Forecasting and Analyzing World Commodity Prices
René Lalonde* Principal Researcher International Department Bank of Canada Zhenhua Zhu Economist Research Department Bank of Canada October 18, 2002 Frédérick Demers** Economist Research Department Bank of Canada

Abstract This paper develops simple econometric models to analyze and forecast three components of the Bank of Canada commodity price index (BCPI), namely non-energy commodity prices (BCNE), the West Texas Intermediate crude oil price (WTI), and other energy prices. In the paper, we present different methodologies to identify transitory and permanent components of movements in these prices. A structural vector autoregressive (SVAR) model is used for real BCNE prices, a multiple structural-break technique is employed for real crude oil prices, and an errorcorrection model is constructed for real prices of other energy components. Then we use these transitory and permanent components to develop forecasting models. We assess our models’ performance in various aspects, and our main results indicate: (a) for real BCNE prices, most of the short-run variation is attributed to demand shocks, (b) the world economic activity and real U.S. dollar effective exchange rate explain much of the cyclical variation of real BCNE prices, (c) real crude oil prices have two structural breaks over the sample period, and their link with the world economic activity is strongest in the most recent regime, (d) real prices of other energy components are highly correlated with the U.S. economic activity, and they are co-integrated with real crude oil prices, (e) our models outperform benchmark models, namely a VAR model, autoregressive (AR) model and a random walk (RW) model, in terms of out-of-sample forecasting, and (f) a 1% positive shock to world economic activity leads to an approximate 7.2% peak response of world commodity