Hierarchical forecasting with a top-down alignment of independent-level forecasts

Hierarchical forecasting with intermittent time series is a challenge in both research and empirical studies. Extensive focuses on improving the accuracy of each hierarchy, especially at bottom levels. Then, hierarchical reconciliation can be used to improve overall performance further. In this paper, we present hierarchical-forecasting-with-alignment approach that treats bottom-level forecasts as mutable ensure higher upper levels hierarchy. We employ pure deep learning approach, N-BEATS, for continuous top levels, widely tree-based algorithm, LightGBM, level. The simple yet effective variant bottom-up method, accounting biases are difficult observe It allows suboptimal lower level retain performance. study was developed by first author during M5 Accuracy competition, ranking second place. method also business orientated facilitate strategic planning.