Izvorni znanstveni članak
https://doi.org/10.2478/bsrj-2026-0023
Balancing Return Maximization and Downside Risk Control in Emerging Markets: A Machine Learning and Deep Learning Approach to Portfolio Optimization
Minh Duc Nguyen
orcid.org/0000-0003-4702-0888
; Department of Economic Information Systems, University of Economics Hue University, Hue, Vietnam
Sažetak
Background: Emerging markets pose challenges for portfolio optimization due to volatility and the non-linear behavior of stock returns. Traditional methods often fail to capture these complexities, highlighting the need for advanced forecasting combined with robust optimization. Objectives: This study develops a framework that integrates machine learning and deep learning forecasting models with optimization strategies to improve stock return prediction and portfolio performance. Methods/Approach: Using VN-100 index stocks, four models, including Support Vector Regression (SVR), Random Forest (RF), Long Short-Term Memory (LSTM), and One-Dimensional Convolutional Neural Networks (1D-CNN), were applied to forecast returns. The top 30 stocks per model were selected. Eight portfolios were constructed by combining these forecasts with two optimization methods: Mean-Variance with Forecasting (MVF) for return maximization and Maximum Drawdown Portfolio (MDP) for downside risk control. Performance was tested over three years using net asset value trajectories and monthly excess returns. Results: MVF portfolios achieved higher mean returns with lower volatility compared to MDP. The RF-MVF portfolio delivered the strongest risk-adjusted performance, while LSTM-based stock sets enhanced portfolio outcomes. Conclusions: Integrating advanced forecasting with optimization improves portfolio construction in volatile markets. The findings show that balancing return maximization and downside risk control yields superior risk-adjusted returns, offering practical guidance for investors in emerging economies.
Ključne riječi
stock return forecasting; portfolio optimization; stock selection; machine learning; deep learning; VN-100
Hrčak ID:
353560
URI
Datum izdavanja:
24.9.2026.
Posjeta: 0 *