A cross-temporal hierarchical framework and deep learning for supply chain forecasting

Organizations require short-term up to long-run aggregated forecasts for making strategic, tactical, and operational decisions for their supply chain management. In supply chain forecasting, the Tt emphasis is primarily on the accuracy while coherency of forecasts often gets ignored. This paper proposes a novel cross-temporal forecasting framework (CTFF) to generate coherent forecasts at all levels of a retail supply chain. A deep learning method, the long-short-term-memory network, is used as the base forecasting method in the CTFF. The performance of the CTFF is evaluated on point-of-sales data from a large multi-channel retail supply chain. Through several performance metrics and statistical tests, we conclude that forecasts from the CTFF are significantly better than the direct forecasts. In addition, improvements are significant and consistent across cross-sectional and temporal levels of a supply chain. Further, it has been observed that bottom-up forecasts are more accurate than top-down forecasts when point-of-sales data is used for forecasting in online and offline retail supply chain.