The role of comovement and time-varying dynamics in forecasting commodity prices

Commodity prices co-move, but the strength of this co-movement changes over time due to structural factors, like changing energy intensity in production and consumption as well as changing composition of underlying shocks. This paper explores whether econometric models that exploit this co-movement and account for parameter instability provide more accurate point and density forecasts of ten major commodity indices viz-a-viz constant-coefficient mo dels. Improvements in point forecast accuracy are small, with predictability varying substantially across forecast horizons and commodity indices, but they are large and significant in terms of density f orecasting. An economic evaluation reveals that allowing for parameter time variation and commonalities leads to higher portfolios returns, and to higher utility values for investors.