This talk has been automatically captioned. You can remove these by pressing CC on the video toolbar.

Name: Olivier Pasche

Talk Title: Causal Modelling of Heavy-Tailed Variables and Confounders

Abstract: Identifying causation is central to understanding the world around us, and the field of causal inference has developed massively in recent decades. In many situations, causal mechanisms manifest themselves only in extreme events or simplify in the tails of distributions. Standard methods of causal inference are ill-suited for the study of such phenomena, and recent research has begun to forge links between causality and extreme value theory. This talk addresses a central challenge: the presence of confounders, which can make it hard or impossible to correctly infer causal relationships. We propose a method that removes or reduces the unwanted effect of a known confounder on an extremal causal analysis, by considering it as a covariate in the modelling, and then present a statistical test for direct causality between two observed extremal variables. This enables causal discovery and inference for a greater variety of situations, as confounders are present in many, if not all, real-world situations. The methodology is illustrated on discharge data from stations in the Rhine and Aare catchments in Switzerland. The work is joint with Valérie Chavez-Demoulin and Anthony Davison.

This talk is a contributed talk at EVA 2021.