M. Zied Babai

On the empirical performance of (T, s, S) heuristics

The periodic (T,s,S) policies have received considerable attention from the academic literature. Determination of the optimal parameters is computationally prohibitive, and a number of heuristic procedures have been put forward. However, these heuristics have never been compared in an extensive empirical study. Such an investigation on 3055 SKUs is carried out in this paper. Our study provides insights into the performance of (T,s,S) heuristics, also in relation to demand forecasting. The results show that Naddors heuristic is best able to minimize the total cost. However, the normal and power approximations achieve more efficient solutions in that backorder volumes are smaller at the same inventory levels, indicating the potentially superior performance of these methods if the balancing of holding and backorder costs can be improved. The results also show that, for all heuristics, the SBA variant of the Croston forecasting method significantly outperforms Croston as well as Single Exponential Smoothing (SES).

Distributional Assumptions for Parametric Forecasting of Intermittent Demand

Parametric approaches to stock control rely upon a demand distributional assumption and the employment of an appropriate forecasting procedure for estimating the moments of such a distribution. For the case of fast demand items the Normality assumption is typically sufficient. However, spare parts typically exhibit intermittent or irregular demand patterns that may not be represented by the Normal distribution. The objective of this work is three-fold: first, we conduct an empirical investigation that enables the analysis of the goodness-of-fit of various continous and discrete, compound and non-compound, two-parameter statistical distributions used in the literature in the context of intermittent demand; second, we crictically link the results to theoretical expectations; third, we provide an agenda for further research in this area. We use three empirical datasets for the purposes of our analysis that collectively constitute the individual demand histories of approximately 13,000 SKUs. Our work allows insights to be gained on the suitability of various distributions in a spare parts context.

The multi-sourcing location inventory problem with stochastic demand

Abstract This paper deals with a multi-period location-inventory optimization problem in a multi-echelon supply chain network characterized by an uncertain demand and a multi-sourcing feature. The aim of the paper is to propose a generic modeling approach to integrate key features of the inventory planning decisions, made under a reorder point order-up-to-level ( s, S ) policy, with the location-allocation design decisions to cope with demand uncertainty. Given the hierarchical structure of the problem, a two-stage stochastic mathematical model that maximizes the total expected supply chain network profit is proposed. This optimization model is intractable due to its non-linearity. Therefore, a linear approximation is proposed and a sample average approximation approach is used to produce near-optimal solutions. Numerical experiments are conducted to validate the proposed modeling and solution approaches. The results show the efficiency of the linear approximation of the ( s, S ) policy at the strategic level to produce robust design solutions under uncertainty. They underline the sensitivity of the design solution to the demand type and the impact of the inventory holding costs and backorder costs, especially under non-stationary processes.

The impact of temporal aggregation on supply chains with ARMA(1,1) demand processes

Various approaches have been considered in the literature to improve demand forecasting in supply chains. Among these approaches, non-overlapping temporal aggregation has been shown to be an effective approach that can improve forecast accuracy. However, the benefit of this approach has been shown only under single exponential smoothing (when it is a non-optimal method) and no theoretical analysis has been conducted to look at the impact of this approach under optimal forecasting. This paper aims to bridge this gap by analysing the impact of temporal aggregation on supply chain demand and orders when optimal forecasting is used. To do so, we consider a two-stage supply chain (e.g. a retailer and a manufacturer) where the retailer faces an autoregressive moving average demand process of order (1,1) -ARMA(1,1)- that is forecasted by using the optimal Minimum Mean Squared Error (MMSE) forecasting method. We derive the analytical expressions of the mean squared forecast error (MSE) at the retailer and the manufacturer levels as well as the bullwhip ratio when the aggregation approach is used. We numerically show that, although the aggregation approach leads to an accuracy loss at the retailers level, it may result in a reduction of the MSE at the manufacturer level up to 90% and a reduction of the bullwhip effect in the supply chain that can reach up to 84% for high lead-times.

Revisiting the value of information sharing in two-stage supply chains

Abstract There is a substantive amount of literature showing that demand information sharing can lead to considerable reduction of the bullwhip effect/inventory costs. The core argument/analysis underlying these results is that the downstream supply-chain member (the retailer) quickly adapts its inventory position to an updated end-customer demand forecast. However, in many real-life situations, retailers adapt slowly rather than quickly to changes in customer demand as they cannot be sure that any change is structural. In this paper, we show that the adaption speed and underlying (unknown) demand process crucially affect the value of information sharing. For the situation with a single upstream supply-chain member (manufacturer) and a single retailer, we consider two demand processes: stationary or random walk. These represent two extremes where a change in customer demand is never or always structural, respectively. The retailer and manufacturer both forecast demand using a moving average, where the manufacturer bases its forecast on retailer demand without information sharing, but on end-customer demand with information sharing. In line with existing results, the value of information turns out to be positive under stationary demand. One contribution, though, is showing that some of the existing papers have overestimated this value by making an unfair comparison. Our most striking and insightful finding is that the value of information is negative when demand follows a random walk and the retailer is slow to react. Slow adaptation is the norm in real-life situations and deserves more attention in future research – exploring when information sharing indeed pays off.

Impact of product characteristics on distribution strategy selection

Nowadays companies must look to develop new distribution strategies in order to achieve the required performance from their supply chain. In this quest, companies wonder about the consistency of their distribution strategies with the types of products they are selling. This article deals with the issue of product segmentation and distribution strategy selection (cross-docking versus traditional warehousing). In the literature many factors have been described that affect product segmentation and supply chain selection. In this article the selection is based on the impact of product shelf space, demand variability, product value, and lead time on the supply chain performance (store inventory, service level). We carry out a simulation investigation to determine the impact of each product characteristic on the supply chain performance for the two different strategies. This article provides a framework to help managers understand the nature of their products and choose the right distribution strategy for each one.

Multi-criteria inventory classification problem: A consensus approach

In this paper, we consider the multicriteria inventory classification problem. We propose a new classification algorithm referred to as Constructive Order Classification Algorithm (COCA). This algorithm is based on some simple priority rules and aims to standardize the classification and provide relative stability in the classification through a consensus process. An illustrative example is used to explain the different steps of the proposed algorithm.