Time Series Methods for Commodity Price Forecasting: An Application to Market Pulp

Abstract

The goal of this research is to assess the usefulness of cointegration analysis and related time series techniques for forecasting commodity prices. The analysis focuses on market pulp, a typical commodity. Important short-term factors in determining pulp prices include capacity utilization, the shipments rate, and inventories. Important longterm factors include investment behavior and costs of production. Autoregressive, moving average (ARMA), vector autoregressive (VAR) and error correction models of price and these variables are developed. Market scope is defined to include ail North American and Scandinavian (Norscan) chemical paper grade pulp, and the sample period is 1976-1991. Out of sample forecasting performance of the error correction models is no better than that of the VAR model, according to the RMSE criterion. However, the forecasts generated by the error correction models have the property that sensible long-term relationships between variables are maintained. In addition the error correction models a.re more amenable to incorporating expert knowledge into the model-based forecasts. It is thus concluded that error correction models are useful for forecasting commodity prices.