The Role of News Sentiment in Predicting the Jakarta Composite Index Using Long Short-Term Memory

This paper investigates the integration of sentiment analysis and historical data to enhance the accuracy of Jakarta Composite Index (JCI) stock return predictions using a Long Short-Term Memory (LSTM) model. The dataset spans January 3, 2014, to August 6, 2024, consisting of 2,647 daily observations enriched with sentiment scores derived from over 10,000 Kompas.com news articles. Sentiment analysis, performed using a Large Language Model (LLM)-based ChatGPT model, classified sentiment into Positive, Neutral, and Negative categories, which were then integrated as predictive features. Five scenarios for incorporating sentiment data were evaluated, with Scenario 2 (sequence of past sentiment scores) yielding the best performance. Specifically, it achieved the lowest Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) across various metrics: 36.4570 and 47.5527 for Absolute Return prediction, 0.005209 and 0.006804 for Relative Return prediction, and 50,725 and 65,653 for Close Price prediction. These findings underscore the significant role of sequential sentiment data in improving prediction accuracy, offering practical recommendations for investors to leverage sentiment analysis in making more informed decisions in the JCI market.