What Matters When Developing Oil Price Volatility Forecasting Frameworks?

Abstract

Forecasting oil price volatility is considered of major importance for numerous stakeholders, including, policy makers, industries and investors. This paper examines and evaluates the main factors that oil price volatility forecasters should take before constructing their forecasting models. Such factors are related to: i) direct vs iterated forecasts, ii) the incorporation of continuous and jump components, iii) the importance of semi variance volatility measures, and iv) OLS vs time-varying parameter (TVP) estimation procedures. Even more, we evaluate the performance of these factors for both realized and implied volatility measures, based on statistical loss functions, as well
as, their economic use. The results show that depending on whether end-users are interested in forecasting the realized or the implied volatility, the factors influencing the accuracy of forecasts are different. In particular, for the realized volatility, direct forecasting based on TVP estimation procedure, as well as, using the information obtained in the semi variance measures are capable of producing significantly superior forecasts. By contrast, separating the continuous and the jump components of the realized volatility does not provide any added value to these forecasts. Turning to the OVX, based on the economic evaluation of our forecasts, the TVP estimation procedure
seems to performbetter. In addition, we find evidence that the continuous component and the semi variance measures of the realized volatility also yield better OVX forecasts in the longer run horizons.

Keywords: HAR model, realized volatility, Oil price implied volatility index, multi-step ahead forecasts, time-varying parameter model