Abstract
Financial time series forecasting presents unique challenges due to the complex temporal dependencies, trends, and reversals inherent in stock prices. Long Short-Term Memory (LSTM) networks, a specialised type of recurrent neural network, have emerged as a powerful tool for modelling sequential data by integrating short-term and long-term dependencies. This research addresses the problem of optimising portfolio management on the Johannesburg Stock Exchange (JSE), where traditional models often fail to account for the volatile and sector-specific risks characteristic of emerging markets. The study aims to develop and evaluate a hybrid LSTM model that incorporates both technical indicators (e.g., moving averages, Bollinger Bands) and fundamental financial metrics (e.g., Price-to-Book ratio, Debt-to-Equity ratio) to enhance predictive accuracy and portfolio allocation.
A comprehensive methodology was employed, including the collection of historical stock data from the JSE, preprocessing steps to address missing values and heteroskedasticity, and the design of an LSTM architecture tailored to financial forecasting. The model’s performance was assessed against benchmark strategies, such as equally weighted and price-weighted portfolios, using metrics like Sharpe ratio, maximum drawdown, and Value at Risk (VaR).
Key findings indicate that the hybrid LSTM model outperforms traditional methods, achieving superior predictive accuracy and delivering improved risk-adjusted returns. The results demonstrate that combining technical and fundamental indicators enables the model to capture both short-term market trends and long-term economic factors, resulting in more robust portfolio strategies.
This study concludes that LSTM networks, when integrated with diverse financial indicators, offer significant advantages in managing portfolio risk and optimising returns in volatile markets. The findings contribute to the growing body of research on machine learning applications in finance and provide practical insights for portfolio managers navigating emerging market dynamics.