Causal Discovery from Temporally Aggregated Time Series

By Mingming Gong, Kun Zhang, Bernhard Schölkopf, Clark Glymour, Dacheng Tao. 2017

65 citations recorded by the source corpus.

Discovering causal structure of a dynamical system from observed time series is a traditional and important problem. In many practical applications, observed data are obtained by applying subsampling or temporally aggregation to the original causal processes, making it difficult to discover the underlying causal relations. Subsampling refers to the procedure that for every k consecutive observations, one is kept, the rest being skipped, and recently some advances have been made in causal discovery from such data. With temporal aggregation, the local averages or sums of k consecutive, non-overlapping observations in the causal process are computed as new observations, and causal discovery from such data is even harder. In this paper, we investigate how to recover causal relations at the original causal frequency from temporally aggregated data when k is known. Assuming the time series at the causal frequency follows a vector autoregressive (VAR) model, we show that the causal structure at the causal frequency is identifiable from aggregated time series if the noise terms are independent and non-Gaussian and some other technical conditions hold. We then present an estimation method based on non-Gaussian state-space modeling and evaluate its performance on both synthetic and real data.

Subfields: Causal Inference & Discovery, Time-Series & Forecasting

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Causal Inference & Discovery

In Time-Series & Forecasting

Citations in the corpus

Toward Causal Representation Learning (2021, background): “Defining objects or variables that can be causally related amounts to coarse-graining of more detailed models of the world, including microscopic structural equation models [208], ordinary differential equations [173], [207], and temporally aggregated time series [79].”
Towards Causal Representation Learning (2021, background): “structural equation models [207], ordinary differential equations [173, 208], and temporally aggregated time series [78].”
Causality for Machine Learning (2019, background): “Subject to appropriate conditions, structural models can arise from coarse-graining of microscopic models, including microscopic structural equation models (Rubenstein et al., 2017), ordinary differential equations (Rubenstein et al., 2018), and temporally aggregated time series (Gong et al., 2017).”
Causal Discovery and Forecasting in Nonstationary Environments with State-Space Models (2019, background): “…that some causal interactions occur in the same time period, and thus it is important to consider contemporaneous causal relations, especially in time series with low temporal resolutions (Hyvärinen et al., 2010; Gong* et al., 2015), in aggregated data (Gong et al., 2017), or in equilibrium data.”
Review of Causal Discovery Methods Based on Graphical Models (2019, methodology): “It has been shown that under suitable assumptions, the true causal relations are still identiﬁable from both subsampled and temporally aggregated data; interested readers may refer to Gong* et al. (2015); Gong et al. (2017) and references therein.”
A Survey of Learning Causality with Data (2018, background): “Although the ground-truth causal graph is often unique [50, 119], many methods output a set of candidate causal graphs.”
A Survey of Learning Causality with Data: Problems and Methods (2018, background): “We aim to learn causal relationships between different variables [48].”