Deep Learning-Based Portfolio Management: Empirical Study of Listed Shares on the Indonesian Stock Exchange

A lack of accurate and effective information causes difficulty in choosing the right shares. Therefore, an accurate analysis model is needed for stock selection management. One algorithm model often used is Deep Learning Long Short Term Memory, which utilizes artificial intelligence (AI) and machine learning technology. This research aims to prove whether the Deep Learning Long Short Term Memory (LSTM) prediction results are close to real prices. The method used is deep learning long short-term memory, which uses six well-established sectors in the Indonesian stock market: large banks, medium banks, digital banks, property, pharmaceuticals, and consumers. The daily stock prices used were taken from the Yahoo Finance API from January 1 2021 – June 6 2023, to predict stock prices on June 16 2023. A trading window simulation for each sector was carried out for one week, starting on June 7, 2023, and liquidated on the 16th. June 2023. The simulation was carried out for one hundred thousand USD invested at the edge of the estimated efficient frontier, optimizing the risk and return tradeoff. The results show that the predicted and actual values are not significantly different, meaning that LSTM can be relied on to predict stock prices. Furthermore, including fundamentals and technicalities and further refinement of LSTM hyperparameters, coupled with genetic algorithms, can increase prediction precision.