Deep recurrent modelling of Granger causality with latent confounding

作者：

Highlights：

• Granger causality tests in the presence of confounding can lead to biased results.

• A neural network Granger causality test is proposed to model nonlinear dynamics.

• Neural networks can infer representations of confounders from available proxies.

摘要

•Granger causality tests in the presence of confounding can lead to biased results.•A neural network Granger causality test is proposed to model nonlinear dynamics.•Neural networks can infer representations of confounders from available proxies.

论文关键词：Latent confounders,Recurrent neural networks,Time series prediction

论文评审过程：Received 6 August 2021, Revised 24 May 2022, Accepted 30 June 2022, Available online 4 July 2022, Version of Record 8 July 2022.

论文官网地址：https://doi.org/10.1016/j.eswa.2022.118036