Modelling and prediction of intermittent demand distributions

The management of spare part inventories is an issue of strategic concern for most industrial firms. However, the demand for spare parts is typically intermittent in nature meaning that orders arrive sporadically and the order sizes may be highly variable. A number of authors have suggested that compound distributions could be used to model intermittent demand patterns. There is however a lack of theoretical analysis and relevant empirical evidence on this issue. In this work, we assess whether compound Poisson distributions provide a good fit for the demand distributions of spare part items. A framework that links demand classification and the distributional properties of demand is proposed and the empirical validity of the framework is assessed by means of experimentation with real data. This study also examines a number of different approaches for managing inventory items with intermittent demand. The literature on inventory management is dominated by the 'frequentist-approach'; this is the term that is being used in this thesis to refer to all the solutions that rely on frequentist inference in order to obtain the demand distribution. The frequentist-based approach is characterised by a reliance on a number of assumptions (including, at a minimum, that the demand distribution and the associated parameters are known). As demonstrated in this study, such assumptions may pose considerable practical problems when demand is intermittent. An argument is being made in favour of other inventory management approaches that rely on fewer, less restrictive, assumptions. The alternative approaches considered in this study include the bootstrapping-based solution proposed by Willemain et al. (1994), a new solution based on the work by Efron (1979) and a new approach that is based on the Bayesian paradigm. A comparison of the stock control performance of these alternatives suggests that non-frequentist approaches may perform as well as the frequentist one