Title: Forecasting commodity prices in a data-rich, unstable environment
Authors: Anastasia Allayioti - University of Warwick (United Kingdom) [presenting]
Fabrizio Venditti - ECB (Germany)
Abstract: Recent research has shown that commodity prices exhibit substantial co-movement, which can be captured by few common factors, broadly related to global demand for commodity shocks, which are pervasive across all commodity prices, and idiosyncratic (commodity-specific) supply shocks. A separate literature has stressed how the composition of underlying structural shocks that drive commodity prices has changed over time, potentially resulting in unstable unconditional correlations. These two findings suggest that (i) forecast accuracy for the price of a given commodity could benefit from the information contained in other commodity prices and that (ii) dealing with potential structural breaks could also improve forecast accuracy. We investigate the merits of constructing forecasts for key commodity prices from models that use large information sets and deal with structural breaks. We consider large TVP-VARs, TVP dynamic factor models (TVP-DFMs) and TVP hierarchical dynamic factor models (TVP-HDFMs), which impose the presence of specific blocks on the factor model structure of commodity prices. Given that standard estimation methods for small-dimensional models fail in a data-rich environment, we adopt non-parametric kernel-based methods and forgetting factor techniques. A distinct contribution of these methods is that, unlike most of previous ones, we evaluate both point and density forecasts.