan Malaysia

As part of a financial institution, the stock market has been an essential factor in the growth and stability of the national economy. Investment in the stock market is risky because of its price complexity and unpredictable nature. Deep learning is an emerging approach in stock market prediction modeling that can learn the non-linearity and complexity of stock market data. To date, not much study on stock market prediction in Malaysia employs the deep learning prediction model, especially in handling univariate and multivariate data. This study aims to develop a univariate and multivariate stock market forecasting model using three deep learning algorithms and compare the performance of those models. The algorithm intends to predict the close price of the Malaysian stock market using the Axiata Group Berhad and Petronas Gas Berhad from Bursa Malaysia, listed in Kuala Lumpur Composite Index (KLCI) datasets. Three deep learning algorithms, Multilayer Perceptron (MLP), Convolutional Neural Network (CNN), and Long Short-Term Memory (LSTM), are used to develop the prediction model. The deep learning models achieved the highest accuracy and outperformed the baseline models in short and long-term forecasts. It also shows that LSTM possessed the best deep learning model for the Malaysian stock market, achieving the lowest prediction error among the other models. Deep learning demonstrates the ability to handle univariate and multivariate data in preserving important information, thus forecasting the stock market. This finding is relatively significant as deep learning works well with high-dimensional datasets.