Essays on extreme value theory in economics and finance

Extreme value methods have been successfully applied in various disciplines with the purpose of estimating tail quantiles. The probabilistic results underlying the inference procedures for the extreme values rely on the assumption of independent and identically distributed (iid) random variables. However, empirical observations often present time variation and violate the iid assumption, thus the development of methods for modelling the extremes of dependent data is currently the subject of ongoing research. This thesis provides original contributions in this direction. Exploiting data on financial asset returns, we address questions regarding the tails of the conditional return distribution and propose models for them. We begin questioning whether extreme returns exhibit seasonal behaviour and develop an approach to uncover this fact. Next, we propose to employ a method based on highfrequency data to pre-whiten the returns, and then apply an extreme value model to the tails of the estimated residuals. We then study the extremal dependence inherent in financial returns, and evaluate the ability of various high-frequency based volatility processes in generating such dependence. Finally, we propose a new class of dynamic extreme value models that exploit high-frequency data to model the tails of the conditional return distribution