useR! 2024

Forecast reconciliation is a post-forecasting approach to ensure the coherence of forecasts across constraints (not just simple aggregation). It harmonizes individual predictions to meet predefined relationships, leading to a consistent and comprehensive picture. This can include ensuring market share forecasts for different brands sum up to the total, or guaranteeing some property (e.g. non negativity). By incorporating these constraints, reconciliation can also improve forecast accuracy by leveraging the individual strengths. This technique finds applications in several fields like finance, supply chain, macroeconomics, load, renewable energy generation, and weather forecasting. The R package FoReco provides a powerful toolset for implementing classical and regression-based forecast reconciliation. It offers a wide range of different approaches to address different types of constraints, including cross-sectional (e.g., market share), temporal (e.g., annual-monthly data), and cross-temporal relationships. This talk presents an overview of the forecast reconciliation process and provides examples using FoReco in real-world applications. https://github.com/danigiro/FoReco