Vector Autoregression, Long Short-Term Memory, dan Xgboost Dalam Kerangka Ensemble

Abstract

Stock market volatility is a crucial indicator for risk assessment and investment decision-making, particularly for the S&P 500 index as a proxy for global financial markets. This study aims to forecast S&P 500 volatility by integrating macroeconomic and technical indicators using an ensemble approach. Daily data from March 18, 2015 to January 8, 2026 are employed. The input variables include macroeconomic indicators such as the Federal Funds Rate, Consumer Price Index, US Dollar Index, crude oil prices, and the Volatility Index (VIX), as well as technical indicators including log returns, realized volatility, moving averages, Relative Strength Index, and Moving Average Convergence Divergence. Realized volatility is computed using a 21-day rolling window and annualized. The proposed framework combines Vector Autoregression, Long Short-Term Memory networks, and XGBoost within an ensemble scheme. Empirical results indicate that the ensemble model produces more stable forecasts and outperforms classical econometric models. These findings highlight the effectiveness of combining macroeconomic and technical information for volatility forecasting and stock market risk management.