Using weather data to improve demand forecasting for seasonal products

In seasonal business, manufacturers need to make major supply decisions up to a year before delivering products to retailers. Traditionally, they make those decisions based on sales forecasts that in turn are based on previous season's sales. In our research, we study whether demand forecasts for the upcoming season could be made more accurate by taking into account the weather of the previous sales season. We use a ten-year dataset of winter sports equipment (e.g. skis, boots, and snowboards) sales in Switzerland and Finland, linked with daily meteorological data, for developing and training a generalised additive model (GAM) to predict demand for the next season. Results show a forecasting error reduction of up to 45% when including meteorological data from the past season. In our case, the value of this reduction in the forecasting error corresponds to around 2% of total sales. The results contribute to the theory of stochastic inventory control by showing that taking into account external disturbances, in this case the fluctuations in weather, improves forecasting accuracy in situations where the lag between ordering and demand is around one year