Forecasting co-movements in energy futures: the role of structural breaks in short and long-run correlation components

To forecast the covariance matrix of energy future returns, this paper investigates the impact of structural breaks on the short and long-run components of correlation dynamics. We consider the DCC MIDAS structure to separate long-run from short-run components. By the specification, we can separately model the smooth transition of unconditional correlations between regimes caused by macroeconomic fundamentals and the abrupt markovian type regime-shifts of conditional correlations. The empirical application investigates gains in forecasting the crack spread. We model the covariances between the front month continuous contract of Crude oil (NYMEX Light Sweet Crude), NY Harbor ULSD (Ultra Low Sulfur Diesel) and RBOB Gasoline to predict their difference, the crack spread, and improve forecasting accuracy of refining margins. The results indicate the benefits of including different types of regime-switch in correlation components both in term of statistical criteria, such as the SPA test and the Model Confidence set, and with respect to optimal portfolio allocation strategies and risk management in crude oil markets.