Statistical and economic performance of combination methods for forecasting crude oil price volatility

This article aims to investigate whether a wide variety of combination methods, ranging from simple averaging approaches to time-varying techniques based on the past performance of the single models and regression tools, improve forecasting accuracy, risk management and economic gains across different forecasting horizons in a crude oil future framework. For this purpose, we implement various combination methods in fourteen HAR-RV models including different characteristics and uncertainty factors and evaluate their performance through statistical, risk management and economic loss functions. The empirical analysis yields some important conclusions: (i) sophisticated combinations improve the forecasting accuracy in all forecasting horizons, (ii) from a risk management perspective, regression-based combinations predict better potential loss of an investment on crude oil futures compared to individual models, (iii) machine learning and non-parametric combinations lead to higher economic gains across all forecasting horizons in a portfolio exercise and (iv) combinations are considered as a better alternative to single models using various robustness checks although the best performer among the combination methods is not stable across applications.