Forecasting crude oil price volatility with uncertainty: New modeling with multivariate selection

This study proposes an approach for the multivariate selection of the GARCH-MIDAS model by integrating genetic algorithm, which overcomes the problems of unidentified weight and parameter hopping encountered in the previous Lasso-based method. Our approach effectively identifies the optimal combination of uncertainty indices for volatility forecasting. Empirical analyses demonstrate that the GARCH-MIDAS model incorporating four uncertainty indices, namely economic policy uncertainty, world uncertainty, energy-related uncertainty, and monetary policy uncertainty, outperforms alternative models in oil price volatility prediction. Out-of-sample forecasts further validate the superior predictive performance of the multivariate model selected through our approach.