Predicting Stock Performance on the MSE Using Deep Learning Models

Abstract The Malawi Stock Exchange (MSE) is the principal market for stocks and bonds in Malawi. This study addresses a gap in market knowledge and the lack of analytical tools that enable investors to make informed decisions based on historical information. We propose the use of deep learning models as guiding tools for forecasting stock market prices. Deep learning models—hybrid LSTM-DNN, Long Short-Term Memory (LSTM), and Deep Neural Networks (DNN)—were identified from literature as suitable models for training and predicting future stock prices for the MSE. The study utilises data accumulated by the MSE from 2009 to 2023 for three counters: National Bank of Malawi (NBM), NICO Holdings (NICO), and Telecom Networks Malawi (TNM). Data from 2009–2021 is used for training, while data from 2022–2023 is used for testing. Results show that the hybrid LSTM-DNN model consistently outperforms both LSTM and DNN models. On the TNM dataset, it achieved an MSE of 0.01253, MAE of 0.002729, and an R² of 0.95019; on the NBM dataset, an MSE of 0.01253, MAE of 0.2829, and an R² of 0.99019; and on the NICO dataset, an MSE of 0.01927, MAE of 0.0483, and an R² of 0.92839, with an average training time of approximately 50 seconds. ANOVA analysis reveals significant differences in R² and training time across models, with the hybrid model demonstrating superior predictive accuracy at the cost of longer training duration. This research contributes to academic knowledge by introducing a novel approach to forecasting stock market prices for local companies in an emerging market and by centralising MSE data into a single repository for future research.