A decoder-only foundation model for time-series forecasting

TL;DR

This paper introduces a decoder-only foundation model for time-series forecasting that achieves near state-of-the-art zero-shot performance across diverse datasets by pretraining on a large corpus, demonstrating versatility in various forecasting settings.

Contribution

The paper presents a novel decoder-only model pretrained on a large time-series corpus, enabling effective zero-shot forecasting across multiple datasets and settings.

Findings

Achieves near state-of-the-art zero-shot accuracy on various datasets.

Performs well across different forecasting history lengths and prediction granularities.

Demonstrates versatility comparable to supervised models without task-specific training.

Abstract

Motivated by recent advances in large language models for Natural Language Processing (NLP), we design a time-series foundation model for forecasting whose out-of-the-box zero-shot performance on a variety of public datasets comes close to the accuracy of state-of-the-art supervised forecasting models for each individual dataset. Our model is based on pretraining a patched-decoder style attention model on a large time-series corpus, and can work well across different forecasting history lengths, prediction lengths and temporal granularities.