Dependence-aware block-maxima inference

Severity, persistence, and design-life levels under serial dependence.

UniBM is a Python package for dependence-aware block-maxima inference in heavy-tailed time series. It keeps severity inference, persistence inference, and design-life levels in one coherent workflow while exposing a small public API under unibm, unibm.evi, unibm.ei, and unibm.cdf.

Start with the package See runnable examples Explore case studies

Guide

Getting Started

Install the package, sync the local environment, and make the first severity or persistence call.

Concepts

Read the conceptual split between the severity branch, the persistence branch, and design-life levels.

Worked Examples

Use short, runnable examples for the public package surface without stepping into the full repo orchestration.

Reading Returned Objects

Interpret the most useful result fields and understand what the fitted objects actually contain.

Statistical benchmark

Compare estimation error, interval score, and coverage against known targets in the EVI and EI benchmark.

Evidence in use

01

Streamflow

Texas and Florida daily mean discharge.

USGS · Raw · EVI + EI 02

NFIP claims

Texas and Florida daily building-claim totals in 2025 dollars.

OpenFEMA · CPI-adjusted · EVI + EI 03

Houston precipitation

Full-calendar daily rainfall divided by its past EWMA level.

GHCN · Normalized · EVI + EI 04

Phoenix hot–dry severity

A derived daily severity index divided by its past EWMA level.

GHCN · Normalized · EVI + EI 05

GOES soft X-rays

Hourly maxima divided by their past EWMA level, with gaps retained.

NOAA · Normalized · EVI only 06

SPY / QQQ losses

Left-tail log losses divided by past EWMA return volatility.

Massive · Normalized · EVI + EI

API Surface

API Overview

Find EVI, EI, and design-life estimators, their defaults, and explicit choices.

EVI Namespace

Severity-side estimation, block-quantile scaling, and design-life helpers.

EI Namespace

Persistence-side procedures for extremal-index estimation under dependence.

CDF Helper

Supporting helper functionality exposed as part of the public package surface.

This site is package-first. Case figures and benchmark summaries are frozen outputs: the browser renders them but never downloads data or fits a model. Repository-level orchestration remains in the root README and justfile.