Forecasting the realized variance of oil-price returns using machine learning: Is there a role for US state-level uncertainty?

Predicting the variance of oil-price returns is of paramount importance for policymakers and investors. Recent research has focused on whether disaggregate measures of economic-policy uncertainty provide better forecasts. Given that the United States (U.S.) is a major player in the international oil market, we extend this line of research by exploring by means of machine-learning techniques whether accounting for U.S. state-level measures of economic-policy uncertainty results in more accurate forecasts. We find improvements in forecast accuracy, especially when we study intermediate and long forecast horizons. This finding is robust to various changes in the model configuration (realized variance vs. realized volatility, sample period, recursive vs. rolling-estimation window, loss function of forecast consumers). Understandably, our findings have important implications for oil traders and policy authorities.

Automated summary

This study investigates the impact of U.S. state-level economic-policy uncertainty measures on the prediction of oil-price return variance, and finds that incorporating these measures improves forecast accuracy, especially in the intermediate and long forecast horizon.