Utilizing LSTM-RNN Algorithm in a Multi-Factor Model for Forecasting Investment Portfolio Returns

Portfolio risk management is a subject of great interest for fund managers and academics alike. The multi-factor method is instrumental in establishing the linear relationship between return and risk in diversified portfolios, despite its notable flaws, which are more evident in real-world scenarios. To address these limitations, machine learning algorithms have emerged as a valuable means to compensate for theoretical model shortcomings. This research examines the Ordinary Least Squares (OLS) and LSTM-RNN machine learning techniques for assessing the Fama-French five-factor model (Long Short-Term Memory Recurrent Neural Network). Hypothetical data for the Ho Chi Minh City market from January 2015 through June 2023 form the basis of this study. The FF5 model based on the LSTM-RNN algorithm outperforms the OLS model when comparing prediction error using the rolling window method and calculating the Root Mean Square Error (RMSE). This underscores the importance of investors and fund managers adopting LSTM-RNN to improve forecasting efficiency in portfolio management.