Stock Investment Portfolio Optimization Using Mean-Variance Model Based on Stock Price Prediction with Long-Short Term Memory

Abstract

Stock investment in the technology sector in Indonesia offers high potential returns. However, like any other investment instruments, the associated risks cannot be overlooked. Therefore, an appropriate portfolio optimization strategy is needed to enable investors to achieve optimal returns while managing risk. In this study, the author combines stock price prediction approaches with portfolio optimization methods to construct an efficient portfolio. The Long-Short Term Memory (LSTM) model is used to predict daily closing stock prices, with model performance evaluated using Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE) metrics. An optimal LSTM model is obtained with a batch size hyperparameter of 16 for ISAT, MTDL, MLPT, and EDGE stocks, and a batch size of 32 for DCII stock. For all stocks, the average prediction error from the actual values falls within the range of 1.53% ≤ MAPE ≤ 3.52%. The optimal portfolio is constructed using the Mean-Variance risk aversion model to maximize expected returns while considering risk. The resulting optimal portfolio composition consists of a weight allocation of 19.7% for ISAT stock, 36.8% for MTDL stock, 34.8% for MLPT stock, 3.6% for EDGE stock, and 15% for DCII stock. This portfolio yields an expected portfolio return of 0.001249 and a portfolio variance of 0.000311.