TEXT ANALYTICS AND LASSO REGRESSION FOR STOCK PRICE MOVEMENTS

Abstract

This study investigates the impact of news events on stock price movements in the IDX Kompas 100 index (JKKM100) by combining machine learning-based text analytics with event study analysis using LASSO regression. News data from January 1, 2019 to July 22, 2024 were collected via web scraping and used as training data for text classification. The trained model was then applied to classify news events in the period from January 1, 2024 to April 30, 2025, which were subsequently used in the event study analysis. Stock price data for the same period (January 1, 2024 to April 30, 2025) were collected to ensure consistency between predictor and response variables. Due to class imbalance, the synthetic minority over-sampling technique (SMOTE) was applied. Several machine learning algorithms were evaluated, and XGBoost achieved the highest accuracy of 72.22%, improving to 79% after hyperparameter tuning. Using weighted abnormal returns as predictors and stock closing prices as response variables, the LASSO regression results show that 13 out of 180 news events significantly influenced stock price movements. The model explains 48.17% of the variance with an RMSE of approximately 5% of the average stock price. Industry-related news contributed the most (43.20%), followed by PESTEL (3.97%) and Investment (1.00%). This study demonstrates that integrating text analytics with LASSO-based event study provides an effective framework for analyzing the impact of news on stock price movements.