What Makes a Good Time-series Forecasting Model? A Causal Perspective

Xingyu Zhang, Zeen Song, Siyu Zhao, Jingyao Wang, Jianqi Zhang, Changwen Zheng, Wenwen Qiang

Keywords: Time-series Forecasting, Causal discovery

Abstract: Generalization is a long-standing challenge in multivariate time series forecasting (MTSF) tasks. Most existing forecasting methods use all available variables in historical series to predict all future variables, assuming that there may be correlations among all variables. From a causal perspective, this reliance on correlated variables can compromise the model’s generalization. To address this, we aim to explore the role of causal relationships in enhancing the generalization of multivariate time series models. We examine how graphical causal models, through conditional independence constraints, can reduce the hypothesis space, thereby improving generalization. Building on this foundation, we introduce a novel causality-based MTSF algorithm CAusal Informed Transformer (CAIFormer). It first constructs a Directed Acyclic Graph (DAG) among variables using causal discovery techniques. Then we build the forecasting model by enforcing the causal constraints informed by the DAG. Empirical evaluations on benchmark datasets demonstrate that our method surpasses traditional approaches in predictive accuracy. Additionally, we present the structural causal models derived for these datasets, underscoring the practical applicability of our causality-driven framework in MTSF.

Primary Area: learning on time series and dynamical systems

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 1782