Causal discovery in heavy-tailed models

Causal questions are omnipresent in many scientific problems. While much progress has been made the analysis of causal relationships between random variables, these methods not well suited if mechanisms only manifest themselves extremes. This work aims to connect two fields inference and extreme value theory. We define tail coefficient that captures asymmetries extremal dependence variables. In population case, is shown reveal structure distribution follows a linear structural model. holds even presence latent common causes have same index as observed Based on consistent estimator coefficient, we propose computationally highly efficient algorithm estimates structure. prove our method consistently recovers order compare it other well-established nonextremal approaches discovery synthetic real data. The code available an open-access R package.