Classical extreme value theory is the only mathematically justified framework for extrapolating into unobserved tail regions, but its standard tools were not designed for the high-dimensional, nonstationary, and causally interconnected systems that define modern risk. This book develops the necessary extensions, integrating foundational EVT with Bayesian nonparametrics, information theory, machine learning, and quantum probability across thirteen chapters in four parts.
The first part develops GEV and GPD foundations with complete measure-theoretic proofs. It extends max-stability to high dimensions using angular measure decompositions that scale to hundreds of risk factors. The second part covers nonparametric tail estimation with boundary bias correction, Bayesian Extreme Learning with entropy-regularized posteriors, Dirichlet process mixtures for heavy-tailed data, and the Extreme Value Information Criterion for tail-focused model selection. The third part introduces dynamic GEV state-space models with particle filtering for nonstationary extremes, threshold-weighted scoring rules for forecast evaluation and combination, Pareto-EVaR as a coherent risk measure that unifies GPD calibration with exponential moment constraints, and portfolio optimization under extremal transfer entropy constraints. The fourth part formalizes causal inference under regularly varying noise, develops tail-adaptive machine learning for extreme quantile estimation, and applies quantum density matrices and quantum copulas to systemic risk detection.
All theoretical results include complete proofs. Methods are implemented in R and Python with reproducible code tested on financial returns, temperature records, flood data, and cryptocurrency markets. The book serves researchers, graduate students, and advanced undergraduates in econometrics, finance, environmental science, and risk management.