Modeling and forecasting commodity price volatility using a common leverage factor

Abstract

We propose a new variant of the heterogeneous autoregressive model, the pseudo leverage HAR model, which exploits the well-known leverage effect to improve forecasting performance. Built on the fact there is an interconnectedness among commodities we employ a common leverage factor in forecasting exercises which is derived by principal component regression. Including this common leverage variable in HAR framework leads to significant improvements in both in-sample estimates and out-of-sample forecasts, suggesting that the factor structure is a valid assumption not just for return and volatility, but for volatility asymmetry too. The robustness tests confirm the usefulness of the common leverage factor, since the model incorporating this variable consistently remains in the model confidence set, implying that the model’s performance independent of the choice of the leverage structure or volatility proxy. Moreover, the portfolio evaluation exercise and the cumulative sum of forecast errors revealed the incremental gain of using the common leverage variable at all forecasting horizons, especially in turbulent periods.