Causal learning captures fast and delayed effects in changing time-series
Causal Representation Learning with Instantaneous and Lagged Relations via Nonstationarity
Machine Learning
Summary
Figuring out hidden factors and how they affect each other over time is tricky, especially when some effects happen quickly and others take time. The authors show how to use extra information like time or conditions to find these hidden states and both immediate and delayed causes, even when things change over time. They introduce a new method named iCReN that learns these hidden factors and their causal connections and prove it works well on made-up data and real forecasting tasks.
What this means in practice
- •For forecasting engineers: Build predictive models that accurately identify hidden factors and their immediate and delayed influences in time-series data under changing conditions.
- •For financial risk analysts: Improve understanding of quickly and slowly acting causes in market data to better assess risks under evolving market regimes.
Authors
Tatsuya Yamada, Hiroshi Morioka, Yoshinobu Kawahara
Abstract
Causal representation learning for time-series data aims to identify latent states and their causal relations from observations. In this setting, an important challenge is to model both lagged causal relations across observation intervals and faster causal effects that appear as instantaneous relations within an interval, while accounting for nonstationarity in time-series data. However, methods that jointly handle these causal relations and nonstationarity remain limited. To address this gap, we establish sufficient conditions for identifying latent states up to component permutation and component-wise invertible transformations, and their instantaneous and lagged causal structures up to the same permutation, using an observed auxiliary variable, such as time or a condition label, associated with changes in transition-noise distributions. Based on these results, we propose iCReN, a framework that uses contrastive learning with discrete or continuous auxiliary variables to learn latent representations and estimate their instantaneous and lagged causal structures. Experiments demonstrate accurate recovery of latent states and both instantaneous and lagged causal structures on synthetic data and the utility of the learned representations for downstream forecasting on real-world data.