Volatility Prediction Research Integrating Macroeconomic Indicators and Social Media Sentiment

Abstract

This study presents a multi-source LightGBM model to predict stock market volatility by combining macroeconomic indicators, Google Trends data, and Twitter sentiment. Using daily S&P 500 data from 2015 to 2024, the model was tested to evaluate how different information sources improve prediction results. The findings show that the multi-source model lowered RMSE by 7.8% compared with the version using only technical indicators. Sentiment and search data made a greater contribution during volatile market periods, showing that they can capture early signals of risk. These results indicate that mixing economic variables with real-time online sentiment can improve volatility forecasts and support financial risk management and investment analysis. Future studies should test shorter time intervals and more markets to confirm the model’s reliability and broader use.