Joint Bottom-up Method for Probabilistic Forecasting of Hierarchical Time Series

In a stark departure from conventional wisdom, Nicolò Bertani, Shane T. Jensen, and Ville A. Satopää’s recently published research article titled “Joint Bottom-up Method for Probabilistic Forecasting of Hierarchical Time Series” dismantles a long-held belief in hierarchical forecasting: that the hierarchical structure itself contains vital information. The joint bottom-up (JBU) method proves otherwise. The authors demonstrate that the sums within a hierarchy—often seen as critical—add no additional information beyond what is contained in the most granular, bottom-level series. By modeling these bottom-level series jointly, JBU leverages their dependencies to deliver probabilistic forecasts that are both coherent and highly accurate across all levels of aggregation. This groundbreaking insight challenges decades of hierarchical forecasting practices. It underscores that upper-level series can be entirely reconstructed from the bottom-level series, rendering the hierarchy redundant. This finding, validated through real-world applications, sets a new standard for forecasting in fields like retail, energy, and tourism. Explore the full study to understand its far-reaching implications.