Title: Functional covariate-adjusted extremal dependence
Authors: Anwar Albdulathem - University of Edinburgh (United Kingdom) [presenting]
Abstract: A method is proposed that tracks how the dependence between the extreme values of a random vector may change conditionally on a random function. Our model can be regarded as a functional covariate regression model, tailored for situations where there is the need of assessing how extremal dependence changes according to a random function. The main target of interest is what we define as the angular manifold, which is a family of angular densities indexed by a functional covariate. The methods are motivated by the need of evaluating how the dependence between extreme losses in two stock markets (e.g. NYSE and NASDAQ) changes according to the shape of a certain random curve (e.g. Daily Treasury Yield Curve). To estimate the family of angular densities on the angular manifold, we follow a similar line of attack as a popular approach for extending the Nadaraya-Watson estimator to the functional context. Our estimator can be regarded as a version of a previous one, and the simulation study suggests that the proposed methods perform well in wealth of simulation scenarios.