The centralisation of inventory and the modelling of demand
The motivation for this research arose from two projects in which the author advised on inventory centralisation. Since it was found that the literature was of limited value, inventory centralisation was identified as a suitable topic for research. Operationalisation is not adequately addressed in the literature on centralisation models. 'Operationalisation' denotes the translation of abstract concepts such as 'inventory service' into measures enabling observations to be recorded. In the literature, 'inventory service' is often equated to 'probability of stock-out' but many other measures are used in practice. This thesis presents a network of relationships between six commonly used measures. This is useful when decentralised depots do not share a common measure, or when the measure changes after centralisation. This thesis argues that, under all circumstances which arise in practice, it will be possible to achieve inventory availability benefits from centralisation with no added investment in stocks. A counter-example to the universal application of this rule have been presented in the literature but it is shown that such counter-examples are artificial. Since savings from centralisation may be offset by increases in transport costs, the reduction in stock-holdings needs to be estimated. The models presented in the literature assume that the estimation of demand variance and the correlation of demand between depots is not problematic. In practice, reliable estimates may be difficult to obtain, particularly for slow-moving items. Consequently, it is difficult to decide which items should be centralised and which, if any, should not This thesis proposes a 'quadratic variance law' approach', linking the variance of demand to its mean. This approach is underpinned by a model which allows correlation effects to be taken into account The 'variance law' approach is a contribution towards the operationalisation of centralisation models, since reliable estimates of mean demand are easier to obtain than estimates of variance and correlation. The 'quadratic variance law' is examined empirically using a sample of 230 stockkeeping units from an engineering supplies company. The approach is shown to be well-supported by the data. All the model assumptions are supported except one. The postulated independence between mean order-size and mean demand is rejected since a weak correlation was found. However, the simpler quadratic law is found to be more robust than a more complex law which would have taken this correlation into account.