extreme-value-theory

Code for the WSDM 2021 paper "FluxEV: A Fast and Effective Unsupervised Framework for Time-Series Anomaly Detection".

Automatic optimal sequential investment decisions. Forecasts made using advanced stochastic processes with Monte Carlo simulation. Dependency is handled with vine copulas.

This project studies the effects of the shape parameter estimator uncertainty at different threshold levels on the value-at-risk confidence interval for quantitative risk management (QRM) using the Generalized Pareto Distribution (GPD) from the Extreme Value Theory (EVT) approach.

[ DPhil project ] Extreme value theory and GANs to generate compound coastal hazards (wind speed + sea level pressure) from ERA5 reanalysis data over the Bay of Bengal. In development...

Crash detection via fat-tail statistics. LPPLS, EVT, Hill, Kappa. Rust core + Python.

Official Python implementation of the Pioneer Detection Method (PDM) — convergence-based expert aggregation and opinion pooling under structural change. Code for Vansteenberghe (2026), The Geneva Papers 51(1).

heavytails is a Python library implementing heavy-tailed probability distributions, built from first principles with NumPy-backed vectorized evaluation. The library provides comprehensive support for continuous and discrete heavy-tailed distributions, tail index estimation methods, and diagnostic utilities

Out-of-sample VaR and expected-shortfall audit for Gaussian, historical, and EVT models

Python package for fitting statistical models using calibrating priors.

EVT-based noise injection toolkit for evaluating time series forecasting robustness

A fat-tail-native quantitative finance toolkit for Python.

Dependence-aware block-maxima inference for severity, persistence, and design-life levels.

Estimate tail parameters of heavy-tailed distributions (including power law exponent gamma) in Python

Potential Height Python packages: runs the experiments for "Finding the potential height of tropical cyclone storm surges in a changing climate using Bayesian optimization"

Reproducible R study of motor insurance pricing on freMTPL2, with separate reserving and excess-of-loss simulation examples.

Extreme value analysis for insurance losses: POT/GPD, GEV with uncertainty, threshold diagnostics, return levels, body-tail splicing.

A deep study of human longevity using demographic data (HLD, IDL) and Extreme Value Theory to assess the potential existence of a theoretical limit to human lifespan

Extreme value theory and heavy-tailed probability in Lean 4: Fisher–Tippett–Gnedenko, regular variation, Karamata, Breiman

Two-stage frequentist framework for fusing sparse observations with dense simulations in spatial extreme value analysis