Organic produce price forecasting at a farm level: criteria, methods, and forecast evaluation

Organic farmers, wholesalers, and retailers need price forecasts. A methodology and protocol to select the best performing method from several time and frequency domain candidates is suggested. Seasonal autoregression, the additive Holt-Winters exponential smoothing, and spectral decomposition are considered. The forecasting methods are evaluated on the basis of an aggregate accuracy measure and several out-of-sample predictive ability tests. The seasonal autoregression is found to be broadly the best performing method. The Holt-Winters method provides better forecasts in the short run; spectral decomposition is preferable for more distant periods. The price-generating process is found to have a strong autoregressive component and a clear but simple seasonal pattern. The role of better price forecasts for the agents who deal in less common organic produce is highlighted. A conrmation for the claim that the organic produce industry needs better farmgate price forecasts to grow is provided. Contracting and diversication are suggested.