Review of short-term demand forecasting methods and selection of the appropriate model for softwood sawmills in Tanzania

Abstract

This paper reviewed and tested forecasting methods for  use-appropriateness in forecasting lumber demands in the sawmill
industry in Tanzania. These methods included naïve or intuitive, simple moving average, weighted average, regression analysis, and exponentially weighted moving average (EWMA). The value of the mean absolute deviation (MAD) was used to test the models forecasting accuracy and the smaller the value of MAD the higher the forecasting accuracy. The best two methods were subjected to further analysis by forming a confidence interval for the difference emanating from the difference between MADs of the two methods. The naïve and EWMA were found to have the smallest MADs. The MADs produced by the two methods were 150.08 and 153.64 m3 for naïve and EWMA methods, respectively. Although these errors were mathematically different, confidence interval [-52.21, 45.09] constructed for MADs’ difference contained a zero indicating that the MADs difference was not significantly different from zero. Since the lumber demand behaviour in the country indicated presence of trend and seasonality, a
phenomenon that cannot be accommodated by the naïve method and
the fact that EWMA managed to capture such phenomenon and yet produced minimum error, it was recommended for use by sawmill managers in Tanzania.

Key words: Sawmill industry, demand forecasting, forecasting accuracy and
lumber production.