Capability of Hybrid Long Short-Term Memory in Stock Price Prediction: A Comprehensive Literature Review

Abstract

Stocks are financial instruments representing ownership in a company. They provide holders with rights to a portion of the company's assets and earnings. The stock market serves as a means for companies to raise capital. By selling shares to the public, companies can obtain funds needed for expansion, research and development, as well as various other investments. Though significant, predicting stock prices poses a challenge for investors due to their unpredictable nature. Stock price prediction is also an intriguing topic in finance and economics due to its potential for significant financial gains. However, manually predicting stock prices is complex and requires in-depth analysis of various factors influencing stock price movements. Moreover, human limitations in processing and interpreting information quickly can lead to prediction errors, while psychological factors such as bias and emotion can also affect investment decisions, reducing prediction objectivity and accuracy. Therefore, machine processing methods become an alternative to expedite and reduce errors in processing large amounts of data. This study attempts to review one of the commonly used prediction algorithms in time series forecasting, namely hybrid LSTM. This approach combines the LSTM model with other methods such as optimization algorithms, statistical techniques, or feature processing to enhance the accuracy of stock price prediction. The results of this literature review indicate that the hybrid LSTM method in stock price prediction shows promise in improving prediction accuracy. The use of optimization algorithms such as GA, AGA, and APSO has successfully produced models with low RMSE values, indicating minimal prediction errors. However, some methods such as LSTM-EMD and LSTM-RNN-LSTM still require further development to improve their performance.