Here is a brief information about the presentations of the CMAF members:

Ritika Aurora

Title: Forecasting and demand planning in the presence of disruptions

Abstract: In recent years, retail companies have experienced several disruptions with potentially large effects on demand, supply, or both. For example, companies in the UK were subject to supply issues, delays due to Brexit, Covid-19 and lockdowns. During these times, many decision makers are operating in unknown terrain and retail forecasting systems they use during normal times do not provide reliable estimates. Being able to still forecast demand as accurately as possible during disruptions and also afterwards if data has been distorted is crucial for businesses to be successful. We propose a shock-smoothing approach to forecasting and demand planning in the presence of disruptions, with an emphasis on improving forecasting performance. We compare our approach to standard and adaptive methods from the literature to show when it outperforms existing methods. Our approach is suitable to help demand planners make data-driven decisions by estimating and planning for demand during and post disruptive events.

Robyn Goldsmith

Title: Intermittent Demand Forecasting for Final Purchase Decisions

Abstract: Across a number of sectors, after-sales services constitute a billion-dollar business. Final purchases are high stake decisions of particular importance for the aftermarket business. They are often made towards the end of the spare part life cycle and need to balance shortages and over-ordering so as to sustain demand for the rest of a part’s life. In the automotive industry, among others, this period can extend for as long as ten years.

Forecasting in order to make effective final purchase decisions is made difficult by the nature of the demand. A large majority of spare part inventories consist of items with intermittent demand structures, where in some periods no demand is observed at all. As a result, standard forecasting methods provide inaccurate forecasts for intermittent items. As final purchase decisions are frequently made to sustain demand several years into the future, modelling procedures should also be designed to include the decay in demand as time passes. A forecasting model that accounts for aspects of both decline and the intermittency of demand are the subject of this talk.

Kandrika Pritularga

Title: What is the value of congruous forecasts across time?

Forecasts of future demand are necessary for inventory management. Typically, the focus is on producing accurate forecasts, which are desirable in the statistical sense. On the other hand, the limited work that looks jointly at forecasting and inventory control has identified low out-of-sample bias to be more important than accuracy. Shrinkage estimators and forecast combination shift the attention from in-sample fit to better generalization in the future. These result in less volatile forecasts, with typically better predictive distributions, specifically at quantiles of interest, and less out-of-sample bias. Moreover, companies often prefer forecasts that may be suboptimal in the statistical sense, but change less across time periods, putting less strain on production planning and inventory management, even though this may harm accuracy, attempting to minimize total costs. Arguably this increased congruous forecasts across time periods points to a different objective than accuracy. This is also reflected in recent views on forecast evaluation, where metrics closer to the relevant decision making are seen as desirable, albeit difficult to operationalize, and has been speculated to relate to the trustworthiness of forecasts.

We are looking forward to the OR66 conference at Bangor University next year!