MODELING FRESH TOMATO MARKETING MARGINS: ECONOMETRICS AND NEURAL NETWORKS

This study compares two methods of estimating a reduced form model of fresh tomato marketing margins: an econometric and an artificial neural network (ANN) approach. Model performance is evaluated by comparing out-of-sample forecasts for the period of January 1992 to December 1994. Parameter estimates using the econometric model fail to reject a dynamic, imperfectly competitive, uncertain relative price spread margin specification, but misspecification tests reject both linearity and log-linearity. This nonlinearity suggests that an inherently nonlinear method, such as a neural network, may be of some value. The neural network is able to forecast with approximately half the mean square error of the econometric model, but both are equally adept at predicting turning points in the time series.