Forecasting Prices at the Dutch Flower Auctions

Prices at the Dutch flower auctions are extremely volatile. Price changes of +/-20 per cent one week to the next represents a normal event, and +/-50 per cent is not uncommon. Since production planning in the flower business offers a complicated variation over the Newsboy Problem, good price forecasts would improve decision making on space allocation; what species to plant; the timing of harvesting, etc. This paper analyses weekly prices for three major species, i.e., roses, chrysanthemums and carnations, 1993–1996. We find that there are strong calendar regularities particularly for roses and chrysanthemums. Establishing a model in which we combine information on seasonal regularities and autoregressive price patterns, we manage to explain a substantial part of the short-term price variability for all three species. The model is tested in an out-of-sample dynamic forecasting experiment during the first 35 weeks of 1997.