Hierarchical Time Series Forecasting Via Latent Mean Encoding

TL;DR

This paper introduces a hierarchical forecasting architecture that encodes target variable behavior at multiple temporal scales, achieving more accurate and coherent predictions across different aggregation levels in real-world datasets.

Contribution

It proposes a novel hierarchical model that learns to encode average behaviors at various temporal levels, improving forecasting coherence and accuracy over existing methods.

Findings

Outperforms TSMixer on the M5 dataset

Achieves coherent forecasts across multiple temporal scales

Demonstrates effectiveness on real-world hierarchical data

Abstract

Coherently forecasting the behaviour of a target variable across both coarse and fine temporal scales is crucial for profit-optimized decision-making in several business applications, and remains an open research problem in temporal hierarchical forecasting. Here, we propose a new hierarchical architecture that tackles this problem by leveraging modules that specialize in forecasting the different temporal aggregation levels of interest. The architecture, which learns to encode the average behaviour of the target variable within its hidden layers, makes accurate and coherent forecasts across the target temporal hierarchies. We validate our architecture on the challenging, real-world M5 dataset and show that it outperforms established methods, such as the TSMixer model.