Integrated Forecasting and Inventory Control for Seasonal Demand

Abstract

We present a data-driven forecasting technique with integrated inventory control for seasonal data, and compare it to the traditional Holt-Winters algorithm in the context of the newsvendor problem. The data-driven approach relies on (i) clustering data points reflecting a similar phase of the demand process, and (ii) computing the optimal order quantity using the critical quantile for the relevant data, i.e., data observed when the demand was in a similar phase to the one forecasted for the next time period. Results indicate that the data-driven approach achieves a 1-5% improvement in the average regret when holding and backorder costs are of the same order of magnitude. For particularly imbalanced cost structures, average regret can be improved by up to 90%. This is because traditional forecasting penalizes under- and over-shooting equally, but penalties at the inventory management level are much more severe in one case (typically, backorder) than the other (typically, holding). This suggests the data-driven approach holds much promise as a cost reduction technique.