TrendDiff: Decoupling Intrinsic and Measurement Trends for Enhanced Time Series Causal Discovery

Pei Zhang, Biwei Huang, Linwei Tian, Kun Zhang

Keywords: Causal discovery, Time trends, Measurement error, Constraint-based algorithm

TL;DR: We present a novel algorithm, termed Trend Differentiator (TrendDiff), which is capable of detecting all trend-influenced variables and differentiating between those affected by measurement trends and those displaying intrinsic trends.

Abstract: Time trends can be classified into intrinsic (real) and measurement (false) trends. There has long been a critical need for techniques to discern them, especially in investment decision-making. In causal discovery, these measurement trends, essentially measurement errors, can significantly impact the performance of algorithms, making it crucial to identify and eliminate them before analysis as well. Recognizing this need, we present a novel algorithm, termed Trend Differentiator (TrendDiff). It is capable of detecting all trend-influenced variables and differentiating between those affected by measurement trends and those displaying intrinsic trends, relying on changing causal module detection and trend-influenced variables’ structural properties, respectively. Extensive experiments on synthetic and real-world data demonstrate the efficacy of this approach.

Primary Area: causal reasoning

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.

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: 9730