Non-stationarity in Sugar Prices

Time series analysis of unstable series such as raw material prices contains several methodological pitfalls. These are illustrated by a survey of the Box-Jenkins method of analysis applied to 12 years of daily sugar prices.Two problems in particular have received insufficient attention in previous analyses. First, the model variance fluctuates substantially over time. The logarithmic transformation of prices is recommended as significantly reducing the fluctuations. Second, it is essential to study the constant term in the Box-Jenkins model, which is shown to differ significantly from zero. This proves the existence of price trends. The commonly accepted random walk model for commodity prices is thus incorrect, at least for sugar.However, these price trends vary over time and this possibility is not covered by the Box-Jenkins models. Consequently, a new model for price movements is proposed, characterized by frequent changes in the daily price trend yet consistent with the results of the autocorrelation analysis.