Forecasting the Indonesian Stock Market Index Using ARIMA–GARCH Models with a Rolling Window Approach

Abstract

This study investigates the forecasting of the Indonesian Stock Market Index (IHSG) using an ARIMA–GARCH hybrid modeling framework combined with a rolling window estimation approach. Stock market indices are characterized by nonstationary behavior and time-varying volatility, which pose challenges for accurate forecasting, particularly in emerging markets. To address these issues, daily index data were transformed into log returns to achieve stationarity. The ARIMA model was employed to capture conditional mean dynamics, while the GARCH model was used to model volatility clustering commonly observed in financial time series. To improve adaptability to changing market conditions, a rolling window estimation with a fixed length of 180 trading days was implemented, allowing model parameters to be updated continuously over time. The empirical results indicate that the optimal specification consists of an AR(1) process for the mean equation and a GARCH(1,1) process for the volatility equation. The forecasting results show that the rolling ARIMA–GARCH model closely follows actual market movements and effectively captures both trend behavior and volatility fluctuations. Forecast accuracy evaluation demonstrates strong predictive performance, with a Root Mean Squared Error of 53.40 and a Mean Absolute Percentage Error of 0.58 percent, indicating a high level of forecasting precision. These findings confirm that the ARIMA–GARCH model combined with rolling window estimation provides a robust and reliable framework for short-term stock market index forecasting in volatile market environments.