Abstract:Hierarchical time series forecasting plays a crucial role in decision-making in various domains while presenting significant challenges for modelling as they involve multiple levels of aggregation, constraints, and availability of information. This study explores the influence of distinct information utilisation on the accuracy of hierarchical forecasts, proposing and evaluating locals and a range of Global Forecasting Models (GFMs). In contrast to local models, which forecast each series independently, we develop GFMs to exploit cross-series and cross-hierarchies information, improving both forecasting performance and computational efficiency. We employ reconciliation methods to ensure coherency in forecasts and use the Mean Absolute Scaled Error (MASE) and Multiple Comparisons with the Best (MCB) tests to assess statistical significance. The findings indicate that GFMs possess significant advantages for hierarchical forecasting, providing more accurate and computationally efficient solutions across different levels in a hierarchy. Two specific GFMs based on LightGBM are introduced, demonstrating superior accuracy and lower model complexity than their counterpart local models and conventional methods such as Exponential Smoothing (ES) and Autoregressive Integrated Moving Average (ARIMA).

What problem does this paper attempt to address?

The problem that this paper attempts to solve is **the challenges in hierarchical time - series forecasting**. Specifically, hierarchical time - series forecasting plays a crucial role in decision - making in multiple fields (such as energy, demand forecasting, inventory control, and finance), but it faces multiple challenges during modeling, including complex issues such as multi - level aggregation, constraints, and information availability. ### Main problems: 1. **Limitations of local models**: Traditional hierarchical time - series forecasting methods rely on developing independent local models for each time series and then generating consistent forecasts through coordination methods. This method is not only computationally expensive, but also the local models cannot fully utilize the information across series and levels, resulting in low information utilization efficiency. 2. **Advantages of global models**: To overcome the limitations of local models, this study proposes and evaluates multiple **global forecasting models (GFMs)**. Unlike local models, GFMs can utilize information across series and levels, thereby improving forecasting performance and computational efficiency. 3. **Impact of information utilization**: The study also explores the impact of different information utilization methods on forecasting accuracy, aiming to determine the most effective information utilization method to handle time - series data at different levels. ### Research objectives: - **Explore the impact of different information utilization methods on hierarchical forecasting accuracy**, especially by developing and evaluating local models and global models (GFMs). - **Verify the advantages of global models in hierarchical time - series forecasting**, especially in terms of forecasting accuracy and computational efficiency. - **Introduce two LightGBM - based global models**, demonstrating their superiority in terms of accuracy and model complexity. Through these studies, the author hopes to provide decision - makers with more reliable and efficient forecasting tools and provide valuable references for future research.

Local vs. Global Models for Hierarchical Forecasting

Ensembles of Localised Models for Time Series Forecasting

Forecasting Hierarchical Time Series

Hierarchically Regularized Deep Forecasting

Hierarchical forecasting with a top-down alignment of independent level forecasts

Model selection in reconciling hierarchical time series

Machine learning applications in time series hierarchical forecasting

Hierarchical Time Series Forecasting with Bayesian Modeling

Hierarchical learning, forecasting coherent spatio-temporal individual and aggregated building loads

End-to-End Modeling Hierarchical Time Series Using Autoregressive Transformer and Conditional Normalizing Flow based Reconciliation

Hierarchical Forecasting at Scale

Efficient Forecasting of Large Scale Hierarchical Time Series via Multilevel Clustering

Hierarchies Everywhere -- Managing & Measuring Uncertainty in Hierarchical Time Series

Handling Concept Drift in Global Time Series Forecasting

A Bayesian Model for Forecasting Hierarchically Structured Time Series

A machine learning approach for forecasting hierarchical time series

FengWu-GHR: Learning the Kilometer-scale Medium-range Global Weather Forecasting

A Novel Closed-Loop Clustering Algorithm for Hierarchical Load Forecasting

Simultaneously Reconciled Quantile Forecasting of Hierarchically Related Time Series

GMP-AR: Granularity Message Passing and Adaptive Reconciliation for Temporal Hierarchy Forecasting

Improving the forecast accuracy of wind power by leveraging multiple hierarchical structure