CFE 2019: Start Registration
View Submission - CMStatistics
Title: What hides behind an extreme currency demand: Bayesian conditionally heteroscedastic extremes Authors:  Miguel de Carvalho - School of Mathematics, University of Edinburgh (Portugal) [presenting]
Junho Lee - University of Edinburgh (United Kingdom)
Antonio Rua - Banco de Portugal (Portugal)
Abstract: Bayesian inference methods will be introduced for conditionally heteroscedastic extremes. The proposed model is based on an tail index regression and on a proportional tails model, and can be used for assessing how the magnitude and frequency of the extreme values can change along with a covariate. We start with the unconditional setting for estimating the tail index and the scedasis function and show that the proposed inference methods for the scedasis density---based on a Bernstein--Dirichlet prior---perform well in Monte Carlo simulation studies, are exact apart from Monte Carlo error, and have full support on the space of all continuous scedasis functions. We then extend the proposed methods to the conditional setting using dependent Bernstein-Dirichlet process. We resort to the proposed methodologies to examine an extreme currency demand in Portugal. The signatures of the fitted scedasis densities of extreme currency demand---over different denominations---reveal some interesting insights on the dynamics governing currency demand over periods of economic stress.