Exploring the impact of cluster bagging-based weather attention on oil market volatility under the GARCH-MIDAS framework

Many studies have confirmed the impact of weather and Google Trends attention on oil markets. However, research examining the predictive influence of weather attention on oil price volatility is lacking. To fill this research gap, we have developed an extended GARCH-MIDAS forecasting model based on TBATS and clustering bagging. We utilize weather-related keywords from Google Trends to obtain weather attention data and assess its predictive capability. By decomposing the data using TBATS and combining the decomposed trend and seasonal components with the remaining portion through clustering bagging and arithmetic mean aggregation, we construct a series of GARCH-MIDAS models, and specifically focus on the clustering bagging-based TBATS decomposition GARCH-MIDAS model. Empirical analysis demonstrates a significant impact of weather attention magnitude on in-sample testing, with the clustering bagging-based TBATS model outperforming other models in out-of-sample forecasting. Our findings provide new evidence that incorporating weather attention information into a GARCH-MIDAS model can improve prediction results and highlight the influence of weather attention on future oil price fluctuations. Finally, robustness analysis confirms the robustness of the proposed model. The proposed extended GARCH-MIDAS model based on TBATS and clustering bagging offers new insights into oil market volatility prediction.