Invariant Causal Prediction for Gaussian Multivariate Regression Graphs

In this work we study a causal framework for linear structural equa- tion models that can be represented by a Gaussian regression chain graph typi- cally used to model multivariate regressions. We propose a methodology based on invariance causal prediction for the identification of the causal parents of a given multivariate response variable and the estimation of the causal effects. Prelimi- nary simulation studies show that, under suitable assumptions, the methodology is able to correctly identify the true causal parents.