ASACT - Data preparation for forecasting: A method to substitute transaction data for unavailable product consumption data

Improving demand forecasting for customers with missing downstream data in intermittent demand supply chains with supervised multivariate clustering

“…However, this procedure does carry the risk of overfitting when there are few observations in the demand time series. Drawbacks to Croston's method are the lack of a proper underlying stochastic model (Shenstone & Hyndman, 2005) and the fact that the transformation can diminish or mask the demand behavior (Murray et al, 2018a).…”

“…For the specific case of forecasting demand under missing POS demand data, Murray et al (2018a) combined the Croston and ADIDA frameworks to propose their Aggregate, Smooth, Aggregate, Convert to Time-series (ASACT) method. Murray et al (2018a) showed that previous intermittent demand forecasting and their new method could also be used to infer missing product consumption data at a customer's point of sale from a supplier's delivery records. This, in turn, improves the demand forecast.…”

“…This, in turn, improves the demand forecast. Essentially, they proposed the often implicitly assumed link that smoothed intermittent demand data offer a reasonable approximation of the real unobserved POS demand by removing noise so as to retain the characteristic behavior of the time series (Murray et al, 2018a; Nikolopoulos et al, 2011). However, due to a lack of POS data, their study did not evaluate its results on the underlying unobserved demand data, instead using simulated data and out‐of‐sample delivery demand observations.…”

“…Demand data can be captured at multiple locations along a supply chain (Holweg et al, 2005), and delivery records may be available for suppliers. These historical delivery records, however, can contain noise from bullwhip ‐type effects or logistic decisions, or have intermittent behavior, and thus may not reflect the actual demand behavior of customers (Murray et al, 2018a; Syntetos et al, 2016). Downstream, suppliers may sometimes have access to customers' demand data at the direct point of sale.…”

“…For the scenario where a customer's POS information is missing, combined forecasting and smoothing approaches have been proposed to remove noise from a supplier's delivery demand time series to derive missing POS demand data to improve demand forecasts (Murray et al, 2018a; Nikolopoulos et al, 2011). However, when evaluating the performance of these methods, the authors did not evaluate their models on downstream POS demand information due to limitations in available data, resorting instead to either using the upstream delivery demand for the out‐of‐sample error or graphical evaluation.…”