Changing determinant driver and oil volatility forecasting: A comprehensive analysis

Abstract

Academic research relies on exogenous drivers to enhance the accuracy of forecasting oil volatility. Following the relevant literature, this study collects 62 exogenous drivers that reflect the movements of oil demand, oil supply, oil inventory, macroeconomic fundamentals, financial indicators, and measures of uncertainty. Our empirical results indicate that dimension reduction regressions, especially principal component analysis regression (PCA), successfully predict both WTI and Brent oil volatility at the one-month ahead forecast horizon. Shrinkage methods, on the other hand, outperform their counterparts for medium- and long-term forecast horizons. Furthermore, the unsupervised learning method (PCA) achieves superior forecasting performance during periods of oil price decrease, whereas supervised learning methods (i.e., shrinkage methods) significantly improve volatility accuracy. Additionally, the empirical results reveal that movements in the Kilian index, World industrial production index, global economic conditions index, U.S. steel production, Chicago Fed national activity index, capacity utilization for manufacturing, U.S. default yield spread, and MSCI emerging market index have a significant impact on oil volatility.