Hierarchical Time Series Forecasting

Hierarchical time series forecasting (HTSF) focuses on accurately predicting multiple interconnected time series organized in a hierarchical structure, ensuring that forecasts at different levels (e.g., total sales and sales by region) are consistent. Current research emphasizes developing robust and scalable models, including probabilistic methods, graph neural networks (GNNs), and autoregressive approaches, often incorporating reconciliation techniques to maintain hierarchical coherence. This field is crucial for various applications, from supply chain optimization and financial forecasting to resource management, as accurate and consistent forecasts across multiple levels are essential for effective decision-making. Improved accuracy and scalability of HTSF models are actively pursued to address challenges posed by large, sparse, and complex real-world datasets.