Introducing the Hierarchical Modeling node in SAS Visual Forecasting
When we deal with hierarchical data in time series forecasting, such as sales data that could be grouped by country, state, region, and product, independent forecasts should be generated at all the levels of the hierarchy and then the results should be reconciled so we end up with one version of the truth. In our example, that would mean that the forecasts of sales by product should add up to the forecasts of sales by region which should in turn add up to the forecasts by state etc.
The choice of hierarchies depends on the needs of an organization and many times the forecasters and planners need access to all hierarchical levels to optimize their forecasting process and make better informed decisions.
Figure 1: Example of hierarchical data where Middle-Out reconciliation is performed
In the first quarter of 2024, the Hierarchical Modeling (HM) node was introduced in SAS Visual Forecasting (VF). While hierarchical forecasting has been available since the initial release of SAS VF, there were limitations in terms of modeling flexibility and configurability for each level of the hierarchy. The new HM node allows users to perform in-depth modeling and analysis at each level of the project hierarchy.
Users now have separate pipelines at each level of the hierarchy and can apply advanced techniques, including ensemble and interactive modeling nodes, to improve the quality and potentially the accuracy of their forecasts.
After each level of the hierarchy has run, reconciliation is automatically performed based on the project hierarchy that the users have set. Reconciliation is an important part of the forecasting process as it ensures that the final forecasts are coherent across the hierarchical structure. With the HM node, users can access, visualize, compare, and save their forecasts at each level of the hierarchy.
In the demo below, SAS forecasting product manager, Joe Katz, showcases the HM node's functionalities and how to use it in practice.
