What is hierarchical forecasting and reconciliation?

Hierarchical forecasting is a method of time series forecasting where data is organized into a hierarchical structure (for example, sales by regions, countries, and worldwide). The goal is to obtain forecasts at different levels of the hierarchy.

Reconciliation is the process of adjusting forecasts at different levels of the hierarchy so that they are consistent with each other. For example, the sum of forecasts for regions should equal the forecast for the country.

Main approaches to reconciliation:

• Bottom-up: forecasts are built at the lowest level and then aggregated upwards.
• Top-down: forecasts are made at the top level and distributed downwards.
• Middle-out: combines both approaches.
• Statistical methods (e.g., MinT), which minimize errors during reconciliation.

Thus, hierarchical forecasting and reconciliation help improve the accuracy and consistency of forecasts in complex hierarchical data structures.