Volatility forecasting of non-ferrous metal futures: Covariances, covariates or combinations?

Abstract

This is the first comprehensive study on the forecasting of the realized volatility of non-ferrous metal futures. Based on 8.5 years of intraday data on copper, zinc, nickel, lead and aluminum, we explore a variety of extensions of the univariate heterogeneous autoregressive (HAR) model and seek to harness the economic linkages among these metals to improve forecasts. A simple approach that augments the models with shocks in other metals’ series appears to outperform more sophisticated specifications, which explicitly model covariances. The results suggest that the information inherent in the volatility series of aluminum is most useful in enhancing the accuracy of forecasts for other metals. While consistently outperforming the original HAR model with an individual model is difficult, combination forecasts, especially with univariate specifications or Bayesian model averaging, are found to conclusively outperform the benchmark.