Forecasting horiticultural prices: A comparison of methods

Forecasts of horticultural prices provide useful information for farmers and retailers. These prices have large fluctuations as a result of seasonal changes in the production and the lagged response relationship between a decision to produce and the availability of the crop. Inaccurate price forecasts may lead to a decision not to harvest the crop, with a significant loss to the farmer. This paper compares different forecasting approaches applied to Portuguese horticultural data prices. The following steps are proposed: (1) spectral analysis for the detection of the dominating and hidden frequencies, (2) modeling the price series using decomposition methods, SARIMA models and causal models followed by an univariate procedure and, finally, (3) diagnostic checking using the last year of available data. Results show that for vegetables produced mainly under protected conditions (greenhouses) simple univeriate models fit well because the planted area has not changed over the last years. When the production comes only from open fields, the use of causal variables representing the planted area, is necessary. Cyclical behaviour is also reported (Cobweb model) in the case of melons, production being dictated by previous season’s price level.