Hierarchical forecasting: forecasting a collection of time series

Overview

Hierarchical forecasting is a method used to generate forecasts at multiple levels of aggregation within a structured hierarchy. This technique is particularly valuable in situations where data can be organised into a hierarchy based on different dimensions, such as geography, product categories, or time. The approach ensures that forecasts at the top levels (e.g. total sales) align with forecasts at the lower levels (e.g. regional or product-level sales), creating a coherent and consistent forecasting process across the entire hierarchy.

In many real-world applications, organisations deal with data that can be naturally broken down into different levels of detail. For instance, a retail company might forecast sales at a national level, then break those forecasts down to regional or store levels. Similarly, in supply chain management, forecasts can be aggregated across different product lines, factories, or distribution centers. Hierarchical forecasting ensures that forecasts at each level are not only accurate but also consistent when summed up or broken down into finer details.

Hierarchical forecasting is crucial in many settings because it provides several advantages including consistency across levels, improved accuracy, making better decisions, etc.

Research activities

Hierarchical forecasting is versatile and can be applied in many industries:

• Retail: forecasting sales by product category, store, region, or other dimensions to optimize inventory and pricing strategies
• Supply chain management: predicting demand at various levels in the supply chain, from global demand to local warehouse needs
• Energy sector: forecasting energy consumption across different sources (e.g. wind, solar, fossil fuels) or across different geographical areas to ensure efficient grid management
• Finance: budget forecasting at different levels, from overall company projections to department-specific forecasts
• Healthcare: forecasting patient admissions or demand for medical supplies across hospitals, regions, and departments.

Outcomes

I have been involved in several projects that apply hierarchical forecasting across a range of industries, including retail and energy management. These projects have demonstrated the value of this approach in achieving improved accuracy and consistency, and they highlight its broad applicability in solving complex forecasting problems. Get in touch for more details if you are interested.

Skills and experience

• Fluent in R.

Scholarships

You may be eligible to apply for a research scholarship.

Explore our research scholarships

Keywords

• forecasting
• machine learning
• statistical modeling

Contact

Contact the supervisor for more information.