An evaluation of alternative forecasting models for natural rubber prices

Abstract

One of the prominent features of the Natural Rubber (NR) market is its price variability, and the aim of this study is to project accurate short-term NR prices. This is accomplished by exploiting the use of forecasting techniques and information sets to seek the combination with the best forecasts, and exploring best ways of combining forecasts. We evaluate the relative performance of 19 models based upon three different forecasting techniques, and four information sets. In addition, we compare their forecasts with 13 other forecasts combined in various different ways, and taking the Naive forecast as benchmark. The generalised autoregressive conditional heteroscedasticity regression (or ARCH-type) models, though more complex, are generally better than the simpler regression models. In general, the performance of the various techniques seems to perform consistently well (or poorly) over the forecasting horizons, with alternations in performance due mainly to the type of information set used. We also adopted a simple trading rule to find out the economic values of our forecasts, and the results are most promising. Importantly, the forecasts generated from the alternative models developed in this study can potentially be beneficial to participants in the NR futures market.