Leveraging LSTM Predictions for Enhanced Portfolio Allocation with Markowitz Mean-Variance Optimization

This research investigates the application of Long Short-Term Memory (LSTM) networks for predicting expected returns and integrating these predictions into the Markowitz Mean-Variance Optimization (MVO) framework. The study utilized historical data from eight Indonesian stocks: BBCA, BBRI, TLKM, EXCL, UNVR, ICBP, ASII, and SMGR. The dataset covered the period from 2018 to 2024. The LSTM model was employed to predict cumulative returns over a 90-day horizon, and its performance was compared to the Exponentially Weighted Moving Average (EWMA) method. The findings indicate that LSTM achieved lower Root Mean Squared Error (RMSE) than EWMA for four stocks (BBCA, BBRI, UNVR, ICBP), while EWMA demonstrated better performance for the remaining four stocks. MVO results revealed that LSTM-based predictions achieved an average return of 4.285%, surpassing EWMA's 1.856% but falling short of the 12.298% obtained using actual returns. These results highlight the potential of LSTM models to enhance portfolio allocation strategies.