An investigation on bootstrapping forecasting methods for intermittent demands

Many bootstrapping approaches have been proposed in the academic literature for non-parametric demand forecasting. Two approaches have been developed to deal particularly with intermittent demands. A first approach that samples demand data by using a Markov chain to switch between no demand and demand periods and a second approach that separately samples demand intervals and demand sizes. The relevant studies have claimed improvements over parametric approaches when estimating the lead-time demand distribution. However, it should be noted that the outperformance of the two bootstrapping approaches has been shown under a limited set of control parameters and assumptions. The purpose of this paper is to broaden the empirical and numerical settings when analyzing the performance of the two bootstrapping approaches and the parametric one. Hence, more exhaustive assumptions (i.e. generated demand distributions) and a wider range of control parameters (i.e. length of demand histories, length of lead-times, demand distribution parameters and cost parameters) are considered in the numerical analysis. More empirical data with a wider range of demand patterns are also used for the purpose of the empirical investigation. The results show that for highly intermittent demands and small values of lead-times, contrary to what has been claimed in the literature, the parametric approach outperforms the bootstrapping approach that separately samples demand intervals and sizes.