Deep Factors for Forecasting

TL;DR

This paper introduces a hybrid deep learning model for probabilistic forecasting of large, dependent time series, combining global neural components with local classical models to improve accuracy, scalability, and uncertainty estimation.

Contribution

A novel hybrid approach that integrates deep neural networks with classical models, enabling scalable, data-efficient probabilistic forecasting with uncertainty quantification.

Findings

Outperforms classical models in accuracy and data efficiency

Scales effectively to large collections of time series

Provides reliable uncertainty estimates

Abstract

Producing probabilistic forecasts for large collections of similar and/or dependent time series is a practically relevant and challenging task. Classical time series models fail to capture complex patterns in the data, and multivariate techniques struggle to scale to large problem sizes. Their reliance on strong structural assumptions makes them data-efficient, and allows them to provide uncertainty estimates. The converse is true for models based on deep neural networks, which can learn complex patterns and dependencies given enough data. In this paper, we propose a hybrid model that incorporates the benefits of both approaches. Our new method is data-driven and scalable via a latent, global, deep component. It also handles uncertainty through a local classical model. We provide both theoretical and empirical evidence for the soundness of our approach through a necessary and sufficient decomposition of exchangeable time series into a global and a local part. Our experiments demonstrate the advantages of our model both in term of data efficiency, accuracy and computational complexity.