Zero-Shot Time Series Forecasting with Covariates via In-Context Learning

2025-06-03

AI Summary

Existing zero-shot time-series pre-trained models generally lack native support for, or struggle to effectively integrate, covariates. Method: We propose COSMIC, the first zero-shot forecasting framework that supports arbitrary covariate inputs without fine-tuning. It achieves this via covariate-aware prompt construction and informative covariate augmentation—enabling covariate dependency modeling during pre-training using only covariate-free data—and leverages in-context learning to dynamically incorporate covariate information at inference, eliminating the need for labeled data with covariates. Contribution/Results: COSMIC establishes new state-of-the-art performance on zero-shot forecasting tasks both with and without covariates, significantly outperforming prior methods. Moreover, it provides an interpretable, plug-and-play mechanism for covariate injection, offering both empirical superiority and analytical transparency.

Technology Category

• Machine Learning: Time-Series/Data Streams
• Computer Vision: Diffusion Models for Vision
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding

Application Category

• Web Mining and Content Analysis: Large pretrained models with web data
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Pretrained time series models, capable of zero-shot forecasting, have demonstrated significant potential in enhancing both the performance and accessibility of time series forecasting. However, existing pretrained models either do not support covariates or fail to incorporate them effectively. We introduce COSMIC, a zero-shot forecasting model that utilizes covariates via in-context learning. To address the challenge of data scarcity, we propose Informative Covariate Augmentation, which enables the training of COSMIC without requiring any datasets that include covariates. COSMIC achieves state-of-the-art performance in zero-shot forecasting, both with and without covariates. Our quantitative and qualitative analysis demonstrates that COSMIC effectively leverages covariates in zero-shot forecasting.

Problem

• Enabling zero-shot time series forecasting with covariates
• Addressing data scarcity via Informative Covariate Augmentation
• Improving covariate utilization in pretrained forecasting models

Innovation

• Uses in-context learning for covariates
• Proposes Informative Covariate Augmentation
• Achieves state-of-the-art zero-shot performance