Investor Sentiment Dynamics and Market Volatility in Indonesia: Hybrid Approach Using GARCH-Midas and Machine Learning

Abstrakt

The Indonesian stock market, represented by the Jakarta Composite Index (IHSG), experiences significant volatility influenced by both domestic and global macroeconomic factors as well as investor sentiment. This study investigates the impact of key macroeconomic variables and investor sentiment indicators, such as the Consumer Confidence Index (CCI) and Trading Volume Activity (TVA), on IHSG volatility. The research applies the GARCH-MIDAS model to capture long-term macroeconomic effects and integrates machine learning techniques, Extreme Gradient Boosting (XGBoost) for short-term volatility prediction. Monthly data on macroeconomic variables and sentiment indicators are combined with daily IHSG return data. Performance metrics like Mean Square Error are used to compare the forecast with realized volatility. The findings show that macroeconomic variables, particularly Inflation and Exchange Rates, significantly affect IHSG volatility, while sentiment indicator CCI play crucial roles and demonstrated the highest predictive power. The hybrid GARCH-MIDAS-XGBoost model outperformed the traditional GARCH-MIDAS, reducing MSE by approximately 20% and improving MAE by 15% during volatile periods. The model also excelled in predicting volatility spikes, especially during market turbulence. This study confirms that both macroeconomic variables and investor sentiment indicators, especially CCI, significantly impact stock market volatility. The hybrid model improves forecasting accuracy, offering valuable insights for investors and policymakers navigating market risks.