Improving stock volatility forecasting with an outlier- corrected GARCH-MIDAS model for enhanced accuracy across major global stock markets

The volatility of stock price indices not only has a profound impact on investors' asset allocation and returns but also threatens the stability of financial markets. To explore this further, we examine six major stock markets: the United States, Japan, China, Hong Kong, Singapore, and Germany. These markets have been selected due to their significant influence on the global financial system. The U.S., with its highly liquid markets, serves as a global benchmark. Japan, the third-largest economy, represents the Asia-Pacific region. China, an emerging market, offers insights into rapid economic growth and volatility. Hong Kong, a financial hub connecting East and West, highlights market integration. Singapore, known for its strong regulatory environment and fintech sector, and Germany, Europe’s largest economy, provide perspectives on stability and volatility in developed markets. These diverse markets offer a comprehensive view of how volatility manifests across different economic and regulatory contexts. Accurately predicting volatility is crucial for academic research and financial practice. While low-frequency macroeconomic variables, such as economic policy uncertainty and exchange rates, improve volatility prediction models, most studies overlook the negative impact of additive outliers (AO) in financial time series. Outliers, caused by extreme market events, data errors, or structural changes, can bias in-sample estimations and reduce out-of-sample prediction accuracy. They distort model parameters, disrupt the detection of volatility clusters, and weaken predictive performance. Although robust statistical methods have advanced outlier processing, their integration into econometric models like GARCH-MIDAS remains limited. Additionally, existing research relies heavily on empirical data, lacking comprehensive theoretical verification using simulated data, which restricts understanding of model mechanisms and robustness. To address these gaps, this study proposes a robust generalized autoregressive conditional heteroskedasticity-mixed- data sampling model (AO-GARCH-MIDAS). By incorporating an outlier correction mechanism into the GARCH-MIDAS framework, the model effectively combines high-frequency volatility dynamics with low-frequency macroeconomic variables while mitigating outlier interference. Through in-sample estimation, out-of-sample prediction, and simulated data experiments, the AO-GARCH-MIDAS model demonstrates superior performance and robustness. The study evaluates predictive performance using simulated data generated by a contaminated AR(1)-GARCH(1,1) process, focusing on the six stock markets. Monte Carlo experiments reveal that the AO-GARCH-MIDAS model outperforms traditional models in key metrics like mean absolute error (MAE), root mean square error (RMSE), and median absolute error (MedAE), particularly in handling large-scale outliers. This underscores its ability to capture extreme market volatility and enhance forecasting accuracy. Empirical analysis incorporates macroeconomic variables, including exchange rates and economic policy uncertainty indices, constructing univariate, bivariate, and multivariate models. The AO-GARCH-MIDAS-RV-EPU(V) framework, combining realized volatility (RV) and economic policy uncertainty (EPU), achieves the best predictive performance, with an average loss value of 0.7783, significantly outperforming other models. Model confidence set (MCS) tests further validate its superiority. In summary, this study provides a new theoretical framework and empirical basis for volatility prediction by introducing the AO-GARCH-MIDAS model, simulation data evaluation, and the prediction method combining RV. It holds important academic value and practical significance.

Text SPE 2025 2 - Full Text.pdf Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (2MB) Text SPE 2025 2.pdf Restricted to Repository staff only Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (3MB)

Actions (login required)

View Item