Machine Learning Return Prediction for Enhanced Investment Portfolio Analysis in Emerging Markets

This study addresses the challenge of improving investment portfolio performance in emerging financial markets, where high volatility and structural instability often limit the effectiveness of traditional forecasting approaches. The main objective is to enhance stock return prediction and portfolio allocation by applying advanced machine learning techniques to the Johannesburg Stock Exchange, Africa’s largest and most liquid equity market. The Johannesburg Stock Exchange is selected due to its representative role among emerging markets and its exposure to sectoral concentration, market inefficiencies, and macroeconomic shocks. The analysis covers the period from 2004 to 2024, allowing the model to capture multiple market cycles, including periods of stress and recovery. The empirical analysis is based on daily stock price and trading volume data for nineteen highly liquid firms, complemented by firm-level financial indicators obtained from established financial databases. The research employs a recurrent neural network framework designed for sequential data, incorporating both market-based indicators and firm fundamentals, alongside rigorous data preprocessing and rolling-window validation. The findings confirm that the proposed hybrid modeling approach improves return predictability and leads to superior risk-adjusted portfolio performance compared with conventional benchmark strategies. In particular, portfolios constructed using the model exhibit higher cumulative returns, improved risk–return trade-offs, and reduced downside risk during volatile periods. The study demonstrates clear scientific novelty by integrating diverse financial information within a unified predictive framework tailored to emerging markets. The results have practical relevance for portfolio managers, institutional investors, and policymakers seeking data-driven tools for investment decision-making in volatile market environments.