Forecasting Agricultural Commodity Prices Using Multivariate Bayesian Machine

The purpose of this paper is to perform multiple predictions for agricultural commodity prices (one, two and three month periods ahead). In order to obtain multiple-time-ahead predictions, this paper applies the Multivariate Relevance Vector Machine (MVRVM) that is based on a Bayesian learning machine approach for regression. The performance of the MVRVM model is compared with the performance of another multiple output model such as Artificial Neural Network (ANN). Bootstrapping methodology is applied to analyze robustness of the MVRVM and ANN