Dynamic Score-Driven Models in Extreme Value Theory

Dynamic score-driven models are time series models that incorporate any underlying probability
distribution, with their dynamics driven by the conditional score of time-varying parameters. In
recent years, these models have gained recognition as a valuable modern methodology for time
series analysis, offering information-theoretic optimality regardless of model misspecification
and broad applicability in econometrics, quantitative finance, and operations research. This
project aims to advance the methodology for modeling extremes by integrating extreme value
theory with dynamic score-driven frameworks. We focus on both a parametric approach, which
models the entire time-varying distribution, and a semiparametric approach, which emphasizes
time-varying quantiles, while also deriving the asymptotic properties of the corresponding
estimators.