Exploiting the heteroskedasticity in measurement error to improve volatility predictions in oil and biofuel feedstock markets

Abstract

We examine whether heteroskedasticity in measurement errors improves the volatility forecasting ability of the Heterogenous Autoregressive (HAR) model in crude oil and biofuel feedstock markets. We also examine the incremental explanatory power of jumps and the investor fear gauge (IFG) over heteroskedasticity in measurement errors in improving the volatility forecasting ability of the HAR model in each of these markets. For the in-sample evaluation, we find that exploiting the heteroskedasticity of measurement errors in the HAR model improves the model's goodness of fit (measured by adjusted R2) by up to 10% depending on the market. IFG has a significant incremental role over heteroskedasticity in measurement errors in improving the fit of the HAR model in both the crude oil and biofuel feedstock markets, while jumps have a significant incremental role in improving the fit of the HAR model in the crude oil market, but not the biofuel feedstock markets. For the out-of-sample forecasts, including regime switching improves volatility predictions in the corn and wheat markets across all forecasting horizons, while for the soybean market, including regime switching improves the performance of multi-step volatility forecasts. In the out-of-sample forecasts the best ranked models almost always include heteroskedasticity of measurement error and IFG.